]>
Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8cdea7c0 BS |
2 | /* memcontrol.c - Memory Controller |
3 | * | |
4 | * Copyright IBM Corporation, 2007 | |
5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
6 | * | |
78fb7466 PE |
7 | * Copyright 2007 OpenVZ SWsoft Inc |
8 | * Author: Pavel Emelianov <xemul@openvz.org> | |
9 | * | |
2e72b634 KS |
10 | * Memory thresholds |
11 | * Copyright (C) 2009 Nokia Corporation | |
12 | * Author: Kirill A. Shutemov | |
13 | * | |
7ae1e1d0 GC |
14 | * Kernel Memory Controller |
15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
16 | * Authors: Glauber Costa and Suleiman Souhlal | |
17 | * | |
1575e68b JW |
18 | * Native page reclaim |
19 | * Charge lifetime sanitation | |
20 | * Lockless page tracking & accounting | |
21 | * Unified hierarchy configuration model | |
22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
6168d0da AS |
23 | * |
24 | * Per memcg lru locking | |
25 | * Copyright (C) 2020 Alibaba, Inc, Alex Shi | |
8cdea7c0 BS |
26 | */ |
27 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
29 | #include <linux/memcontrol.h> |
30 | #include <linux/cgroup.h> | |
a520110e | 31 | #include <linux/pagewalk.h> |
6e84f315 | 32 | #include <linux/sched/mm.h> |
3a4f8a0b | 33 | #include <linux/shmem_fs.h> |
4ffef5fe | 34 | #include <linux/hugetlb.h> |
d13d1443 | 35 | #include <linux/pagemap.h> |
1ff9e6e1 | 36 | #include <linux/vm_event_item.h> |
d52aa412 | 37 | #include <linux/smp.h> |
8a9f3ccd | 38 | #include <linux/page-flags.h> |
66e1707b | 39 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
40 | #include <linux/bit_spinlock.h> |
41 | #include <linux/rcupdate.h> | |
e222432b | 42 | #include <linux/limits.h> |
b9e15baf | 43 | #include <linux/export.h> |
8c7c6e34 | 44 | #include <linux/mutex.h> |
bb4cc1a8 | 45 | #include <linux/rbtree.h> |
b6ac57d5 | 46 | #include <linux/slab.h> |
66e1707b | 47 | #include <linux/swap.h> |
02491447 | 48 | #include <linux/swapops.h> |
66e1707b | 49 | #include <linux/spinlock.h> |
2e72b634 | 50 | #include <linux/eventfd.h> |
79bd9814 | 51 | #include <linux/poll.h> |
2e72b634 | 52 | #include <linux/sort.h> |
66e1707b | 53 | #include <linux/fs.h> |
d2ceb9b7 | 54 | #include <linux/seq_file.h> |
70ddf637 | 55 | #include <linux/vmpressure.h> |
dc90f084 | 56 | #include <linux/memremap.h> |
b69408e8 | 57 | #include <linux/mm_inline.h> |
5d1ea48b | 58 | #include <linux/swap_cgroup.h> |
cdec2e42 | 59 | #include <linux/cpu.h> |
158e0a2d | 60 | #include <linux/oom.h> |
0056f4e6 | 61 | #include <linux/lockdep.h> |
79bd9814 | 62 | #include <linux/file.h> |
03248add | 63 | #include <linux/resume_user_mode.h> |
0e4b01df | 64 | #include <linux/psi.h> |
c8713d0b | 65 | #include <linux/seq_buf.h> |
08e552c6 | 66 | #include "internal.h" |
d1a4c0b3 | 67 | #include <net/sock.h> |
4bd2c1ee | 68 | #include <net/ip.h> |
f35c3a8e | 69 | #include "slab.h" |
014bb1de | 70 | #include "swap.h" |
8cdea7c0 | 71 | |
7c0f6ba6 | 72 | #include <linux/uaccess.h> |
8697d331 | 73 | |
cc8e970c KM |
74 | #include <trace/events/vmscan.h> |
75 | ||
073219e9 TH |
76 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
77 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 78 | |
7d828602 JW |
79 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
80 | ||
37d5985c RG |
81 | /* Active memory cgroup to use from an interrupt context */ |
82 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); | |
c74d40e8 | 83 | EXPORT_PER_CPU_SYMBOL_GPL(int_active_memcg); |
37d5985c | 84 | |
f7e1cb6e | 85 | /* Socket memory accounting disabled? */ |
0f0cace3 | 86 | static bool cgroup_memory_nosocket __ro_after_init; |
f7e1cb6e | 87 | |
04823c83 | 88 | /* Kernel memory accounting disabled? */ |
17c17367 | 89 | static bool cgroup_memory_nokmem __ro_after_init; |
04823c83 | 90 | |
21afa38e | 91 | /* Whether the swap controller is active */ |
c255a458 | 92 | #ifdef CONFIG_MEMCG_SWAP |
c1b8fdae KS |
93 | static bool cgroup_memory_noswap __initdata; |
94 | ||
95 | static DEFINE_STATIC_KEY_FALSE(memcg_swap_enabled_key); | |
96 | static inline bool memcg_swap_enabled(void) | |
97 | { | |
98 | return static_branch_likely(&memcg_swap_enabled_key); | |
99 | } | |
c077719b | 100 | #else |
c1b8fdae KS |
101 | static inline bool memcg_swap_enabled(void) |
102 | { | |
103 | return false; | |
104 | } | |
2d1c4980 | 105 | #endif |
c077719b | 106 | |
97b27821 TH |
107 | #ifdef CONFIG_CGROUP_WRITEBACK |
108 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
109 | #endif | |
110 | ||
7941d214 JW |
111 | /* Whether legacy memory+swap accounting is active */ |
112 | static bool do_memsw_account(void) | |
113 | { | |
c1b8fdae | 114 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg_swap_enabled(); |
7941d214 JW |
115 | } |
116 | ||
a0db00fc KS |
117 | #define THRESHOLDS_EVENTS_TARGET 128 |
118 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 119 | |
bb4cc1a8 AM |
120 | /* |
121 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
122 | * their hierarchy representation | |
123 | */ | |
124 | ||
ef8f2327 | 125 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 126 | struct rb_root rb_root; |
fa90b2fd | 127 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
128 | spinlock_t lock; |
129 | }; | |
130 | ||
bb4cc1a8 AM |
131 | struct mem_cgroup_tree { |
132 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
133 | }; | |
134 | ||
135 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
136 | ||
9490ff27 KH |
137 | /* for OOM */ |
138 | struct mem_cgroup_eventfd_list { | |
139 | struct list_head list; | |
140 | struct eventfd_ctx *eventfd; | |
141 | }; | |
2e72b634 | 142 | |
79bd9814 TH |
143 | /* |
144 | * cgroup_event represents events which userspace want to receive. | |
145 | */ | |
3bc942f3 | 146 | struct mem_cgroup_event { |
79bd9814 | 147 | /* |
59b6f873 | 148 | * memcg which the event belongs to. |
79bd9814 | 149 | */ |
59b6f873 | 150 | struct mem_cgroup *memcg; |
79bd9814 TH |
151 | /* |
152 | * eventfd to signal userspace about the event. | |
153 | */ | |
154 | struct eventfd_ctx *eventfd; | |
155 | /* | |
156 | * Each of these stored in a list by the cgroup. | |
157 | */ | |
158 | struct list_head list; | |
fba94807 TH |
159 | /* |
160 | * register_event() callback will be used to add new userspace | |
161 | * waiter for changes related to this event. Use eventfd_signal() | |
162 | * on eventfd to send notification to userspace. | |
163 | */ | |
59b6f873 | 164 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 165 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
166 | /* |
167 | * unregister_event() callback will be called when userspace closes | |
168 | * the eventfd or on cgroup removing. This callback must be set, | |
169 | * if you want provide notification functionality. | |
170 | */ | |
59b6f873 | 171 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 172 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
173 | /* |
174 | * All fields below needed to unregister event when | |
175 | * userspace closes eventfd. | |
176 | */ | |
177 | poll_table pt; | |
178 | wait_queue_head_t *wqh; | |
ac6424b9 | 179 | wait_queue_entry_t wait; |
79bd9814 TH |
180 | struct work_struct remove; |
181 | }; | |
182 | ||
c0ff4b85 R |
183 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
184 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 185 | |
7dc74be0 DN |
186 | /* Stuffs for move charges at task migration. */ |
187 | /* | |
1dfab5ab | 188 | * Types of charges to be moved. |
7dc74be0 | 189 | */ |
1dfab5ab JW |
190 | #define MOVE_ANON 0x1U |
191 | #define MOVE_FILE 0x2U | |
192 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 193 | |
4ffef5fe DN |
194 | /* "mc" and its members are protected by cgroup_mutex */ |
195 | static struct move_charge_struct { | |
b1dd693e | 196 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 197 | struct mm_struct *mm; |
4ffef5fe DN |
198 | struct mem_cgroup *from; |
199 | struct mem_cgroup *to; | |
1dfab5ab | 200 | unsigned long flags; |
4ffef5fe | 201 | unsigned long precharge; |
854ffa8d | 202 | unsigned long moved_charge; |
483c30b5 | 203 | unsigned long moved_swap; |
8033b97c DN |
204 | struct task_struct *moving_task; /* a task moving charges */ |
205 | wait_queue_head_t waitq; /* a waitq for other context */ | |
206 | } mc = { | |
2bd9bb20 | 207 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
208 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
209 | }; | |
4ffef5fe | 210 | |
4e416953 BS |
211 | /* |
212 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
213 | * limit reclaim to prevent infinite loops, if they ever occur. | |
214 | */ | |
a0db00fc | 215 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 216 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 217 | |
8c7c6e34 | 218 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
219 | enum res_type { |
220 | _MEM, | |
221 | _MEMSWAP, | |
510fc4e1 | 222 | _KMEM, |
d55f90bf | 223 | _TCP, |
86ae53e1 GC |
224 | }; |
225 | ||
a0db00fc KS |
226 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
227 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 KH |
228 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
229 | ||
b05706f1 KT |
230 | /* |
231 | * Iteration constructs for visiting all cgroups (under a tree). If | |
232 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
233 | * be used for reference counting. | |
234 | */ | |
235 | #define for_each_mem_cgroup_tree(iter, root) \ | |
236 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
237 | iter != NULL; \ | |
238 | iter = mem_cgroup_iter(root, iter, NULL)) | |
239 | ||
240 | #define for_each_mem_cgroup(iter) \ | |
241 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
242 | iter != NULL; \ | |
243 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
244 | ||
a4ebf1b6 | 245 | static inline bool task_is_dying(void) |
7775face TH |
246 | { |
247 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
248 | (current->flags & PF_EXITING); | |
249 | } | |
250 | ||
70ddf637 AV |
251 | /* Some nice accessors for the vmpressure. */ |
252 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
253 | { | |
254 | if (!memcg) | |
255 | memcg = root_mem_cgroup; | |
256 | return &memcg->vmpressure; | |
257 | } | |
258 | ||
9647875b | 259 | struct mem_cgroup *vmpressure_to_memcg(struct vmpressure *vmpr) |
70ddf637 | 260 | { |
9647875b | 261 | return container_of(vmpr, struct mem_cgroup, vmpressure); |
70ddf637 AV |
262 | } |
263 | ||
84c07d11 | 264 | #ifdef CONFIG_MEMCG_KMEM |
0764db9b | 265 | static DEFINE_SPINLOCK(objcg_lock); |
bf4f0599 | 266 | |
4d5c8aed RG |
267 | bool mem_cgroup_kmem_disabled(void) |
268 | { | |
269 | return cgroup_memory_nokmem; | |
270 | } | |
271 | ||
f1286fae MS |
272 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
273 | unsigned int nr_pages); | |
c1a660de | 274 | |
bf4f0599 RG |
275 | static void obj_cgroup_release(struct percpu_ref *ref) |
276 | { | |
277 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
bf4f0599 RG |
278 | unsigned int nr_bytes; |
279 | unsigned int nr_pages; | |
280 | unsigned long flags; | |
281 | ||
282 | /* | |
283 | * At this point all allocated objects are freed, and | |
284 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
285 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
286 | * | |
287 | * The following sequence can lead to it: | |
288 | * 1) CPU0: objcg == stock->cached_objcg | |
289 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
290 | * PAGE_SIZE bytes are charged | |
291 | * 3) CPU1: a process from another memcg is allocating something, | |
292 | * the stock if flushed, | |
293 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
294 | * 5) CPU0: we do release this object, | |
295 | * 92 bytes are added to stock->nr_bytes | |
296 | * 6) CPU0: stock is flushed, | |
297 | * 92 bytes are added to objcg->nr_charged_bytes | |
298 | * | |
299 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
300 | * This page will be uncharged in obj_cgroup_release(). | |
301 | */ | |
302 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
303 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
304 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
305 | ||
bf4f0599 | 306 | if (nr_pages) |
f1286fae | 307 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
271dd6b1 | 308 | |
0764db9b | 309 | spin_lock_irqsave(&objcg_lock, flags); |
bf4f0599 | 310 | list_del(&objcg->list); |
0764db9b | 311 | spin_unlock_irqrestore(&objcg_lock, flags); |
bf4f0599 RG |
312 | |
313 | percpu_ref_exit(ref); | |
314 | kfree_rcu(objcg, rcu); | |
315 | } | |
316 | ||
317 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
318 | { | |
319 | struct obj_cgroup *objcg; | |
320 | int ret; | |
321 | ||
322 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
323 | if (!objcg) | |
324 | return NULL; | |
325 | ||
326 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
327 | GFP_KERNEL); | |
328 | if (ret) { | |
329 | kfree(objcg); | |
330 | return NULL; | |
331 | } | |
332 | INIT_LIST_HEAD(&objcg->list); | |
333 | return objcg; | |
334 | } | |
335 | ||
336 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
337 | struct mem_cgroup *parent) | |
338 | { | |
339 | struct obj_cgroup *objcg, *iter; | |
340 | ||
341 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
342 | ||
0764db9b | 343 | spin_lock_irq(&objcg_lock); |
bf4f0599 | 344 | |
9838354e MS |
345 | /* 1) Ready to reparent active objcg. */ |
346 | list_add(&objcg->list, &memcg->objcg_list); | |
347 | /* 2) Reparent active objcg and already reparented objcgs to parent. */ | |
348 | list_for_each_entry(iter, &memcg->objcg_list, list) | |
349 | WRITE_ONCE(iter->memcg, parent); | |
350 | /* 3) Move already reparented objcgs to the parent's list */ | |
bf4f0599 RG |
351 | list_splice(&memcg->objcg_list, &parent->objcg_list); |
352 | ||
0764db9b | 353 | spin_unlock_irq(&objcg_lock); |
bf4f0599 RG |
354 | |
355 | percpu_ref_kill(&objcg->refcnt); | |
356 | } | |
357 | ||
d7f25f8a GC |
358 | /* |
359 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 360 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
361 | * conditional to this static branch, we'll have to allow modules that does |
362 | * kmem_cache_alloc and the such to see this symbol as well | |
363 | */ | |
ef12947c | 364 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 365 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 366 | #endif |
17cc4dfe | 367 | |
ad7fa852 TH |
368 | /** |
369 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
370 | * @page: page of interest | |
371 | * | |
372 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
373 | * with @page is returned. The returned css remains associated with @page | |
374 | * until it is released. | |
375 | * | |
376 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
377 | * is returned. | |
ad7fa852 TH |
378 | */ |
379 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
380 | { | |
381 | struct mem_cgroup *memcg; | |
382 | ||
bcfe06bf | 383 | memcg = page_memcg(page); |
ad7fa852 | 384 | |
9e10a130 | 385 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
386 | memcg = root_mem_cgroup; |
387 | ||
ad7fa852 TH |
388 | return &memcg->css; |
389 | } | |
390 | ||
2fc04524 VD |
391 | /** |
392 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
393 | * @page: the page | |
394 | * | |
395 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
396 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
397 | * is safe to call this function without holding a reference to @page. | |
398 | * | |
399 | * Note, this function is inherently racy, because there is nothing to prevent | |
400 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
401 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
402 | * do not care (such as procfs interfaces). | |
403 | */ | |
404 | ino_t page_cgroup_ino(struct page *page) | |
405 | { | |
406 | struct mem_cgroup *memcg; | |
407 | unsigned long ino = 0; | |
408 | ||
409 | rcu_read_lock(); | |
bcfe06bf | 410 | memcg = page_memcg_check(page); |
286e04b8 | 411 | |
2fc04524 VD |
412 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
413 | memcg = parent_mem_cgroup(memcg); | |
414 | if (memcg) | |
415 | ino = cgroup_ino(memcg->css.cgroup); | |
416 | rcu_read_unlock(); | |
417 | return ino; | |
418 | } | |
419 | ||
ef8f2327 MG |
420 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
421 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 422 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
423 | { |
424 | struct rb_node **p = &mctz->rb_root.rb_node; | |
425 | struct rb_node *parent = NULL; | |
ef8f2327 | 426 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 427 | bool rightmost = true; |
bb4cc1a8 AM |
428 | |
429 | if (mz->on_tree) | |
430 | return; | |
431 | ||
432 | mz->usage_in_excess = new_usage_in_excess; | |
433 | if (!mz->usage_in_excess) | |
434 | return; | |
435 | while (*p) { | |
436 | parent = *p; | |
ef8f2327 | 437 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 438 | tree_node); |
fa90b2fd | 439 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 440 | p = &(*p)->rb_left; |
fa90b2fd | 441 | rightmost = false; |
378876b0 | 442 | } else { |
bb4cc1a8 | 443 | p = &(*p)->rb_right; |
378876b0 | 444 | } |
bb4cc1a8 | 445 | } |
fa90b2fd DB |
446 | |
447 | if (rightmost) | |
448 | mctz->rb_rightmost = &mz->tree_node; | |
449 | ||
bb4cc1a8 AM |
450 | rb_link_node(&mz->tree_node, parent, p); |
451 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
452 | mz->on_tree = true; | |
453 | } | |
454 | ||
ef8f2327 MG |
455 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
456 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
457 | { |
458 | if (!mz->on_tree) | |
459 | return; | |
fa90b2fd DB |
460 | |
461 | if (&mz->tree_node == mctz->rb_rightmost) | |
462 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
463 | ||
bb4cc1a8 AM |
464 | rb_erase(&mz->tree_node, &mctz->rb_root); |
465 | mz->on_tree = false; | |
466 | } | |
467 | ||
ef8f2327 MG |
468 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
469 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 470 | { |
0a31bc97 JW |
471 | unsigned long flags; |
472 | ||
473 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 474 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 475 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
476 | } |
477 | ||
3e32cb2e JW |
478 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
479 | { | |
480 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 481 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
482 | unsigned long excess = 0; |
483 | ||
484 | if (nr_pages > soft_limit) | |
485 | excess = nr_pages - soft_limit; | |
486 | ||
487 | return excess; | |
488 | } | |
bb4cc1a8 | 489 | |
658b69c9 | 490 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, int nid) |
bb4cc1a8 | 491 | { |
3e32cb2e | 492 | unsigned long excess; |
ef8f2327 MG |
493 | struct mem_cgroup_per_node *mz; |
494 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 495 | |
2ab082ba | 496 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
497 | if (!mctz) |
498 | return; | |
bb4cc1a8 AM |
499 | /* |
500 | * Necessary to update all ancestors when hierarchy is used. | |
501 | * because their event counter is not touched. | |
502 | */ | |
503 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
658b69c9 | 504 | mz = memcg->nodeinfo[nid]; |
3e32cb2e | 505 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
506 | /* |
507 | * We have to update the tree if mz is on RB-tree or | |
508 | * mem is over its softlimit. | |
509 | */ | |
510 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
511 | unsigned long flags; |
512 | ||
513 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
514 | /* if on-tree, remove it */ |
515 | if (mz->on_tree) | |
cf2c8127 | 516 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
517 | /* |
518 | * Insert again. mz->usage_in_excess will be updated. | |
519 | * If excess is 0, no tree ops. | |
520 | */ | |
cf2c8127 | 521 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 522 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
523 | } |
524 | } | |
525 | } | |
526 | ||
527 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
528 | { | |
ef8f2327 MG |
529 | struct mem_cgroup_tree_per_node *mctz; |
530 | struct mem_cgroup_per_node *mz; | |
531 | int nid; | |
bb4cc1a8 | 532 | |
e231875b | 533 | for_each_node(nid) { |
a3747b53 | 534 | mz = memcg->nodeinfo[nid]; |
2ab082ba | 535 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
536 | if (mctz) |
537 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
538 | } |
539 | } | |
540 | ||
ef8f2327 MG |
541 | static struct mem_cgroup_per_node * |
542 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 543 | { |
ef8f2327 | 544 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
545 | |
546 | retry: | |
547 | mz = NULL; | |
fa90b2fd | 548 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
549 | goto done; /* Nothing to reclaim from */ |
550 | ||
fa90b2fd DB |
551 | mz = rb_entry(mctz->rb_rightmost, |
552 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
553 | /* |
554 | * Remove the node now but someone else can add it back, | |
555 | * we will to add it back at the end of reclaim to its correct | |
556 | * position in the tree. | |
557 | */ | |
cf2c8127 | 558 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 559 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 560 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
561 | goto retry; |
562 | done: | |
563 | return mz; | |
564 | } | |
565 | ||
ef8f2327 MG |
566 | static struct mem_cgroup_per_node * |
567 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 568 | { |
ef8f2327 | 569 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 570 | |
0a31bc97 | 571 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 572 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 573 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
574 | return mz; |
575 | } | |
576 | ||
11192d9c SB |
577 | /* |
578 | * memcg and lruvec stats flushing | |
579 | * | |
580 | * Many codepaths leading to stats update or read are performance sensitive and | |
581 | * adding stats flushing in such codepaths is not desirable. So, to optimize the | |
582 | * flushing the kernel does: | |
583 | * | |
584 | * 1) Periodically and asynchronously flush the stats every 2 seconds to not let | |
585 | * rstat update tree grow unbounded. | |
586 | * | |
587 | * 2) Flush the stats synchronously on reader side only when there are more than | |
588 | * (MEMCG_CHARGE_BATCH * nr_cpus) update events. Though this optimization | |
589 | * will let stats be out of sync by atmost (MEMCG_CHARGE_BATCH * nr_cpus) but | |
590 | * only for 2 seconds due to (1). | |
591 | */ | |
592 | static void flush_memcg_stats_dwork(struct work_struct *w); | |
593 | static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork); | |
594 | static DEFINE_SPINLOCK(stats_flush_lock); | |
595 | static DEFINE_PER_CPU(unsigned int, stats_updates); | |
596 | static atomic_t stats_flush_threshold = ATOMIC_INIT(0); | |
9b301615 SB |
597 | static u64 flush_next_time; |
598 | ||
599 | #define FLUSH_TIME (2UL*HZ) | |
11192d9c | 600 | |
be3e67b5 SAS |
601 | /* |
602 | * Accessors to ensure that preemption is disabled on PREEMPT_RT because it can | |
603 | * not rely on this as part of an acquired spinlock_t lock. These functions are | |
604 | * never used in hardirq context on PREEMPT_RT and therefore disabling preemtion | |
605 | * is sufficient. | |
606 | */ | |
607 | static void memcg_stats_lock(void) | |
608 | { | |
609 | #ifdef CONFIG_PREEMPT_RT | |
610 | preempt_disable(); | |
611 | #else | |
612 | VM_BUG_ON(!irqs_disabled()); | |
613 | #endif | |
614 | } | |
615 | ||
616 | static void __memcg_stats_lock(void) | |
617 | { | |
618 | #ifdef CONFIG_PREEMPT_RT | |
619 | preempt_disable(); | |
620 | #endif | |
621 | } | |
622 | ||
623 | static void memcg_stats_unlock(void) | |
624 | { | |
625 | #ifdef CONFIG_PREEMPT_RT | |
626 | preempt_enable(); | |
627 | #endif | |
628 | } | |
629 | ||
5b3be698 | 630 | static inline void memcg_rstat_updated(struct mem_cgroup *memcg, int val) |
11192d9c | 631 | { |
5b3be698 SB |
632 | unsigned int x; |
633 | ||
11192d9c | 634 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); |
5b3be698 SB |
635 | |
636 | x = __this_cpu_add_return(stats_updates, abs(val)); | |
637 | if (x > MEMCG_CHARGE_BATCH) { | |
873f64b7 JS |
638 | /* |
639 | * If stats_flush_threshold exceeds the threshold | |
640 | * (>num_online_cpus()), cgroup stats update will be triggered | |
641 | * in __mem_cgroup_flush_stats(). Increasing this var further | |
642 | * is redundant and simply adds overhead in atomic update. | |
643 | */ | |
644 | if (atomic_read(&stats_flush_threshold) <= num_online_cpus()) | |
645 | atomic_add(x / MEMCG_CHARGE_BATCH, &stats_flush_threshold); | |
5b3be698 SB |
646 | __this_cpu_write(stats_updates, 0); |
647 | } | |
11192d9c SB |
648 | } |
649 | ||
650 | static void __mem_cgroup_flush_stats(void) | |
651 | { | |
fd25a9e0 SB |
652 | unsigned long flag; |
653 | ||
654 | if (!spin_trylock_irqsave(&stats_flush_lock, flag)) | |
11192d9c SB |
655 | return; |
656 | ||
9b301615 | 657 | flush_next_time = jiffies_64 + 2*FLUSH_TIME; |
11192d9c SB |
658 | cgroup_rstat_flush_irqsafe(root_mem_cgroup->css.cgroup); |
659 | atomic_set(&stats_flush_threshold, 0); | |
fd25a9e0 | 660 | spin_unlock_irqrestore(&stats_flush_lock, flag); |
11192d9c SB |
661 | } |
662 | ||
663 | void mem_cgroup_flush_stats(void) | |
664 | { | |
665 | if (atomic_read(&stats_flush_threshold) > num_online_cpus()) | |
666 | __mem_cgroup_flush_stats(); | |
667 | } | |
668 | ||
9b301615 SB |
669 | void mem_cgroup_flush_stats_delayed(void) |
670 | { | |
671 | if (time_after64(jiffies_64, flush_next_time)) | |
672 | mem_cgroup_flush_stats(); | |
673 | } | |
674 | ||
11192d9c SB |
675 | static void flush_memcg_stats_dwork(struct work_struct *w) |
676 | { | |
5b3be698 | 677 | __mem_cgroup_flush_stats(); |
9b301615 | 678 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, FLUSH_TIME); |
11192d9c SB |
679 | } |
680 | ||
d396def5 SB |
681 | /* Subset of vm_event_item to report for memcg event stats */ |
682 | static const unsigned int memcg_vm_event_stat[] = { | |
8278f1c7 SB |
683 | PGPGIN, |
684 | PGPGOUT, | |
d396def5 SB |
685 | PGSCAN_KSWAPD, |
686 | PGSCAN_DIRECT, | |
687 | PGSTEAL_KSWAPD, | |
688 | PGSTEAL_DIRECT, | |
689 | PGFAULT, | |
690 | PGMAJFAULT, | |
691 | PGREFILL, | |
692 | PGACTIVATE, | |
693 | PGDEACTIVATE, | |
694 | PGLAZYFREE, | |
695 | PGLAZYFREED, | |
696 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) | |
697 | ZSWPIN, | |
698 | ZSWPOUT, | |
699 | #endif | |
700 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
701 | THP_FAULT_ALLOC, | |
702 | THP_COLLAPSE_ALLOC, | |
703 | #endif | |
704 | }; | |
705 | ||
8278f1c7 SB |
706 | #define NR_MEMCG_EVENTS ARRAY_SIZE(memcg_vm_event_stat) |
707 | static int mem_cgroup_events_index[NR_VM_EVENT_ITEMS] __read_mostly; | |
708 | ||
709 | static void init_memcg_events(void) | |
710 | { | |
711 | int i; | |
712 | ||
713 | for (i = 0; i < NR_MEMCG_EVENTS; ++i) | |
714 | mem_cgroup_events_index[memcg_vm_event_stat[i]] = i + 1; | |
715 | } | |
716 | ||
717 | static inline int memcg_events_index(enum vm_event_item idx) | |
718 | { | |
719 | return mem_cgroup_events_index[idx] - 1; | |
720 | } | |
721 | ||
410f8e82 SB |
722 | struct memcg_vmstats_percpu { |
723 | /* Local (CPU and cgroup) page state & events */ | |
724 | long state[MEMCG_NR_STAT]; | |
8278f1c7 | 725 | unsigned long events[NR_MEMCG_EVENTS]; |
410f8e82 SB |
726 | |
727 | /* Delta calculation for lockless upward propagation */ | |
728 | long state_prev[MEMCG_NR_STAT]; | |
8278f1c7 | 729 | unsigned long events_prev[NR_MEMCG_EVENTS]; |
410f8e82 SB |
730 | |
731 | /* Cgroup1: threshold notifications & softlimit tree updates */ | |
732 | unsigned long nr_page_events; | |
733 | unsigned long targets[MEM_CGROUP_NTARGETS]; | |
734 | }; | |
735 | ||
736 | struct memcg_vmstats { | |
737 | /* Aggregated (CPU and subtree) page state & events */ | |
738 | long state[MEMCG_NR_STAT]; | |
8278f1c7 | 739 | unsigned long events[NR_MEMCG_EVENTS]; |
410f8e82 SB |
740 | |
741 | /* Pending child counts during tree propagation */ | |
742 | long state_pending[MEMCG_NR_STAT]; | |
8278f1c7 | 743 | unsigned long events_pending[NR_MEMCG_EVENTS]; |
410f8e82 SB |
744 | }; |
745 | ||
746 | unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) | |
747 | { | |
748 | long x = READ_ONCE(memcg->vmstats->state[idx]); | |
749 | #ifdef CONFIG_SMP | |
750 | if (x < 0) | |
751 | x = 0; | |
752 | #endif | |
753 | return x; | |
754 | } | |
755 | ||
db9adbcb JW |
756 | /** |
757 | * __mod_memcg_state - update cgroup memory statistics | |
758 | * @memcg: the memory cgroup | |
759 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
760 | * @val: delta to add to the counter, can be negative | |
761 | */ | |
762 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
763 | { | |
db9adbcb JW |
764 | if (mem_cgroup_disabled()) |
765 | return; | |
766 | ||
2d146aa3 | 767 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
5b3be698 | 768 | memcg_rstat_updated(memcg, val); |
db9adbcb JW |
769 | } |
770 | ||
2d146aa3 | 771 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
a18e6e6e JW |
772 | static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx) |
773 | { | |
774 | long x = 0; | |
775 | int cpu; | |
776 | ||
777 | for_each_possible_cpu(cpu) | |
2d146aa3 | 778 | x += per_cpu(memcg->vmstats_percpu->state[idx], cpu); |
a18e6e6e JW |
779 | #ifdef CONFIG_SMP |
780 | if (x < 0) | |
781 | x = 0; | |
782 | #endif | |
783 | return x; | |
784 | } | |
785 | ||
eedc4e5a RG |
786 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
787 | int val) | |
db9adbcb JW |
788 | { |
789 | struct mem_cgroup_per_node *pn; | |
42a30035 | 790 | struct mem_cgroup *memcg; |
db9adbcb | 791 | |
db9adbcb | 792 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 793 | memcg = pn->memcg; |
db9adbcb | 794 | |
be3e67b5 SAS |
795 | /* |
796 | * The caller from rmap relay on disabled preemption becase they never | |
797 | * update their counter from in-interrupt context. For these two | |
798 | * counters we check that the update is never performed from an | |
799 | * interrupt context while other caller need to have disabled interrupt. | |
800 | */ | |
801 | __memcg_stats_lock(); | |
802 | if (IS_ENABLED(CONFIG_DEBUG_VM) && !IS_ENABLED(CONFIG_PREEMPT_RT)) { | |
803 | switch (idx) { | |
804 | case NR_ANON_MAPPED: | |
805 | case NR_FILE_MAPPED: | |
806 | case NR_ANON_THPS: | |
807 | case NR_SHMEM_PMDMAPPED: | |
808 | case NR_FILE_PMDMAPPED: | |
809 | WARN_ON_ONCE(!in_task()); | |
810 | break; | |
811 | default: | |
812 | WARN_ON_ONCE(!irqs_disabled()); | |
813 | } | |
814 | } | |
815 | ||
db9adbcb | 816 | /* Update memcg */ |
11192d9c | 817 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
db9adbcb | 818 | |
b4c46484 | 819 | /* Update lruvec */ |
7e1c0d6f | 820 | __this_cpu_add(pn->lruvec_stats_percpu->state[idx], val); |
11192d9c | 821 | |
5b3be698 | 822 | memcg_rstat_updated(memcg, val); |
be3e67b5 | 823 | memcg_stats_unlock(); |
db9adbcb JW |
824 | } |
825 | ||
eedc4e5a RG |
826 | /** |
827 | * __mod_lruvec_state - update lruvec memory statistics | |
828 | * @lruvec: the lruvec | |
829 | * @idx: the stat item | |
830 | * @val: delta to add to the counter, can be negative | |
831 | * | |
832 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
833 | * function updates the all three counters that are affected by a | |
834 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
835 | */ | |
836 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
837 | int val) | |
838 | { | |
839 | /* Update node */ | |
840 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
841 | ||
842 | /* Update memcg and lruvec */ | |
843 | if (!mem_cgroup_disabled()) | |
844 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
845 | } | |
846 | ||
c47d5032 SB |
847 | void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx, |
848 | int val) | |
849 | { | |
850 | struct page *head = compound_head(page); /* rmap on tail pages */ | |
b4e0b68f | 851 | struct mem_cgroup *memcg; |
c47d5032 SB |
852 | pg_data_t *pgdat = page_pgdat(page); |
853 | struct lruvec *lruvec; | |
854 | ||
b4e0b68f MS |
855 | rcu_read_lock(); |
856 | memcg = page_memcg(head); | |
c47d5032 | 857 | /* Untracked pages have no memcg, no lruvec. Update only the node */ |
d635a69d | 858 | if (!memcg) { |
b4e0b68f | 859 | rcu_read_unlock(); |
c47d5032 SB |
860 | __mod_node_page_state(pgdat, idx, val); |
861 | return; | |
862 | } | |
863 | ||
d635a69d | 864 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 | 865 | __mod_lruvec_state(lruvec, idx, val); |
b4e0b68f | 866 | rcu_read_unlock(); |
c47d5032 | 867 | } |
f0c0c115 | 868 | EXPORT_SYMBOL(__mod_lruvec_page_state); |
c47d5032 | 869 | |
da3ceeff | 870 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 871 | { |
4f103c63 | 872 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
873 | struct mem_cgroup *memcg; |
874 | struct lruvec *lruvec; | |
875 | ||
876 | rcu_read_lock(); | |
fc4db90f | 877 | memcg = mem_cgroup_from_slab_obj(p); |
ec9f0238 | 878 | |
8faeb1ff MS |
879 | /* |
880 | * Untracked pages have no memcg, no lruvec. Update only the | |
881 | * node. If we reparent the slab objects to the root memcg, | |
882 | * when we free the slab object, we need to update the per-memcg | |
883 | * vmstats to keep it correct for the root memcg. | |
884 | */ | |
885 | if (!memcg) { | |
ec9f0238 RG |
886 | __mod_node_page_state(pgdat, idx, val); |
887 | } else { | |
867e5e1d | 888 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
889 | __mod_lruvec_state(lruvec, idx, val); |
890 | } | |
891 | rcu_read_unlock(); | |
892 | } | |
893 | ||
db9adbcb JW |
894 | /** |
895 | * __count_memcg_events - account VM events in a cgroup | |
896 | * @memcg: the memory cgroup | |
897 | * @idx: the event item | |
f0953a1b | 898 | * @count: the number of events that occurred |
db9adbcb JW |
899 | */ |
900 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
901 | unsigned long count) | |
902 | { | |
8278f1c7 SB |
903 | int index = memcg_events_index(idx); |
904 | ||
905 | if (mem_cgroup_disabled() || index < 0) | |
db9adbcb JW |
906 | return; |
907 | ||
be3e67b5 | 908 | memcg_stats_lock(); |
8278f1c7 | 909 | __this_cpu_add(memcg->vmstats_percpu->events[index], count); |
5b3be698 | 910 | memcg_rstat_updated(memcg, count); |
be3e67b5 | 911 | memcg_stats_unlock(); |
db9adbcb JW |
912 | } |
913 | ||
42a30035 | 914 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 915 | { |
8278f1c7 SB |
916 | int index = memcg_events_index(event); |
917 | ||
918 | if (index < 0) | |
919 | return 0; | |
920 | return READ_ONCE(memcg->vmstats->events[index]); | |
e9f8974f JW |
921 | } |
922 | ||
42a30035 JW |
923 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
924 | { | |
815744d7 JW |
925 | long x = 0; |
926 | int cpu; | |
8278f1c7 SB |
927 | int index = memcg_events_index(event); |
928 | ||
929 | if (index < 0) | |
930 | return 0; | |
815744d7 JW |
931 | |
932 | for_each_possible_cpu(cpu) | |
8278f1c7 | 933 | x += per_cpu(memcg->vmstats_percpu->events[index], cpu); |
815744d7 | 934 | return x; |
42a30035 JW |
935 | } |
936 | ||
c0ff4b85 | 937 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
3fba69a5 | 938 | int nr_pages) |
d52aa412 | 939 | { |
e401f176 KH |
940 | /* pagein of a big page is an event. So, ignore page size */ |
941 | if (nr_pages > 0) | |
c9019e9b | 942 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 943 | else { |
c9019e9b | 944 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
945 | nr_pages = -nr_pages; /* for event */ |
946 | } | |
e401f176 | 947 | |
871789d4 | 948 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
949 | } |
950 | ||
f53d7ce3 JW |
951 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
952 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
953 | { |
954 | unsigned long val, next; | |
955 | ||
871789d4 CD |
956 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
957 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 958 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 959 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
960 | switch (target) { |
961 | case MEM_CGROUP_TARGET_THRESH: | |
962 | next = val + THRESHOLDS_EVENTS_TARGET; | |
963 | break; | |
bb4cc1a8 AM |
964 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
965 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
966 | break; | |
f53d7ce3 JW |
967 | default: |
968 | break; | |
969 | } | |
871789d4 | 970 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 971 | return true; |
7a159cc9 | 972 | } |
f53d7ce3 | 973 | return false; |
d2265e6f KH |
974 | } |
975 | ||
976 | /* | |
977 | * Check events in order. | |
978 | * | |
979 | */ | |
8e88bd2d | 980 | static void memcg_check_events(struct mem_cgroup *memcg, int nid) |
d2265e6f | 981 | { |
2343e88d SAS |
982 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
983 | return; | |
984 | ||
d2265e6f | 985 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
986 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
987 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 988 | bool do_softlimit; |
f53d7ce3 | 989 | |
bb4cc1a8 AM |
990 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
991 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 992 | mem_cgroup_threshold(memcg); |
bb4cc1a8 | 993 | if (unlikely(do_softlimit)) |
8e88bd2d | 994 | mem_cgroup_update_tree(memcg, nid); |
0a31bc97 | 995 | } |
d2265e6f KH |
996 | } |
997 | ||
cf475ad2 | 998 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 999 | { |
31a78f23 BS |
1000 | /* |
1001 | * mm_update_next_owner() may clear mm->owner to NULL | |
1002 | * if it races with swapoff, page migration, etc. | |
1003 | * So this can be called with p == NULL. | |
1004 | */ | |
1005 | if (unlikely(!p)) | |
1006 | return NULL; | |
1007 | ||
073219e9 | 1008 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1009 | } |
33398cf2 | 1010 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1011 | |
04f94e3f DS |
1012 | static __always_inline struct mem_cgroup *active_memcg(void) |
1013 | { | |
55a68c82 | 1014 | if (!in_task()) |
04f94e3f DS |
1015 | return this_cpu_read(int_active_memcg); |
1016 | else | |
1017 | return current->active_memcg; | |
1018 | } | |
1019 | ||
d46eb14b SB |
1020 | /** |
1021 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1022 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1023 | * | |
04f94e3f DS |
1024 | * Obtain a reference on mm->memcg and returns it if successful. If mm |
1025 | * is NULL, then the memcg is chosen as follows: | |
1026 | * 1) The active memcg, if set. | |
1027 | * 2) current->mm->memcg, if available | |
1028 | * 3) root memcg | |
1029 | * If mem_cgroup is disabled, NULL is returned. | |
d46eb14b SB |
1030 | */ |
1031 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1032 | { |
d46eb14b SB |
1033 | struct mem_cgroup *memcg; |
1034 | ||
1035 | if (mem_cgroup_disabled()) | |
1036 | return NULL; | |
0b7f569e | 1037 | |
2884b6b7 MS |
1038 | /* |
1039 | * Page cache insertions can happen without an | |
1040 | * actual mm context, e.g. during disk probing | |
1041 | * on boot, loopback IO, acct() writes etc. | |
1042 | * | |
1043 | * No need to css_get on root memcg as the reference | |
1044 | * counting is disabled on the root level in the | |
1045 | * cgroup core. See CSS_NO_REF. | |
1046 | */ | |
04f94e3f DS |
1047 | if (unlikely(!mm)) { |
1048 | memcg = active_memcg(); | |
1049 | if (unlikely(memcg)) { | |
1050 | /* remote memcg must hold a ref */ | |
1051 | css_get(&memcg->css); | |
1052 | return memcg; | |
1053 | } | |
1054 | mm = current->mm; | |
1055 | if (unlikely(!mm)) | |
1056 | return root_mem_cgroup; | |
1057 | } | |
2884b6b7 | 1058 | |
54595fe2 KH |
1059 | rcu_read_lock(); |
1060 | do { | |
2884b6b7 MS |
1061 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1062 | if (unlikely(!memcg)) | |
df381975 | 1063 | memcg = root_mem_cgroup; |
00d484f3 | 1064 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1065 | rcu_read_unlock(); |
c0ff4b85 | 1066 | return memcg; |
54595fe2 | 1067 | } |
d46eb14b SB |
1068 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1069 | ||
4127c650 RG |
1070 | static __always_inline bool memcg_kmem_bypass(void) |
1071 | { | |
1072 | /* Allow remote memcg charging from any context. */ | |
1073 | if (unlikely(active_memcg())) | |
1074 | return false; | |
1075 | ||
1076 | /* Memcg to charge can't be determined. */ | |
6126891c | 1077 | if (!in_task() || !current->mm || (current->flags & PF_KTHREAD)) |
4127c650 RG |
1078 | return true; |
1079 | ||
1080 | return false; | |
1081 | } | |
1082 | ||
5660048c JW |
1083 | /** |
1084 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1085 | * @root: hierarchy root | |
1086 | * @prev: previously returned memcg, NULL on first invocation | |
1087 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1088 | * | |
1089 | * Returns references to children of the hierarchy below @root, or | |
1090 | * @root itself, or %NULL after a full round-trip. | |
1091 | * | |
1092 | * Caller must pass the return value in @prev on subsequent | |
1093 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1094 | * to cancel a hierarchy walk before the round-trip is complete. | |
1095 | * | |
05bdc520 ML |
1096 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1097 | * in the hierarchy among all concurrent reclaimers operating on the | |
1098 | * same node. | |
5660048c | 1099 | */ |
694fbc0f | 1100 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1101 | struct mem_cgroup *prev, |
694fbc0f | 1102 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1103 | { |
3f649ab7 | 1104 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1105 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1106 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1107 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1108 | |
694fbc0f AM |
1109 | if (mem_cgroup_disabled()) |
1110 | return NULL; | |
5660048c | 1111 | |
9f3a0d09 JW |
1112 | if (!root) |
1113 | root = root_mem_cgroup; | |
7d74b06f | 1114 | |
542f85f9 | 1115 | rcu_read_lock(); |
5f578161 | 1116 | |
5ac8fb31 | 1117 | if (reclaim) { |
ef8f2327 | 1118 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1119 | |
a3747b53 | 1120 | mz = root->nodeinfo[reclaim->pgdat->node_id]; |
9da83f3f | 1121 | iter = &mz->iter; |
5ac8fb31 | 1122 | |
a9320aae WY |
1123 | /* |
1124 | * On start, join the current reclaim iteration cycle. | |
1125 | * Exit when a concurrent walker completes it. | |
1126 | */ | |
1127 | if (!prev) | |
1128 | reclaim->generation = iter->generation; | |
1129 | else if (reclaim->generation != iter->generation) | |
5ac8fb31 JW |
1130 | goto out_unlock; |
1131 | ||
6df38689 | 1132 | while (1) { |
4db0c3c2 | 1133 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1134 | if (!pos || css_tryget(&pos->css)) |
1135 | break; | |
5ac8fb31 | 1136 | /* |
6df38689 VD |
1137 | * css reference reached zero, so iter->position will |
1138 | * be cleared by ->css_released. However, we should not | |
1139 | * rely on this happening soon, because ->css_released | |
1140 | * is called from a work queue, and by busy-waiting we | |
1141 | * might block it. So we clear iter->position right | |
1142 | * away. | |
5ac8fb31 | 1143 | */ |
6df38689 VD |
1144 | (void)cmpxchg(&iter->position, pos, NULL); |
1145 | } | |
89d8330c WY |
1146 | } else if (prev) { |
1147 | pos = prev; | |
5ac8fb31 JW |
1148 | } |
1149 | ||
1150 | if (pos) | |
1151 | css = &pos->css; | |
1152 | ||
1153 | for (;;) { | |
1154 | css = css_next_descendant_pre(css, &root->css); | |
1155 | if (!css) { | |
1156 | /* | |
1157 | * Reclaimers share the hierarchy walk, and a | |
1158 | * new one might jump in right at the end of | |
1159 | * the hierarchy - make sure they see at least | |
1160 | * one group and restart from the beginning. | |
1161 | */ | |
1162 | if (!prev) | |
1163 | continue; | |
1164 | break; | |
527a5ec9 | 1165 | } |
7d74b06f | 1166 | |
5ac8fb31 JW |
1167 | /* |
1168 | * Verify the css and acquire a reference. The root | |
1169 | * is provided by the caller, so we know it's alive | |
1170 | * and kicking, and don't take an extra reference. | |
1171 | */ | |
41555dad WY |
1172 | if (css == &root->css || css_tryget(css)) { |
1173 | memcg = mem_cgroup_from_css(css); | |
0b8f73e1 | 1174 | break; |
41555dad | 1175 | } |
9f3a0d09 | 1176 | } |
5ac8fb31 JW |
1177 | |
1178 | if (reclaim) { | |
5ac8fb31 | 1179 | /* |
6df38689 VD |
1180 | * The position could have already been updated by a competing |
1181 | * thread, so check that the value hasn't changed since we read | |
1182 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1183 | */ |
6df38689 VD |
1184 | (void)cmpxchg(&iter->position, pos, memcg); |
1185 | ||
5ac8fb31 JW |
1186 | if (pos) |
1187 | css_put(&pos->css); | |
1188 | ||
1189 | if (!memcg) | |
1190 | iter->generation++; | |
9f3a0d09 | 1191 | } |
5ac8fb31 | 1192 | |
542f85f9 MH |
1193 | out_unlock: |
1194 | rcu_read_unlock(); | |
c40046f3 MH |
1195 | if (prev && prev != root) |
1196 | css_put(&prev->css); | |
1197 | ||
9f3a0d09 | 1198 | return memcg; |
14067bb3 | 1199 | } |
7d74b06f | 1200 | |
5660048c JW |
1201 | /** |
1202 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1203 | * @root: hierarchy root | |
1204 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1205 | */ | |
1206 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1207 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1208 | { |
1209 | if (!root) | |
1210 | root = root_mem_cgroup; | |
1211 | if (prev && prev != root) | |
1212 | css_put(&prev->css); | |
1213 | } | |
7d74b06f | 1214 | |
54a83d6b MC |
1215 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1216 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1217 | { |
6df38689 | 1218 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1219 | struct mem_cgroup_per_node *mz; |
1220 | int nid; | |
6df38689 | 1221 | |
54a83d6b | 1222 | for_each_node(nid) { |
a3747b53 | 1223 | mz = from->nodeinfo[nid]; |
9da83f3f YS |
1224 | iter = &mz->iter; |
1225 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1226 | } |
1227 | } | |
1228 | ||
54a83d6b MC |
1229 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1230 | { | |
1231 | struct mem_cgroup *memcg = dead_memcg; | |
1232 | struct mem_cgroup *last; | |
1233 | ||
1234 | do { | |
1235 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1236 | last = memcg; | |
1237 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1238 | ||
1239 | /* | |
b8dd3ee9 | 1240 | * When cgroup1 non-hierarchy mode is used, |
54a83d6b MC |
1241 | * parent_mem_cgroup() does not walk all the way up to the |
1242 | * cgroup root (root_mem_cgroup). So we have to handle | |
1243 | * dead_memcg from cgroup root separately. | |
1244 | */ | |
1245 | if (last != root_mem_cgroup) | |
1246 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1247 | dead_memcg); | |
1248 | } | |
1249 | ||
7c5f64f8 VD |
1250 | /** |
1251 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1252 | * @memcg: hierarchy root | |
1253 | * @fn: function to call for each task | |
1254 | * @arg: argument passed to @fn | |
1255 | * | |
1256 | * This function iterates over tasks attached to @memcg or to any of its | |
1257 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1258 | * value, the function breaks the iteration loop and returns the value. | |
1259 | * Otherwise, it will iterate over all tasks and return 0. | |
1260 | * | |
1261 | * This function must not be called for the root memory cgroup. | |
1262 | */ | |
1263 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1264 | int (*fn)(struct task_struct *, void *), void *arg) | |
1265 | { | |
1266 | struct mem_cgroup *iter; | |
1267 | int ret = 0; | |
1268 | ||
1269 | BUG_ON(memcg == root_mem_cgroup); | |
1270 | ||
1271 | for_each_mem_cgroup_tree(iter, memcg) { | |
1272 | struct css_task_iter it; | |
1273 | struct task_struct *task; | |
1274 | ||
f168a9a5 | 1275 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1276 | while (!ret && (task = css_task_iter_next(&it))) |
1277 | ret = fn(task, arg); | |
1278 | css_task_iter_end(&it); | |
1279 | if (ret) { | |
1280 | mem_cgroup_iter_break(memcg, iter); | |
1281 | break; | |
1282 | } | |
1283 | } | |
1284 | return ret; | |
1285 | } | |
1286 | ||
6168d0da | 1287 | #ifdef CONFIG_DEBUG_VM |
e809c3fe | 1288 | void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio) |
6168d0da AS |
1289 | { |
1290 | struct mem_cgroup *memcg; | |
1291 | ||
1292 | if (mem_cgroup_disabled()) | |
1293 | return; | |
1294 | ||
e809c3fe | 1295 | memcg = folio_memcg(folio); |
6168d0da AS |
1296 | |
1297 | if (!memcg) | |
e809c3fe | 1298 | VM_BUG_ON_FOLIO(lruvec_memcg(lruvec) != root_mem_cgroup, folio); |
6168d0da | 1299 | else |
e809c3fe | 1300 | VM_BUG_ON_FOLIO(lruvec_memcg(lruvec) != memcg, folio); |
6168d0da AS |
1301 | } |
1302 | #endif | |
1303 | ||
6168d0da | 1304 | /** |
e809c3fe MWO |
1305 | * folio_lruvec_lock - Lock the lruvec for a folio. |
1306 | * @folio: Pointer to the folio. | |
6168d0da | 1307 | * |
d7e3aba5 | 1308 | * These functions are safe to use under any of the following conditions: |
e809c3fe MWO |
1309 | * - folio locked |
1310 | * - folio_test_lru false | |
1311 | * - folio_memcg_lock() | |
1312 | * - folio frozen (refcount of 0) | |
1313 | * | |
1314 | * Return: The lruvec this folio is on with its lock held. | |
6168d0da | 1315 | */ |
e809c3fe | 1316 | struct lruvec *folio_lruvec_lock(struct folio *folio) |
6168d0da | 1317 | { |
e809c3fe | 1318 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1319 | |
6168d0da | 1320 | spin_lock(&lruvec->lru_lock); |
e809c3fe | 1321 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1322 | |
1323 | return lruvec; | |
1324 | } | |
1325 | ||
e809c3fe MWO |
1326 | /** |
1327 | * folio_lruvec_lock_irq - Lock the lruvec for a folio. | |
1328 | * @folio: Pointer to the folio. | |
1329 | * | |
1330 | * These functions are safe to use under any of the following conditions: | |
1331 | * - folio locked | |
1332 | * - folio_test_lru false | |
1333 | * - folio_memcg_lock() | |
1334 | * - folio frozen (refcount of 0) | |
1335 | * | |
1336 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1337 | * disabled. | |
1338 | */ | |
1339 | struct lruvec *folio_lruvec_lock_irq(struct folio *folio) | |
6168d0da | 1340 | { |
e809c3fe | 1341 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1342 | |
6168d0da | 1343 | spin_lock_irq(&lruvec->lru_lock); |
e809c3fe | 1344 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1345 | |
1346 | return lruvec; | |
1347 | } | |
1348 | ||
e809c3fe MWO |
1349 | /** |
1350 | * folio_lruvec_lock_irqsave - Lock the lruvec for a folio. | |
1351 | * @folio: Pointer to the folio. | |
1352 | * @flags: Pointer to irqsave flags. | |
1353 | * | |
1354 | * These functions are safe to use under any of the following conditions: | |
1355 | * - folio locked | |
1356 | * - folio_test_lru false | |
1357 | * - folio_memcg_lock() | |
1358 | * - folio frozen (refcount of 0) | |
1359 | * | |
1360 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1361 | * disabled. | |
1362 | */ | |
1363 | struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio, | |
1364 | unsigned long *flags) | |
6168d0da | 1365 | { |
e809c3fe | 1366 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1367 | |
6168d0da | 1368 | spin_lock_irqsave(&lruvec->lru_lock, *flags); |
e809c3fe | 1369 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1370 | |
1371 | return lruvec; | |
1372 | } | |
1373 | ||
925b7673 | 1374 | /** |
fa9add64 HD |
1375 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1376 | * @lruvec: mem_cgroup per zone lru vector | |
1377 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1378 | * @zid: zone id of the accounted pages |
fa9add64 | 1379 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1380 | * |
ca707239 | 1381 | * This function must be called under lru_lock, just before a page is added |
07ca7606 | 1382 | * to or just after a page is removed from an lru list. |
3f58a829 | 1383 | */ |
fa9add64 | 1384 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1385 | int zid, int nr_pages) |
3f58a829 | 1386 | { |
ef8f2327 | 1387 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1388 | unsigned long *lru_size; |
ca707239 | 1389 | long size; |
3f58a829 MK |
1390 | |
1391 | if (mem_cgroup_disabled()) | |
1392 | return; | |
1393 | ||
ef8f2327 | 1394 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1395 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1396 | |
1397 | if (nr_pages < 0) | |
1398 | *lru_size += nr_pages; | |
1399 | ||
1400 | size = *lru_size; | |
b4536f0c MH |
1401 | if (WARN_ONCE(size < 0, |
1402 | "%s(%p, %d, %d): lru_size %ld\n", | |
1403 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1404 | VM_BUG_ON(1); |
1405 | *lru_size = 0; | |
1406 | } | |
1407 | ||
1408 | if (nr_pages > 0) | |
1409 | *lru_size += nr_pages; | |
08e552c6 | 1410 | } |
544122e5 | 1411 | |
19942822 | 1412 | /** |
9d11ea9f | 1413 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1414 | * @memcg: the memory cgroup |
19942822 | 1415 | * |
9d11ea9f | 1416 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1417 | * pages. |
19942822 | 1418 | */ |
c0ff4b85 | 1419 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1420 | { |
3e32cb2e JW |
1421 | unsigned long margin = 0; |
1422 | unsigned long count; | |
1423 | unsigned long limit; | |
9d11ea9f | 1424 | |
3e32cb2e | 1425 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1426 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1427 | if (count < limit) |
1428 | margin = limit - count; | |
1429 | ||
7941d214 | 1430 | if (do_memsw_account()) { |
3e32cb2e | 1431 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1432 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1433 | if (count < limit) |
3e32cb2e | 1434 | margin = min(margin, limit - count); |
cbedbac3 LR |
1435 | else |
1436 | margin = 0; | |
3e32cb2e JW |
1437 | } |
1438 | ||
1439 | return margin; | |
19942822 JW |
1440 | } |
1441 | ||
32047e2a | 1442 | /* |
bdcbb659 | 1443 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1444 | * |
bdcbb659 QH |
1445 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1446 | * moving cgroups. This is for waiting at high-memory pressure | |
1447 | * caused by "move". | |
32047e2a | 1448 | */ |
c0ff4b85 | 1449 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1450 | { |
2bd9bb20 KH |
1451 | struct mem_cgroup *from; |
1452 | struct mem_cgroup *to; | |
4b534334 | 1453 | bool ret = false; |
2bd9bb20 KH |
1454 | /* |
1455 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1456 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1457 | */ | |
1458 | spin_lock(&mc.lock); | |
1459 | from = mc.from; | |
1460 | to = mc.to; | |
1461 | if (!from) | |
1462 | goto unlock; | |
3e92041d | 1463 | |
2314b42d JW |
1464 | ret = mem_cgroup_is_descendant(from, memcg) || |
1465 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1466 | unlock: |
1467 | spin_unlock(&mc.lock); | |
4b534334 KH |
1468 | return ret; |
1469 | } | |
1470 | ||
c0ff4b85 | 1471 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1472 | { |
1473 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1474 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1475 | DEFINE_WAIT(wait); |
1476 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1477 | /* moving charge context might have finished. */ | |
1478 | if (mc.moving_task) | |
1479 | schedule(); | |
1480 | finish_wait(&mc.waitq, &wait); | |
1481 | return true; | |
1482 | } | |
1483 | } | |
1484 | return false; | |
1485 | } | |
1486 | ||
5f9a4f4a MS |
1487 | struct memory_stat { |
1488 | const char *name; | |
5f9a4f4a MS |
1489 | unsigned int idx; |
1490 | }; | |
1491 | ||
57b2847d | 1492 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1493 | { "anon", NR_ANON_MAPPED }, |
1494 | { "file", NR_FILE_PAGES }, | |
a8c49af3 | 1495 | { "kernel", MEMCG_KMEM }, |
fff66b79 MS |
1496 | { "kernel_stack", NR_KERNEL_STACK_KB }, |
1497 | { "pagetables", NR_PAGETABLE }, | |
1498 | { "percpu", MEMCG_PERCPU_B }, | |
1499 | { "sock", MEMCG_SOCK }, | |
4e5aa1f4 | 1500 | { "vmalloc", MEMCG_VMALLOC }, |
fff66b79 | 1501 | { "shmem", NR_SHMEM }, |
f4840ccf JW |
1502 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
1503 | { "zswap", MEMCG_ZSWAP_B }, | |
1504 | { "zswapped", MEMCG_ZSWAPPED }, | |
1505 | #endif | |
fff66b79 MS |
1506 | { "file_mapped", NR_FILE_MAPPED }, |
1507 | { "file_dirty", NR_FILE_DIRTY }, | |
1508 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1509 | #ifdef CONFIG_SWAP |
1510 | { "swapcached", NR_SWAPCACHE }, | |
1511 | #endif | |
5f9a4f4a | 1512 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1513 | { "anon_thp", NR_ANON_THPS }, |
1514 | { "file_thp", NR_FILE_THPS }, | |
1515 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1516 | #endif |
fff66b79 MS |
1517 | { "inactive_anon", NR_INACTIVE_ANON }, |
1518 | { "active_anon", NR_ACTIVE_ANON }, | |
1519 | { "inactive_file", NR_INACTIVE_FILE }, | |
1520 | { "active_file", NR_ACTIVE_FILE }, | |
1521 | { "unevictable", NR_UNEVICTABLE }, | |
1522 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1523 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1524 | |
1525 | /* The memory events */ | |
fff66b79 MS |
1526 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1527 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1528 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1529 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1530 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1531 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1532 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1533 | }; |
1534 | ||
fff66b79 MS |
1535 | /* Translate stat items to the correct unit for memory.stat output */ |
1536 | static int memcg_page_state_unit(int item) | |
1537 | { | |
1538 | switch (item) { | |
1539 | case MEMCG_PERCPU_B: | |
f4840ccf | 1540 | case MEMCG_ZSWAP_B: |
fff66b79 MS |
1541 | case NR_SLAB_RECLAIMABLE_B: |
1542 | case NR_SLAB_UNRECLAIMABLE_B: | |
1543 | case WORKINGSET_REFAULT_ANON: | |
1544 | case WORKINGSET_REFAULT_FILE: | |
1545 | case WORKINGSET_ACTIVATE_ANON: | |
1546 | case WORKINGSET_ACTIVATE_FILE: | |
1547 | case WORKINGSET_RESTORE_ANON: | |
1548 | case WORKINGSET_RESTORE_FILE: | |
1549 | case WORKINGSET_NODERECLAIM: | |
1550 | return 1; | |
1551 | case NR_KERNEL_STACK_KB: | |
1552 | return SZ_1K; | |
1553 | default: | |
1554 | return PAGE_SIZE; | |
1555 | } | |
1556 | } | |
1557 | ||
1558 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1559 | int item) | |
1560 | { | |
1561 | return memcg_page_state(memcg, item) * memcg_page_state_unit(item); | |
1562 | } | |
1563 | ||
68aaee14 | 1564 | static void memory_stat_format(struct mem_cgroup *memcg, char *buf, int bufsize) |
c8713d0b JW |
1565 | { |
1566 | struct seq_buf s; | |
1567 | int i; | |
71cd3113 | 1568 | |
68aaee14 | 1569 | seq_buf_init(&s, buf, bufsize); |
c8713d0b JW |
1570 | |
1571 | /* | |
1572 | * Provide statistics on the state of the memory subsystem as | |
1573 | * well as cumulative event counters that show past behavior. | |
1574 | * | |
1575 | * This list is ordered following a combination of these gradients: | |
1576 | * 1) generic big picture -> specifics and details | |
1577 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1578 | * | |
1579 | * Current memory state: | |
1580 | */ | |
fd25a9e0 | 1581 | mem_cgroup_flush_stats(); |
c8713d0b | 1582 | |
5f9a4f4a MS |
1583 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1584 | u64 size; | |
c8713d0b | 1585 | |
fff66b79 | 1586 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5f9a4f4a | 1587 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1588 | |
5f9a4f4a | 1589 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1590 | size += memcg_page_state_output(memcg, |
1591 | NR_SLAB_RECLAIMABLE_B); | |
5f9a4f4a MS |
1592 | seq_buf_printf(&s, "slab %llu\n", size); |
1593 | } | |
1594 | } | |
c8713d0b JW |
1595 | |
1596 | /* Accumulated memory events */ | |
c8713d0b JW |
1597 | seq_buf_printf(&s, "pgscan %lu\n", |
1598 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1599 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1600 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1601 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1602 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
c8713d0b | 1603 | |
8278f1c7 SB |
1604 | for (i = 0; i < ARRAY_SIZE(memcg_vm_event_stat); i++) { |
1605 | if (memcg_vm_event_stat[i] == PGPGIN || | |
1606 | memcg_vm_event_stat[i] == PGPGOUT) | |
1607 | continue; | |
1608 | ||
673520f8 QZ |
1609 | seq_buf_printf(&s, "%s %lu\n", |
1610 | vm_event_name(memcg_vm_event_stat[i]), | |
1611 | memcg_events(memcg, memcg_vm_event_stat[i])); | |
8278f1c7 | 1612 | } |
c8713d0b JW |
1613 | |
1614 | /* The above should easily fit into one page */ | |
1615 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
c8713d0b | 1616 | } |
71cd3113 | 1617 | |
58cf188e | 1618 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1619 | /** |
f0c867d9 | 1620 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1621 | * memory controller. | |
e222432b BS |
1622 | * @memcg: The memory cgroup that went over limit |
1623 | * @p: Task that is going to be killed | |
1624 | * | |
1625 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1626 | * enabled | |
1627 | */ | |
f0c867d9 | 1628 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1629 | { |
e222432b BS |
1630 | rcu_read_lock(); |
1631 | ||
f0c867d9 | 1632 | if (memcg) { |
1633 | pr_cont(",oom_memcg="); | |
1634 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1635 | } else | |
1636 | pr_cont(",global_oom"); | |
2415b9f5 | 1637 | if (p) { |
f0c867d9 | 1638 | pr_cont(",task_memcg="); |
2415b9f5 | 1639 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1640 | } |
e222432b | 1641 | rcu_read_unlock(); |
f0c867d9 | 1642 | } |
1643 | ||
1644 | /** | |
1645 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1646 | * memory controller. | |
1647 | * @memcg: The memory cgroup that went over limit | |
1648 | */ | |
1649 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1650 | { | |
68aaee14 TH |
1651 | /* Use static buffer, for the caller is holding oom_lock. */ |
1652 | static char buf[PAGE_SIZE]; | |
1653 | ||
1654 | lockdep_assert_held(&oom_lock); | |
e222432b | 1655 | |
3e32cb2e JW |
1656 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1657 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1658 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1659 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1660 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1661 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1662 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1663 | else { |
1664 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1665 | K((u64)page_counter_read(&memcg->memsw)), | |
1666 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1667 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1668 | K((u64)page_counter_read(&memcg->kmem)), | |
1669 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1670 | } |
c8713d0b JW |
1671 | |
1672 | pr_info("Memory cgroup stats for "); | |
1673 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1674 | pr_cont(":"); | |
68aaee14 | 1675 | memory_stat_format(memcg, buf, sizeof(buf)); |
c8713d0b | 1676 | pr_info("%s", buf); |
e222432b BS |
1677 | } |
1678 | ||
a63d83f4 DR |
1679 | /* |
1680 | * Return the memory (and swap, if configured) limit for a memcg. | |
1681 | */ | |
bbec2e15 | 1682 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1683 | { |
8d387a5f WL |
1684 | unsigned long max = READ_ONCE(memcg->memory.max); |
1685 | ||
1686 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1687 | if (mem_cgroup_swappiness(memcg)) | |
1688 | max += min(READ_ONCE(memcg->swap.max), | |
1689 | (unsigned long)total_swap_pages); | |
1690 | } else { /* v1 */ | |
1691 | if (mem_cgroup_swappiness(memcg)) { | |
1692 | /* Calculate swap excess capacity from memsw limit */ | |
1693 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1694 | ||
1695 | max += min(swap, (unsigned long)total_swap_pages); | |
1696 | } | |
9a5a8f19 | 1697 | } |
bbec2e15 | 1698 | return max; |
a63d83f4 DR |
1699 | } |
1700 | ||
9783aa99 CD |
1701 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1702 | { | |
1703 | return page_counter_read(&memcg->memory); | |
1704 | } | |
1705 | ||
b6e6edcf | 1706 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1707 | int order) |
9cbb78bb | 1708 | { |
6e0fc46d DR |
1709 | struct oom_control oc = { |
1710 | .zonelist = NULL, | |
1711 | .nodemask = NULL, | |
2a966b77 | 1712 | .memcg = memcg, |
6e0fc46d DR |
1713 | .gfp_mask = gfp_mask, |
1714 | .order = order, | |
6e0fc46d | 1715 | }; |
1378b37d | 1716 | bool ret = true; |
9cbb78bb | 1717 | |
7775face TH |
1718 | if (mutex_lock_killable(&oom_lock)) |
1719 | return true; | |
1378b37d YS |
1720 | |
1721 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1722 | goto unlock; | |
1723 | ||
7775face TH |
1724 | /* |
1725 | * A few threads which were not waiting at mutex_lock_killable() can | |
1726 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1727 | */ | |
a4ebf1b6 | 1728 | ret = task_is_dying() || out_of_memory(&oc); |
1378b37d YS |
1729 | |
1730 | unlock: | |
dc56401f | 1731 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1732 | return ret; |
9cbb78bb DR |
1733 | } |
1734 | ||
0608f43d | 1735 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1736 | pg_data_t *pgdat, |
0608f43d AM |
1737 | gfp_t gfp_mask, |
1738 | unsigned long *total_scanned) | |
1739 | { | |
1740 | struct mem_cgroup *victim = NULL; | |
1741 | int total = 0; | |
1742 | int loop = 0; | |
1743 | unsigned long excess; | |
1744 | unsigned long nr_scanned; | |
1745 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1746 | .pgdat = pgdat, |
0608f43d AM |
1747 | }; |
1748 | ||
3e32cb2e | 1749 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1750 | |
1751 | while (1) { | |
1752 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1753 | if (!victim) { | |
1754 | loop++; | |
1755 | if (loop >= 2) { | |
1756 | /* | |
1757 | * If we have not been able to reclaim | |
1758 | * anything, it might because there are | |
1759 | * no reclaimable pages under this hierarchy | |
1760 | */ | |
1761 | if (!total) | |
1762 | break; | |
1763 | /* | |
1764 | * We want to do more targeted reclaim. | |
1765 | * excess >> 2 is not to excessive so as to | |
1766 | * reclaim too much, nor too less that we keep | |
1767 | * coming back to reclaim from this cgroup | |
1768 | */ | |
1769 | if (total >= (excess >> 2) || | |
1770 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1771 | break; | |
1772 | } | |
1773 | continue; | |
1774 | } | |
a9dd0a83 | 1775 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1776 | pgdat, &nr_scanned); |
0608f43d | 1777 | *total_scanned += nr_scanned; |
3e32cb2e | 1778 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1779 | break; |
6d61ef40 | 1780 | } |
0608f43d AM |
1781 | mem_cgroup_iter_break(root_memcg, victim); |
1782 | return total; | |
6d61ef40 BS |
1783 | } |
1784 | ||
0056f4e6 JW |
1785 | #ifdef CONFIG_LOCKDEP |
1786 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1787 | .name = "memcg_oom_lock", | |
1788 | }; | |
1789 | #endif | |
1790 | ||
fb2a6fc5 JW |
1791 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1792 | ||
867578cb KH |
1793 | /* |
1794 | * Check OOM-Killer is already running under our hierarchy. | |
1795 | * If someone is running, return false. | |
1796 | */ | |
fb2a6fc5 | 1797 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1798 | { |
79dfdacc | 1799 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1800 | |
fb2a6fc5 JW |
1801 | spin_lock(&memcg_oom_lock); |
1802 | ||
9f3a0d09 | 1803 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1804 | if (iter->oom_lock) { |
79dfdacc MH |
1805 | /* |
1806 | * this subtree of our hierarchy is already locked | |
1807 | * so we cannot give a lock. | |
1808 | */ | |
79dfdacc | 1809 | failed = iter; |
9f3a0d09 JW |
1810 | mem_cgroup_iter_break(memcg, iter); |
1811 | break; | |
23751be0 JW |
1812 | } else |
1813 | iter->oom_lock = true; | |
7d74b06f | 1814 | } |
867578cb | 1815 | |
fb2a6fc5 JW |
1816 | if (failed) { |
1817 | /* | |
1818 | * OK, we failed to lock the whole subtree so we have | |
1819 | * to clean up what we set up to the failing subtree | |
1820 | */ | |
1821 | for_each_mem_cgroup_tree(iter, memcg) { | |
1822 | if (iter == failed) { | |
1823 | mem_cgroup_iter_break(memcg, iter); | |
1824 | break; | |
1825 | } | |
1826 | iter->oom_lock = false; | |
79dfdacc | 1827 | } |
0056f4e6 JW |
1828 | } else |
1829 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1830 | |
1831 | spin_unlock(&memcg_oom_lock); | |
1832 | ||
1833 | return !failed; | |
a636b327 | 1834 | } |
0b7f569e | 1835 | |
fb2a6fc5 | 1836 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1837 | { |
7d74b06f KH |
1838 | struct mem_cgroup *iter; |
1839 | ||
fb2a6fc5 | 1840 | spin_lock(&memcg_oom_lock); |
5facae4f | 1841 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1842 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1843 | iter->oom_lock = false; |
fb2a6fc5 | 1844 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1845 | } |
1846 | ||
c0ff4b85 | 1847 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1848 | { |
1849 | struct mem_cgroup *iter; | |
1850 | ||
c2b42d3c | 1851 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1852 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1853 | iter->under_oom++; |
1854 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1855 | } |
1856 | ||
c0ff4b85 | 1857 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1858 | { |
1859 | struct mem_cgroup *iter; | |
1860 | ||
867578cb | 1861 | /* |
f0953a1b | 1862 | * Be careful about under_oom underflows because a child memcg |
7a52d4d8 | 1863 | * could have been added after mem_cgroup_mark_under_oom. |
867578cb | 1864 | */ |
c2b42d3c | 1865 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1866 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1867 | if (iter->under_oom > 0) |
1868 | iter->under_oom--; | |
1869 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1870 | } |
1871 | ||
867578cb KH |
1872 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1873 | ||
dc98df5a | 1874 | struct oom_wait_info { |
d79154bb | 1875 | struct mem_cgroup *memcg; |
ac6424b9 | 1876 | wait_queue_entry_t wait; |
dc98df5a KH |
1877 | }; |
1878 | ||
ac6424b9 | 1879 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1880 | unsigned mode, int sync, void *arg) |
1881 | { | |
d79154bb HD |
1882 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1883 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1884 | struct oom_wait_info *oom_wait_info; |
1885 | ||
1886 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1887 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1888 | |
2314b42d JW |
1889 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1890 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1891 | return 0; |
dc98df5a KH |
1892 | return autoremove_wake_function(wait, mode, sync, arg); |
1893 | } | |
1894 | ||
c0ff4b85 | 1895 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1896 | { |
c2b42d3c TH |
1897 | /* |
1898 | * For the following lockless ->under_oom test, the only required | |
1899 | * guarantee is that it must see the state asserted by an OOM when | |
1900 | * this function is called as a result of userland actions | |
1901 | * triggered by the notification of the OOM. This is trivially | |
1902 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1903 | * triggering notification. | |
1904 | */ | |
1905 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1906 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1907 | } |
1908 | ||
becdf89d SB |
1909 | /* |
1910 | * Returns true if successfully killed one or more processes. Though in some | |
1911 | * corner cases it can return true even without killing any process. | |
1912 | */ | |
1913 | static bool mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1914 | { |
becdf89d | 1915 | bool locked, ret; |
7056d3a3 | 1916 | |
29ef680a | 1917 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
becdf89d | 1918 | return false; |
29ef680a | 1919 | |
7a1adfdd RG |
1920 | memcg_memory_event(memcg, MEMCG_OOM); |
1921 | ||
867578cb | 1922 | /* |
49426420 JW |
1923 | * We are in the middle of the charge context here, so we |
1924 | * don't want to block when potentially sitting on a callstack | |
1925 | * that holds all kinds of filesystem and mm locks. | |
1926 | * | |
29ef680a MH |
1927 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1928 | * handling until the charge can succeed; remember the context and put | |
1929 | * the task to sleep at the end of the page fault when all locks are | |
1930 | * released. | |
49426420 | 1931 | * |
29ef680a MH |
1932 | * On the other hand, in-kernel OOM killer allows for an async victim |
1933 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1934 | * relying on the oom victim to make a forward progress and we can | |
1935 | * invoke the oom killer here. | |
1936 | * | |
1937 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1938 | * victim and then we have to bail out from the charge path. | |
867578cb | 1939 | */ |
29ef680a | 1940 | if (memcg->oom_kill_disable) { |
becdf89d SB |
1941 | if (current->in_user_fault) { |
1942 | css_get(&memcg->css); | |
1943 | current->memcg_in_oom = memcg; | |
1944 | current->memcg_oom_gfp_mask = mask; | |
1945 | current->memcg_oom_order = order; | |
1946 | } | |
1947 | return false; | |
29ef680a MH |
1948 | } |
1949 | ||
7056d3a3 MH |
1950 | mem_cgroup_mark_under_oom(memcg); |
1951 | ||
1952 | locked = mem_cgroup_oom_trylock(memcg); | |
1953 | ||
1954 | if (locked) | |
1955 | mem_cgroup_oom_notify(memcg); | |
1956 | ||
1957 | mem_cgroup_unmark_under_oom(memcg); | |
becdf89d | 1958 | ret = mem_cgroup_out_of_memory(memcg, mask, order); |
7056d3a3 MH |
1959 | |
1960 | if (locked) | |
1961 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1962 | |
7056d3a3 | 1963 | return ret; |
3812c8c8 JW |
1964 | } |
1965 | ||
1966 | /** | |
1967 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1968 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1969 | * |
49426420 JW |
1970 | * This has to be called at the end of a page fault if the memcg OOM |
1971 | * handler was enabled. | |
3812c8c8 | 1972 | * |
49426420 | 1973 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1974 | * sleep on a waitqueue until the userspace task resolves the |
1975 | * situation. Sleeping directly in the charge context with all kinds | |
1976 | * of locks held is not a good idea, instead we remember an OOM state | |
1977 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1978 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1979 | * |
1980 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1981 | * completed, %false otherwise. |
3812c8c8 | 1982 | */ |
49426420 | 1983 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1984 | { |
626ebc41 | 1985 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1986 | struct oom_wait_info owait; |
49426420 | 1987 | bool locked; |
3812c8c8 JW |
1988 | |
1989 | /* OOM is global, do not handle */ | |
3812c8c8 | 1990 | if (!memcg) |
49426420 | 1991 | return false; |
3812c8c8 | 1992 | |
7c5f64f8 | 1993 | if (!handle) |
49426420 | 1994 | goto cleanup; |
3812c8c8 JW |
1995 | |
1996 | owait.memcg = memcg; | |
1997 | owait.wait.flags = 0; | |
1998 | owait.wait.func = memcg_oom_wake_function; | |
1999 | owait.wait.private = current; | |
2055da97 | 2000 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 2001 | |
3812c8c8 | 2002 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2003 | mem_cgroup_mark_under_oom(memcg); |
2004 | ||
2005 | locked = mem_cgroup_oom_trylock(memcg); | |
2006 | ||
2007 | if (locked) | |
2008 | mem_cgroup_oom_notify(memcg); | |
2009 | ||
2010 | if (locked && !memcg->oom_kill_disable) { | |
2011 | mem_cgroup_unmark_under_oom(memcg); | |
2012 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
2013 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
2014 | current->memcg_oom_order); | |
49426420 | 2015 | } else { |
3812c8c8 | 2016 | schedule(); |
49426420 JW |
2017 | mem_cgroup_unmark_under_oom(memcg); |
2018 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2019 | } | |
2020 | ||
2021 | if (locked) { | |
fb2a6fc5 JW |
2022 | mem_cgroup_oom_unlock(memcg); |
2023 | /* | |
2024 | * There is no guarantee that an OOM-lock contender | |
2025 | * sees the wakeups triggered by the OOM kill | |
f0953a1b | 2026 | * uncharges. Wake any sleepers explicitly. |
fb2a6fc5 JW |
2027 | */ |
2028 | memcg_oom_recover(memcg); | |
2029 | } | |
49426420 | 2030 | cleanup: |
626ebc41 | 2031 | current->memcg_in_oom = NULL; |
3812c8c8 | 2032 | css_put(&memcg->css); |
867578cb | 2033 | return true; |
0b7f569e KH |
2034 | } |
2035 | ||
3d8b38eb RG |
2036 | /** |
2037 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2038 | * @victim: task to be killed by the OOM killer | |
2039 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2040 | * | |
2041 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2042 | * by killing all belonging OOM-killable tasks. | |
2043 | * | |
2044 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2045 | */ | |
2046 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2047 | struct mem_cgroup *oom_domain) | |
2048 | { | |
2049 | struct mem_cgroup *oom_group = NULL; | |
2050 | struct mem_cgroup *memcg; | |
2051 | ||
2052 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2053 | return NULL; | |
2054 | ||
2055 | if (!oom_domain) | |
2056 | oom_domain = root_mem_cgroup; | |
2057 | ||
2058 | rcu_read_lock(); | |
2059 | ||
2060 | memcg = mem_cgroup_from_task(victim); | |
2061 | if (memcg == root_mem_cgroup) | |
2062 | goto out; | |
2063 | ||
48fe267c RG |
2064 | /* |
2065 | * If the victim task has been asynchronously moved to a different | |
2066 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2067 | * In this case it's better to ignore memory.group.oom. | |
2068 | */ | |
2069 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2070 | goto out; | |
2071 | ||
3d8b38eb RG |
2072 | /* |
2073 | * Traverse the memory cgroup hierarchy from the victim task's | |
2074 | * cgroup up to the OOMing cgroup (or root) to find the | |
2075 | * highest-level memory cgroup with oom.group set. | |
2076 | */ | |
2077 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2078 | if (memcg->oom_group) | |
2079 | oom_group = memcg; | |
2080 | ||
2081 | if (memcg == oom_domain) | |
2082 | break; | |
2083 | } | |
2084 | ||
2085 | if (oom_group) | |
2086 | css_get(&oom_group->css); | |
2087 | out: | |
2088 | rcu_read_unlock(); | |
2089 | ||
2090 | return oom_group; | |
2091 | } | |
2092 | ||
2093 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2094 | { | |
2095 | pr_info("Tasks in "); | |
2096 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2097 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2098 | } | |
2099 | ||
d7365e78 | 2100 | /** |
f70ad448 MWO |
2101 | * folio_memcg_lock - Bind a folio to its memcg. |
2102 | * @folio: The folio. | |
32047e2a | 2103 | * |
f70ad448 | 2104 | * This function prevents unlocked LRU folios from being moved to |
739f79fc JW |
2105 | * another cgroup. |
2106 | * | |
f70ad448 MWO |
2107 | * It ensures lifetime of the bound memcg. The caller is responsible |
2108 | * for the lifetime of the folio. | |
d69b042f | 2109 | */ |
f70ad448 | 2110 | void folio_memcg_lock(struct folio *folio) |
89c06bd5 KH |
2111 | { |
2112 | struct mem_cgroup *memcg; | |
6de22619 | 2113 | unsigned long flags; |
89c06bd5 | 2114 | |
6de22619 JW |
2115 | /* |
2116 | * The RCU lock is held throughout the transaction. The fast | |
2117 | * path can get away without acquiring the memcg->move_lock | |
2118 | * because page moving starts with an RCU grace period. | |
739f79fc | 2119 | */ |
d7365e78 JW |
2120 | rcu_read_lock(); |
2121 | ||
2122 | if (mem_cgroup_disabled()) | |
1c824a68 | 2123 | return; |
89c06bd5 | 2124 | again: |
f70ad448 | 2125 | memcg = folio_memcg(folio); |
29833315 | 2126 | if (unlikely(!memcg)) |
1c824a68 | 2127 | return; |
d7365e78 | 2128 | |
20ad50d6 AS |
2129 | #ifdef CONFIG_PROVE_LOCKING |
2130 | local_irq_save(flags); | |
2131 | might_lock(&memcg->move_lock); | |
2132 | local_irq_restore(flags); | |
2133 | #endif | |
2134 | ||
bdcbb659 | 2135 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 2136 | return; |
89c06bd5 | 2137 | |
6de22619 | 2138 | spin_lock_irqsave(&memcg->move_lock, flags); |
f70ad448 | 2139 | if (memcg != folio_memcg(folio)) { |
6de22619 | 2140 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2141 | goto again; |
2142 | } | |
6de22619 JW |
2143 | |
2144 | /* | |
1c824a68 JW |
2145 | * When charge migration first begins, we can have multiple |
2146 | * critical sections holding the fast-path RCU lock and one | |
2147 | * holding the slowpath move_lock. Track the task who has the | |
2148 | * move_lock for unlock_page_memcg(). | |
6de22619 JW |
2149 | */ |
2150 | memcg->move_lock_task = current; | |
2151 | memcg->move_lock_flags = flags; | |
89c06bd5 | 2152 | } |
f70ad448 MWO |
2153 | |
2154 | void lock_page_memcg(struct page *page) | |
2155 | { | |
2156 | folio_memcg_lock(page_folio(page)); | |
2157 | } | |
89c06bd5 | 2158 | |
f70ad448 | 2159 | static void __folio_memcg_unlock(struct mem_cgroup *memcg) |
89c06bd5 | 2160 | { |
6de22619 JW |
2161 | if (memcg && memcg->move_lock_task == current) { |
2162 | unsigned long flags = memcg->move_lock_flags; | |
2163 | ||
2164 | memcg->move_lock_task = NULL; | |
2165 | memcg->move_lock_flags = 0; | |
2166 | ||
2167 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2168 | } | |
89c06bd5 | 2169 | |
d7365e78 | 2170 | rcu_read_unlock(); |
89c06bd5 | 2171 | } |
739f79fc JW |
2172 | |
2173 | /** | |
f70ad448 MWO |
2174 | * folio_memcg_unlock - Release the binding between a folio and its memcg. |
2175 | * @folio: The folio. | |
2176 | * | |
2177 | * This releases the binding created by folio_memcg_lock(). This does | |
2178 | * not change the accounting of this folio to its memcg, but it does | |
2179 | * permit others to change it. | |
739f79fc | 2180 | */ |
f70ad448 | 2181 | void folio_memcg_unlock(struct folio *folio) |
739f79fc | 2182 | { |
f70ad448 MWO |
2183 | __folio_memcg_unlock(folio_memcg(folio)); |
2184 | } | |
9da7b521 | 2185 | |
f70ad448 MWO |
2186 | void unlock_page_memcg(struct page *page) |
2187 | { | |
2188 | folio_memcg_unlock(page_folio(page)); | |
739f79fc | 2189 | } |
89c06bd5 | 2190 | |
fead2b86 | 2191 | struct memcg_stock_pcp { |
56751146 | 2192 | local_lock_t stock_lock; |
fead2b86 MH |
2193 | struct mem_cgroup *cached; /* this never be root cgroup */ |
2194 | unsigned int nr_pages; | |
2195 | ||
bf4f0599 RG |
2196 | #ifdef CONFIG_MEMCG_KMEM |
2197 | struct obj_cgroup *cached_objcg; | |
68ac5b3c | 2198 | struct pglist_data *cached_pgdat; |
bf4f0599 | 2199 | unsigned int nr_bytes; |
68ac5b3c WL |
2200 | int nr_slab_reclaimable_b; |
2201 | int nr_slab_unreclaimable_b; | |
bf4f0599 RG |
2202 | #endif |
2203 | ||
cdec2e42 | 2204 | struct work_struct work; |
26fe6168 | 2205 | unsigned long flags; |
a0db00fc | 2206 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 | 2207 | }; |
56751146 SAS |
2208 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = { |
2209 | .stock_lock = INIT_LOCAL_LOCK(stock_lock), | |
2210 | }; | |
9f50fad6 | 2211 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2212 | |
bf4f0599 | 2213 | #ifdef CONFIG_MEMCG_KMEM |
56751146 | 2214 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock); |
bf4f0599 RG |
2215 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
2216 | struct mem_cgroup *root_memcg); | |
a8c49af3 | 2217 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages); |
bf4f0599 RG |
2218 | |
2219 | #else | |
56751146 | 2220 | static inline struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 2221 | { |
56751146 | 2222 | return NULL; |
bf4f0599 RG |
2223 | } |
2224 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2225 | struct mem_cgroup *root_memcg) | |
2226 | { | |
2227 | return false; | |
2228 | } | |
a8c49af3 YA |
2229 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
2230 | { | |
2231 | } | |
bf4f0599 RG |
2232 | #endif |
2233 | ||
a0956d54 SS |
2234 | /** |
2235 | * consume_stock: Try to consume stocked charge on this cpu. | |
2236 | * @memcg: memcg to consume from. | |
2237 | * @nr_pages: how many pages to charge. | |
2238 | * | |
2239 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2240 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2241 | * service an allocation will refill the stock. | |
2242 | * | |
2243 | * returns true if successful, false otherwise. | |
cdec2e42 | 2244 | */ |
a0956d54 | 2245 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2246 | { |
2247 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2248 | unsigned long flags; |
3e32cb2e | 2249 | bool ret = false; |
cdec2e42 | 2250 | |
a983b5eb | 2251 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2252 | return ret; |
a0956d54 | 2253 | |
56751146 | 2254 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2255 | |
2256 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2257 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2258 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2259 | ret = true; |
2260 | } | |
db2ba40c | 2261 | |
56751146 | 2262 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
db2ba40c | 2263 | |
cdec2e42 KH |
2264 | return ret; |
2265 | } | |
2266 | ||
2267 | /* | |
3e32cb2e | 2268 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2269 | */ |
2270 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2271 | { | |
2272 | struct mem_cgroup *old = stock->cached; | |
2273 | ||
1a3e1f40 JW |
2274 | if (!old) |
2275 | return; | |
2276 | ||
11c9ea4e | 2277 | if (stock->nr_pages) { |
3e32cb2e | 2278 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2279 | if (do_memsw_account()) |
3e32cb2e | 2280 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2281 | stock->nr_pages = 0; |
cdec2e42 | 2282 | } |
1a3e1f40 JW |
2283 | |
2284 | css_put(&old->css); | |
cdec2e42 | 2285 | stock->cached = NULL; |
cdec2e42 KH |
2286 | } |
2287 | ||
cdec2e42 KH |
2288 | static void drain_local_stock(struct work_struct *dummy) |
2289 | { | |
db2ba40c | 2290 | struct memcg_stock_pcp *stock; |
56751146 | 2291 | struct obj_cgroup *old = NULL; |
db2ba40c JW |
2292 | unsigned long flags; |
2293 | ||
72f0184c | 2294 | /* |
5c49cf9a MH |
2295 | * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs. |
2296 | * drain_stock races is that we always operate on local CPU stock | |
2297 | * here with IRQ disabled | |
72f0184c | 2298 | */ |
56751146 | 2299 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2300 | |
2301 | stock = this_cpu_ptr(&memcg_stock); | |
56751146 | 2302 | old = drain_obj_stock(stock); |
cdec2e42 | 2303 | drain_stock(stock); |
26fe6168 | 2304 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c | 2305 | |
56751146 SAS |
2306 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
2307 | if (old) | |
2308 | obj_cgroup_put(old); | |
cdec2e42 KH |
2309 | } |
2310 | ||
2311 | /* | |
3e32cb2e | 2312 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2313 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2314 | */ |
af9a3b69 | 2315 | static void __refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2316 | { |
db2ba40c | 2317 | struct memcg_stock_pcp *stock; |
cdec2e42 | 2318 | |
db2ba40c | 2319 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2320 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2321 | drain_stock(stock); |
1a3e1f40 | 2322 | css_get(&memcg->css); |
c0ff4b85 | 2323 | stock->cached = memcg; |
cdec2e42 | 2324 | } |
11c9ea4e | 2325 | stock->nr_pages += nr_pages; |
db2ba40c | 2326 | |
a983b5eb | 2327 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 | 2328 | drain_stock(stock); |
af9a3b69 JW |
2329 | } |
2330 | ||
2331 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) | |
2332 | { | |
2333 | unsigned long flags; | |
475d0487 | 2334 | |
56751146 | 2335 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
af9a3b69 | 2336 | __refill_stock(memcg, nr_pages); |
56751146 | 2337 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
cdec2e42 KH |
2338 | } |
2339 | ||
2340 | /* | |
c0ff4b85 | 2341 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2342 | * of the hierarchy under it. |
cdec2e42 | 2343 | */ |
6d3d6aa2 | 2344 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2345 | { |
26fe6168 | 2346 | int cpu, curcpu; |
d38144b7 | 2347 | |
6d3d6aa2 JW |
2348 | /* If someone's already draining, avoid adding running more workers. */ |
2349 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2350 | return; | |
72f0184c MH |
2351 | /* |
2352 | * Notify other cpus that system-wide "drain" is running | |
2353 | * We do not care about races with the cpu hotplug because cpu down | |
2354 | * as well as workers from this path always operate on the local | |
2355 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2356 | */ | |
0790ed62 SAS |
2357 | migrate_disable(); |
2358 | curcpu = smp_processor_id(); | |
cdec2e42 KH |
2359 | for_each_online_cpu(cpu) { |
2360 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2361 | struct mem_cgroup *memcg; |
e1a366be | 2362 | bool flush = false; |
26fe6168 | 2363 | |
e1a366be | 2364 | rcu_read_lock(); |
c0ff4b85 | 2365 | memcg = stock->cached; |
e1a366be RG |
2366 | if (memcg && stock->nr_pages && |
2367 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2368 | flush = true; | |
27fb0956 | 2369 | else if (obj_stock_flush_required(stock, root_memcg)) |
bf4f0599 | 2370 | flush = true; |
e1a366be RG |
2371 | rcu_read_unlock(); |
2372 | ||
2373 | if (flush && | |
2374 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2375 | if (cpu == curcpu) |
2376 | drain_local_stock(&stock->work); | |
2377 | else | |
2378 | schedule_work_on(cpu, &stock->work); | |
2379 | } | |
cdec2e42 | 2380 | } |
0790ed62 | 2381 | migrate_enable(); |
9f50fad6 | 2382 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2383 | } |
2384 | ||
2cd21c89 JW |
2385 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
2386 | { | |
2387 | struct memcg_stock_pcp *stock; | |
a3d4c05a | 2388 | |
2cd21c89 JW |
2389 | stock = &per_cpu(memcg_stock, cpu); |
2390 | drain_stock(stock); | |
a3d4c05a | 2391 | |
308167fc | 2392 | return 0; |
cdec2e42 KH |
2393 | } |
2394 | ||
b3ff9291 CD |
2395 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2396 | unsigned int nr_pages, | |
2397 | gfp_t gfp_mask) | |
f7e1cb6e | 2398 | { |
b3ff9291 CD |
2399 | unsigned long nr_reclaimed = 0; |
2400 | ||
f7e1cb6e | 2401 | do { |
e22c6ed9 JW |
2402 | unsigned long pflags; |
2403 | ||
d1663a90 JK |
2404 | if (page_counter_read(&memcg->memory) <= |
2405 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2406 | continue; |
e22c6ed9 | 2407 | |
e27be240 | 2408 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2409 | |
2410 | psi_memstall_enter(&pflags); | |
b3ff9291 | 2411 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
73b73bac YA |
2412 | gfp_mask, |
2413 | MEMCG_RECLAIM_MAY_SWAP); | |
e22c6ed9 | 2414 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2415 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2416 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2417 | |
2418 | return nr_reclaimed; | |
f7e1cb6e JW |
2419 | } |
2420 | ||
2421 | static void high_work_func(struct work_struct *work) | |
2422 | { | |
2423 | struct mem_cgroup *memcg; | |
2424 | ||
2425 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2426 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2427 | } |
2428 | ||
0e4b01df CD |
2429 | /* |
2430 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2431 | * enough to still cause a significant slowdown in most cases, while still | |
2432 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2433 | */ | |
2434 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2435 | ||
2436 | /* | |
2437 | * When calculating the delay, we use these either side of the exponentiation to | |
2438 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2439 | * below. | |
2440 | * | |
2441 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2442 | * overage ratio to a delay. | |
ac5ddd0f | 2443 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2444 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2445 | * to produce a reasonable delay curve. | |
2446 | * | |
2447 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2448 | * reasonable delay curve compared to precision-adjusted overage, not | |
2449 | * penalising heavily at first, but still making sure that growth beyond the | |
2450 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2451 | * example, with a high of 100 megabytes: | |
2452 | * | |
2453 | * +-------+------------------------+ | |
2454 | * | usage | time to allocate in ms | | |
2455 | * +-------+------------------------+ | |
2456 | * | 100M | 0 | | |
2457 | * | 101M | 6 | | |
2458 | * | 102M | 25 | | |
2459 | * | 103M | 57 | | |
2460 | * | 104M | 102 | | |
2461 | * | 105M | 159 | | |
2462 | * | 106M | 230 | | |
2463 | * | 107M | 313 | | |
2464 | * | 108M | 409 | | |
2465 | * | 109M | 518 | | |
2466 | * | 110M | 639 | | |
2467 | * | 111M | 774 | | |
2468 | * | 112M | 921 | | |
2469 | * | 113M | 1081 | | |
2470 | * | 114M | 1254 | | |
2471 | * | 115M | 1439 | | |
2472 | * | 116M | 1638 | | |
2473 | * | 117M | 1849 | | |
2474 | * | 118M | 2000 | | |
2475 | * | 119M | 2000 | | |
2476 | * | 120M | 2000 | | |
2477 | * +-------+------------------------+ | |
2478 | */ | |
2479 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2480 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2481 | ||
8a5dbc65 | 2482 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2483 | { |
8a5dbc65 | 2484 | u64 overage; |
b23afb93 | 2485 | |
8a5dbc65 JK |
2486 | if (usage <= high) |
2487 | return 0; | |
e26733e0 | 2488 | |
8a5dbc65 JK |
2489 | /* |
2490 | * Prevent division by 0 in overage calculation by acting as if | |
2491 | * it was a threshold of 1 page | |
2492 | */ | |
2493 | high = max(high, 1UL); | |
9b8b1754 | 2494 | |
8a5dbc65 JK |
2495 | overage = usage - high; |
2496 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2497 | return div64_u64(overage, high); | |
2498 | } | |
e26733e0 | 2499 | |
8a5dbc65 JK |
2500 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2501 | { | |
2502 | u64 overage, max_overage = 0; | |
e26733e0 | 2503 | |
8a5dbc65 JK |
2504 | do { |
2505 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2506 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2507 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2508 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2509 | !mem_cgroup_is_root(memcg)); | |
2510 | ||
8a5dbc65 JK |
2511 | return max_overage; |
2512 | } | |
2513 | ||
4b82ab4f JK |
2514 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2515 | { | |
2516 | u64 overage, max_overage = 0; | |
2517 | ||
2518 | do { | |
2519 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2520 | READ_ONCE(memcg->swap.high)); | |
2521 | if (overage) | |
2522 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2523 | max_overage = max(overage, max_overage); | |
2524 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2525 | !mem_cgroup_is_root(memcg)); | |
2526 | ||
2527 | return max_overage; | |
2528 | } | |
2529 | ||
8a5dbc65 JK |
2530 | /* |
2531 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2532 | * is exceeding its memory.high by checking both it and its ancestors. | |
2533 | */ | |
2534 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2535 | unsigned int nr_pages, | |
2536 | u64 max_overage) | |
2537 | { | |
2538 | unsigned long penalty_jiffies; | |
2539 | ||
e26733e0 CD |
2540 | if (!max_overage) |
2541 | return 0; | |
0e4b01df CD |
2542 | |
2543 | /* | |
0e4b01df CD |
2544 | * We use overage compared to memory.high to calculate the number of |
2545 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2546 | * fairly lenient on small overages, and increasingly harsh when the | |
2547 | * memcg in question makes it clear that it has no intention of stopping | |
2548 | * its crazy behaviour, so we exponentially increase the delay based on | |
2549 | * overage amount. | |
2550 | */ | |
e26733e0 CD |
2551 | penalty_jiffies = max_overage * max_overage * HZ; |
2552 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2553 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2554 | |
2555 | /* | |
2556 | * Factor in the task's own contribution to the overage, such that four | |
2557 | * N-sized allocations are throttled approximately the same as one | |
2558 | * 4N-sized allocation. | |
2559 | * | |
2560 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2561 | * larger the current charge patch is than that. | |
2562 | */ | |
ff144e69 | 2563 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2564 | } |
2565 | ||
2566 | /* | |
2567 | * Scheduled by try_charge() to be executed from the userland return path | |
2568 | * and reclaims memory over the high limit. | |
2569 | */ | |
2570 | void mem_cgroup_handle_over_high(void) | |
2571 | { | |
2572 | unsigned long penalty_jiffies; | |
2573 | unsigned long pflags; | |
b3ff9291 | 2574 | unsigned long nr_reclaimed; |
e26733e0 | 2575 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2576 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2577 | struct mem_cgroup *memcg; |
b3ff9291 | 2578 | bool in_retry = false; |
e26733e0 CD |
2579 | |
2580 | if (likely(!nr_pages)) | |
2581 | return; | |
2582 | ||
2583 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2584 | current->memcg_nr_pages_over_high = 0; |
2585 | ||
b3ff9291 CD |
2586 | retry_reclaim: |
2587 | /* | |
2588 | * The allocating task should reclaim at least the batch size, but for | |
2589 | * subsequent retries we only want to do what's necessary to prevent oom | |
2590 | * or breaching resource isolation. | |
2591 | * | |
2592 | * This is distinct from memory.max or page allocator behaviour because | |
2593 | * memory.high is currently batched, whereas memory.max and the page | |
2594 | * allocator run every time an allocation is made. | |
2595 | */ | |
2596 | nr_reclaimed = reclaim_high(memcg, | |
2597 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2598 | GFP_KERNEL); | |
2599 | ||
e26733e0 CD |
2600 | /* |
2601 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2602 | * allocators proactively to slow down excessive growth. | |
2603 | */ | |
8a5dbc65 JK |
2604 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2605 | mem_find_max_overage(memcg)); | |
0e4b01df | 2606 | |
4b82ab4f JK |
2607 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2608 | swap_find_max_overage(memcg)); | |
2609 | ||
ff144e69 JK |
2610 | /* |
2611 | * Clamp the max delay per usermode return so as to still keep the | |
2612 | * application moving forwards and also permit diagnostics, albeit | |
2613 | * extremely slowly. | |
2614 | */ | |
2615 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2616 | ||
0e4b01df CD |
2617 | /* |
2618 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2619 | * that it's not even worth doing, in an attempt to be nice to those who | |
2620 | * go only a small amount over their memory.high value and maybe haven't | |
2621 | * been aggressively reclaimed enough yet. | |
2622 | */ | |
2623 | if (penalty_jiffies <= HZ / 100) | |
2624 | goto out; | |
2625 | ||
b3ff9291 CD |
2626 | /* |
2627 | * If reclaim is making forward progress but we're still over | |
2628 | * memory.high, we want to encourage that rather than doing allocator | |
2629 | * throttling. | |
2630 | */ | |
2631 | if (nr_reclaimed || nr_retries--) { | |
2632 | in_retry = true; | |
2633 | goto retry_reclaim; | |
2634 | } | |
2635 | ||
0e4b01df CD |
2636 | /* |
2637 | * If we exit early, we're guaranteed to die (since | |
2638 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2639 | * need to account for any ill-begotten jiffies to pay them off later. | |
2640 | */ | |
2641 | psi_memstall_enter(&pflags); | |
2642 | schedule_timeout_killable(penalty_jiffies); | |
2643 | psi_memstall_leave(&pflags); | |
2644 | ||
2645 | out: | |
2646 | css_put(&memcg->css); | |
b23afb93 TH |
2647 | } |
2648 | ||
c5c8b16b MS |
2649 | static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2650 | unsigned int nr_pages) | |
8a9f3ccd | 2651 | { |
a983b5eb | 2652 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2653 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2654 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2655 | struct page_counter *counter; |
6539cc05 | 2656 | unsigned long nr_reclaimed; |
a4ebf1b6 | 2657 | bool passed_oom = false; |
73b73bac | 2658 | unsigned int reclaim_options = MEMCG_RECLAIM_MAY_SWAP; |
b70a2a21 | 2659 | bool drained = false; |
d6e103a7 | 2660 | bool raised_max_event = false; |
e22c6ed9 | 2661 | unsigned long pflags; |
a636b327 | 2662 | |
6539cc05 | 2663 | retry: |
b6b6cc72 | 2664 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2665 | return 0; |
8a9f3ccd | 2666 | |
7941d214 | 2667 | if (!do_memsw_account() || |
6071ca52 JW |
2668 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2669 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2670 | goto done_restock; |
7941d214 | 2671 | if (do_memsw_account()) |
3e32cb2e JW |
2672 | page_counter_uncharge(&memcg->memsw, batch); |
2673 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2674 | } else { |
3e32cb2e | 2675 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
73b73bac | 2676 | reclaim_options &= ~MEMCG_RECLAIM_MAY_SWAP; |
3fbe7244 | 2677 | } |
7a81b88c | 2678 | |
6539cc05 JW |
2679 | if (batch > nr_pages) { |
2680 | batch = nr_pages; | |
2681 | goto retry; | |
2682 | } | |
6d61ef40 | 2683 | |
89a28483 JW |
2684 | /* |
2685 | * Prevent unbounded recursion when reclaim operations need to | |
2686 | * allocate memory. This might exceed the limits temporarily, | |
2687 | * but we prefer facilitating memory reclaim and getting back | |
2688 | * under the limit over triggering OOM kills in these cases. | |
2689 | */ | |
2690 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2691 | goto force; | |
2692 | ||
06b078fc JW |
2693 | if (unlikely(task_in_memcg_oom(current))) |
2694 | goto nomem; | |
2695 | ||
d0164adc | 2696 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2697 | goto nomem; |
4b534334 | 2698 | |
e27be240 | 2699 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
d6e103a7 | 2700 | raised_max_event = true; |
241994ed | 2701 | |
e22c6ed9 | 2702 | psi_memstall_enter(&pflags); |
b70a2a21 | 2703 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
73b73bac | 2704 | gfp_mask, reclaim_options); |
e22c6ed9 | 2705 | psi_memstall_leave(&pflags); |
6539cc05 | 2706 | |
61e02c74 | 2707 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2708 | goto retry; |
28c34c29 | 2709 | |
b70a2a21 | 2710 | if (!drained) { |
6d3d6aa2 | 2711 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2712 | drained = true; |
2713 | goto retry; | |
2714 | } | |
2715 | ||
28c34c29 JW |
2716 | if (gfp_mask & __GFP_NORETRY) |
2717 | goto nomem; | |
6539cc05 JW |
2718 | /* |
2719 | * Even though the limit is exceeded at this point, reclaim | |
2720 | * may have been able to free some pages. Retry the charge | |
2721 | * before killing the task. | |
2722 | * | |
2723 | * Only for regular pages, though: huge pages are rather | |
2724 | * unlikely to succeed so close to the limit, and we fall back | |
2725 | * to regular pages anyway in case of failure. | |
2726 | */ | |
61e02c74 | 2727 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2728 | goto retry; |
2729 | /* | |
2730 | * At task move, charge accounts can be doubly counted. So, it's | |
2731 | * better to wait until the end of task_move if something is going on. | |
2732 | */ | |
2733 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2734 | goto retry; | |
2735 | ||
9b130619 JW |
2736 | if (nr_retries--) |
2737 | goto retry; | |
2738 | ||
38d38493 | 2739 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2740 | goto nomem; |
2741 | ||
a4ebf1b6 VA |
2742 | /* Avoid endless loop for tasks bypassed by the oom killer */ |
2743 | if (passed_oom && task_is_dying()) | |
2744 | goto nomem; | |
6539cc05 | 2745 | |
29ef680a MH |
2746 | /* |
2747 | * keep retrying as long as the memcg oom killer is able to make | |
2748 | * a forward progress or bypass the charge if the oom killer | |
2749 | * couldn't make any progress. | |
2750 | */ | |
becdf89d SB |
2751 | if (mem_cgroup_oom(mem_over_limit, gfp_mask, |
2752 | get_order(nr_pages * PAGE_SIZE))) { | |
a4ebf1b6 | 2753 | passed_oom = true; |
d977aa93 | 2754 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a | 2755 | goto retry; |
29ef680a | 2756 | } |
7a81b88c | 2757 | nomem: |
1461e8c2 SB |
2758 | /* |
2759 | * Memcg doesn't have a dedicated reserve for atomic | |
2760 | * allocations. But like the global atomic pool, we need to | |
2761 | * put the burden of reclaim on regular allocation requests | |
2762 | * and let these go through as privileged allocations. | |
2763 | */ | |
2764 | if (!(gfp_mask & (__GFP_NOFAIL | __GFP_HIGH))) | |
3168ecbe | 2765 | return -ENOMEM; |
10d53c74 | 2766 | force: |
d6e103a7 RG |
2767 | /* |
2768 | * If the allocation has to be enforced, don't forget to raise | |
2769 | * a MEMCG_MAX event. | |
2770 | */ | |
2771 | if (!raised_max_event) | |
2772 | memcg_memory_event(mem_over_limit, MEMCG_MAX); | |
2773 | ||
10d53c74 TH |
2774 | /* |
2775 | * The allocation either can't fail or will lead to more memory | |
2776 | * being freed very soon. Allow memory usage go over the limit | |
2777 | * temporarily by force charging it. | |
2778 | */ | |
2779 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2780 | if (do_memsw_account()) |
10d53c74 | 2781 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2782 | |
2783 | return 0; | |
6539cc05 JW |
2784 | |
2785 | done_restock: | |
2786 | if (batch > nr_pages) | |
2787 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2788 | |
241994ed | 2789 | /* |
b23afb93 TH |
2790 | * If the hierarchy is above the normal consumption range, schedule |
2791 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2792 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2793 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2794 | * not recorded as it most likely matches current's and won't | |
2795 | * change in the meantime. As high limit is checked again before | |
2796 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2797 | */ |
2798 | do { | |
4b82ab4f JK |
2799 | bool mem_high, swap_high; |
2800 | ||
2801 | mem_high = page_counter_read(&memcg->memory) > | |
2802 | READ_ONCE(memcg->memory.high); | |
2803 | swap_high = page_counter_read(&memcg->swap) > | |
2804 | READ_ONCE(memcg->swap.high); | |
2805 | ||
2806 | /* Don't bother a random interrupted task */ | |
086f694a | 2807 | if (!in_task()) { |
4b82ab4f | 2808 | if (mem_high) { |
f7e1cb6e JW |
2809 | schedule_work(&memcg->high_work); |
2810 | break; | |
2811 | } | |
4b82ab4f JK |
2812 | continue; |
2813 | } | |
2814 | ||
2815 | if (mem_high || swap_high) { | |
2816 | /* | |
2817 | * The allocating tasks in this cgroup will need to do | |
2818 | * reclaim or be throttled to prevent further growth | |
2819 | * of the memory or swap footprints. | |
2820 | * | |
2821 | * Target some best-effort fairness between the tasks, | |
2822 | * and distribute reclaim work and delay penalties | |
2823 | * based on how much each task is actually allocating. | |
2824 | */ | |
9516a18a | 2825 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2826 | set_notify_resume(current); |
2827 | break; | |
2828 | } | |
241994ed | 2829 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 | 2830 | |
c9afe31e SB |
2831 | if (current->memcg_nr_pages_over_high > MEMCG_CHARGE_BATCH && |
2832 | !(current->flags & PF_MEMALLOC) && | |
2833 | gfpflags_allow_blocking(gfp_mask)) { | |
2834 | mem_cgroup_handle_over_high(); | |
2835 | } | |
10d53c74 | 2836 | return 0; |
7a81b88c | 2837 | } |
8a9f3ccd | 2838 | |
c5c8b16b MS |
2839 | static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2840 | unsigned int nr_pages) | |
2841 | { | |
2842 | if (mem_cgroup_is_root(memcg)) | |
2843 | return 0; | |
2844 | ||
2845 | return try_charge_memcg(memcg, gfp_mask, nr_pages); | |
2846 | } | |
2847 | ||
58056f77 | 2848 | static inline void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2849 | { |
ce00a967 JW |
2850 | if (mem_cgroup_is_root(memcg)) |
2851 | return; | |
2852 | ||
3e32cb2e | 2853 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2854 | if (do_memsw_account()) |
3e32cb2e | 2855 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f KH |
2856 | } |
2857 | ||
118f2875 | 2858 | static void commit_charge(struct folio *folio, struct mem_cgroup *memcg) |
0a31bc97 | 2859 | { |
118f2875 | 2860 | VM_BUG_ON_FOLIO(folio_memcg(folio), folio); |
0a31bc97 | 2861 | /* |
a5eb011a | 2862 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2863 | * |
a0b5b414 JW |
2864 | * - the page lock |
2865 | * - LRU isolation | |
2866 | * - lock_page_memcg() | |
2867 | * - exclusive reference | |
018ee47f | 2868 | * - mem_cgroup_trylock_pages() |
0a31bc97 | 2869 | */ |
118f2875 | 2870 | folio->memcg_data = (unsigned long)memcg; |
7a81b88c | 2871 | } |
66e1707b | 2872 | |
84c07d11 | 2873 | #ifdef CONFIG_MEMCG_KMEM |
41eb5df1 WL |
2874 | /* |
2875 | * The allocated objcg pointers array is not accounted directly. | |
2876 | * Moreover, it should not come from DMA buffer and is not readily | |
2877 | * reclaimable. So those GFP bits should be masked off. | |
2878 | */ | |
2879 | #define OBJCGS_CLEAR_MASK (__GFP_DMA | __GFP_RECLAIMABLE | __GFP_ACCOUNT) | |
2880 | ||
a7ebf564 WL |
2881 | /* |
2882 | * mod_objcg_mlstate() may be called with irq enabled, so | |
2883 | * mod_memcg_lruvec_state() should be used. | |
2884 | */ | |
2885 | static inline void mod_objcg_mlstate(struct obj_cgroup *objcg, | |
2886 | struct pglist_data *pgdat, | |
2887 | enum node_stat_item idx, int nr) | |
2888 | { | |
2889 | struct mem_cgroup *memcg; | |
2890 | struct lruvec *lruvec; | |
2891 | ||
2892 | rcu_read_lock(); | |
2893 | memcg = obj_cgroup_memcg(objcg); | |
2894 | lruvec = mem_cgroup_lruvec(memcg, pgdat); | |
2895 | mod_memcg_lruvec_state(lruvec, idx, nr); | |
2896 | rcu_read_unlock(); | |
2897 | } | |
2898 | ||
4b5f8d9a VB |
2899 | int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s, |
2900 | gfp_t gfp, bool new_slab) | |
10befea9 | 2901 | { |
4b5f8d9a | 2902 | unsigned int objects = objs_per_slab(s, slab); |
2e9bd483 | 2903 | unsigned long memcg_data; |
10befea9 RG |
2904 | void *vec; |
2905 | ||
41eb5df1 | 2906 | gfp &= ~OBJCGS_CLEAR_MASK; |
10befea9 | 2907 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, |
4b5f8d9a | 2908 | slab_nid(slab)); |
10befea9 RG |
2909 | if (!vec) |
2910 | return -ENOMEM; | |
2911 | ||
2e9bd483 | 2912 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
4b5f8d9a | 2913 | if (new_slab) { |
2e9bd483 | 2914 | /* |
4b5f8d9a VB |
2915 | * If the slab is brand new and nobody can yet access its |
2916 | * memcg_data, no synchronization is required and memcg_data can | |
2917 | * be simply assigned. | |
2e9bd483 | 2918 | */ |
4b5f8d9a VB |
2919 | slab->memcg_data = memcg_data; |
2920 | } else if (cmpxchg(&slab->memcg_data, 0, memcg_data)) { | |
2e9bd483 | 2921 | /* |
4b5f8d9a VB |
2922 | * If the slab is already in use, somebody can allocate and |
2923 | * assign obj_cgroups in parallel. In this case the existing | |
2e9bd483 RG |
2924 | * objcg vector should be reused. |
2925 | */ | |
10befea9 | 2926 | kfree(vec); |
2e9bd483 RG |
2927 | return 0; |
2928 | } | |
10befea9 | 2929 | |
2e9bd483 | 2930 | kmemleak_not_leak(vec); |
10befea9 RG |
2931 | return 0; |
2932 | } | |
2933 | ||
fc4db90f RG |
2934 | static __always_inline |
2935 | struct mem_cgroup *mem_cgroup_from_obj_folio(struct folio *folio, void *p) | |
8380ce47 | 2936 | { |
8380ce47 | 2937 | /* |
9855609b RG |
2938 | * Slab objects are accounted individually, not per-page. |
2939 | * Memcg membership data for each individual object is saved in | |
4b5f8d9a | 2940 | * slab->memcg_data. |
8380ce47 | 2941 | */ |
4b5f8d9a VB |
2942 | if (folio_test_slab(folio)) { |
2943 | struct obj_cgroup **objcgs; | |
2944 | struct slab *slab; | |
9855609b RG |
2945 | unsigned int off; |
2946 | ||
4b5f8d9a VB |
2947 | slab = folio_slab(folio); |
2948 | objcgs = slab_objcgs(slab); | |
2949 | if (!objcgs) | |
2950 | return NULL; | |
2951 | ||
2952 | off = obj_to_index(slab->slab_cache, slab, p); | |
2953 | if (objcgs[off]) | |
2954 | return obj_cgroup_memcg(objcgs[off]); | |
10befea9 RG |
2955 | |
2956 | return NULL; | |
9855609b | 2957 | } |
8380ce47 | 2958 | |
bcfe06bf | 2959 | /* |
4b5f8d9a VB |
2960 | * page_memcg_check() is used here, because in theory we can encounter |
2961 | * a folio where the slab flag has been cleared already, but | |
2962 | * slab->memcg_data has not been freed yet | |
bcfe06bf RG |
2963 | * page_memcg_check(page) will guarantee that a proper memory |
2964 | * cgroup pointer or NULL will be returned. | |
2965 | */ | |
4b5f8d9a | 2966 | return page_memcg_check(folio_page(folio, 0)); |
8380ce47 RG |
2967 | } |
2968 | ||
fc4db90f RG |
2969 | /* |
2970 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2971 | * | |
2972 | * A passed kernel object can be a slab object, vmalloc object or a generic | |
2973 | * kernel page, so different mechanisms for getting the memory cgroup pointer | |
2974 | * should be used. | |
2975 | * | |
2976 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
2977 | * can not know for sure how the kernel object is implemented. | |
2978 | * mem_cgroup_from_obj() can be safely used in such cases. | |
2979 | * | |
2980 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2981 | * cgroup_mutex, etc. | |
2982 | */ | |
2983 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2984 | { | |
2985 | struct folio *folio; | |
2986 | ||
2987 | if (mem_cgroup_disabled()) | |
2988 | return NULL; | |
2989 | ||
2990 | if (unlikely(is_vmalloc_addr(p))) | |
2991 | folio = page_folio(vmalloc_to_page(p)); | |
2992 | else | |
2993 | folio = virt_to_folio(p); | |
2994 | ||
2995 | return mem_cgroup_from_obj_folio(folio, p); | |
2996 | } | |
2997 | ||
2998 | /* | |
2999 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
3000 | * Similar to mem_cgroup_from_obj(), but faster and not suitable for objects, | |
3001 | * allocated using vmalloc(). | |
3002 | * | |
3003 | * A passed kernel object must be a slab object or a generic kernel page. | |
3004 | * | |
3005 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
3006 | * cgroup_mutex, etc. | |
3007 | */ | |
3008 | struct mem_cgroup *mem_cgroup_from_slab_obj(void *p) | |
3009 | { | |
3010 | if (mem_cgroup_disabled()) | |
3011 | return NULL; | |
3012 | ||
3013 | return mem_cgroup_from_obj_folio(virt_to_folio(p), p); | |
3014 | } | |
3015 | ||
f4840ccf JW |
3016 | static struct obj_cgroup *__get_obj_cgroup_from_memcg(struct mem_cgroup *memcg) |
3017 | { | |
3018 | struct obj_cgroup *objcg = NULL; | |
3019 | ||
3020 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
3021 | objcg = rcu_dereference(memcg->objcg); | |
3022 | if (objcg && obj_cgroup_tryget(objcg)) | |
3023 | break; | |
3024 | objcg = NULL; | |
3025 | } | |
3026 | return objcg; | |
3027 | } | |
3028 | ||
bf4f0599 RG |
3029 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
3030 | { | |
3031 | struct obj_cgroup *objcg = NULL; | |
3032 | struct mem_cgroup *memcg; | |
3033 | ||
279c3393 RG |
3034 | if (memcg_kmem_bypass()) |
3035 | return NULL; | |
3036 | ||
bf4f0599 | 3037 | rcu_read_lock(); |
37d5985c RG |
3038 | if (unlikely(active_memcg())) |
3039 | memcg = active_memcg(); | |
bf4f0599 RG |
3040 | else |
3041 | memcg = mem_cgroup_from_task(current); | |
f4840ccf | 3042 | objcg = __get_obj_cgroup_from_memcg(memcg); |
bf4f0599 | 3043 | rcu_read_unlock(); |
f4840ccf JW |
3044 | return objcg; |
3045 | } | |
3046 | ||
3047 | struct obj_cgroup *get_obj_cgroup_from_page(struct page *page) | |
3048 | { | |
3049 | struct obj_cgroup *objcg; | |
3050 | ||
3051 | if (!memcg_kmem_enabled() || memcg_kmem_bypass()) | |
3052 | return NULL; | |
3053 | ||
3054 | if (PageMemcgKmem(page)) { | |
3055 | objcg = __folio_objcg(page_folio(page)); | |
3056 | obj_cgroup_get(objcg); | |
3057 | } else { | |
3058 | struct mem_cgroup *memcg; | |
bf4f0599 | 3059 | |
f4840ccf JW |
3060 | rcu_read_lock(); |
3061 | memcg = __folio_memcg(page_folio(page)); | |
3062 | if (memcg) | |
3063 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
3064 | else | |
3065 | objcg = NULL; | |
3066 | rcu_read_unlock(); | |
3067 | } | |
bf4f0599 RG |
3068 | return objcg; |
3069 | } | |
3070 | ||
a8c49af3 YA |
3071 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
3072 | { | |
3073 | mod_memcg_state(memcg, MEMCG_KMEM, nr_pages); | |
3074 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
3075 | if (nr_pages > 0) | |
3076 | page_counter_charge(&memcg->kmem, nr_pages); | |
3077 | else | |
3078 | page_counter_uncharge(&memcg->kmem, -nr_pages); | |
3079 | } | |
3080 | } | |
3081 | ||
3082 | ||
f1286fae MS |
3083 | /* |
3084 | * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg | |
3085 | * @objcg: object cgroup to uncharge | |
3086 | * @nr_pages: number of pages to uncharge | |
3087 | */ | |
e74d2259 MS |
3088 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
3089 | unsigned int nr_pages) | |
3090 | { | |
3091 | struct mem_cgroup *memcg; | |
3092 | ||
3093 | memcg = get_mem_cgroup_from_objcg(objcg); | |
e74d2259 | 3094 | |
a8c49af3 | 3095 | memcg_account_kmem(memcg, -nr_pages); |
f1286fae | 3096 | refill_stock(memcg, nr_pages); |
e74d2259 | 3097 | |
e74d2259 | 3098 | css_put(&memcg->css); |
e74d2259 MS |
3099 | } |
3100 | ||
f1286fae MS |
3101 | /* |
3102 | * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg | |
3103 | * @objcg: object cgroup to charge | |
45264778 | 3104 | * @gfp: reclaim mode |
92d0510c | 3105 | * @nr_pages: number of pages to charge |
45264778 VD |
3106 | * |
3107 | * Returns 0 on success, an error code on failure. | |
3108 | */ | |
f1286fae MS |
3109 | static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp, |
3110 | unsigned int nr_pages) | |
7ae1e1d0 | 3111 | { |
f1286fae | 3112 | struct mem_cgroup *memcg; |
7ae1e1d0 GC |
3113 | int ret; |
3114 | ||
f1286fae MS |
3115 | memcg = get_mem_cgroup_from_objcg(objcg); |
3116 | ||
c5c8b16b | 3117 | ret = try_charge_memcg(memcg, gfp, nr_pages); |
52c29b04 | 3118 | if (ret) |
f1286fae | 3119 | goto out; |
52c29b04 | 3120 | |
a8c49af3 | 3121 | memcg_account_kmem(memcg, nr_pages); |
f1286fae MS |
3122 | out: |
3123 | css_put(&memcg->css); | |
4b13f64d | 3124 | |
f1286fae | 3125 | return ret; |
4b13f64d RG |
3126 | } |
3127 | ||
45264778 | 3128 | /** |
f4b00eab | 3129 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3130 | * @page: page to charge |
3131 | * @gfp: reclaim mode | |
3132 | * @order: allocation order | |
3133 | * | |
3134 | * Returns 0 on success, an error code on failure. | |
3135 | */ | |
f4b00eab | 3136 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3137 | { |
b4e0b68f | 3138 | struct obj_cgroup *objcg; |
fcff7d7e | 3139 | int ret = 0; |
7ae1e1d0 | 3140 | |
b4e0b68f MS |
3141 | objcg = get_obj_cgroup_from_current(); |
3142 | if (objcg) { | |
3143 | ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order); | |
4d96ba35 | 3144 | if (!ret) { |
b4e0b68f | 3145 | page->memcg_data = (unsigned long)objcg | |
18b2db3b | 3146 | MEMCG_DATA_KMEM; |
1a3e1f40 | 3147 | return 0; |
4d96ba35 | 3148 | } |
b4e0b68f | 3149 | obj_cgroup_put(objcg); |
c4159a75 | 3150 | } |
d05e83a6 | 3151 | return ret; |
7ae1e1d0 | 3152 | } |
49a18eae | 3153 | |
45264778 | 3154 | /** |
f4b00eab | 3155 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3156 | * @page: page to uncharge |
3157 | * @order: allocation order | |
3158 | */ | |
f4b00eab | 3159 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3160 | { |
1b7e4464 | 3161 | struct folio *folio = page_folio(page); |
b4e0b68f | 3162 | struct obj_cgroup *objcg; |
f3ccb2c4 | 3163 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3164 | |
1b7e4464 | 3165 | if (!folio_memcg_kmem(folio)) |
7ae1e1d0 GC |
3166 | return; |
3167 | ||
1b7e4464 | 3168 | objcg = __folio_objcg(folio); |
b4e0b68f | 3169 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
1b7e4464 | 3170 | folio->memcg_data = 0; |
b4e0b68f | 3171 | obj_cgroup_put(objcg); |
60d3fd32 | 3172 | } |
bf4f0599 | 3173 | |
68ac5b3c WL |
3174 | void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
3175 | enum node_stat_item idx, int nr) | |
3176 | { | |
fead2b86 | 3177 | struct memcg_stock_pcp *stock; |
56751146 | 3178 | struct obj_cgroup *old = NULL; |
68ac5b3c WL |
3179 | unsigned long flags; |
3180 | int *bytes; | |
3181 | ||
56751146 | 3182 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3183 | stock = this_cpu_ptr(&memcg_stock); |
3184 | ||
68ac5b3c WL |
3185 | /* |
3186 | * Save vmstat data in stock and skip vmstat array update unless | |
3187 | * accumulating over a page of vmstat data or when pgdat or idx | |
3188 | * changes. | |
3189 | */ | |
3190 | if (stock->cached_objcg != objcg) { | |
56751146 | 3191 | old = drain_obj_stock(stock); |
68ac5b3c WL |
3192 | obj_cgroup_get(objcg); |
3193 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) | |
3194 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3195 | stock->cached_objcg = objcg; | |
3196 | stock->cached_pgdat = pgdat; | |
3197 | } else if (stock->cached_pgdat != pgdat) { | |
3198 | /* Flush the existing cached vmstat data */ | |
7fa0dacb WL |
3199 | struct pglist_data *oldpg = stock->cached_pgdat; |
3200 | ||
68ac5b3c | 3201 | if (stock->nr_slab_reclaimable_b) { |
7fa0dacb | 3202 | mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B, |
68ac5b3c WL |
3203 | stock->nr_slab_reclaimable_b); |
3204 | stock->nr_slab_reclaimable_b = 0; | |
3205 | } | |
3206 | if (stock->nr_slab_unreclaimable_b) { | |
7fa0dacb | 3207 | mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B, |
68ac5b3c WL |
3208 | stock->nr_slab_unreclaimable_b); |
3209 | stock->nr_slab_unreclaimable_b = 0; | |
3210 | } | |
3211 | stock->cached_pgdat = pgdat; | |
3212 | } | |
3213 | ||
3214 | bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b | |
3215 | : &stock->nr_slab_unreclaimable_b; | |
3216 | /* | |
3217 | * Even for large object >= PAGE_SIZE, the vmstat data will still be | |
3218 | * cached locally at least once before pushing it out. | |
3219 | */ | |
3220 | if (!*bytes) { | |
3221 | *bytes = nr; | |
3222 | nr = 0; | |
3223 | } else { | |
3224 | *bytes += nr; | |
3225 | if (abs(*bytes) > PAGE_SIZE) { | |
3226 | nr = *bytes; | |
3227 | *bytes = 0; | |
3228 | } else { | |
3229 | nr = 0; | |
3230 | } | |
3231 | } | |
3232 | if (nr) | |
3233 | mod_objcg_mlstate(objcg, pgdat, idx, nr); | |
3234 | ||
56751146 SAS |
3235 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
3236 | if (old) | |
3237 | obj_cgroup_put(old); | |
68ac5b3c WL |
3238 | } |
3239 | ||
bf4f0599 RG |
3240 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
3241 | { | |
fead2b86 | 3242 | struct memcg_stock_pcp *stock; |
bf4f0599 RG |
3243 | unsigned long flags; |
3244 | bool ret = false; | |
3245 | ||
56751146 | 3246 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3247 | |
3248 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 RG |
3249 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { |
3250 | stock->nr_bytes -= nr_bytes; | |
3251 | ret = true; | |
3252 | } | |
3253 | ||
56751146 | 3254 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
bf4f0599 RG |
3255 | |
3256 | return ret; | |
3257 | } | |
3258 | ||
56751146 | 3259 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 RG |
3260 | { |
3261 | struct obj_cgroup *old = stock->cached_objcg; | |
3262 | ||
3263 | if (!old) | |
56751146 | 3264 | return NULL; |
bf4f0599 RG |
3265 | |
3266 | if (stock->nr_bytes) { | |
3267 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3268 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3269 | ||
af9a3b69 JW |
3270 | if (nr_pages) { |
3271 | struct mem_cgroup *memcg; | |
3272 | ||
3273 | memcg = get_mem_cgroup_from_objcg(old); | |
3274 | ||
3275 | memcg_account_kmem(memcg, -nr_pages); | |
3276 | __refill_stock(memcg, nr_pages); | |
3277 | ||
3278 | css_put(&memcg->css); | |
3279 | } | |
bf4f0599 RG |
3280 | |
3281 | /* | |
3282 | * The leftover is flushed to the centralized per-memcg value. | |
3283 | * On the next attempt to refill obj stock it will be moved | |
3284 | * to a per-cpu stock (probably, on an other CPU), see | |
3285 | * refill_obj_stock(). | |
3286 | * | |
3287 | * How often it's flushed is a trade-off between the memory | |
3288 | * limit enforcement accuracy and potential CPU contention, | |
3289 | * so it might be changed in the future. | |
3290 | */ | |
3291 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3292 | stock->nr_bytes = 0; | |
3293 | } | |
3294 | ||
68ac5b3c WL |
3295 | /* |
3296 | * Flush the vmstat data in current stock | |
3297 | */ | |
3298 | if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) { | |
3299 | if (stock->nr_slab_reclaimable_b) { | |
3300 | mod_objcg_mlstate(old, stock->cached_pgdat, | |
3301 | NR_SLAB_RECLAIMABLE_B, | |
3302 | stock->nr_slab_reclaimable_b); | |
3303 | stock->nr_slab_reclaimable_b = 0; | |
3304 | } | |
3305 | if (stock->nr_slab_unreclaimable_b) { | |
3306 | mod_objcg_mlstate(old, stock->cached_pgdat, | |
3307 | NR_SLAB_UNRECLAIMABLE_B, | |
3308 | stock->nr_slab_unreclaimable_b); | |
3309 | stock->nr_slab_unreclaimable_b = 0; | |
3310 | } | |
3311 | stock->cached_pgdat = NULL; | |
3312 | } | |
3313 | ||
bf4f0599 | 3314 | stock->cached_objcg = NULL; |
56751146 SAS |
3315 | /* |
3316 | * The `old' objects needs to be released by the caller via | |
3317 | * obj_cgroup_put() outside of memcg_stock_pcp::stock_lock. | |
3318 | */ | |
3319 | return old; | |
bf4f0599 RG |
3320 | } |
3321 | ||
3322 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3323 | struct mem_cgroup *root_memcg) | |
3324 | { | |
3325 | struct mem_cgroup *memcg; | |
3326 | ||
fead2b86 MH |
3327 | if (stock->cached_objcg) { |
3328 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
bf4f0599 RG |
3329 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) |
3330 | return true; | |
3331 | } | |
3332 | ||
3333 | return false; | |
3334 | } | |
3335 | ||
5387c904 WL |
3336 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes, |
3337 | bool allow_uncharge) | |
bf4f0599 | 3338 | { |
fead2b86 | 3339 | struct memcg_stock_pcp *stock; |
56751146 | 3340 | struct obj_cgroup *old = NULL; |
bf4f0599 | 3341 | unsigned long flags; |
5387c904 | 3342 | unsigned int nr_pages = 0; |
bf4f0599 | 3343 | |
56751146 | 3344 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3345 | |
3346 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 3347 | if (stock->cached_objcg != objcg) { /* reset if necessary */ |
56751146 | 3348 | old = drain_obj_stock(stock); |
bf4f0599 RG |
3349 | obj_cgroup_get(objcg); |
3350 | stock->cached_objcg = objcg; | |
5387c904 WL |
3351 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) |
3352 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3353 | allow_uncharge = true; /* Allow uncharge when objcg changes */ | |
bf4f0599 RG |
3354 | } |
3355 | stock->nr_bytes += nr_bytes; | |
3356 | ||
5387c904 WL |
3357 | if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) { |
3358 | nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3359 | stock->nr_bytes &= (PAGE_SIZE - 1); | |
3360 | } | |
bf4f0599 | 3361 | |
56751146 SAS |
3362 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
3363 | if (old) | |
3364 | obj_cgroup_put(old); | |
5387c904 WL |
3365 | |
3366 | if (nr_pages) | |
3367 | obj_cgroup_uncharge_pages(objcg, nr_pages); | |
bf4f0599 RG |
3368 | } |
3369 | ||
3370 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3371 | { | |
bf4f0599 RG |
3372 | unsigned int nr_pages, nr_bytes; |
3373 | int ret; | |
3374 | ||
3375 | if (consume_obj_stock(objcg, size)) | |
3376 | return 0; | |
3377 | ||
3378 | /* | |
5387c904 | 3379 | * In theory, objcg->nr_charged_bytes can have enough |
bf4f0599 | 3380 | * pre-charged bytes to satisfy the allocation. However, |
5387c904 WL |
3381 | * flushing objcg->nr_charged_bytes requires two atomic |
3382 | * operations, and objcg->nr_charged_bytes can't be big. | |
3383 | * The shared objcg->nr_charged_bytes can also become a | |
3384 | * performance bottleneck if all tasks of the same memcg are | |
3385 | * trying to update it. So it's better to ignore it and try | |
3386 | * grab some new pages. The stock's nr_bytes will be flushed to | |
3387 | * objcg->nr_charged_bytes later on when objcg changes. | |
3388 | * | |
3389 | * The stock's nr_bytes may contain enough pre-charged bytes | |
3390 | * to allow one less page from being charged, but we can't rely | |
3391 | * on the pre-charged bytes not being changed outside of | |
3392 | * consume_obj_stock() or refill_obj_stock(). So ignore those | |
3393 | * pre-charged bytes as well when charging pages. To avoid a | |
3394 | * page uncharge right after a page charge, we set the | |
3395 | * allow_uncharge flag to false when calling refill_obj_stock() | |
3396 | * to temporarily allow the pre-charged bytes to exceed the page | |
3397 | * size limit. The maximum reachable value of the pre-charged | |
3398 | * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data | |
3399 | * race. | |
bf4f0599 | 3400 | */ |
bf4f0599 RG |
3401 | nr_pages = size >> PAGE_SHIFT; |
3402 | nr_bytes = size & (PAGE_SIZE - 1); | |
3403 | ||
3404 | if (nr_bytes) | |
3405 | nr_pages += 1; | |
3406 | ||
e74d2259 | 3407 | ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages); |
bf4f0599 | 3408 | if (!ret && nr_bytes) |
5387c904 | 3409 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false); |
bf4f0599 | 3410 | |
bf4f0599 RG |
3411 | return ret; |
3412 | } | |
3413 | ||
3414 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3415 | { | |
5387c904 | 3416 | refill_obj_stock(objcg, size, true); |
bf4f0599 RG |
3417 | } |
3418 | ||
84c07d11 | 3419 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3420 | |
ca3e0214 | 3421 | /* |
be6c8982 | 3422 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3423 | */ |
be6c8982 | 3424 | void split_page_memcg(struct page *head, unsigned int nr) |
ca3e0214 | 3425 | { |
1b7e4464 MWO |
3426 | struct folio *folio = page_folio(head); |
3427 | struct mem_cgroup *memcg = folio_memcg(folio); | |
e94c8a9c | 3428 | int i; |
ca3e0214 | 3429 | |
be6c8982 | 3430 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3431 | return; |
b070e65c | 3432 | |
be6c8982 | 3433 | for (i = 1; i < nr; i++) |
1b7e4464 | 3434 | folio_page(folio, i)->memcg_data = folio->memcg_data; |
b4e0b68f | 3435 | |
1b7e4464 MWO |
3436 | if (folio_memcg_kmem(folio)) |
3437 | obj_cgroup_get_many(__folio_objcg(folio), nr - 1); | |
b4e0b68f MS |
3438 | else |
3439 | css_get_many(&memcg->css, nr - 1); | |
ca3e0214 | 3440 | } |
ca3e0214 | 3441 | |
c255a458 | 3442 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3443 | /** |
3444 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3445 | * @entry: swap entry to be moved | |
3446 | * @from: mem_cgroup which the entry is moved from | |
3447 | * @to: mem_cgroup which the entry is moved to | |
3448 | * | |
3449 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3450 | * as the mem_cgroup's id of @from. | |
3451 | * | |
3452 | * Returns 0 on success, -EINVAL on failure. | |
3453 | * | |
3e32cb2e | 3454 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3455 | * both res and memsw, and called css_get(). |
3456 | */ | |
3457 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3458 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3459 | { |
3460 | unsigned short old_id, new_id; | |
3461 | ||
34c00c31 LZ |
3462 | old_id = mem_cgroup_id(from); |
3463 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3464 | |
3465 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3466 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3467 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3468 | return 0; |
3469 | } | |
3470 | return -EINVAL; | |
3471 | } | |
3472 | #else | |
3473 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3474 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3475 | { |
3476 | return -EINVAL; | |
3477 | } | |
8c7c6e34 | 3478 | #endif |
d13d1443 | 3479 | |
bbec2e15 | 3480 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3481 | |
bbec2e15 RG |
3482 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3483 | unsigned long max, bool memsw) | |
628f4235 | 3484 | { |
3e32cb2e | 3485 | bool enlarge = false; |
bb4a7ea2 | 3486 | bool drained = false; |
3e32cb2e | 3487 | int ret; |
c054a78c YZ |
3488 | bool limits_invariant; |
3489 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3490 | |
3e32cb2e | 3491 | do { |
628f4235 KH |
3492 | if (signal_pending(current)) { |
3493 | ret = -EINTR; | |
3494 | break; | |
3495 | } | |
3e32cb2e | 3496 | |
bbec2e15 | 3497 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3498 | /* |
3499 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3500 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3501 | */ |
15b42562 | 3502 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3503 | max <= memcg->memsw.max; |
c054a78c | 3504 | if (!limits_invariant) { |
bbec2e15 | 3505 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3506 | ret = -EINVAL; |
8c7c6e34 KH |
3507 | break; |
3508 | } | |
bbec2e15 | 3509 | if (max > counter->max) |
3e32cb2e | 3510 | enlarge = true; |
bbec2e15 RG |
3511 | ret = page_counter_set_max(counter, max); |
3512 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3513 | |
3514 | if (!ret) | |
3515 | break; | |
3516 | ||
bb4a7ea2 SB |
3517 | if (!drained) { |
3518 | drain_all_stock(memcg); | |
3519 | drained = true; | |
3520 | continue; | |
3521 | } | |
3522 | ||
73b73bac YA |
3523 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
3524 | memsw ? 0 : MEMCG_RECLAIM_MAY_SWAP)) { | |
1ab5c056 AR |
3525 | ret = -EBUSY; |
3526 | break; | |
3527 | } | |
3528 | } while (true); | |
3e32cb2e | 3529 | |
3c11ecf4 KH |
3530 | if (!ret && enlarge) |
3531 | memcg_oom_recover(memcg); | |
3e32cb2e | 3532 | |
628f4235 KH |
3533 | return ret; |
3534 | } | |
3535 | ||
ef8f2327 | 3536 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3537 | gfp_t gfp_mask, |
3538 | unsigned long *total_scanned) | |
3539 | { | |
3540 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3541 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3542 | unsigned long reclaimed; |
3543 | int loop = 0; | |
ef8f2327 | 3544 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3545 | unsigned long excess; |
0608f43d AM |
3546 | |
3547 | if (order > 0) | |
3548 | return 0; | |
3549 | ||
2ab082ba | 3550 | mctz = soft_limit_tree.rb_tree_per_node[pgdat->node_id]; |
d6507ff5 MH |
3551 | |
3552 | /* | |
3553 | * Do not even bother to check the largest node if the root | |
3554 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3555 | * are acceptable as soft limit is best effort anyway. | |
3556 | */ | |
bfc7228b | 3557 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3558 | return 0; |
3559 | ||
0608f43d AM |
3560 | /* |
3561 | * This loop can run a while, specially if mem_cgroup's continuously | |
3562 | * keep exceeding their soft limit and putting the system under | |
3563 | * pressure | |
3564 | */ | |
3565 | do { | |
3566 | if (next_mz) | |
3567 | mz = next_mz; | |
3568 | else | |
3569 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3570 | if (!mz) | |
3571 | break; | |
3572 | ||
ef8f2327 | 3573 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
d8f65338 | 3574 | gfp_mask, total_scanned); |
0608f43d | 3575 | nr_reclaimed += reclaimed; |
0a31bc97 | 3576 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3577 | |
3578 | /* | |
3579 | * If we failed to reclaim anything from this memory cgroup | |
3580 | * it is time to move on to the next cgroup | |
3581 | */ | |
3582 | next_mz = NULL; | |
bc2f2e7f VD |
3583 | if (!reclaimed) |
3584 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3585 | ||
3e32cb2e | 3586 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3587 | /* |
3588 | * One school of thought says that we should not add | |
3589 | * back the node to the tree if reclaim returns 0. | |
3590 | * But our reclaim could return 0, simply because due | |
3591 | * to priority we are exposing a smaller subset of | |
3592 | * memory to reclaim from. Consider this as a longer | |
3593 | * term TODO. | |
3594 | */ | |
3595 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3596 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3597 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3598 | css_put(&mz->memcg->css); |
3599 | loop++; | |
3600 | /* | |
3601 | * Could not reclaim anything and there are no more | |
3602 | * mem cgroups to try or we seem to be looping without | |
3603 | * reclaiming anything. | |
3604 | */ | |
3605 | if (!nr_reclaimed && | |
3606 | (next_mz == NULL || | |
3607 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3608 | break; | |
3609 | } while (!nr_reclaimed); | |
3610 | if (next_mz) | |
3611 | css_put(&next_mz->memcg->css); | |
3612 | return nr_reclaimed; | |
3613 | } | |
3614 | ||
c26251f9 | 3615 | /* |
51038171 | 3616 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3617 | * |
3618 | * Caller is responsible for holding css reference for memcg. | |
3619 | */ | |
3620 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3621 | { | |
d977aa93 | 3622 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3623 | |
c1e862c1 KH |
3624 | /* we call try-to-free pages for make this cgroup empty */ |
3625 | lru_add_drain_all(); | |
d12c60f6 JS |
3626 | |
3627 | drain_all_stock(memcg); | |
3628 | ||
f817ed48 | 3629 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3630 | while (nr_retries && page_counter_read(&memcg->memory)) { |
c26251f9 MH |
3631 | if (signal_pending(current)) |
3632 | return -EINTR; | |
3633 | ||
73b73bac YA |
3634 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
3635 | MEMCG_RECLAIM_MAY_SWAP)) | |
f817ed48 | 3636 | nr_retries--; |
f817ed48 | 3637 | } |
ab5196c2 MH |
3638 | |
3639 | return 0; | |
cc847582 KH |
3640 | } |
3641 | ||
6770c64e TH |
3642 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3643 | char *buf, size_t nbytes, | |
3644 | loff_t off) | |
c1e862c1 | 3645 | { |
6770c64e | 3646 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3647 | |
d8423011 MH |
3648 | if (mem_cgroup_is_root(memcg)) |
3649 | return -EINVAL; | |
6770c64e | 3650 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3651 | } |
3652 | ||
182446d0 TH |
3653 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3654 | struct cftype *cft) | |
18f59ea7 | 3655 | { |
bef8620c | 3656 | return 1; |
18f59ea7 BS |
3657 | } |
3658 | ||
182446d0 TH |
3659 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3660 | struct cftype *cft, u64 val) | |
18f59ea7 | 3661 | { |
bef8620c | 3662 | if (val == 1) |
0b8f73e1 | 3663 | return 0; |
567fb435 | 3664 | |
bef8620c RG |
3665 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3666 | "Please report your usecase to linux-mm@kvack.org if you " | |
3667 | "depend on this functionality.\n"); | |
567fb435 | 3668 | |
bef8620c | 3669 | return -EINVAL; |
18f59ea7 BS |
3670 | } |
3671 | ||
6f646156 | 3672 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3673 | { |
42a30035 | 3674 | unsigned long val; |
ce00a967 | 3675 | |
3e32cb2e | 3676 | if (mem_cgroup_is_root(memcg)) { |
fd25a9e0 | 3677 | mem_cgroup_flush_stats(); |
0d1c2072 | 3678 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3679 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3680 | if (swap) |
3681 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3682 | } else { |
ce00a967 | 3683 | if (!swap) |
3e32cb2e | 3684 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3685 | else |
3e32cb2e | 3686 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3687 | } |
c12176d3 | 3688 | return val; |
ce00a967 JW |
3689 | } |
3690 | ||
3e32cb2e JW |
3691 | enum { |
3692 | RES_USAGE, | |
3693 | RES_LIMIT, | |
3694 | RES_MAX_USAGE, | |
3695 | RES_FAILCNT, | |
3696 | RES_SOFT_LIMIT, | |
3697 | }; | |
ce00a967 | 3698 | |
791badbd | 3699 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3700 | struct cftype *cft) |
8cdea7c0 | 3701 | { |
182446d0 | 3702 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3703 | struct page_counter *counter; |
af36f906 | 3704 | |
3e32cb2e | 3705 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3706 | case _MEM: |
3e32cb2e JW |
3707 | counter = &memcg->memory; |
3708 | break; | |
8c7c6e34 | 3709 | case _MEMSWAP: |
3e32cb2e JW |
3710 | counter = &memcg->memsw; |
3711 | break; | |
510fc4e1 | 3712 | case _KMEM: |
3e32cb2e | 3713 | counter = &memcg->kmem; |
510fc4e1 | 3714 | break; |
d55f90bf | 3715 | case _TCP: |
0db15298 | 3716 | counter = &memcg->tcpmem; |
d55f90bf | 3717 | break; |
8c7c6e34 KH |
3718 | default: |
3719 | BUG(); | |
8c7c6e34 | 3720 | } |
3e32cb2e JW |
3721 | |
3722 | switch (MEMFILE_ATTR(cft->private)) { | |
3723 | case RES_USAGE: | |
3724 | if (counter == &memcg->memory) | |
c12176d3 | 3725 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3726 | if (counter == &memcg->memsw) |
c12176d3 | 3727 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3728 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3729 | case RES_LIMIT: | |
bbec2e15 | 3730 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3731 | case RES_MAX_USAGE: |
3732 | return (u64)counter->watermark * PAGE_SIZE; | |
3733 | case RES_FAILCNT: | |
3734 | return counter->failcnt; | |
3735 | case RES_SOFT_LIMIT: | |
3736 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3737 | default: | |
3738 | BUG(); | |
3739 | } | |
8cdea7c0 | 3740 | } |
510fc4e1 | 3741 | |
84c07d11 | 3742 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3743 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3744 | { |
bf4f0599 | 3745 | struct obj_cgroup *objcg; |
d6441637 | 3746 | |
9c94bef9 | 3747 | if (mem_cgroup_kmem_disabled()) |
b313aeee VD |
3748 | return 0; |
3749 | ||
da0efe30 MS |
3750 | if (unlikely(mem_cgroup_is_root(memcg))) |
3751 | return 0; | |
d6441637 | 3752 | |
bf4f0599 | 3753 | objcg = obj_cgroup_alloc(); |
f9c69d63 | 3754 | if (!objcg) |
bf4f0599 | 3755 | return -ENOMEM; |
f9c69d63 | 3756 | |
bf4f0599 RG |
3757 | objcg->memcg = memcg; |
3758 | rcu_assign_pointer(memcg->objcg, objcg); | |
3759 | ||
d648bcc7 RG |
3760 | static_branch_enable(&memcg_kmem_enabled_key); |
3761 | ||
f9c69d63 | 3762 | memcg->kmemcg_id = memcg->id.id; |
0b8f73e1 JW |
3763 | |
3764 | return 0; | |
d6441637 VD |
3765 | } |
3766 | ||
8e0a8912 JW |
3767 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3768 | { | |
64268868 | 3769 | struct mem_cgroup *parent; |
8e0a8912 | 3770 | |
9c94bef9 | 3771 | if (mem_cgroup_kmem_disabled()) |
da0efe30 MS |
3772 | return; |
3773 | ||
3774 | if (unlikely(mem_cgroup_is_root(memcg))) | |
8e0a8912 | 3775 | return; |
9855609b | 3776 | |
8e0a8912 JW |
3777 | parent = parent_mem_cgroup(memcg); |
3778 | if (!parent) | |
3779 | parent = root_mem_cgroup; | |
3780 | ||
bf4f0599 | 3781 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a | 3782 | |
8e0a8912 | 3783 | /* |
64268868 MS |
3784 | * After we have finished memcg_reparent_objcgs(), all list_lrus |
3785 | * corresponding to this cgroup are guaranteed to remain empty. | |
3786 | * The ordering is imposed by list_lru_node->lock taken by | |
1f391eb2 | 3787 | * memcg_reparent_list_lrus(). |
8e0a8912 | 3788 | */ |
1f391eb2 | 3789 | memcg_reparent_list_lrus(memcg, parent); |
8e0a8912 | 3790 | } |
d6441637 | 3791 | #else |
0b8f73e1 | 3792 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3793 | { |
3794 | return 0; | |
3795 | } | |
3796 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3797 | { | |
3798 | } | |
84c07d11 | 3799 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3800 | |
bbec2e15 | 3801 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3802 | { |
3803 | int ret; | |
3804 | ||
bbec2e15 | 3805 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3806 | |
bbec2e15 | 3807 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3808 | if (ret) |
3809 | goto out; | |
3810 | ||
0db15298 | 3811 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3812 | /* |
3813 | * The active flag needs to be written after the static_key | |
3814 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3815 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3816 | * for details, and note that we don't mark any socket as | |
3817 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3818 | * |
3819 | * We need to do this, because static_keys will span multiple | |
3820 | * sites, but we can't control their order. If we mark a socket | |
3821 | * as accounted, but the accounting functions are not patched in | |
3822 | * yet, we'll lose accounting. | |
3823 | * | |
2d758073 | 3824 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3825 | * because when this value change, the code to process it is not |
3826 | * patched in yet. | |
3827 | */ | |
3828 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3829 | memcg->tcpmem_active = true; |
d55f90bf VD |
3830 | } |
3831 | out: | |
bbec2e15 | 3832 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3833 | return ret; |
3834 | } | |
d55f90bf | 3835 | |
628f4235 KH |
3836 | /* |
3837 | * The user of this function is... | |
3838 | * RES_LIMIT. | |
3839 | */ | |
451af504 TH |
3840 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3841 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3842 | { |
451af504 | 3843 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3844 | unsigned long nr_pages; |
628f4235 KH |
3845 | int ret; |
3846 | ||
451af504 | 3847 | buf = strstrip(buf); |
650c5e56 | 3848 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3849 | if (ret) |
3850 | return ret; | |
af36f906 | 3851 | |
3e32cb2e | 3852 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3853 | case RES_LIMIT: |
4b3bde4c BS |
3854 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3855 | ret = -EINVAL; | |
3856 | break; | |
3857 | } | |
3e32cb2e JW |
3858 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3859 | case _MEM: | |
bbec2e15 | 3860 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3861 | break; |
3e32cb2e | 3862 | case _MEMSWAP: |
bbec2e15 | 3863 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3864 | break; |
3e32cb2e | 3865 | case _KMEM: |
58056f77 SB |
3866 | /* kmem.limit_in_bytes is deprecated. */ |
3867 | ret = -EOPNOTSUPP; | |
3e32cb2e | 3868 | break; |
d55f90bf | 3869 | case _TCP: |
bbec2e15 | 3870 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3871 | break; |
3e32cb2e | 3872 | } |
296c81d8 | 3873 | break; |
3e32cb2e | 3874 | case RES_SOFT_LIMIT: |
2343e88d SAS |
3875 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
3876 | ret = -EOPNOTSUPP; | |
3877 | } else { | |
3878 | memcg->soft_limit = nr_pages; | |
3879 | ret = 0; | |
3880 | } | |
628f4235 KH |
3881 | break; |
3882 | } | |
451af504 | 3883 | return ret ?: nbytes; |
8cdea7c0 BS |
3884 | } |
3885 | ||
6770c64e TH |
3886 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3887 | size_t nbytes, loff_t off) | |
c84872e1 | 3888 | { |
6770c64e | 3889 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3890 | struct page_counter *counter; |
c84872e1 | 3891 | |
3e32cb2e JW |
3892 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3893 | case _MEM: | |
3894 | counter = &memcg->memory; | |
3895 | break; | |
3896 | case _MEMSWAP: | |
3897 | counter = &memcg->memsw; | |
3898 | break; | |
3899 | case _KMEM: | |
3900 | counter = &memcg->kmem; | |
3901 | break; | |
d55f90bf | 3902 | case _TCP: |
0db15298 | 3903 | counter = &memcg->tcpmem; |
d55f90bf | 3904 | break; |
3e32cb2e JW |
3905 | default: |
3906 | BUG(); | |
3907 | } | |
af36f906 | 3908 | |
3e32cb2e | 3909 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3910 | case RES_MAX_USAGE: |
3e32cb2e | 3911 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3912 | break; |
3913 | case RES_FAILCNT: | |
3e32cb2e | 3914 | counter->failcnt = 0; |
29f2a4da | 3915 | break; |
3e32cb2e JW |
3916 | default: |
3917 | BUG(); | |
29f2a4da | 3918 | } |
f64c3f54 | 3919 | |
6770c64e | 3920 | return nbytes; |
c84872e1 PE |
3921 | } |
3922 | ||
182446d0 | 3923 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3924 | struct cftype *cft) |
3925 | { | |
182446d0 | 3926 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3927 | } |
3928 | ||
02491447 | 3929 | #ifdef CONFIG_MMU |
182446d0 | 3930 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3931 | struct cftype *cft, u64 val) |
3932 | { | |
182446d0 | 3933 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3934 | |
1dfab5ab | 3935 | if (val & ~MOVE_MASK) |
7dc74be0 | 3936 | return -EINVAL; |
ee5e8472 | 3937 | |
7dc74be0 | 3938 | /* |
ee5e8472 GC |
3939 | * No kind of locking is needed in here, because ->can_attach() will |
3940 | * check this value once in the beginning of the process, and then carry | |
3941 | * on with stale data. This means that changes to this value will only | |
3942 | * affect task migrations starting after the change. | |
7dc74be0 | 3943 | */ |
c0ff4b85 | 3944 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3945 | return 0; |
3946 | } | |
02491447 | 3947 | #else |
182446d0 | 3948 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3949 | struct cftype *cft, u64 val) |
3950 | { | |
3951 | return -ENOSYS; | |
3952 | } | |
3953 | #endif | |
7dc74be0 | 3954 | |
406eb0c9 | 3955 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3956 | |
3957 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3958 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3959 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3960 | ||
3961 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3962 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3963 | { |
867e5e1d | 3964 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3965 | unsigned long nr = 0; |
3966 | enum lru_list lru; | |
3967 | ||
3968 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3969 | ||
3970 | for_each_lru(lru) { | |
3971 | if (!(BIT(lru) & lru_mask)) | |
3972 | continue; | |
dd8657b6 SB |
3973 | if (tree) |
3974 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3975 | else | |
3976 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3977 | } |
3978 | return nr; | |
3979 | } | |
3980 | ||
3981 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3982 | unsigned int lru_mask, |
3983 | bool tree) | |
113b7dfd JW |
3984 | { |
3985 | unsigned long nr = 0; | |
3986 | enum lru_list lru; | |
3987 | ||
3988 | for_each_lru(lru) { | |
3989 | if (!(BIT(lru) & lru_mask)) | |
3990 | continue; | |
dd8657b6 SB |
3991 | if (tree) |
3992 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3993 | else | |
3994 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3995 | } |
3996 | return nr; | |
3997 | } | |
3998 | ||
2da8ca82 | 3999 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 4000 | { |
25485de6 GT |
4001 | struct numa_stat { |
4002 | const char *name; | |
4003 | unsigned int lru_mask; | |
4004 | }; | |
4005 | ||
4006 | static const struct numa_stat stats[] = { | |
4007 | { "total", LRU_ALL }, | |
4008 | { "file", LRU_ALL_FILE }, | |
4009 | { "anon", LRU_ALL_ANON }, | |
4010 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4011 | }; | |
4012 | const struct numa_stat *stat; | |
406eb0c9 | 4013 | int nid; |
aa9694bb | 4014 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 4015 | |
fd25a9e0 | 4016 | mem_cgroup_flush_stats(); |
2d146aa3 | 4017 | |
25485de6 | 4018 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4019 | seq_printf(m, "%s=%lu", stat->name, |
4020 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4021 | false)); | |
4022 | for_each_node_state(nid, N_MEMORY) | |
4023 | seq_printf(m, " N%d=%lu", nid, | |
4024 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4025 | stat->lru_mask, false)); | |
25485de6 | 4026 | seq_putc(m, '\n'); |
406eb0c9 | 4027 | } |
406eb0c9 | 4028 | |
071aee13 | 4029 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4030 | |
4031 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
4032 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4033 | true)); | |
4034 | for_each_node_state(nid, N_MEMORY) | |
4035 | seq_printf(m, " N%d=%lu", nid, | |
4036 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4037 | stat->lru_mask, true)); | |
071aee13 | 4038 | seq_putc(m, '\n'); |
406eb0c9 | 4039 | } |
406eb0c9 | 4040 | |
406eb0c9 YH |
4041 | return 0; |
4042 | } | |
4043 | #endif /* CONFIG_NUMA */ | |
4044 | ||
c8713d0b | 4045 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4046 | NR_FILE_PAGES, |
be5d0a74 | 4047 | NR_ANON_MAPPED, |
468c3982 JW |
4048 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4049 | NR_ANON_THPS, | |
4050 | #endif | |
c8713d0b JW |
4051 | NR_SHMEM, |
4052 | NR_FILE_MAPPED, | |
4053 | NR_FILE_DIRTY, | |
4054 | NR_WRITEBACK, | |
e09b0b61 YS |
4055 | WORKINGSET_REFAULT_ANON, |
4056 | WORKINGSET_REFAULT_FILE, | |
c8713d0b JW |
4057 | MEMCG_SWAP, |
4058 | }; | |
4059 | ||
4060 | static const char *const memcg1_stat_names[] = { | |
4061 | "cache", | |
4062 | "rss", | |
468c3982 | 4063 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4064 | "rss_huge", |
468c3982 | 4065 | #endif |
c8713d0b JW |
4066 | "shmem", |
4067 | "mapped_file", | |
4068 | "dirty", | |
4069 | "writeback", | |
e09b0b61 YS |
4070 | "workingset_refault_anon", |
4071 | "workingset_refault_file", | |
c8713d0b JW |
4072 | "swap", |
4073 | }; | |
4074 | ||
df0e53d0 | 4075 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4076 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4077 | PGPGIN, |
4078 | PGPGOUT, | |
4079 | PGFAULT, | |
4080 | PGMAJFAULT, | |
4081 | }; | |
4082 | ||
2da8ca82 | 4083 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4084 | { |
aa9694bb | 4085 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4086 | unsigned long memory, memsw; |
af7c4b0e JW |
4087 | struct mem_cgroup *mi; |
4088 | unsigned int i; | |
406eb0c9 | 4089 | |
71cd3113 | 4090 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4091 | |
fd25a9e0 | 4092 | mem_cgroup_flush_stats(); |
2d146aa3 | 4093 | |
71cd3113 | 4094 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4095 | unsigned long nr; |
4096 | ||
71cd3113 | 4097 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4098 | continue; |
468c3982 | 4099 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
e09b0b61 YS |
4100 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
4101 | nr * memcg_page_state_unit(memcg1_stats[i])); | |
1dd3a273 | 4102 | } |
7b854121 | 4103 | |
df0e53d0 | 4104 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4105 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4106 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4107 | |
4108 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4109 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4110 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4111 | PAGE_SIZE); |
af7c4b0e | 4112 | |
14067bb3 | 4113 | /* Hierarchical information */ |
3e32cb2e JW |
4114 | memory = memsw = PAGE_COUNTER_MAX; |
4115 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4116 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4117 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4118 | } |
3e32cb2e JW |
4119 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4120 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4121 | if (do_memsw_account()) |
3e32cb2e JW |
4122 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4123 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4124 | |
8de7ecc6 | 4125 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4126 | unsigned long nr; |
4127 | ||
71cd3113 | 4128 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4129 | continue; |
7de2e9f1 | 4130 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
8de7ecc6 | 4131 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
e09b0b61 | 4132 | (u64)nr * memcg_page_state_unit(memcg1_stats[i])); |
af7c4b0e JW |
4133 | } |
4134 | ||
8de7ecc6 | 4135 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4136 | seq_printf(m, "total_%s %llu\n", |
4137 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4138 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4139 | |
8de7ecc6 | 4140 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4141 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4142 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4143 | PAGE_SIZE); | |
14067bb3 | 4144 | |
7f016ee8 | 4145 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4146 | { |
ef8f2327 MG |
4147 | pg_data_t *pgdat; |
4148 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4149 | unsigned long anon_cost = 0; |
4150 | unsigned long file_cost = 0; | |
7f016ee8 | 4151 | |
ef8f2327 | 4152 | for_each_online_pgdat(pgdat) { |
a3747b53 | 4153 | mz = memcg->nodeinfo[pgdat->node_id]; |
7f016ee8 | 4154 | |
1431d4d1 JW |
4155 | anon_cost += mz->lruvec.anon_cost; |
4156 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4157 | } |
1431d4d1 JW |
4158 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4159 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4160 | } |
4161 | #endif | |
4162 | ||
d2ceb9b7 KH |
4163 | return 0; |
4164 | } | |
4165 | ||
182446d0 TH |
4166 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4167 | struct cftype *cft) | |
a7885eb8 | 4168 | { |
182446d0 | 4169 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4170 | |
1f4c025b | 4171 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4172 | } |
4173 | ||
182446d0 TH |
4174 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4175 | struct cftype *cft, u64 val) | |
a7885eb8 | 4176 | { |
182446d0 | 4177 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4178 | |
37bc3cb9 | 4179 | if (val > 200) |
a7885eb8 KM |
4180 | return -EINVAL; |
4181 | ||
a4792030 | 4182 | if (!mem_cgroup_is_root(memcg)) |
3dae7fec JW |
4183 | memcg->swappiness = val; |
4184 | else | |
4185 | vm_swappiness = val; | |
068b38c1 | 4186 | |
a7885eb8 KM |
4187 | return 0; |
4188 | } | |
4189 | ||
2e72b634 KS |
4190 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4191 | { | |
4192 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4193 | unsigned long usage; |
2e72b634 KS |
4194 | int i; |
4195 | ||
4196 | rcu_read_lock(); | |
4197 | if (!swap) | |
2c488db2 | 4198 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4199 | else |
2c488db2 | 4200 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4201 | |
4202 | if (!t) | |
4203 | goto unlock; | |
4204 | ||
ce00a967 | 4205 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4206 | |
4207 | /* | |
748dad36 | 4208 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4209 | * If it's not true, a threshold was crossed after last |
4210 | * call of __mem_cgroup_threshold(). | |
4211 | */ | |
5407a562 | 4212 | i = t->current_threshold; |
2e72b634 KS |
4213 | |
4214 | /* | |
4215 | * Iterate backward over array of thresholds starting from | |
4216 | * current_threshold and check if a threshold is crossed. | |
4217 | * If none of thresholds below usage is crossed, we read | |
4218 | * only one element of the array here. | |
4219 | */ | |
4220 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4221 | eventfd_signal(t->entries[i].eventfd, 1); | |
4222 | ||
4223 | /* i = current_threshold + 1 */ | |
4224 | i++; | |
4225 | ||
4226 | /* | |
4227 | * Iterate forward over array of thresholds starting from | |
4228 | * current_threshold+1 and check if a threshold is crossed. | |
4229 | * If none of thresholds above usage is crossed, we read | |
4230 | * only one element of the array here. | |
4231 | */ | |
4232 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4233 | eventfd_signal(t->entries[i].eventfd, 1); | |
4234 | ||
4235 | /* Update current_threshold */ | |
5407a562 | 4236 | t->current_threshold = i - 1; |
2e72b634 KS |
4237 | unlock: |
4238 | rcu_read_unlock(); | |
4239 | } | |
4240 | ||
4241 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4242 | { | |
ad4ca5f4 KS |
4243 | while (memcg) { |
4244 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4245 | if (do_memsw_account()) |
ad4ca5f4 KS |
4246 | __mem_cgroup_threshold(memcg, true); |
4247 | ||
4248 | memcg = parent_mem_cgroup(memcg); | |
4249 | } | |
2e72b634 KS |
4250 | } |
4251 | ||
4252 | static int compare_thresholds(const void *a, const void *b) | |
4253 | { | |
4254 | const struct mem_cgroup_threshold *_a = a; | |
4255 | const struct mem_cgroup_threshold *_b = b; | |
4256 | ||
2bff24a3 GT |
4257 | if (_a->threshold > _b->threshold) |
4258 | return 1; | |
4259 | ||
4260 | if (_a->threshold < _b->threshold) | |
4261 | return -1; | |
4262 | ||
4263 | return 0; | |
2e72b634 KS |
4264 | } |
4265 | ||
c0ff4b85 | 4266 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4267 | { |
4268 | struct mem_cgroup_eventfd_list *ev; | |
4269 | ||
2bcf2e92 MH |
4270 | spin_lock(&memcg_oom_lock); |
4271 | ||
c0ff4b85 | 4272 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4273 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4274 | |
4275 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4276 | return 0; |
4277 | } | |
4278 | ||
c0ff4b85 | 4279 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4280 | { |
7d74b06f KH |
4281 | struct mem_cgroup *iter; |
4282 | ||
c0ff4b85 | 4283 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4284 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4285 | } |
4286 | ||
59b6f873 | 4287 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4288 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4289 | { |
2c488db2 KS |
4290 | struct mem_cgroup_thresholds *thresholds; |
4291 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4292 | unsigned long threshold; |
4293 | unsigned long usage; | |
2c488db2 | 4294 | int i, size, ret; |
2e72b634 | 4295 | |
650c5e56 | 4296 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4297 | if (ret) |
4298 | return ret; | |
4299 | ||
4300 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4301 | |
05b84301 | 4302 | if (type == _MEM) { |
2c488db2 | 4303 | thresholds = &memcg->thresholds; |
ce00a967 | 4304 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4305 | } else if (type == _MEMSWAP) { |
2c488db2 | 4306 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4307 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4308 | } else |
2e72b634 KS |
4309 | BUG(); |
4310 | ||
2e72b634 | 4311 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4312 | if (thresholds->primary) |
2e72b634 KS |
4313 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4314 | ||
2c488db2 | 4315 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4316 | |
4317 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4318 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4319 | if (!new) { |
2e72b634 KS |
4320 | ret = -ENOMEM; |
4321 | goto unlock; | |
4322 | } | |
2c488db2 | 4323 | new->size = size; |
2e72b634 KS |
4324 | |
4325 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4326 | if (thresholds->primary) |
4327 | memcpy(new->entries, thresholds->primary->entries, | |
4328 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4329 | |
2e72b634 | 4330 | /* Add new threshold */ |
2c488db2 KS |
4331 | new->entries[size - 1].eventfd = eventfd; |
4332 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4333 | |
4334 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4335 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4336 | compare_thresholds, NULL); |
4337 | ||
4338 | /* Find current threshold */ | |
2c488db2 | 4339 | new->current_threshold = -1; |
2e72b634 | 4340 | for (i = 0; i < size; i++) { |
748dad36 | 4341 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4342 | /* |
2c488db2 KS |
4343 | * new->current_threshold will not be used until |
4344 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4345 | * it here. |
4346 | */ | |
2c488db2 | 4347 | ++new->current_threshold; |
748dad36 SZ |
4348 | } else |
4349 | break; | |
2e72b634 KS |
4350 | } |
4351 | ||
2c488db2 KS |
4352 | /* Free old spare buffer and save old primary buffer as spare */ |
4353 | kfree(thresholds->spare); | |
4354 | thresholds->spare = thresholds->primary; | |
4355 | ||
4356 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4357 | |
907860ed | 4358 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4359 | synchronize_rcu(); |
4360 | ||
2e72b634 KS |
4361 | unlock: |
4362 | mutex_unlock(&memcg->thresholds_lock); | |
4363 | ||
4364 | return ret; | |
4365 | } | |
4366 | ||
59b6f873 | 4367 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4368 | struct eventfd_ctx *eventfd, const char *args) |
4369 | { | |
59b6f873 | 4370 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4371 | } |
4372 | ||
59b6f873 | 4373 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4374 | struct eventfd_ctx *eventfd, const char *args) |
4375 | { | |
59b6f873 | 4376 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4377 | } |
4378 | ||
59b6f873 | 4379 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4380 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4381 | { |
2c488db2 KS |
4382 | struct mem_cgroup_thresholds *thresholds; |
4383 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4384 | unsigned long usage; |
7d36665a | 4385 | int i, j, size, entries; |
2e72b634 KS |
4386 | |
4387 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4388 | |
4389 | if (type == _MEM) { | |
2c488db2 | 4390 | thresholds = &memcg->thresholds; |
ce00a967 | 4391 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4392 | } else if (type == _MEMSWAP) { |
2c488db2 | 4393 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4394 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4395 | } else |
2e72b634 KS |
4396 | BUG(); |
4397 | ||
371528ca AV |
4398 | if (!thresholds->primary) |
4399 | goto unlock; | |
4400 | ||
2e72b634 KS |
4401 | /* Check if a threshold crossed before removing */ |
4402 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4403 | ||
4404 | /* Calculate new number of threshold */ | |
7d36665a | 4405 | size = entries = 0; |
2c488db2 KS |
4406 | for (i = 0; i < thresholds->primary->size; i++) { |
4407 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4408 | size++; |
7d36665a CX |
4409 | else |
4410 | entries++; | |
2e72b634 KS |
4411 | } |
4412 | ||
2c488db2 | 4413 | new = thresholds->spare; |
907860ed | 4414 | |
7d36665a CX |
4415 | /* If no items related to eventfd have been cleared, nothing to do */ |
4416 | if (!entries) | |
4417 | goto unlock; | |
4418 | ||
2e72b634 KS |
4419 | /* Set thresholds array to NULL if we don't have thresholds */ |
4420 | if (!size) { | |
2c488db2 KS |
4421 | kfree(new); |
4422 | new = NULL; | |
907860ed | 4423 | goto swap_buffers; |
2e72b634 KS |
4424 | } |
4425 | ||
2c488db2 | 4426 | new->size = size; |
2e72b634 KS |
4427 | |
4428 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4429 | new->current_threshold = -1; |
4430 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4431 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4432 | continue; |
4433 | ||
2c488db2 | 4434 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4435 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4436 | /* |
2c488db2 | 4437 | * new->current_threshold will not be used |
2e72b634 KS |
4438 | * until rcu_assign_pointer(), so it's safe to increment |
4439 | * it here. | |
4440 | */ | |
2c488db2 | 4441 | ++new->current_threshold; |
2e72b634 KS |
4442 | } |
4443 | j++; | |
4444 | } | |
4445 | ||
907860ed | 4446 | swap_buffers: |
2c488db2 KS |
4447 | /* Swap primary and spare array */ |
4448 | thresholds->spare = thresholds->primary; | |
8c757763 | 4449 | |
2c488db2 | 4450 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4451 | |
907860ed | 4452 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4453 | synchronize_rcu(); |
6611d8d7 MC |
4454 | |
4455 | /* If all events are unregistered, free the spare array */ | |
4456 | if (!new) { | |
4457 | kfree(thresholds->spare); | |
4458 | thresholds->spare = NULL; | |
4459 | } | |
371528ca | 4460 | unlock: |
2e72b634 | 4461 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4462 | } |
c1e862c1 | 4463 | |
59b6f873 | 4464 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4465 | struct eventfd_ctx *eventfd) |
4466 | { | |
59b6f873 | 4467 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4468 | } |
4469 | ||
59b6f873 | 4470 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4471 | struct eventfd_ctx *eventfd) |
4472 | { | |
59b6f873 | 4473 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4474 | } |
4475 | ||
59b6f873 | 4476 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4477 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4478 | { |
9490ff27 | 4479 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4480 | |
9490ff27 KH |
4481 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4482 | if (!event) | |
4483 | return -ENOMEM; | |
4484 | ||
1af8efe9 | 4485 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4486 | |
4487 | event->eventfd = eventfd; | |
4488 | list_add(&event->list, &memcg->oom_notify); | |
4489 | ||
4490 | /* already in OOM ? */ | |
c2b42d3c | 4491 | if (memcg->under_oom) |
9490ff27 | 4492 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4493 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4494 | |
4495 | return 0; | |
4496 | } | |
4497 | ||
59b6f873 | 4498 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4499 | struct eventfd_ctx *eventfd) |
9490ff27 | 4500 | { |
9490ff27 | 4501 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4502 | |
1af8efe9 | 4503 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4504 | |
c0ff4b85 | 4505 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4506 | if (ev->eventfd == eventfd) { |
4507 | list_del(&ev->list); | |
4508 | kfree(ev); | |
4509 | } | |
4510 | } | |
4511 | ||
1af8efe9 | 4512 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4513 | } |
4514 | ||
2da8ca82 | 4515 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4516 | { |
aa9694bb | 4517 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4518 | |
791badbd | 4519 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4520 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4521 | seq_printf(sf, "oom_kill %lu\n", |
4522 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4523 | return 0; |
4524 | } | |
4525 | ||
182446d0 | 4526 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4527 | struct cftype *cft, u64 val) |
4528 | { | |
182446d0 | 4529 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4530 | |
4531 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4532 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4533 | return -EINVAL; |
4534 | ||
c0ff4b85 | 4535 | memcg->oom_kill_disable = val; |
4d845ebf | 4536 | if (!val) |
c0ff4b85 | 4537 | memcg_oom_recover(memcg); |
3dae7fec | 4538 | |
3c11ecf4 KH |
4539 | return 0; |
4540 | } | |
4541 | ||
52ebea74 TH |
4542 | #ifdef CONFIG_CGROUP_WRITEBACK |
4543 | ||
3a8e9ac8 TH |
4544 | #include <trace/events/writeback.h> |
4545 | ||
841710aa TH |
4546 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4547 | { | |
4548 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4549 | } | |
4550 | ||
4551 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4552 | { | |
4553 | wb_domain_exit(&memcg->cgwb_domain); | |
4554 | } | |
4555 | ||
2529bb3a TH |
4556 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4557 | { | |
4558 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4559 | } | |
4560 | ||
841710aa TH |
4561 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4562 | { | |
4563 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4564 | ||
4565 | if (!memcg->css.parent) | |
4566 | return NULL; | |
4567 | ||
4568 | return &memcg->cgwb_domain; | |
4569 | } | |
4570 | ||
c2aa723a TH |
4571 | /** |
4572 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4573 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4574 | * @pfilepages: out parameter for number of file pages |
4575 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4576 | * @pdirty: out parameter for number of dirty pages |
4577 | * @pwriteback: out parameter for number of pages under writeback | |
4578 | * | |
c5edf9cd TH |
4579 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4580 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4581 | * is a bit more involved. | |
c2aa723a | 4582 | * |
c5edf9cd TH |
4583 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4584 | * headroom is calculated as the lowest headroom of itself and the | |
4585 | * ancestors. Note that this doesn't consider the actual amount of | |
4586 | * available memory in the system. The caller should further cap | |
4587 | * *@pheadroom accordingly. | |
c2aa723a | 4588 | */ |
c5edf9cd TH |
4589 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4590 | unsigned long *pheadroom, unsigned long *pdirty, | |
4591 | unsigned long *pwriteback) | |
c2aa723a TH |
4592 | { |
4593 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4594 | struct mem_cgroup *parent; | |
c2aa723a | 4595 | |
fd25a9e0 | 4596 | mem_cgroup_flush_stats(); |
c2aa723a | 4597 | |
2d146aa3 JW |
4598 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
4599 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); | |
4600 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + | |
4601 | memcg_page_state(memcg, NR_ACTIVE_FILE); | |
c2aa723a | 4602 | |
2d146aa3 | 4603 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4604 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4605 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4606 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4607 | unsigned long used = page_counter_read(&memcg->memory); |
4608 | ||
c5edf9cd | 4609 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4610 | memcg = parent; |
4611 | } | |
c2aa723a TH |
4612 | } |
4613 | ||
97b27821 TH |
4614 | /* |
4615 | * Foreign dirty flushing | |
4616 | * | |
4617 | * There's an inherent mismatch between memcg and writeback. The former | |
f0953a1b | 4618 | * tracks ownership per-page while the latter per-inode. This was a |
97b27821 TH |
4619 | * deliberate design decision because honoring per-page ownership in the |
4620 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4621 | * and deemed unnecessary given that write-sharing an inode across | |
4622 | * different cgroups isn't a common use-case. | |
4623 | * | |
4624 | * Combined with inode majority-writer ownership switching, this works well | |
4625 | * enough in most cases but there are some pathological cases. For | |
4626 | * example, let's say there are two cgroups A and B which keep writing to | |
4627 | * different but confined parts of the same inode. B owns the inode and | |
4628 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4629 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4630 | * triggering background writeback. A will be slowed down without a way to | |
4631 | * make writeback of the dirty pages happen. | |
4632 | * | |
f0953a1b | 4633 | * Conditions like the above can lead to a cgroup getting repeatedly and |
97b27821 | 4634 | * severely throttled after making some progress after each |
f0953a1b | 4635 | * dirty_expire_interval while the underlying IO device is almost |
97b27821 TH |
4636 | * completely idle. |
4637 | * | |
4638 | * Solving this problem completely requires matching the ownership tracking | |
4639 | * granularities between memcg and writeback in either direction. However, | |
4640 | * the more egregious behaviors can be avoided by simply remembering the | |
4641 | * most recent foreign dirtying events and initiating remote flushes on | |
4642 | * them when local writeback isn't enough to keep the memory clean enough. | |
4643 | * | |
4644 | * The following two functions implement such mechanism. When a foreign | |
4645 | * page - a page whose memcg and writeback ownerships don't match - is | |
4646 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4647 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4648 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4649 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4650 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4651 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4652 | * limited to MEMCG_CGWB_FRN_CNT. | |
4653 | * | |
4654 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4655 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4656 | * records are lockless and racy. | |
4657 | */ | |
9d8053fc | 4658 | void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio, |
97b27821 TH |
4659 | struct bdi_writeback *wb) |
4660 | { | |
9d8053fc | 4661 | struct mem_cgroup *memcg = folio_memcg(folio); |
97b27821 TH |
4662 | struct memcg_cgwb_frn *frn; |
4663 | u64 now = get_jiffies_64(); | |
4664 | u64 oldest_at = now; | |
4665 | int oldest = -1; | |
4666 | int i; | |
4667 | ||
9d8053fc | 4668 | trace_track_foreign_dirty(folio, wb); |
3a8e9ac8 | 4669 | |
97b27821 TH |
4670 | /* |
4671 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4672 | * using it. If not replace the oldest one which isn't being | |
4673 | * written out. | |
4674 | */ | |
4675 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4676 | frn = &memcg->cgwb_frn[i]; | |
4677 | if (frn->bdi_id == wb->bdi->id && | |
4678 | frn->memcg_id == wb->memcg_css->id) | |
4679 | break; | |
4680 | if (time_before64(frn->at, oldest_at) && | |
4681 | atomic_read(&frn->done.cnt) == 1) { | |
4682 | oldest = i; | |
4683 | oldest_at = frn->at; | |
4684 | } | |
4685 | } | |
4686 | ||
4687 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4688 | /* | |
4689 | * Re-using an existing one. Update timestamp lazily to | |
4690 | * avoid making the cacheline hot. We want them to be | |
4691 | * reasonably up-to-date and significantly shorter than | |
4692 | * dirty_expire_interval as that's what expires the record. | |
4693 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4694 | */ | |
4695 | unsigned long update_intv = | |
4696 | min_t(unsigned long, HZ, | |
4697 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4698 | ||
4699 | if (time_before64(frn->at, now - update_intv)) | |
4700 | frn->at = now; | |
4701 | } else if (oldest >= 0) { | |
4702 | /* replace the oldest free one */ | |
4703 | frn = &memcg->cgwb_frn[oldest]; | |
4704 | frn->bdi_id = wb->bdi->id; | |
4705 | frn->memcg_id = wb->memcg_css->id; | |
4706 | frn->at = now; | |
4707 | } | |
4708 | } | |
4709 | ||
4710 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4711 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4712 | { | |
4713 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4714 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4715 | u64 now = jiffies_64; | |
4716 | int i; | |
4717 | ||
4718 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4719 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4720 | ||
4721 | /* | |
4722 | * If the record is older than dirty_expire_interval, | |
4723 | * writeback on it has already started. No need to kick it | |
4724 | * off again. Also, don't start a new one if there's | |
4725 | * already one in flight. | |
4726 | */ | |
4727 | if (time_after64(frn->at, now - intv) && | |
4728 | atomic_read(&frn->done.cnt) == 1) { | |
4729 | frn->at = 0; | |
3a8e9ac8 | 4730 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
7490a2d2 | 4731 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, |
97b27821 TH |
4732 | WB_REASON_FOREIGN_FLUSH, |
4733 | &frn->done); | |
4734 | } | |
4735 | } | |
4736 | } | |
4737 | ||
841710aa TH |
4738 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4739 | ||
4740 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4741 | { | |
4742 | return 0; | |
4743 | } | |
4744 | ||
4745 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4746 | { | |
4747 | } | |
4748 | ||
2529bb3a TH |
4749 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4750 | { | |
4751 | } | |
4752 | ||
52ebea74 TH |
4753 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4754 | ||
3bc942f3 TH |
4755 | /* |
4756 | * DO NOT USE IN NEW FILES. | |
4757 | * | |
4758 | * "cgroup.event_control" implementation. | |
4759 | * | |
4760 | * This is way over-engineered. It tries to support fully configurable | |
4761 | * events for each user. Such level of flexibility is completely | |
4762 | * unnecessary especially in the light of the planned unified hierarchy. | |
4763 | * | |
4764 | * Please deprecate this and replace with something simpler if at all | |
4765 | * possible. | |
4766 | */ | |
4767 | ||
79bd9814 TH |
4768 | /* |
4769 | * Unregister event and free resources. | |
4770 | * | |
4771 | * Gets called from workqueue. | |
4772 | */ | |
3bc942f3 | 4773 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4774 | { |
3bc942f3 TH |
4775 | struct mem_cgroup_event *event = |
4776 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4777 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4778 | |
4779 | remove_wait_queue(event->wqh, &event->wait); | |
4780 | ||
59b6f873 | 4781 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4782 | |
4783 | /* Notify userspace the event is going away. */ | |
4784 | eventfd_signal(event->eventfd, 1); | |
4785 | ||
4786 | eventfd_ctx_put(event->eventfd); | |
4787 | kfree(event); | |
59b6f873 | 4788 | css_put(&memcg->css); |
79bd9814 TH |
4789 | } |
4790 | ||
4791 | /* | |
a9a08845 | 4792 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4793 | * |
4794 | * Called with wqh->lock held and interrupts disabled. | |
4795 | */ | |
ac6424b9 | 4796 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4797 | int sync, void *key) |
79bd9814 | 4798 | { |
3bc942f3 TH |
4799 | struct mem_cgroup_event *event = |
4800 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4801 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4802 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4803 | |
a9a08845 | 4804 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4805 | /* |
4806 | * If the event has been detached at cgroup removal, we | |
4807 | * can simply return knowing the other side will cleanup | |
4808 | * for us. | |
4809 | * | |
4810 | * We can't race against event freeing since the other | |
4811 | * side will require wqh->lock via remove_wait_queue(), | |
4812 | * which we hold. | |
4813 | */ | |
fba94807 | 4814 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4815 | if (!list_empty(&event->list)) { |
4816 | list_del_init(&event->list); | |
4817 | /* | |
4818 | * We are in atomic context, but cgroup_event_remove() | |
4819 | * may sleep, so we have to call it in workqueue. | |
4820 | */ | |
4821 | schedule_work(&event->remove); | |
4822 | } | |
fba94807 | 4823 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4824 | } |
4825 | ||
4826 | return 0; | |
4827 | } | |
4828 | ||
3bc942f3 | 4829 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4830 | wait_queue_head_t *wqh, poll_table *pt) |
4831 | { | |
3bc942f3 TH |
4832 | struct mem_cgroup_event *event = |
4833 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4834 | |
4835 | event->wqh = wqh; | |
4836 | add_wait_queue(wqh, &event->wait); | |
4837 | } | |
4838 | ||
4839 | /* | |
3bc942f3 TH |
4840 | * DO NOT USE IN NEW FILES. |
4841 | * | |
79bd9814 TH |
4842 | * Parse input and register new cgroup event handler. |
4843 | * | |
4844 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4845 | * Interpretation of args is defined by control file implementation. | |
4846 | */ | |
451af504 TH |
4847 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4848 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4849 | { |
451af504 | 4850 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4851 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4852 | struct mem_cgroup_event *event; |
79bd9814 TH |
4853 | struct cgroup_subsys_state *cfile_css; |
4854 | unsigned int efd, cfd; | |
4855 | struct fd efile; | |
4856 | struct fd cfile; | |
fba94807 | 4857 | const char *name; |
79bd9814 TH |
4858 | char *endp; |
4859 | int ret; | |
4860 | ||
2343e88d SAS |
4861 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
4862 | return -EOPNOTSUPP; | |
4863 | ||
451af504 TH |
4864 | buf = strstrip(buf); |
4865 | ||
4866 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4867 | if (*endp != ' ') |
4868 | return -EINVAL; | |
451af504 | 4869 | buf = endp + 1; |
79bd9814 | 4870 | |
451af504 | 4871 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4872 | if ((*endp != ' ') && (*endp != '\0')) |
4873 | return -EINVAL; | |
451af504 | 4874 | buf = endp + 1; |
79bd9814 TH |
4875 | |
4876 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4877 | if (!event) | |
4878 | return -ENOMEM; | |
4879 | ||
59b6f873 | 4880 | event->memcg = memcg; |
79bd9814 | 4881 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4882 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4883 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4884 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4885 | |
4886 | efile = fdget(efd); | |
4887 | if (!efile.file) { | |
4888 | ret = -EBADF; | |
4889 | goto out_kfree; | |
4890 | } | |
4891 | ||
4892 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4893 | if (IS_ERR(event->eventfd)) { | |
4894 | ret = PTR_ERR(event->eventfd); | |
4895 | goto out_put_efile; | |
4896 | } | |
4897 | ||
4898 | cfile = fdget(cfd); | |
4899 | if (!cfile.file) { | |
4900 | ret = -EBADF; | |
4901 | goto out_put_eventfd; | |
4902 | } | |
4903 | ||
4904 | /* the process need read permission on control file */ | |
4905 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 4906 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
4907 | if (ret < 0) |
4908 | goto out_put_cfile; | |
4909 | ||
fba94807 TH |
4910 | /* |
4911 | * Determine the event callbacks and set them in @event. This used | |
4912 | * to be done via struct cftype but cgroup core no longer knows | |
4913 | * about these events. The following is crude but the whole thing | |
4914 | * is for compatibility anyway. | |
3bc942f3 TH |
4915 | * |
4916 | * DO NOT ADD NEW FILES. | |
fba94807 | 4917 | */ |
b583043e | 4918 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4919 | |
4920 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4921 | event->register_event = mem_cgroup_usage_register_event; | |
4922 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4923 | } else if (!strcmp(name, "memory.oom_control")) { | |
4924 | event->register_event = mem_cgroup_oom_register_event; | |
4925 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4926 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4927 | event->register_event = vmpressure_register_event; | |
4928 | event->unregister_event = vmpressure_unregister_event; | |
4929 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4930 | event->register_event = memsw_cgroup_usage_register_event; |
4931 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4932 | } else { |
4933 | ret = -EINVAL; | |
4934 | goto out_put_cfile; | |
4935 | } | |
4936 | ||
79bd9814 | 4937 | /* |
b5557c4c TH |
4938 | * Verify @cfile should belong to @css. Also, remaining events are |
4939 | * automatically removed on cgroup destruction but the removal is | |
4940 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4941 | */ |
b583043e | 4942 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4943 | &memory_cgrp_subsys); |
79bd9814 | 4944 | ret = -EINVAL; |
5a17f543 | 4945 | if (IS_ERR(cfile_css)) |
79bd9814 | 4946 | goto out_put_cfile; |
5a17f543 TH |
4947 | if (cfile_css != css) { |
4948 | css_put(cfile_css); | |
79bd9814 | 4949 | goto out_put_cfile; |
5a17f543 | 4950 | } |
79bd9814 | 4951 | |
451af504 | 4952 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4953 | if (ret) |
4954 | goto out_put_css; | |
4955 | ||
9965ed17 | 4956 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4957 | |
4ba9515d | 4958 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 4959 | list_add(&event->list, &memcg->event_list); |
4ba9515d | 4960 | spin_unlock_irq(&memcg->event_list_lock); |
79bd9814 TH |
4961 | |
4962 | fdput(cfile); | |
4963 | fdput(efile); | |
4964 | ||
451af504 | 4965 | return nbytes; |
79bd9814 TH |
4966 | |
4967 | out_put_css: | |
b5557c4c | 4968 | css_put(css); |
79bd9814 TH |
4969 | out_put_cfile: |
4970 | fdput(cfile); | |
4971 | out_put_eventfd: | |
4972 | eventfd_ctx_put(event->eventfd); | |
4973 | out_put_efile: | |
4974 | fdput(efile); | |
4975 | out_kfree: | |
4976 | kfree(event); | |
4977 | ||
4978 | return ret; | |
4979 | } | |
4980 | ||
c29b5b3d MS |
4981 | #if defined(CONFIG_MEMCG_KMEM) && (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) |
4982 | static int mem_cgroup_slab_show(struct seq_file *m, void *p) | |
4983 | { | |
4984 | /* | |
4985 | * Deprecated. | |
df4ae285 | 4986 | * Please, take a look at tools/cgroup/memcg_slabinfo.py . |
c29b5b3d MS |
4987 | */ |
4988 | return 0; | |
4989 | } | |
4990 | #endif | |
4991 | ||
241994ed | 4992 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4993 | { |
0eea1030 | 4994 | .name = "usage_in_bytes", |
8c7c6e34 | 4995 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4996 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4997 | }, |
c84872e1 PE |
4998 | { |
4999 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5000 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5001 | .write = mem_cgroup_reset, |
791badbd | 5002 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5003 | }, |
8cdea7c0 | 5004 | { |
0eea1030 | 5005 | .name = "limit_in_bytes", |
8c7c6e34 | 5006 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5007 | .write = mem_cgroup_write, |
791badbd | 5008 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5009 | }, |
296c81d8 BS |
5010 | { |
5011 | .name = "soft_limit_in_bytes", | |
5012 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5013 | .write = mem_cgroup_write, |
791badbd | 5014 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5015 | }, |
8cdea7c0 BS |
5016 | { |
5017 | .name = "failcnt", | |
8c7c6e34 | 5018 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5019 | .write = mem_cgroup_reset, |
791badbd | 5020 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5021 | }, |
d2ceb9b7 KH |
5022 | { |
5023 | .name = "stat", | |
2da8ca82 | 5024 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5025 | }, |
c1e862c1 KH |
5026 | { |
5027 | .name = "force_empty", | |
6770c64e | 5028 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5029 | }, |
18f59ea7 BS |
5030 | { |
5031 | .name = "use_hierarchy", | |
5032 | .write_u64 = mem_cgroup_hierarchy_write, | |
5033 | .read_u64 = mem_cgroup_hierarchy_read, | |
5034 | }, | |
79bd9814 | 5035 | { |
3bc942f3 | 5036 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5037 | .write = memcg_write_event_control, |
7dbdb199 | 5038 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 5039 | }, |
a7885eb8 KM |
5040 | { |
5041 | .name = "swappiness", | |
5042 | .read_u64 = mem_cgroup_swappiness_read, | |
5043 | .write_u64 = mem_cgroup_swappiness_write, | |
5044 | }, | |
7dc74be0 DN |
5045 | { |
5046 | .name = "move_charge_at_immigrate", | |
5047 | .read_u64 = mem_cgroup_move_charge_read, | |
5048 | .write_u64 = mem_cgroup_move_charge_write, | |
5049 | }, | |
9490ff27 KH |
5050 | { |
5051 | .name = "oom_control", | |
2da8ca82 | 5052 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5053 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 | 5054 | }, |
70ddf637 AV |
5055 | { |
5056 | .name = "pressure_level", | |
70ddf637 | 5057 | }, |
406eb0c9 YH |
5058 | #ifdef CONFIG_NUMA |
5059 | { | |
5060 | .name = "numa_stat", | |
2da8ca82 | 5061 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5062 | }, |
5063 | #endif | |
510fc4e1 GC |
5064 | { |
5065 | .name = "kmem.limit_in_bytes", | |
5066 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5067 | .write = mem_cgroup_write, |
791badbd | 5068 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5069 | }, |
5070 | { | |
5071 | .name = "kmem.usage_in_bytes", | |
5072 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5073 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5074 | }, |
5075 | { | |
5076 | .name = "kmem.failcnt", | |
5077 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5078 | .write = mem_cgroup_reset, |
791badbd | 5079 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5080 | }, |
5081 | { | |
5082 | .name = "kmem.max_usage_in_bytes", | |
5083 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5084 | .write = mem_cgroup_reset, |
791badbd | 5085 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5086 | }, |
a87425a3 YS |
5087 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5088 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5089 | { |
5090 | .name = "kmem.slabinfo", | |
c29b5b3d | 5091 | .seq_show = mem_cgroup_slab_show, |
749c5415 GC |
5092 | }, |
5093 | #endif | |
d55f90bf VD |
5094 | { |
5095 | .name = "kmem.tcp.limit_in_bytes", | |
5096 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5097 | .write = mem_cgroup_write, | |
5098 | .read_u64 = mem_cgroup_read_u64, | |
5099 | }, | |
5100 | { | |
5101 | .name = "kmem.tcp.usage_in_bytes", | |
5102 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5103 | .read_u64 = mem_cgroup_read_u64, | |
5104 | }, | |
5105 | { | |
5106 | .name = "kmem.tcp.failcnt", | |
5107 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5108 | .write = mem_cgroup_reset, | |
5109 | .read_u64 = mem_cgroup_read_u64, | |
5110 | }, | |
5111 | { | |
5112 | .name = "kmem.tcp.max_usage_in_bytes", | |
5113 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5114 | .write = mem_cgroup_reset, | |
5115 | .read_u64 = mem_cgroup_read_u64, | |
5116 | }, | |
6bc10349 | 5117 | { }, /* terminate */ |
af36f906 | 5118 | }; |
8c7c6e34 | 5119 | |
73f576c0 JW |
5120 | /* |
5121 | * Private memory cgroup IDR | |
5122 | * | |
5123 | * Swap-out records and page cache shadow entries need to store memcg | |
5124 | * references in constrained space, so we maintain an ID space that is | |
5125 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5126 | * memory-controlled cgroups to 64k. | |
5127 | * | |
b8f2935f | 5128 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5129 | * the cgroup has been destroyed, such as page cache or reclaimable |
5130 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5131 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5132 | * relatively small ID space and prevent the creation of new cgroups | |
5133 | * even when there are much fewer than 64k cgroups - possibly none. | |
5134 | * | |
5135 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5136 | * be freed and recycled when it's no longer needed, which is usually | |
5137 | * when the CSS is offlined. | |
5138 | * | |
5139 | * The only exception to that are records of swapped out tmpfs/shmem | |
5140 | * pages that need to be attributed to live ancestors on swapin. But | |
5141 | * those references are manageable from userspace. | |
5142 | */ | |
5143 | ||
5144 | static DEFINE_IDR(mem_cgroup_idr); | |
5145 | ||
7e97de0b KT |
5146 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5147 | { | |
5148 | if (memcg->id.id > 0) { | |
5149 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5150 | memcg->id.id = 0; | |
5151 | } | |
5152 | } | |
5153 | ||
c1514c0a VF |
5154 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5155 | unsigned int n) | |
73f576c0 | 5156 | { |
1c2d479a | 5157 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5158 | } |
5159 | ||
615d66c3 | 5160 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5161 | { |
1c2d479a | 5162 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5163 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5164 | |
5165 | /* Memcg ID pins CSS */ | |
5166 | css_put(&memcg->css); | |
5167 | } | |
5168 | } | |
5169 | ||
615d66c3 VD |
5170 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5171 | { | |
5172 | mem_cgroup_id_put_many(memcg, 1); | |
5173 | } | |
5174 | ||
73f576c0 JW |
5175 | /** |
5176 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5177 | * @id: the memcg id to look up | |
5178 | * | |
5179 | * Caller must hold rcu_read_lock(). | |
5180 | */ | |
5181 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5182 | { | |
5183 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5184 | return idr_find(&mem_cgroup_idr, id); | |
5185 | } | |
5186 | ||
c15187a4 RG |
5187 | #ifdef CONFIG_SHRINKER_DEBUG |
5188 | struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino) | |
5189 | { | |
5190 | struct cgroup *cgrp; | |
5191 | struct cgroup_subsys_state *css; | |
5192 | struct mem_cgroup *memcg; | |
5193 | ||
5194 | cgrp = cgroup_get_from_id(ino); | |
5195 | if (!cgrp) | |
5196 | return ERR_PTR(-ENOENT); | |
5197 | ||
5198 | css = cgroup_get_e_css(cgrp, &memory_cgrp_subsys); | |
5199 | if (css) | |
5200 | memcg = container_of(css, struct mem_cgroup, css); | |
5201 | else | |
5202 | memcg = ERR_PTR(-ENOENT); | |
5203 | ||
5204 | cgroup_put(cgrp); | |
5205 | ||
5206 | return memcg; | |
5207 | } | |
5208 | #endif | |
5209 | ||
ef8f2327 | 5210 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5211 | { |
5212 | struct mem_cgroup_per_node *pn; | |
8c9bb398 WY |
5213 | |
5214 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
6d12e2d8 KH |
5215 | if (!pn) |
5216 | return 1; | |
1ecaab2b | 5217 | |
7e1c0d6f SB |
5218 | pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu, |
5219 | GFP_KERNEL_ACCOUNT); | |
5220 | if (!pn->lruvec_stats_percpu) { | |
00f3ca2c JW |
5221 | kfree(pn); |
5222 | return 1; | |
5223 | } | |
5224 | ||
ef8f2327 | 5225 | lruvec_init(&pn->lruvec); |
ef8f2327 MG |
5226 | pn->memcg = memcg; |
5227 | ||
54f72fe0 | 5228 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5229 | return 0; |
5230 | } | |
5231 | ||
ef8f2327 | 5232 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5233 | { |
00f3ca2c JW |
5234 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5235 | ||
4eaf431f MH |
5236 | if (!pn) |
5237 | return; | |
5238 | ||
7e1c0d6f | 5239 | free_percpu(pn->lruvec_stats_percpu); |
00f3ca2c | 5240 | kfree(pn); |
1ecaab2b KH |
5241 | } |
5242 | ||
40e952f9 | 5243 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5244 | { |
c8b2a36f | 5245 | int node; |
59927fb9 | 5246 | |
c8b2a36f | 5247 | for_each_node(node) |
ef8f2327 | 5248 | free_mem_cgroup_per_node_info(memcg, node); |
410f8e82 | 5249 | kfree(memcg->vmstats); |
871789d4 | 5250 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 5251 | kfree(memcg); |
59927fb9 | 5252 | } |
3afe36b1 | 5253 | |
40e952f9 TE |
5254 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5255 | { | |
ec1c86b2 | 5256 | lru_gen_exit_memcg(memcg); |
40e952f9 TE |
5257 | memcg_wb_domain_exit(memcg); |
5258 | __mem_cgroup_free(memcg); | |
5259 | } | |
5260 | ||
0b8f73e1 | 5261 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5262 | { |
d142e3e6 | 5263 | struct mem_cgroup *memcg; |
6d12e2d8 | 5264 | int node; |
97b27821 | 5265 | int __maybe_unused i; |
11d67612 | 5266 | long error = -ENOMEM; |
8cdea7c0 | 5267 | |
06b2c3b0 | 5268 | memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL); |
c0ff4b85 | 5269 | if (!memcg) |
11d67612 | 5270 | return ERR_PTR(error); |
0b8f73e1 | 5271 | |
73f576c0 | 5272 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
be740503 | 5273 | 1, MEM_CGROUP_ID_MAX + 1, GFP_KERNEL); |
11d67612 YS |
5274 | if (memcg->id.id < 0) { |
5275 | error = memcg->id.id; | |
73f576c0 | 5276 | goto fail; |
11d67612 | 5277 | } |
73f576c0 | 5278 | |
410f8e82 SB |
5279 | memcg->vmstats = kzalloc(sizeof(struct memcg_vmstats), GFP_KERNEL); |
5280 | if (!memcg->vmstats) | |
5281 | goto fail; | |
5282 | ||
3e38e0aa RG |
5283 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5284 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5285 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5286 | goto fail; |
78fb7466 | 5287 | |
3ed28fa1 | 5288 | for_each_node(node) |
ef8f2327 | 5289 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5290 | goto fail; |
f64c3f54 | 5291 | |
0b8f73e1 JW |
5292 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5293 | goto fail; | |
28dbc4b6 | 5294 | |
f7e1cb6e | 5295 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5296 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5297 | mutex_init(&memcg->thresholds_lock); |
5298 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5299 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5300 | INIT_LIST_HEAD(&memcg->event_list); |
5301 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5302 | memcg->socket_pressure = jiffies; |
84c07d11 | 5303 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5304 | memcg->kmemcg_id = -1; |
bf4f0599 | 5305 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5306 | #endif |
52ebea74 TH |
5307 | #ifdef CONFIG_CGROUP_WRITEBACK |
5308 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5309 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5310 | memcg->cgwb_frn[i].done = | |
5311 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5312 | #endif |
5313 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5314 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5315 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5316 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5317 | #endif |
73f576c0 | 5318 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
ec1c86b2 | 5319 | lru_gen_init_memcg(memcg); |
0b8f73e1 JW |
5320 | return memcg; |
5321 | fail: | |
7e97de0b | 5322 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5323 | __mem_cgroup_free(memcg); |
11d67612 | 5324 | return ERR_PTR(error); |
d142e3e6 GC |
5325 | } |
5326 | ||
0b8f73e1 JW |
5327 | static struct cgroup_subsys_state * __ref |
5328 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5329 | { |
0b8f73e1 | 5330 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5331 | struct mem_cgroup *memcg, *old_memcg; |
d142e3e6 | 5332 | |
b87d8cef | 5333 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5334 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5335 | set_active_memcg(old_memcg); |
11d67612 YS |
5336 | if (IS_ERR(memcg)) |
5337 | return ERR_CAST(memcg); | |
d142e3e6 | 5338 | |
d1663a90 | 5339 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5340 | memcg->soft_limit = PAGE_COUNTER_MAX; |
f4840ccf JW |
5341 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
5342 | memcg->zswap_max = PAGE_COUNTER_MAX; | |
5343 | #endif | |
4b82ab4f | 5344 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5345 | if (parent) { |
5346 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5347 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
bef8620c | 5348 | |
3e32cb2e | 5349 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5350 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5351 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5352 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5353 | } else { |
8278f1c7 | 5354 | init_memcg_events(); |
bef8620c RG |
5355 | page_counter_init(&memcg->memory, NULL); |
5356 | page_counter_init(&memcg->swap, NULL); | |
5357 | page_counter_init(&memcg->kmem, NULL); | |
5358 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5359 | |
0b8f73e1 JW |
5360 | root_mem_cgroup = memcg; |
5361 | return &memcg->css; | |
5362 | } | |
5363 | ||
f7e1cb6e | 5364 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5365 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5366 | |
0b8f73e1 | 5367 | return &memcg->css; |
0b8f73e1 JW |
5368 | } |
5369 | ||
73f576c0 | 5370 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5371 | { |
58fa2a55 VD |
5372 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5373 | ||
da0efe30 MS |
5374 | if (memcg_online_kmem(memcg)) |
5375 | goto remove_id; | |
5376 | ||
0a4465d3 | 5377 | /* |
e4262c4f | 5378 | * A memcg must be visible for expand_shrinker_info() |
0a4465d3 KT |
5379 | * by the time the maps are allocated. So, we allocate maps |
5380 | * here, when for_each_mem_cgroup() can't skip it. | |
5381 | */ | |
da0efe30 MS |
5382 | if (alloc_shrinker_info(memcg)) |
5383 | goto offline_kmem; | |
0a4465d3 | 5384 | |
73f576c0 | 5385 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5386 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5387 | css_get(css); |
aa48e47e SB |
5388 | |
5389 | if (unlikely(mem_cgroup_is_root(memcg))) | |
5390 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, | |
5391 | 2UL*HZ); | |
2f7dd7a4 | 5392 | return 0; |
da0efe30 MS |
5393 | offline_kmem: |
5394 | memcg_offline_kmem(memcg); | |
5395 | remove_id: | |
5396 | mem_cgroup_id_remove(memcg); | |
5397 | return -ENOMEM; | |
8cdea7c0 BS |
5398 | } |
5399 | ||
eb95419b | 5400 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5401 | { |
eb95419b | 5402 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5403 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5404 | |
5405 | /* | |
5406 | * Unregister events and notify userspace. | |
5407 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5408 | * directory to avoid race between userspace and kernelspace. | |
5409 | */ | |
4ba9515d | 5410 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5411 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
79bd9814 TH |
5412 | list_del_init(&event->list); |
5413 | schedule_work(&event->remove); | |
5414 | } | |
4ba9515d | 5415 | spin_unlock_irq(&memcg->event_list_lock); |
ec64f515 | 5416 | |
bf8d5d52 | 5417 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5418 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5419 | |
567e9ab2 | 5420 | memcg_offline_kmem(memcg); |
a178015c | 5421 | reparent_shrinker_deferred(memcg); |
52ebea74 | 5422 | wb_memcg_offline(memcg); |
73f576c0 | 5423 | |
591edfb1 RG |
5424 | drain_all_stock(memcg); |
5425 | ||
73f576c0 | 5426 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5427 | } |
5428 | ||
6df38689 VD |
5429 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5430 | { | |
5431 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5432 | ||
5433 | invalidate_reclaim_iterators(memcg); | |
5434 | } | |
5435 | ||
eb95419b | 5436 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5437 | { |
eb95419b | 5438 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5439 | int __maybe_unused i; |
c268e994 | 5440 | |
97b27821 TH |
5441 | #ifdef CONFIG_CGROUP_WRITEBACK |
5442 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5443 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5444 | #endif | |
f7e1cb6e | 5445 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5446 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5447 | |
0db15298 | 5448 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5449 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5450 | |
0b8f73e1 JW |
5451 | vmpressure_cleanup(&memcg->vmpressure); |
5452 | cancel_work_sync(&memcg->high_work); | |
5453 | mem_cgroup_remove_from_trees(memcg); | |
e4262c4f | 5454 | free_shrinker_info(memcg); |
0b8f73e1 | 5455 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5456 | } |
5457 | ||
1ced953b TH |
5458 | /** |
5459 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5460 | * @css: the target css | |
5461 | * | |
5462 | * Reset the states of the mem_cgroup associated with @css. This is | |
5463 | * invoked when the userland requests disabling on the default hierarchy | |
5464 | * but the memcg is pinned through dependency. The memcg should stop | |
5465 | * applying policies and should revert to the vanilla state as it may be | |
5466 | * made visible again. | |
5467 | * | |
5468 | * The current implementation only resets the essential configurations. | |
5469 | * This needs to be expanded to cover all the visible parts. | |
5470 | */ | |
5471 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5472 | { | |
5473 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5474 | ||
bbec2e15 RG |
5475 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5476 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5477 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5478 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5479 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5480 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5481 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5482 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5483 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5484 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5485 | } |
5486 | ||
2d146aa3 JW |
5487 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
5488 | { | |
5489 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5490 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5491 | struct memcg_vmstats_percpu *statc; | |
5492 | long delta, v; | |
7e1c0d6f | 5493 | int i, nid; |
2d146aa3 JW |
5494 | |
5495 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5496 | ||
5497 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
5498 | /* | |
5499 | * Collect the aggregated propagation counts of groups | |
5500 | * below us. We're in a per-cpu loop here and this is | |
5501 | * a global counter, so the first cycle will get them. | |
5502 | */ | |
410f8e82 | 5503 | delta = memcg->vmstats->state_pending[i]; |
2d146aa3 | 5504 | if (delta) |
410f8e82 | 5505 | memcg->vmstats->state_pending[i] = 0; |
2d146aa3 JW |
5506 | |
5507 | /* Add CPU changes on this level since the last flush */ | |
5508 | v = READ_ONCE(statc->state[i]); | |
5509 | if (v != statc->state_prev[i]) { | |
5510 | delta += v - statc->state_prev[i]; | |
5511 | statc->state_prev[i] = v; | |
5512 | } | |
5513 | ||
5514 | if (!delta) | |
5515 | continue; | |
5516 | ||
5517 | /* Aggregate counts on this level and propagate upwards */ | |
410f8e82 | 5518 | memcg->vmstats->state[i] += delta; |
2d146aa3 | 5519 | if (parent) |
410f8e82 | 5520 | parent->vmstats->state_pending[i] += delta; |
2d146aa3 JW |
5521 | } |
5522 | ||
8278f1c7 | 5523 | for (i = 0; i < NR_MEMCG_EVENTS; i++) { |
410f8e82 | 5524 | delta = memcg->vmstats->events_pending[i]; |
2d146aa3 | 5525 | if (delta) |
410f8e82 | 5526 | memcg->vmstats->events_pending[i] = 0; |
2d146aa3 JW |
5527 | |
5528 | v = READ_ONCE(statc->events[i]); | |
5529 | if (v != statc->events_prev[i]) { | |
5530 | delta += v - statc->events_prev[i]; | |
5531 | statc->events_prev[i] = v; | |
5532 | } | |
5533 | ||
5534 | if (!delta) | |
5535 | continue; | |
5536 | ||
410f8e82 | 5537 | memcg->vmstats->events[i] += delta; |
2d146aa3 | 5538 | if (parent) |
410f8e82 | 5539 | parent->vmstats->events_pending[i] += delta; |
2d146aa3 | 5540 | } |
7e1c0d6f SB |
5541 | |
5542 | for_each_node_state(nid, N_MEMORY) { | |
5543 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; | |
5544 | struct mem_cgroup_per_node *ppn = NULL; | |
5545 | struct lruvec_stats_percpu *lstatc; | |
5546 | ||
5547 | if (parent) | |
5548 | ppn = parent->nodeinfo[nid]; | |
5549 | ||
5550 | lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu); | |
5551 | ||
5552 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { | |
5553 | delta = pn->lruvec_stats.state_pending[i]; | |
5554 | if (delta) | |
5555 | pn->lruvec_stats.state_pending[i] = 0; | |
5556 | ||
5557 | v = READ_ONCE(lstatc->state[i]); | |
5558 | if (v != lstatc->state_prev[i]) { | |
5559 | delta += v - lstatc->state_prev[i]; | |
5560 | lstatc->state_prev[i] = v; | |
5561 | } | |
5562 | ||
5563 | if (!delta) | |
5564 | continue; | |
5565 | ||
5566 | pn->lruvec_stats.state[i] += delta; | |
5567 | if (ppn) | |
5568 | ppn->lruvec_stats.state_pending[i] += delta; | |
5569 | } | |
5570 | } | |
2d146aa3 JW |
5571 | } |
5572 | ||
02491447 | 5573 | #ifdef CONFIG_MMU |
7dc74be0 | 5574 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5575 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5576 | { |
05b84301 | 5577 | int ret; |
9476db97 | 5578 | |
d0164adc MG |
5579 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5580 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5581 | if (!ret) { |
854ffa8d | 5582 | mc.precharge += count; |
854ffa8d DN |
5583 | return ret; |
5584 | } | |
9476db97 | 5585 | |
3674534b | 5586 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5587 | while (count--) { |
3674534b | 5588 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5589 | if (ret) |
38c5d72f | 5590 | return ret; |
854ffa8d | 5591 | mc.precharge++; |
9476db97 | 5592 | cond_resched(); |
854ffa8d | 5593 | } |
9476db97 | 5594 | return 0; |
4ffef5fe DN |
5595 | } |
5596 | ||
4ffef5fe DN |
5597 | union mc_target { |
5598 | struct page *page; | |
02491447 | 5599 | swp_entry_t ent; |
4ffef5fe DN |
5600 | }; |
5601 | ||
4ffef5fe | 5602 | enum mc_target_type { |
8d32ff84 | 5603 | MC_TARGET_NONE = 0, |
4ffef5fe | 5604 | MC_TARGET_PAGE, |
02491447 | 5605 | MC_TARGET_SWAP, |
c733a828 | 5606 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5607 | }; |
5608 | ||
90254a65 DN |
5609 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5610 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5611 | { |
25b2995a | 5612 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5613 | |
90254a65 DN |
5614 | if (!page || !page_mapped(page)) |
5615 | return NULL; | |
5616 | if (PageAnon(page)) { | |
1dfab5ab | 5617 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5618 | return NULL; |
1dfab5ab JW |
5619 | } else { |
5620 | if (!(mc.flags & MOVE_FILE)) | |
5621 | return NULL; | |
5622 | } | |
90254a65 DN |
5623 | if (!get_page_unless_zero(page)) |
5624 | return NULL; | |
5625 | ||
5626 | return page; | |
5627 | } | |
5628 | ||
c733a828 | 5629 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5630 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5631 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5632 | { |
90254a65 DN |
5633 | struct page *page = NULL; |
5634 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5635 | ||
9a137153 | 5636 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5637 | return NULL; |
c733a828 JG |
5638 | |
5639 | /* | |
27674ef6 CH |
5640 | * Handle device private pages that are not accessible by the CPU, but |
5641 | * stored as special swap entries in the page table. | |
c733a828 JG |
5642 | */ |
5643 | if (is_device_private_entry(ent)) { | |
af5cdaf8 | 5644 | page = pfn_swap_entry_to_page(ent); |
27674ef6 | 5645 | if (!get_page_unless_zero(page)) |
c733a828 JG |
5646 | return NULL; |
5647 | return page; | |
5648 | } | |
5649 | ||
9a137153 RC |
5650 | if (non_swap_entry(ent)) |
5651 | return NULL; | |
5652 | ||
4b91355e | 5653 | /* |
cb691e2f | 5654 | * Because swap_cache_get_folio() updates some statistics counter, |
4b91355e KH |
5655 | * we call find_get_page() with swapper_space directly. |
5656 | */ | |
f6ab1f7f | 5657 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5658 | entry->val = ent.val; |
90254a65 DN |
5659 | |
5660 | return page; | |
5661 | } | |
4b91355e KH |
5662 | #else |
5663 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5664 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5665 | { |
5666 | return NULL; | |
5667 | } | |
5668 | #endif | |
90254a65 | 5669 | |
87946a72 | 5670 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
48384b0b | 5671 | unsigned long addr, pte_t ptent) |
87946a72 | 5672 | { |
87946a72 DN |
5673 | if (!vma->vm_file) /* anonymous vma */ |
5674 | return NULL; | |
1dfab5ab | 5675 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5676 | return NULL; |
5677 | ||
87946a72 | 5678 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5679 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5680 | return find_get_incore_page(vma->vm_file->f_mapping, |
5681 | linear_page_index(vma, addr)); | |
87946a72 DN |
5682 | } |
5683 | ||
b1b0deab CG |
5684 | /** |
5685 | * mem_cgroup_move_account - move account of the page | |
5686 | * @page: the page | |
25843c2b | 5687 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5688 | * @from: mem_cgroup which the page is moved from. |
5689 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5690 | * | |
3ac808fd | 5691 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5692 | * |
5693 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5694 | * from old cgroup. | |
5695 | */ | |
5696 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5697 | bool compound, |
b1b0deab CG |
5698 | struct mem_cgroup *from, |
5699 | struct mem_cgroup *to) | |
5700 | { | |
fcce4672 | 5701 | struct folio *folio = page_folio(page); |
ae8af438 KK |
5702 | struct lruvec *from_vec, *to_vec; |
5703 | struct pglist_data *pgdat; | |
fcce4672 | 5704 | unsigned int nr_pages = compound ? folio_nr_pages(folio) : 1; |
8e88bd2d | 5705 | int nid, ret; |
b1b0deab CG |
5706 | |
5707 | VM_BUG_ON(from == to); | |
fcce4672 | 5708 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
9c325215 | 5709 | VM_BUG_ON(compound && !folio_test_large(folio)); |
b1b0deab CG |
5710 | |
5711 | /* | |
6a93ca8f | 5712 | * Prevent mem_cgroup_migrate() from looking at |
bcfe06bf | 5713 | * page's memory cgroup of its source page while we change it. |
b1b0deab | 5714 | */ |
f627c2f5 | 5715 | ret = -EBUSY; |
fcce4672 | 5716 | if (!folio_trylock(folio)) |
b1b0deab CG |
5717 | goto out; |
5718 | ||
5719 | ret = -EINVAL; | |
fcce4672 | 5720 | if (folio_memcg(folio) != from) |
b1b0deab CG |
5721 | goto out_unlock; |
5722 | ||
fcce4672 | 5723 | pgdat = folio_pgdat(folio); |
867e5e1d JW |
5724 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5725 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5726 | |
fcce4672 | 5727 | folio_memcg_lock(folio); |
b1b0deab | 5728 | |
fcce4672 MWO |
5729 | if (folio_test_anon(folio)) { |
5730 | if (folio_mapped(folio)) { | |
be5d0a74 JW |
5731 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); |
5732 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
fcce4672 | 5733 | if (folio_test_transhuge(folio)) { |
69473e5d MS |
5734 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
5735 | -nr_pages); | |
5736 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5737 | nr_pages); | |
468c3982 | 5738 | } |
be5d0a74 JW |
5739 | } |
5740 | } else { | |
0d1c2072 JW |
5741 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5742 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5743 | ||
fcce4672 | 5744 | if (folio_test_swapbacked(folio)) { |
0d1c2072 JW |
5745 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); |
5746 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5747 | } | |
5748 | ||
fcce4672 | 5749 | if (folio_mapped(folio)) { |
49e50d27 JW |
5750 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
5751 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5752 | } | |
b1b0deab | 5753 | |
fcce4672 MWO |
5754 | if (folio_test_dirty(folio)) { |
5755 | struct address_space *mapping = folio_mapping(folio); | |
c4843a75 | 5756 | |
f56753ac | 5757 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5758 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5759 | -nr_pages); | |
5760 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5761 | nr_pages); | |
5762 | } | |
c4843a75 GT |
5763 | } |
5764 | } | |
5765 | ||
fcce4672 | 5766 | if (folio_test_writeback(folio)) { |
ae8af438 KK |
5767 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5768 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5769 | } |
5770 | ||
5771 | /* | |
abb242f5 JW |
5772 | * All state has been migrated, let's switch to the new memcg. |
5773 | * | |
bcfe06bf | 5774 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
5775 | * is referenced, charged, isolated, and locked: we can't race |
5776 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 5777 | * that would rely on a stable page's memory cgroup. |
abb242f5 JW |
5778 | * |
5779 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
bcfe06bf | 5780 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
5781 | * new memcg that isn't locked, the above state can change |
5782 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5783 | */ |
abb242f5 | 5784 | smp_mb(); |
b1b0deab | 5785 | |
1a3e1f40 JW |
5786 | css_get(&to->css); |
5787 | css_put(&from->css); | |
5788 | ||
fcce4672 | 5789 | folio->memcg_data = (unsigned long)to; |
87eaceb3 | 5790 | |
f70ad448 | 5791 | __folio_memcg_unlock(from); |
b1b0deab CG |
5792 | |
5793 | ret = 0; | |
fcce4672 | 5794 | nid = folio_nid(folio); |
b1b0deab CG |
5795 | |
5796 | local_irq_disable(); | |
6e0110c2 | 5797 | mem_cgroup_charge_statistics(to, nr_pages); |
8e88bd2d | 5798 | memcg_check_events(to, nid); |
6e0110c2 | 5799 | mem_cgroup_charge_statistics(from, -nr_pages); |
8e88bd2d | 5800 | memcg_check_events(from, nid); |
b1b0deab CG |
5801 | local_irq_enable(); |
5802 | out_unlock: | |
fcce4672 | 5803 | folio_unlock(folio); |
b1b0deab CG |
5804 | out: |
5805 | return ret; | |
5806 | } | |
5807 | ||
7cf7806c LR |
5808 | /** |
5809 | * get_mctgt_type - get target type of moving charge | |
5810 | * @vma: the vma the pte to be checked belongs | |
5811 | * @addr: the address corresponding to the pte to be checked | |
5812 | * @ptent: the pte to be checked | |
5813 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5814 | * | |
5815 | * Returns | |
5816 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5817 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5818 | * move charge. if @target is not NULL, the page is stored in target->page | |
5819 | * with extra refcnt got(Callers should handle it). | |
5820 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5821 | * target for charge migration. if @target is not NULL, the entry is stored | |
5822 | * in target->ent. | |
f25cbb7a AS |
5823 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is device memory and |
5824 | * thus not on the lru. | |
df6ad698 JG |
5825 | * For now we such page is charge like a regular page would be as for all |
5826 | * intent and purposes it is just special memory taking the place of a | |
5827 | * regular page. | |
c733a828 JG |
5828 | * |
5829 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5830 | * |
5831 | * Called with pte lock held. | |
5832 | */ | |
5833 | ||
8d32ff84 | 5834 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5835 | unsigned long addr, pte_t ptent, union mc_target *target) |
5836 | { | |
5837 | struct page *page = NULL; | |
8d32ff84 | 5838 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5839 | swp_entry_t ent = { .val = 0 }; |
5840 | ||
5841 | if (pte_present(ptent)) | |
5842 | page = mc_handle_present_pte(vma, addr, ptent); | |
5c041f5d PX |
5843 | else if (pte_none_mostly(ptent)) |
5844 | /* | |
5845 | * PTE markers should be treated as a none pte here, separated | |
5846 | * from other swap handling below. | |
5847 | */ | |
5848 | page = mc_handle_file_pte(vma, addr, ptent); | |
90254a65 | 5849 | else if (is_swap_pte(ptent)) |
48406ef8 | 5850 | page = mc_handle_swap_pte(vma, ptent, &ent); |
90254a65 DN |
5851 | |
5852 | if (!page && !ent.val) | |
8d32ff84 | 5853 | return ret; |
02491447 | 5854 | if (page) { |
02491447 | 5855 | /* |
0a31bc97 | 5856 | * Do only loose check w/o serialization. |
1306a85a | 5857 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5858 | * not under LRU exclusion. |
02491447 | 5859 | */ |
bcfe06bf | 5860 | if (page_memcg(page) == mc.from) { |
02491447 | 5861 | ret = MC_TARGET_PAGE; |
f25cbb7a AS |
5862 | if (is_device_private_page(page) || |
5863 | is_device_coherent_page(page)) | |
c733a828 | 5864 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5865 | if (target) |
5866 | target->page = page; | |
5867 | } | |
5868 | if (!ret || !target) | |
5869 | put_page(page); | |
5870 | } | |
3e14a57b HY |
5871 | /* |
5872 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5873 | * But we cannot move a tail-page in a THP. | |
5874 | */ | |
5875 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5876 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5877 | ret = MC_TARGET_SWAP; |
5878 | if (target) | |
5879 | target->ent = ent; | |
4ffef5fe | 5880 | } |
4ffef5fe DN |
5881 | return ret; |
5882 | } | |
5883 | ||
12724850 NH |
5884 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5885 | /* | |
d6810d73 HY |
5886 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5887 | * not support them for now. | |
12724850 NH |
5888 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5889 | */ | |
5890 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5891 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5892 | { | |
5893 | struct page *page = NULL; | |
12724850 NH |
5894 | enum mc_target_type ret = MC_TARGET_NONE; |
5895 | ||
84c3fc4e ZY |
5896 | if (unlikely(is_swap_pmd(pmd))) { |
5897 | VM_BUG_ON(thp_migration_supported() && | |
5898 | !is_pmd_migration_entry(pmd)); | |
5899 | return ret; | |
5900 | } | |
12724850 | 5901 | page = pmd_page(pmd); |
309381fe | 5902 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5903 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5904 | return ret; |
bcfe06bf | 5905 | if (page_memcg(page) == mc.from) { |
12724850 NH |
5906 | ret = MC_TARGET_PAGE; |
5907 | if (target) { | |
5908 | get_page(page); | |
5909 | target->page = page; | |
5910 | } | |
5911 | } | |
5912 | return ret; | |
5913 | } | |
5914 | #else | |
5915 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5916 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5917 | { | |
5918 | return MC_TARGET_NONE; | |
5919 | } | |
5920 | #endif | |
5921 | ||
4ffef5fe DN |
5922 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5923 | unsigned long addr, unsigned long end, | |
5924 | struct mm_walk *walk) | |
5925 | { | |
26bcd64a | 5926 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5927 | pte_t *pte; |
5928 | spinlock_t *ptl; | |
5929 | ||
b6ec57f4 KS |
5930 | ptl = pmd_trans_huge_lock(pmd, vma); |
5931 | if (ptl) { | |
c733a828 JG |
5932 | /* |
5933 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5934 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5935 | * this might change. | |
c733a828 | 5936 | */ |
12724850 NH |
5937 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5938 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5939 | spin_unlock(ptl); |
1a5a9906 | 5940 | return 0; |
12724850 | 5941 | } |
03319327 | 5942 | |
45f83cef AA |
5943 | if (pmd_trans_unstable(pmd)) |
5944 | return 0; | |
4ffef5fe DN |
5945 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5946 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5947 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5948 | mc.precharge++; /* increment precharge temporarily */ |
5949 | pte_unmap_unlock(pte - 1, ptl); | |
5950 | cond_resched(); | |
5951 | ||
7dc74be0 DN |
5952 | return 0; |
5953 | } | |
5954 | ||
7b86ac33 CH |
5955 | static const struct mm_walk_ops precharge_walk_ops = { |
5956 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5957 | }; | |
5958 | ||
4ffef5fe DN |
5959 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5960 | { | |
5961 | unsigned long precharge; | |
4ffef5fe | 5962 | |
d8ed45c5 | 5963 | mmap_read_lock(mm); |
ba0aff8e | 5964 | walk_page_range(mm, 0, ULONG_MAX, &precharge_walk_ops, NULL); |
d8ed45c5 | 5965 | mmap_read_unlock(mm); |
4ffef5fe DN |
5966 | |
5967 | precharge = mc.precharge; | |
5968 | mc.precharge = 0; | |
5969 | ||
5970 | return precharge; | |
5971 | } | |
5972 | ||
4ffef5fe DN |
5973 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5974 | { | |
dfe076b0 DN |
5975 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5976 | ||
5977 | VM_BUG_ON(mc.moving_task); | |
5978 | mc.moving_task = current; | |
5979 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5980 | } |
5981 | ||
dfe076b0 DN |
5982 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5983 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5984 | { |
2bd9bb20 KH |
5985 | struct mem_cgroup *from = mc.from; |
5986 | struct mem_cgroup *to = mc.to; | |
5987 | ||
4ffef5fe | 5988 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5989 | if (mc.precharge) { |
00501b53 | 5990 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5991 | mc.precharge = 0; |
5992 | } | |
5993 | /* | |
5994 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5995 | * we must uncharge here. | |
5996 | */ | |
5997 | if (mc.moved_charge) { | |
00501b53 | 5998 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5999 | mc.moved_charge = 0; |
4ffef5fe | 6000 | } |
483c30b5 DN |
6001 | /* we must fixup refcnts and charges */ |
6002 | if (mc.moved_swap) { | |
483c30b5 | 6003 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 6004 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 6005 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 6006 | |
615d66c3 VD |
6007 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
6008 | ||
05b84301 | 6009 | /* |
3e32cb2e JW |
6010 | * we charged both to->memory and to->memsw, so we |
6011 | * should uncharge to->memory. | |
05b84301 | 6012 | */ |
ce00a967 | 6013 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
6014 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
6015 | ||
483c30b5 DN |
6016 | mc.moved_swap = 0; |
6017 | } | |
dfe076b0 DN |
6018 | memcg_oom_recover(from); |
6019 | memcg_oom_recover(to); | |
6020 | wake_up_all(&mc.waitq); | |
6021 | } | |
6022 | ||
6023 | static void mem_cgroup_clear_mc(void) | |
6024 | { | |
264a0ae1 TH |
6025 | struct mm_struct *mm = mc.mm; |
6026 | ||
dfe076b0 DN |
6027 | /* |
6028 | * we must clear moving_task before waking up waiters at the end of | |
6029 | * task migration. | |
6030 | */ | |
6031 | mc.moving_task = NULL; | |
6032 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6033 | spin_lock(&mc.lock); |
4ffef5fe DN |
6034 | mc.from = NULL; |
6035 | mc.to = NULL; | |
264a0ae1 | 6036 | mc.mm = NULL; |
2bd9bb20 | 6037 | spin_unlock(&mc.lock); |
264a0ae1 TH |
6038 | |
6039 | mmput(mm); | |
4ffef5fe DN |
6040 | } |
6041 | ||
1f7dd3e5 | 6042 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6043 | { |
1f7dd3e5 | 6044 | struct cgroup_subsys_state *css; |
eed67d75 | 6045 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 6046 | struct mem_cgroup *from; |
4530eddb | 6047 | struct task_struct *leader, *p; |
9f2115f9 | 6048 | struct mm_struct *mm; |
1dfab5ab | 6049 | unsigned long move_flags; |
9f2115f9 | 6050 | int ret = 0; |
7dc74be0 | 6051 | |
1f7dd3e5 TH |
6052 | /* charge immigration isn't supported on the default hierarchy */ |
6053 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
6054 | return 0; |
6055 | ||
4530eddb TH |
6056 | /* |
6057 | * Multi-process migrations only happen on the default hierarchy | |
6058 | * where charge immigration is not used. Perform charge | |
6059 | * immigration if @tset contains a leader and whine if there are | |
6060 | * multiple. | |
6061 | */ | |
6062 | p = NULL; | |
1f7dd3e5 | 6063 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
6064 | WARN_ON_ONCE(p); |
6065 | p = leader; | |
1f7dd3e5 | 6066 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
6067 | } |
6068 | if (!p) | |
6069 | return 0; | |
6070 | ||
1f7dd3e5 | 6071 | /* |
f0953a1b | 6072 | * We are now committed to this value whatever it is. Changes in this |
1f7dd3e5 TH |
6073 | * tunable will only affect upcoming migrations, not the current one. |
6074 | * So we need to save it, and keep it going. | |
6075 | */ | |
6076 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
6077 | if (!move_flags) | |
6078 | return 0; | |
6079 | ||
9f2115f9 TH |
6080 | from = mem_cgroup_from_task(p); |
6081 | ||
6082 | VM_BUG_ON(from == memcg); | |
6083 | ||
6084 | mm = get_task_mm(p); | |
6085 | if (!mm) | |
6086 | return 0; | |
6087 | /* We move charges only when we move a owner of the mm */ | |
6088 | if (mm->owner == p) { | |
6089 | VM_BUG_ON(mc.from); | |
6090 | VM_BUG_ON(mc.to); | |
6091 | VM_BUG_ON(mc.precharge); | |
6092 | VM_BUG_ON(mc.moved_charge); | |
6093 | VM_BUG_ON(mc.moved_swap); | |
6094 | ||
6095 | spin_lock(&mc.lock); | |
264a0ae1 | 6096 | mc.mm = mm; |
9f2115f9 TH |
6097 | mc.from = from; |
6098 | mc.to = memcg; | |
6099 | mc.flags = move_flags; | |
6100 | spin_unlock(&mc.lock); | |
6101 | /* We set mc.moving_task later */ | |
6102 | ||
6103 | ret = mem_cgroup_precharge_mc(mm); | |
6104 | if (ret) | |
6105 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6106 | } else { |
6107 | mmput(mm); | |
7dc74be0 DN |
6108 | } |
6109 | return ret; | |
6110 | } | |
6111 | ||
1f7dd3e5 | 6112 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6113 | { |
4e2f245d JW |
6114 | if (mc.to) |
6115 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6116 | } |
6117 | ||
4ffef5fe DN |
6118 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6119 | unsigned long addr, unsigned long end, | |
6120 | struct mm_walk *walk) | |
7dc74be0 | 6121 | { |
4ffef5fe | 6122 | int ret = 0; |
26bcd64a | 6123 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6124 | pte_t *pte; |
6125 | spinlock_t *ptl; | |
12724850 NH |
6126 | enum mc_target_type target_type; |
6127 | union mc_target target; | |
6128 | struct page *page; | |
4ffef5fe | 6129 | |
b6ec57f4 KS |
6130 | ptl = pmd_trans_huge_lock(pmd, vma); |
6131 | if (ptl) { | |
62ade86a | 6132 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6133 | spin_unlock(ptl); |
12724850 NH |
6134 | return 0; |
6135 | } | |
6136 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6137 | if (target_type == MC_TARGET_PAGE) { | |
6138 | page = target.page; | |
6139 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6140 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6141 | mc.from, mc.to)) { |
12724850 NH |
6142 | mc.precharge -= HPAGE_PMD_NR; |
6143 | mc.moved_charge += HPAGE_PMD_NR; | |
6144 | } | |
6145 | putback_lru_page(page); | |
6146 | } | |
6147 | put_page(page); | |
c733a828 JG |
6148 | } else if (target_type == MC_TARGET_DEVICE) { |
6149 | page = target.page; | |
6150 | if (!mem_cgroup_move_account(page, true, | |
6151 | mc.from, mc.to)) { | |
6152 | mc.precharge -= HPAGE_PMD_NR; | |
6153 | mc.moved_charge += HPAGE_PMD_NR; | |
6154 | } | |
6155 | put_page(page); | |
12724850 | 6156 | } |
bf929152 | 6157 | spin_unlock(ptl); |
1a5a9906 | 6158 | return 0; |
12724850 NH |
6159 | } |
6160 | ||
45f83cef AA |
6161 | if (pmd_trans_unstable(pmd)) |
6162 | return 0; | |
4ffef5fe DN |
6163 | retry: |
6164 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6165 | for (; addr != end; addr += PAGE_SIZE) { | |
6166 | pte_t ptent = *(pte++); | |
c733a828 | 6167 | bool device = false; |
02491447 | 6168 | swp_entry_t ent; |
4ffef5fe DN |
6169 | |
6170 | if (!mc.precharge) | |
6171 | break; | |
6172 | ||
8d32ff84 | 6173 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6174 | case MC_TARGET_DEVICE: |
6175 | device = true; | |
e4a9bc58 | 6176 | fallthrough; |
4ffef5fe DN |
6177 | case MC_TARGET_PAGE: |
6178 | page = target.page; | |
53f9263b KS |
6179 | /* |
6180 | * We can have a part of the split pmd here. Moving it | |
6181 | * can be done but it would be too convoluted so simply | |
6182 | * ignore such a partial THP and keep it in original | |
6183 | * memcg. There should be somebody mapping the head. | |
6184 | */ | |
6185 | if (PageTransCompound(page)) | |
6186 | goto put; | |
c733a828 | 6187 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6188 | goto put; |
f627c2f5 KS |
6189 | if (!mem_cgroup_move_account(page, false, |
6190 | mc.from, mc.to)) { | |
4ffef5fe | 6191 | mc.precharge--; |
854ffa8d DN |
6192 | /* we uncharge from mc.from later. */ |
6193 | mc.moved_charge++; | |
4ffef5fe | 6194 | } |
c733a828 JG |
6195 | if (!device) |
6196 | putback_lru_page(page); | |
8d32ff84 | 6197 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6198 | put_page(page); |
6199 | break; | |
02491447 DN |
6200 | case MC_TARGET_SWAP: |
6201 | ent = target.ent; | |
e91cbb42 | 6202 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6203 | mc.precharge--; |
8d22a935 HD |
6204 | mem_cgroup_id_get_many(mc.to, 1); |
6205 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6206 | mc.moved_swap++; |
6207 | } | |
02491447 | 6208 | break; |
4ffef5fe DN |
6209 | default: |
6210 | break; | |
6211 | } | |
6212 | } | |
6213 | pte_unmap_unlock(pte - 1, ptl); | |
6214 | cond_resched(); | |
6215 | ||
6216 | if (addr != end) { | |
6217 | /* | |
6218 | * We have consumed all precharges we got in can_attach(). | |
6219 | * We try charge one by one, but don't do any additional | |
6220 | * charges to mc.to if we have failed in charge once in attach() | |
6221 | * phase. | |
6222 | */ | |
854ffa8d | 6223 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6224 | if (!ret) |
6225 | goto retry; | |
6226 | } | |
6227 | ||
6228 | return ret; | |
6229 | } | |
6230 | ||
7b86ac33 CH |
6231 | static const struct mm_walk_ops charge_walk_ops = { |
6232 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6233 | }; | |
6234 | ||
264a0ae1 | 6235 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6236 | { |
4ffef5fe | 6237 | lru_add_drain_all(); |
312722cb | 6238 | /* |
81f8c3a4 JW |
6239 | * Signal lock_page_memcg() to take the memcg's move_lock |
6240 | * while we're moving its pages to another memcg. Then wait | |
6241 | * for already started RCU-only updates to finish. | |
312722cb JW |
6242 | */ |
6243 | atomic_inc(&mc.from->moving_account); | |
6244 | synchronize_rcu(); | |
dfe076b0 | 6245 | retry: |
d8ed45c5 | 6246 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6247 | /* |
c1e8d7c6 | 6248 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6249 | * waitq. So we cancel all extra charges, wake up all waiters, |
6250 | * and retry. Because we cancel precharges, we might not be able | |
6251 | * to move enough charges, but moving charge is a best-effort | |
6252 | * feature anyway, so it wouldn't be a big problem. | |
6253 | */ | |
6254 | __mem_cgroup_clear_mc(); | |
6255 | cond_resched(); | |
6256 | goto retry; | |
6257 | } | |
26bcd64a NH |
6258 | /* |
6259 | * When we have consumed all precharges and failed in doing | |
6260 | * additional charge, the page walk just aborts. | |
6261 | */ | |
ba0aff8e | 6262 | walk_page_range(mc.mm, 0, ULONG_MAX, &charge_walk_ops, NULL); |
d8ed45c5 | 6263 | mmap_read_unlock(mc.mm); |
312722cb | 6264 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6265 | } |
6266 | ||
264a0ae1 | 6267 | static void mem_cgroup_move_task(void) |
67e465a7 | 6268 | { |
264a0ae1 TH |
6269 | if (mc.to) { |
6270 | mem_cgroup_move_charge(); | |
a433658c | 6271 | mem_cgroup_clear_mc(); |
264a0ae1 | 6272 | } |
67e465a7 | 6273 | } |
5cfb80a7 | 6274 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6275 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6276 | { |
6277 | return 0; | |
6278 | } | |
1f7dd3e5 | 6279 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6280 | { |
6281 | } | |
264a0ae1 | 6282 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6283 | { |
6284 | } | |
6285 | #endif | |
67e465a7 | 6286 | |
bd74fdae YZ |
6287 | #ifdef CONFIG_LRU_GEN |
6288 | static void mem_cgroup_attach(struct cgroup_taskset *tset) | |
6289 | { | |
6290 | struct task_struct *task; | |
6291 | struct cgroup_subsys_state *css; | |
6292 | ||
6293 | /* find the first leader if there is any */ | |
6294 | cgroup_taskset_for_each_leader(task, css, tset) | |
6295 | break; | |
6296 | ||
6297 | if (!task) | |
6298 | return; | |
6299 | ||
6300 | task_lock(task); | |
6301 | if (task->mm && READ_ONCE(task->mm->owner) == task) | |
6302 | lru_gen_migrate_mm(task->mm); | |
6303 | task_unlock(task); | |
6304 | } | |
6305 | #else | |
6306 | static void mem_cgroup_attach(struct cgroup_taskset *tset) | |
6307 | { | |
6308 | } | |
6309 | #endif /* CONFIG_LRU_GEN */ | |
6310 | ||
677dc973 CD |
6311 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6312 | { | |
6313 | if (value == PAGE_COUNTER_MAX) | |
6314 | seq_puts(m, "max\n"); | |
6315 | else | |
6316 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6317 | ||
6318 | return 0; | |
6319 | } | |
6320 | ||
241994ed JW |
6321 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6322 | struct cftype *cft) | |
6323 | { | |
f5fc3c5d JW |
6324 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6325 | ||
6326 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6327 | } |
6328 | ||
8e20d4b3 GR |
6329 | static u64 memory_peak_read(struct cgroup_subsys_state *css, |
6330 | struct cftype *cft) | |
6331 | { | |
6332 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6333 | ||
6334 | return (u64)memcg->memory.watermark * PAGE_SIZE; | |
6335 | } | |
6336 | ||
bf8d5d52 RG |
6337 | static int memory_min_show(struct seq_file *m, void *v) |
6338 | { | |
677dc973 CD |
6339 | return seq_puts_memcg_tunable(m, |
6340 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6341 | } |
6342 | ||
6343 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6344 | char *buf, size_t nbytes, loff_t off) | |
6345 | { | |
6346 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6347 | unsigned long min; | |
6348 | int err; | |
6349 | ||
6350 | buf = strstrip(buf); | |
6351 | err = page_counter_memparse(buf, "max", &min); | |
6352 | if (err) | |
6353 | return err; | |
6354 | ||
6355 | page_counter_set_min(&memcg->memory, min); | |
6356 | ||
6357 | return nbytes; | |
6358 | } | |
6359 | ||
241994ed JW |
6360 | static int memory_low_show(struct seq_file *m, void *v) |
6361 | { | |
677dc973 CD |
6362 | return seq_puts_memcg_tunable(m, |
6363 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6364 | } |
6365 | ||
6366 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6367 | char *buf, size_t nbytes, loff_t off) | |
6368 | { | |
6369 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6370 | unsigned long low; | |
6371 | int err; | |
6372 | ||
6373 | buf = strstrip(buf); | |
d2973697 | 6374 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6375 | if (err) |
6376 | return err; | |
6377 | ||
23067153 | 6378 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6379 | |
6380 | return nbytes; | |
6381 | } | |
6382 | ||
6383 | static int memory_high_show(struct seq_file *m, void *v) | |
6384 | { | |
d1663a90 JK |
6385 | return seq_puts_memcg_tunable(m, |
6386 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6387 | } |
6388 | ||
6389 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6390 | char *buf, size_t nbytes, loff_t off) | |
6391 | { | |
6392 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6393 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6394 | bool drained = false; |
241994ed JW |
6395 | unsigned long high; |
6396 | int err; | |
6397 | ||
6398 | buf = strstrip(buf); | |
d2973697 | 6399 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6400 | if (err) |
6401 | return err; | |
6402 | ||
e82553c1 JW |
6403 | page_counter_set_high(&memcg->memory, high); |
6404 | ||
8c8c383c JW |
6405 | for (;;) { |
6406 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6407 | unsigned long reclaimed; | |
6408 | ||
6409 | if (nr_pages <= high) | |
6410 | break; | |
6411 | ||
6412 | if (signal_pending(current)) | |
6413 | break; | |
6414 | ||
6415 | if (!drained) { | |
6416 | drain_all_stock(memcg); | |
6417 | drained = true; | |
6418 | continue; | |
6419 | } | |
6420 | ||
6421 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
73b73bac | 6422 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP); |
8c8c383c JW |
6423 | |
6424 | if (!reclaimed && !nr_retries--) | |
6425 | break; | |
6426 | } | |
588083bb | 6427 | |
19ce33ac | 6428 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6429 | return nbytes; |
6430 | } | |
6431 | ||
6432 | static int memory_max_show(struct seq_file *m, void *v) | |
6433 | { | |
677dc973 CD |
6434 | return seq_puts_memcg_tunable(m, |
6435 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6436 | } |
6437 | ||
6438 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6439 | char *buf, size_t nbytes, loff_t off) | |
6440 | { | |
6441 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6442 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6443 | bool drained = false; |
241994ed JW |
6444 | unsigned long max; |
6445 | int err; | |
6446 | ||
6447 | buf = strstrip(buf); | |
d2973697 | 6448 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6449 | if (err) |
6450 | return err; | |
6451 | ||
bbec2e15 | 6452 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6453 | |
6454 | for (;;) { | |
6455 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6456 | ||
6457 | if (nr_pages <= max) | |
6458 | break; | |
6459 | ||
7249c9f0 | 6460 | if (signal_pending(current)) |
b6e6edcf | 6461 | break; |
b6e6edcf JW |
6462 | |
6463 | if (!drained) { | |
6464 | drain_all_stock(memcg); | |
6465 | drained = true; | |
6466 | continue; | |
6467 | } | |
6468 | ||
6469 | if (nr_reclaims) { | |
6470 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
73b73bac | 6471 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP)) |
b6e6edcf JW |
6472 | nr_reclaims--; |
6473 | continue; | |
6474 | } | |
6475 | ||
e27be240 | 6476 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6477 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6478 | break; | |
6479 | } | |
241994ed | 6480 | |
2529bb3a | 6481 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6482 | return nbytes; |
6483 | } | |
6484 | ||
1e577f97 SB |
6485 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6486 | { | |
6487 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6488 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6489 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6490 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6491 | seq_printf(m, "oom_kill %lu\n", | |
6492 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
b6bf9abb DS |
6493 | seq_printf(m, "oom_group_kill %lu\n", |
6494 | atomic_long_read(&events[MEMCG_OOM_GROUP_KILL])); | |
1e577f97 SB |
6495 | } |
6496 | ||
241994ed JW |
6497 | static int memory_events_show(struct seq_file *m, void *v) |
6498 | { | |
aa9694bb | 6499 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6500 | |
1e577f97 SB |
6501 | __memory_events_show(m, memcg->memory_events); |
6502 | return 0; | |
6503 | } | |
6504 | ||
6505 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6506 | { | |
6507 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6508 | |
1e577f97 | 6509 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6510 | return 0; |
6511 | } | |
6512 | ||
587d9f72 JW |
6513 | static int memory_stat_show(struct seq_file *m, void *v) |
6514 | { | |
aa9694bb | 6515 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
68aaee14 | 6516 | char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
1ff9e6e1 | 6517 | |
c8713d0b JW |
6518 | if (!buf) |
6519 | return -ENOMEM; | |
68aaee14 | 6520 | memory_stat_format(memcg, buf, PAGE_SIZE); |
c8713d0b JW |
6521 | seq_puts(m, buf); |
6522 | kfree(buf); | |
587d9f72 JW |
6523 | return 0; |
6524 | } | |
6525 | ||
5f9a4f4a | 6526 | #ifdef CONFIG_NUMA |
fff66b79 MS |
6527 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
6528 | int item) | |
6529 | { | |
6530 | return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item); | |
6531 | } | |
6532 | ||
5f9a4f4a MS |
6533 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
6534 | { | |
6535 | int i; | |
6536 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6537 | ||
fd25a9e0 | 6538 | mem_cgroup_flush_stats(); |
7e1c0d6f | 6539 | |
5f9a4f4a MS |
6540 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
6541 | int nid; | |
6542 | ||
6543 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6544 | continue; | |
6545 | ||
6546 | seq_printf(m, "%s", memory_stats[i].name); | |
6547 | for_each_node_state(nid, N_MEMORY) { | |
6548 | u64 size; | |
6549 | struct lruvec *lruvec; | |
6550 | ||
6551 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
6552 | size = lruvec_page_state_output(lruvec, |
6553 | memory_stats[i].idx); | |
5f9a4f4a MS |
6554 | seq_printf(m, " N%d=%llu", nid, size); |
6555 | } | |
6556 | seq_putc(m, '\n'); | |
6557 | } | |
6558 | ||
6559 | return 0; | |
6560 | } | |
6561 | #endif | |
6562 | ||
3d8b38eb RG |
6563 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6564 | { | |
aa9694bb | 6565 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6566 | |
6567 | seq_printf(m, "%d\n", memcg->oom_group); | |
6568 | ||
6569 | return 0; | |
6570 | } | |
6571 | ||
6572 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6573 | char *buf, size_t nbytes, loff_t off) | |
6574 | { | |
6575 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6576 | int ret, oom_group; | |
6577 | ||
6578 | buf = strstrip(buf); | |
6579 | if (!buf) | |
6580 | return -EINVAL; | |
6581 | ||
6582 | ret = kstrtoint(buf, 0, &oom_group); | |
6583 | if (ret) | |
6584 | return ret; | |
6585 | ||
6586 | if (oom_group != 0 && oom_group != 1) | |
6587 | return -EINVAL; | |
6588 | ||
6589 | memcg->oom_group = oom_group; | |
6590 | ||
6591 | return nbytes; | |
6592 | } | |
6593 | ||
94968384 SB |
6594 | static ssize_t memory_reclaim(struct kernfs_open_file *of, char *buf, |
6595 | size_t nbytes, loff_t off) | |
6596 | { | |
6597 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6598 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; | |
6599 | unsigned long nr_to_reclaim, nr_reclaimed = 0; | |
73b73bac | 6600 | unsigned int reclaim_options; |
94968384 SB |
6601 | int err; |
6602 | ||
6603 | buf = strstrip(buf); | |
6604 | err = page_counter_memparse(buf, "", &nr_to_reclaim); | |
6605 | if (err) | |
6606 | return err; | |
6607 | ||
73b73bac | 6608 | reclaim_options = MEMCG_RECLAIM_MAY_SWAP | MEMCG_RECLAIM_PROACTIVE; |
94968384 SB |
6609 | while (nr_reclaimed < nr_to_reclaim) { |
6610 | unsigned long reclaimed; | |
6611 | ||
6612 | if (signal_pending(current)) | |
6613 | return -EINTR; | |
6614 | ||
6615 | /* | |
6616 | * This is the final attempt, drain percpu lru caches in the | |
6617 | * hope of introducing more evictable pages for | |
6618 | * try_to_free_mem_cgroup_pages(). | |
6619 | */ | |
6620 | if (!nr_retries) | |
6621 | lru_add_drain_all(); | |
6622 | ||
6623 | reclaimed = try_to_free_mem_cgroup_pages(memcg, | |
6624 | nr_to_reclaim - nr_reclaimed, | |
73b73bac | 6625 | GFP_KERNEL, reclaim_options); |
94968384 SB |
6626 | |
6627 | if (!reclaimed && !nr_retries--) | |
6628 | return -EAGAIN; | |
6629 | ||
6630 | nr_reclaimed += reclaimed; | |
6631 | } | |
6632 | ||
6633 | return nbytes; | |
6634 | } | |
6635 | ||
241994ed JW |
6636 | static struct cftype memory_files[] = { |
6637 | { | |
6638 | .name = "current", | |
f5fc3c5d | 6639 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6640 | .read_u64 = memory_current_read, |
6641 | }, | |
8e20d4b3 GR |
6642 | { |
6643 | .name = "peak", | |
6644 | .flags = CFTYPE_NOT_ON_ROOT, | |
6645 | .read_u64 = memory_peak_read, | |
6646 | }, | |
bf8d5d52 RG |
6647 | { |
6648 | .name = "min", | |
6649 | .flags = CFTYPE_NOT_ON_ROOT, | |
6650 | .seq_show = memory_min_show, | |
6651 | .write = memory_min_write, | |
6652 | }, | |
241994ed JW |
6653 | { |
6654 | .name = "low", | |
6655 | .flags = CFTYPE_NOT_ON_ROOT, | |
6656 | .seq_show = memory_low_show, | |
6657 | .write = memory_low_write, | |
6658 | }, | |
6659 | { | |
6660 | .name = "high", | |
6661 | .flags = CFTYPE_NOT_ON_ROOT, | |
6662 | .seq_show = memory_high_show, | |
6663 | .write = memory_high_write, | |
6664 | }, | |
6665 | { | |
6666 | .name = "max", | |
6667 | .flags = CFTYPE_NOT_ON_ROOT, | |
6668 | .seq_show = memory_max_show, | |
6669 | .write = memory_max_write, | |
6670 | }, | |
6671 | { | |
6672 | .name = "events", | |
6673 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6674 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6675 | .seq_show = memory_events_show, |
6676 | }, | |
1e577f97 SB |
6677 | { |
6678 | .name = "events.local", | |
6679 | .flags = CFTYPE_NOT_ON_ROOT, | |
6680 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6681 | .seq_show = memory_events_local_show, | |
6682 | }, | |
587d9f72 JW |
6683 | { |
6684 | .name = "stat", | |
587d9f72 JW |
6685 | .seq_show = memory_stat_show, |
6686 | }, | |
5f9a4f4a MS |
6687 | #ifdef CONFIG_NUMA |
6688 | { | |
6689 | .name = "numa_stat", | |
6690 | .seq_show = memory_numa_stat_show, | |
6691 | }, | |
6692 | #endif | |
3d8b38eb RG |
6693 | { |
6694 | .name = "oom.group", | |
6695 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6696 | .seq_show = memory_oom_group_show, | |
6697 | .write = memory_oom_group_write, | |
6698 | }, | |
94968384 SB |
6699 | { |
6700 | .name = "reclaim", | |
6701 | .flags = CFTYPE_NS_DELEGATABLE, | |
6702 | .write = memory_reclaim, | |
6703 | }, | |
241994ed JW |
6704 | { } /* terminate */ |
6705 | }; | |
6706 | ||
073219e9 | 6707 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6708 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6709 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6710 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6711 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6712 | .css_free = mem_cgroup_css_free, |
1ced953b | 6713 | .css_reset = mem_cgroup_css_reset, |
2d146aa3 | 6714 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
7dc74be0 | 6715 | .can_attach = mem_cgroup_can_attach, |
bd74fdae | 6716 | .attach = mem_cgroup_attach, |
7dc74be0 | 6717 | .cancel_attach = mem_cgroup_cancel_attach, |
264a0ae1 | 6718 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6719 | .dfl_cftypes = memory_files, |
6720 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6721 | .early_init = 0, |
8cdea7c0 | 6722 | }; |
c077719b | 6723 | |
bc50bcc6 JW |
6724 | /* |
6725 | * This function calculates an individual cgroup's effective | |
6726 | * protection which is derived from its own memory.min/low, its | |
6727 | * parent's and siblings' settings, as well as the actual memory | |
6728 | * distribution in the tree. | |
6729 | * | |
6730 | * The following rules apply to the effective protection values: | |
6731 | * | |
6732 | * 1. At the first level of reclaim, effective protection is equal to | |
6733 | * the declared protection in memory.min and memory.low. | |
6734 | * | |
6735 | * 2. To enable safe delegation of the protection configuration, at | |
6736 | * subsequent levels the effective protection is capped to the | |
6737 | * parent's effective protection. | |
6738 | * | |
6739 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6740 | * user is allowed to overcommit the declared protection at a given | |
6741 | * level. If that is the case, the parent's effective protection is | |
6742 | * distributed to the children in proportion to how much protection | |
6743 | * they have declared and how much of it they are utilizing. | |
6744 | * | |
6745 | * This makes distribution proportional, but also work-conserving: | |
6746 | * if one cgroup claims much more protection than it uses memory, | |
6747 | * the unused remainder is available to its siblings. | |
6748 | * | |
6749 | * 4. Conversely, when the declared protection is undercommitted at a | |
6750 | * given level, the distribution of the larger parental protection | |
6751 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6752 | * is capped to its own memory.min/low setting. | |
6753 | * | |
8a931f80 JW |
6754 | * 5. However, to allow protecting recursive subtrees from each other |
6755 | * without having to declare each individual cgroup's fixed share | |
6756 | * of the ancestor's claim to protection, any unutilized - | |
6757 | * "floating" - protection from up the tree is distributed in | |
6758 | * proportion to each cgroup's *usage*. This makes the protection | |
6759 | * neutral wrt sibling cgroups and lets them compete freely over | |
6760 | * the shared parental protection budget, but it protects the | |
6761 | * subtree as a whole from neighboring subtrees. | |
6762 | * | |
6763 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6764 | * against immediate siblings whereas 5. is about protecting against | |
6765 | * neighboring subtrees. | |
bc50bcc6 JW |
6766 | */ |
6767 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6768 | unsigned long parent_usage, |
bc50bcc6 JW |
6769 | unsigned long setting, |
6770 | unsigned long parent_effective, | |
6771 | unsigned long siblings_protected) | |
6772 | { | |
6773 | unsigned long protected; | |
8a931f80 | 6774 | unsigned long ep; |
bc50bcc6 JW |
6775 | |
6776 | protected = min(usage, setting); | |
6777 | /* | |
6778 | * If all cgroups at this level combined claim and use more | |
6779 | * protection then what the parent affords them, distribute | |
6780 | * shares in proportion to utilization. | |
6781 | * | |
6782 | * We are using actual utilization rather than the statically | |
6783 | * claimed protection in order to be work-conserving: claimed | |
6784 | * but unused protection is available to siblings that would | |
6785 | * otherwise get a smaller chunk than what they claimed. | |
6786 | */ | |
6787 | if (siblings_protected > parent_effective) | |
6788 | return protected * parent_effective / siblings_protected; | |
6789 | ||
6790 | /* | |
6791 | * Ok, utilized protection of all children is within what the | |
6792 | * parent affords them, so we know whatever this child claims | |
6793 | * and utilizes is effectively protected. | |
6794 | * | |
6795 | * If there is unprotected usage beyond this value, reclaim | |
6796 | * will apply pressure in proportion to that amount. | |
6797 | * | |
6798 | * If there is unutilized protection, the cgroup will be fully | |
6799 | * shielded from reclaim, but we do return a smaller value for | |
6800 | * protection than what the group could enjoy in theory. This | |
6801 | * is okay. With the overcommit distribution above, effective | |
6802 | * protection is always dependent on how memory is actually | |
6803 | * consumed among the siblings anyway. | |
6804 | */ | |
8a931f80 JW |
6805 | ep = protected; |
6806 | ||
6807 | /* | |
6808 | * If the children aren't claiming (all of) the protection | |
6809 | * afforded to them by the parent, distribute the remainder in | |
6810 | * proportion to the (unprotected) memory of each cgroup. That | |
6811 | * way, cgroups that aren't explicitly prioritized wrt each | |
6812 | * other compete freely over the allowance, but they are | |
6813 | * collectively protected from neighboring trees. | |
6814 | * | |
6815 | * We're using unprotected memory for the weight so that if | |
6816 | * some cgroups DO claim explicit protection, we don't protect | |
6817 | * the same bytes twice. | |
cd324edc JW |
6818 | * |
6819 | * Check both usage and parent_usage against the respective | |
6820 | * protected values. One should imply the other, but they | |
6821 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6822 | */ |
6823 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6824 | return ep; | |
cd324edc JW |
6825 | if (parent_effective > siblings_protected && |
6826 | parent_usage > siblings_protected && | |
6827 | usage > protected) { | |
8a931f80 JW |
6828 | unsigned long unclaimed; |
6829 | ||
6830 | unclaimed = parent_effective - siblings_protected; | |
6831 | unclaimed *= usage - protected; | |
6832 | unclaimed /= parent_usage - siblings_protected; | |
6833 | ||
6834 | ep += unclaimed; | |
6835 | } | |
6836 | ||
6837 | return ep; | |
bc50bcc6 JW |
6838 | } |
6839 | ||
241994ed | 6840 | /** |
05395718 | 6841 | * mem_cgroup_calculate_protection - check if memory consumption is in the normal range |
34c81057 | 6842 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6843 | * @memcg: the memory cgroup to check |
6844 | * | |
23067153 RG |
6845 | * WARNING: This function is not stateless! It can only be used as part |
6846 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6847 | */ |
45c7f7e1 CD |
6848 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6849 | struct mem_cgroup *memcg) | |
241994ed | 6850 | { |
8a931f80 | 6851 | unsigned long usage, parent_usage; |
23067153 RG |
6852 | struct mem_cgroup *parent; |
6853 | ||
241994ed | 6854 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6855 | return; |
241994ed | 6856 | |
34c81057 SC |
6857 | if (!root) |
6858 | root = root_mem_cgroup; | |
22f7496f YS |
6859 | |
6860 | /* | |
6861 | * Effective values of the reclaim targets are ignored so they | |
6862 | * can be stale. Have a look at mem_cgroup_protection for more | |
6863 | * details. | |
6864 | * TODO: calculation should be more robust so that we do not need | |
6865 | * that special casing. | |
6866 | */ | |
34c81057 | 6867 | if (memcg == root) |
45c7f7e1 | 6868 | return; |
241994ed | 6869 | |
23067153 | 6870 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6871 | if (!usage) |
45c7f7e1 | 6872 | return; |
bf8d5d52 | 6873 | |
bf8d5d52 | 6874 | parent = parent_mem_cgroup(memcg); |
df2a4196 | 6875 | |
bc50bcc6 | 6876 | if (parent == root) { |
c3d53200 | 6877 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6878 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6879 | return; |
bf8d5d52 RG |
6880 | } |
6881 | ||
8a931f80 JW |
6882 | parent_usage = page_counter_read(&parent->memory); |
6883 | ||
b3a7822e | 6884 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6885 | READ_ONCE(memcg->memory.min), |
6886 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6887 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6888 | |
b3a7822e | 6889 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6890 | READ_ONCE(memcg->memory.low), |
6891 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6892 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6893 | } |
6894 | ||
8f425e4e MWO |
6895 | static int charge_memcg(struct folio *folio, struct mem_cgroup *memcg, |
6896 | gfp_t gfp) | |
0add0c77 | 6897 | { |
118f2875 | 6898 | long nr_pages = folio_nr_pages(folio); |
0add0c77 SB |
6899 | int ret; |
6900 | ||
6901 | ret = try_charge(memcg, gfp, nr_pages); | |
6902 | if (ret) | |
6903 | goto out; | |
6904 | ||
6905 | css_get(&memcg->css); | |
118f2875 | 6906 | commit_charge(folio, memcg); |
0add0c77 SB |
6907 | |
6908 | local_irq_disable(); | |
6e0110c2 | 6909 | mem_cgroup_charge_statistics(memcg, nr_pages); |
8f425e4e | 6910 | memcg_check_events(memcg, folio_nid(folio)); |
0add0c77 SB |
6911 | local_irq_enable(); |
6912 | out: | |
6913 | return ret; | |
6914 | } | |
6915 | ||
8f425e4e | 6916 | int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp) |
00501b53 | 6917 | { |
0add0c77 SB |
6918 | struct mem_cgroup *memcg; |
6919 | int ret; | |
00501b53 | 6920 | |
0add0c77 | 6921 | memcg = get_mem_cgroup_from_mm(mm); |
8f425e4e | 6922 | ret = charge_memcg(folio, memcg, gfp); |
0add0c77 | 6923 | css_put(&memcg->css); |
2d1c4980 | 6924 | |
0add0c77 SB |
6925 | return ret; |
6926 | } | |
e993d905 | 6927 | |
0add0c77 | 6928 | /** |
65995918 MWO |
6929 | * mem_cgroup_swapin_charge_folio - Charge a newly allocated folio for swapin. |
6930 | * @folio: folio to charge. | |
0add0c77 SB |
6931 | * @mm: mm context of the victim |
6932 | * @gfp: reclaim mode | |
65995918 | 6933 | * @entry: swap entry for which the folio is allocated |
0add0c77 | 6934 | * |
65995918 MWO |
6935 | * This function charges a folio allocated for swapin. Please call this before |
6936 | * adding the folio to the swapcache. | |
0add0c77 SB |
6937 | * |
6938 | * Returns 0 on success. Otherwise, an error code is returned. | |
6939 | */ | |
65995918 | 6940 | int mem_cgroup_swapin_charge_folio(struct folio *folio, struct mm_struct *mm, |
0add0c77 SB |
6941 | gfp_t gfp, swp_entry_t entry) |
6942 | { | |
6943 | struct mem_cgroup *memcg; | |
6944 | unsigned short id; | |
6945 | int ret; | |
00501b53 | 6946 | |
0add0c77 SB |
6947 | if (mem_cgroup_disabled()) |
6948 | return 0; | |
00501b53 | 6949 | |
0add0c77 SB |
6950 | id = lookup_swap_cgroup_id(entry); |
6951 | rcu_read_lock(); | |
6952 | memcg = mem_cgroup_from_id(id); | |
6953 | if (!memcg || !css_tryget_online(&memcg->css)) | |
6954 | memcg = get_mem_cgroup_from_mm(mm); | |
6955 | rcu_read_unlock(); | |
00501b53 | 6956 | |
8f425e4e | 6957 | ret = charge_memcg(folio, memcg, gfp); |
6abb5a86 | 6958 | |
0add0c77 SB |
6959 | css_put(&memcg->css); |
6960 | return ret; | |
6961 | } | |
00501b53 | 6962 | |
0add0c77 SB |
6963 | /* |
6964 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot | |
6965 | * @entry: swap entry for which the page is charged | |
6966 | * | |
6967 | * Call this function after successfully adding the charged page to swapcache. | |
6968 | * | |
6969 | * Note: This function assumes the page for which swap slot is being uncharged | |
6970 | * is order 0 page. | |
6971 | */ | |
6972 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) | |
6973 | { | |
cae3af62 MS |
6974 | /* |
6975 | * Cgroup1's unified memory+swap counter has been charged with the | |
6976 | * new swapcache page, finish the transfer by uncharging the swap | |
6977 | * slot. The swap slot would also get uncharged when it dies, but | |
6978 | * it can stick around indefinitely and we'd count the page twice | |
6979 | * the entire time. | |
6980 | * | |
6981 | * Cgroup2 has separate resource counters for memory and swap, | |
6982 | * so this is a non-issue here. Memory and swap charge lifetimes | |
6983 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
6984 | * page to memory here, and uncharge swap when the slot is freed. | |
6985 | */ | |
0add0c77 | 6986 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
00501b53 JW |
6987 | /* |
6988 | * The swap entry might not get freed for a long time, | |
6989 | * let's not wait for it. The page already received a | |
6990 | * memory+swap charge, drop the swap entry duplicate. | |
6991 | */ | |
0add0c77 | 6992 | mem_cgroup_uncharge_swap(entry, 1); |
00501b53 | 6993 | } |
3fea5a49 JW |
6994 | } |
6995 | ||
a9d5adee JG |
6996 | struct uncharge_gather { |
6997 | struct mem_cgroup *memcg; | |
b4e0b68f | 6998 | unsigned long nr_memory; |
a9d5adee | 6999 | unsigned long pgpgout; |
a9d5adee | 7000 | unsigned long nr_kmem; |
8e88bd2d | 7001 | int nid; |
a9d5adee JG |
7002 | }; |
7003 | ||
7004 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 7005 | { |
a9d5adee JG |
7006 | memset(ug, 0, sizeof(*ug)); |
7007 | } | |
7008 | ||
7009 | static void uncharge_batch(const struct uncharge_gather *ug) | |
7010 | { | |
747db954 JW |
7011 | unsigned long flags; |
7012 | ||
b4e0b68f MS |
7013 | if (ug->nr_memory) { |
7014 | page_counter_uncharge(&ug->memcg->memory, ug->nr_memory); | |
7941d214 | 7015 | if (do_memsw_account()) |
b4e0b68f | 7016 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory); |
a8c49af3 YA |
7017 | if (ug->nr_kmem) |
7018 | memcg_account_kmem(ug->memcg, -ug->nr_kmem); | |
a9d5adee | 7019 | memcg_oom_recover(ug->memcg); |
ce00a967 | 7020 | } |
747db954 JW |
7021 | |
7022 | local_irq_save(flags); | |
c9019e9b | 7023 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
b4e0b68f | 7024 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory); |
8e88bd2d | 7025 | memcg_check_events(ug->memcg, ug->nid); |
747db954 | 7026 | local_irq_restore(flags); |
f1796544 | 7027 | |
c4ed6ebf | 7028 | /* drop reference from uncharge_folio */ |
f1796544 | 7029 | css_put(&ug->memcg->css); |
a9d5adee JG |
7030 | } |
7031 | ||
c4ed6ebf | 7032 | static void uncharge_folio(struct folio *folio, struct uncharge_gather *ug) |
a9d5adee | 7033 | { |
c4ed6ebf | 7034 | long nr_pages; |
b4e0b68f MS |
7035 | struct mem_cgroup *memcg; |
7036 | struct obj_cgroup *objcg; | |
9f762dbe | 7037 | |
c4ed6ebf | 7038 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
a9d5adee | 7039 | |
a9d5adee JG |
7040 | /* |
7041 | * Nobody should be changing or seriously looking at | |
c4ed6ebf MWO |
7042 | * folio memcg or objcg at this point, we have fully |
7043 | * exclusive access to the folio. | |
a9d5adee | 7044 | */ |
fead2b86 | 7045 | if (folio_memcg_kmem(folio)) { |
1b7e4464 | 7046 | objcg = __folio_objcg(folio); |
b4e0b68f MS |
7047 | /* |
7048 | * This get matches the put at the end of the function and | |
7049 | * kmem pages do not hold memcg references anymore. | |
7050 | */ | |
7051 | memcg = get_mem_cgroup_from_objcg(objcg); | |
7052 | } else { | |
1b7e4464 | 7053 | memcg = __folio_memcg(folio); |
b4e0b68f | 7054 | } |
a9d5adee | 7055 | |
b4e0b68f MS |
7056 | if (!memcg) |
7057 | return; | |
7058 | ||
7059 | if (ug->memcg != memcg) { | |
a9d5adee JG |
7060 | if (ug->memcg) { |
7061 | uncharge_batch(ug); | |
7062 | uncharge_gather_clear(ug); | |
7063 | } | |
b4e0b68f | 7064 | ug->memcg = memcg; |
c4ed6ebf | 7065 | ug->nid = folio_nid(folio); |
f1796544 MH |
7066 | |
7067 | /* pairs with css_put in uncharge_batch */ | |
b4e0b68f | 7068 | css_get(&memcg->css); |
a9d5adee JG |
7069 | } |
7070 | ||
c4ed6ebf | 7071 | nr_pages = folio_nr_pages(folio); |
a9d5adee | 7072 | |
fead2b86 | 7073 | if (folio_memcg_kmem(folio)) { |
b4e0b68f | 7074 | ug->nr_memory += nr_pages; |
9f762dbe | 7075 | ug->nr_kmem += nr_pages; |
b4e0b68f | 7076 | |
c4ed6ebf | 7077 | folio->memcg_data = 0; |
b4e0b68f MS |
7078 | obj_cgroup_put(objcg); |
7079 | } else { | |
7080 | /* LRU pages aren't accounted at the root level */ | |
7081 | if (!mem_cgroup_is_root(memcg)) | |
7082 | ug->nr_memory += nr_pages; | |
18b2db3b | 7083 | ug->pgpgout++; |
a9d5adee | 7084 | |
c4ed6ebf | 7085 | folio->memcg_data = 0; |
b4e0b68f MS |
7086 | } |
7087 | ||
7088 | css_put(&memcg->css); | |
747db954 JW |
7089 | } |
7090 | ||
bbc6b703 | 7091 | void __mem_cgroup_uncharge(struct folio *folio) |
0a31bc97 | 7092 | { |
a9d5adee JG |
7093 | struct uncharge_gather ug; |
7094 | ||
bbc6b703 MWO |
7095 | /* Don't touch folio->lru of any random page, pre-check: */ |
7096 | if (!folio_memcg(folio)) | |
0a31bc97 JW |
7097 | return; |
7098 | ||
a9d5adee | 7099 | uncharge_gather_clear(&ug); |
bbc6b703 | 7100 | uncharge_folio(folio, &ug); |
a9d5adee | 7101 | uncharge_batch(&ug); |
747db954 | 7102 | } |
0a31bc97 | 7103 | |
747db954 | 7104 | /** |
2c8d8f97 | 7105 | * __mem_cgroup_uncharge_list - uncharge a list of page |
747db954 JW |
7106 | * @page_list: list of pages to uncharge |
7107 | * | |
7108 | * Uncharge a list of pages previously charged with | |
2c8d8f97 | 7109 | * __mem_cgroup_charge(). |
747db954 | 7110 | */ |
2c8d8f97 | 7111 | void __mem_cgroup_uncharge_list(struct list_head *page_list) |
747db954 | 7112 | { |
c41a40b6 | 7113 | struct uncharge_gather ug; |
c4ed6ebf | 7114 | struct folio *folio; |
c41a40b6 | 7115 | |
c41a40b6 | 7116 | uncharge_gather_clear(&ug); |
c4ed6ebf MWO |
7117 | list_for_each_entry(folio, page_list, lru) |
7118 | uncharge_folio(folio, &ug); | |
c41a40b6 MS |
7119 | if (ug.memcg) |
7120 | uncharge_batch(&ug); | |
0a31bc97 JW |
7121 | } |
7122 | ||
7123 | /** | |
d21bba2b MWO |
7124 | * mem_cgroup_migrate - Charge a folio's replacement. |
7125 | * @old: Currently circulating folio. | |
7126 | * @new: Replacement folio. | |
0a31bc97 | 7127 | * |
d21bba2b | 7128 | * Charge @new as a replacement folio for @old. @old will |
6a93ca8f | 7129 | * be uncharged upon free. |
0a31bc97 | 7130 | * |
d21bba2b | 7131 | * Both folios must be locked, @new->mapping must be set up. |
0a31bc97 | 7132 | */ |
d21bba2b | 7133 | void mem_cgroup_migrate(struct folio *old, struct folio *new) |
0a31bc97 | 7134 | { |
29833315 | 7135 | struct mem_cgroup *memcg; |
d21bba2b | 7136 | long nr_pages = folio_nr_pages(new); |
d93c4130 | 7137 | unsigned long flags; |
0a31bc97 | 7138 | |
d21bba2b MWO |
7139 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); |
7140 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7141 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7142 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != nr_pages, new); | |
0a31bc97 JW |
7143 | |
7144 | if (mem_cgroup_disabled()) | |
7145 | return; | |
7146 | ||
d21bba2b MWO |
7147 | /* Page cache replacement: new folio already charged? */ |
7148 | if (folio_memcg(new)) | |
0a31bc97 JW |
7149 | return; |
7150 | ||
d21bba2b MWO |
7151 | memcg = folio_memcg(old); |
7152 | VM_WARN_ON_ONCE_FOLIO(!memcg, old); | |
29833315 | 7153 | if (!memcg) |
0a31bc97 JW |
7154 | return; |
7155 | ||
44b7a8d3 | 7156 | /* Force-charge the new page. The old one will be freed soon */ |
8dc87c7d MS |
7157 | if (!mem_cgroup_is_root(memcg)) { |
7158 | page_counter_charge(&memcg->memory, nr_pages); | |
7159 | if (do_memsw_account()) | |
7160 | page_counter_charge(&memcg->memsw, nr_pages); | |
7161 | } | |
0a31bc97 | 7162 | |
1a3e1f40 | 7163 | css_get(&memcg->css); |
d21bba2b | 7164 | commit_charge(new, memcg); |
44b7a8d3 | 7165 | |
d93c4130 | 7166 | local_irq_save(flags); |
6e0110c2 | 7167 | mem_cgroup_charge_statistics(memcg, nr_pages); |
d21bba2b | 7168 | memcg_check_events(memcg, folio_nid(new)); |
d93c4130 | 7169 | local_irq_restore(flags); |
0a31bc97 JW |
7170 | } |
7171 | ||
ef12947c | 7172 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
7173 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
7174 | ||
2d758073 | 7175 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
7176 | { |
7177 | struct mem_cgroup *memcg; | |
7178 | ||
2d758073 JW |
7179 | if (!mem_cgroup_sockets_enabled) |
7180 | return; | |
7181 | ||
e876ecc6 | 7182 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
086f694a | 7183 | if (!in_task()) |
e876ecc6 SB |
7184 | return; |
7185 | ||
11092087 JW |
7186 | rcu_read_lock(); |
7187 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
7188 | if (memcg == root_mem_cgroup) |
7189 | goto out; | |
0db15298 | 7190 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 7191 | goto out; |
8965aa28 | 7192 | if (css_tryget(&memcg->css)) |
11092087 | 7193 | sk->sk_memcg = memcg; |
f7e1cb6e | 7194 | out: |
11092087 JW |
7195 | rcu_read_unlock(); |
7196 | } | |
11092087 | 7197 | |
2d758073 | 7198 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 7199 | { |
2d758073 JW |
7200 | if (sk->sk_memcg) |
7201 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7202 | } |
7203 | ||
7204 | /** | |
7205 | * mem_cgroup_charge_skmem - charge socket memory | |
7206 | * @memcg: memcg to charge | |
7207 | * @nr_pages: number of pages to charge | |
4b1327be | 7208 | * @gfp_mask: reclaim mode |
11092087 JW |
7209 | * |
7210 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
4b1327be | 7211 | * @memcg's configured limit, %false if it doesn't. |
11092087 | 7212 | */ |
4b1327be WW |
7213 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, |
7214 | gfp_t gfp_mask) | |
11092087 | 7215 | { |
f7e1cb6e | 7216 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7217 | struct page_counter *fail; |
f7e1cb6e | 7218 | |
0db15298 JW |
7219 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7220 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7221 | return true; |
7222 | } | |
0db15298 | 7223 | memcg->tcpmem_pressure = 1; |
4b1327be WW |
7224 | if (gfp_mask & __GFP_NOFAIL) { |
7225 | page_counter_charge(&memcg->tcpmem, nr_pages); | |
7226 | return true; | |
7227 | } | |
f7e1cb6e | 7228 | return false; |
11092087 | 7229 | } |
d886f4e4 | 7230 | |
4b1327be WW |
7231 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) { |
7232 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); | |
f7e1cb6e | 7233 | return true; |
4b1327be | 7234 | } |
f7e1cb6e | 7235 | |
11092087 JW |
7236 | return false; |
7237 | } | |
7238 | ||
7239 | /** | |
7240 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7241 | * @memcg: memcg to uncharge |
7242 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7243 | */ |
7244 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7245 | { | |
f7e1cb6e | 7246 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7247 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7248 | return; |
7249 | } | |
d886f4e4 | 7250 | |
c9019e9b | 7251 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7252 | |
475d0487 | 7253 | refill_stock(memcg, nr_pages); |
11092087 JW |
7254 | } |
7255 | ||
f7e1cb6e JW |
7256 | static int __init cgroup_memory(char *s) |
7257 | { | |
7258 | char *token; | |
7259 | ||
7260 | while ((token = strsep(&s, ",")) != NULL) { | |
7261 | if (!*token) | |
7262 | continue; | |
7263 | if (!strcmp(token, "nosocket")) | |
7264 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7265 | if (!strcmp(token, "nokmem")) |
7266 | cgroup_memory_nokmem = true; | |
f7e1cb6e | 7267 | } |
460a79e1 | 7268 | return 1; |
f7e1cb6e JW |
7269 | } |
7270 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7271 | |
2d11085e | 7272 | /* |
1081312f MH |
7273 | * subsys_initcall() for memory controller. |
7274 | * | |
308167fc SAS |
7275 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7276 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7277 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7278 | * should be initialized from here. | |
2d11085e MH |
7279 | */ |
7280 | static int __init mem_cgroup_init(void) | |
7281 | { | |
95a045f6 JW |
7282 | int cpu, node; |
7283 | ||
f3344adf MS |
7284 | /* |
7285 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7286 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7287 | * to work fine, we should make sure that the overfill threshold can't | |
7288 | * exceed S32_MAX / PAGE_SIZE. | |
7289 | */ | |
7290 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7291 | ||
308167fc SAS |
7292 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7293 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7294 | |
7295 | for_each_possible_cpu(cpu) | |
7296 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7297 | drain_local_stock); | |
7298 | ||
7299 | for_each_node(node) { | |
7300 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7301 | |
7302 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7303 | node_online(node) ? node : NUMA_NO_NODE); | |
7304 | ||
ef8f2327 | 7305 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7306 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7307 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7308 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7309 | } | |
7310 | ||
2d11085e MH |
7311 | return 0; |
7312 | } | |
7313 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7314 | |
7315 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7316 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7317 | { | |
1c2d479a | 7318 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7319 | /* |
7320 | * The root cgroup cannot be destroyed, so it's refcount must | |
7321 | * always be >= 1. | |
7322 | */ | |
7323 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7324 | VM_BUG_ON(1); | |
7325 | break; | |
7326 | } | |
7327 | memcg = parent_mem_cgroup(memcg); | |
7328 | if (!memcg) | |
7329 | memcg = root_mem_cgroup; | |
7330 | } | |
7331 | return memcg; | |
7332 | } | |
7333 | ||
21afa38e JW |
7334 | /** |
7335 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
3ecb0087 | 7336 | * @folio: folio whose memsw charge to transfer |
21afa38e JW |
7337 | * @entry: swap entry to move the charge to |
7338 | * | |
3ecb0087 | 7339 | * Transfer the memsw charge of @folio to @entry. |
21afa38e | 7340 | */ |
3ecb0087 | 7341 | void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry) |
21afa38e | 7342 | { |
1f47b61f | 7343 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7344 | unsigned int nr_entries; |
21afa38e JW |
7345 | unsigned short oldid; |
7346 | ||
3ecb0087 MWO |
7347 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
7348 | VM_BUG_ON_FOLIO(folio_ref_count(folio), folio); | |
21afa38e | 7349 | |
76358ab5 AS |
7350 | if (mem_cgroup_disabled()) |
7351 | return; | |
7352 | ||
2d1c4980 | 7353 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7354 | return; |
7355 | ||
3ecb0087 | 7356 | memcg = folio_memcg(folio); |
21afa38e | 7357 | |
3ecb0087 | 7358 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
21afa38e JW |
7359 | if (!memcg) |
7360 | return; | |
7361 | ||
1f47b61f VD |
7362 | /* |
7363 | * In case the memcg owning these pages has been offlined and doesn't | |
7364 | * have an ID allocated to it anymore, charge the closest online | |
7365 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7366 | */ | |
7367 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
3ecb0087 | 7368 | nr_entries = folio_nr_pages(folio); |
d6810d73 HY |
7369 | /* Get references for the tail pages, too */ |
7370 | if (nr_entries > 1) | |
7371 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7372 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7373 | nr_entries); | |
3ecb0087 | 7374 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 7375 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 7376 | |
3ecb0087 | 7377 | folio->memcg_data = 0; |
21afa38e JW |
7378 | |
7379 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7380 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7381 | |
c1b8fdae | 7382 | if (memcg_swap_enabled() && memcg != swap_memcg) { |
1f47b61f | 7383 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7384 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7385 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7386 | } |
7387 | ||
ce9ce665 SAS |
7388 | /* |
7389 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7390 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7391 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7392 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 | 7393 | */ |
be3e67b5 | 7394 | memcg_stats_lock(); |
6e0110c2 | 7395 | mem_cgroup_charge_statistics(memcg, -nr_entries); |
be3e67b5 | 7396 | memcg_stats_unlock(); |
3ecb0087 | 7397 | memcg_check_events(memcg, folio_nid(folio)); |
73f576c0 | 7398 | |
1a3e1f40 | 7399 | css_put(&memcg->css); |
21afa38e JW |
7400 | } |
7401 | ||
38d8b4e6 | 7402 | /** |
e2e3fdc7 MWO |
7403 | * __mem_cgroup_try_charge_swap - try charging swap space for a folio |
7404 | * @folio: folio being added to swap | |
37e84351 VD |
7405 | * @entry: swap entry to charge |
7406 | * | |
e2e3fdc7 | 7407 | * Try to charge @folio's memcg for the swap space at @entry. |
37e84351 VD |
7408 | * |
7409 | * Returns 0 on success, -ENOMEM on failure. | |
7410 | */ | |
e2e3fdc7 | 7411 | int __mem_cgroup_try_charge_swap(struct folio *folio, swp_entry_t entry) |
37e84351 | 7412 | { |
e2e3fdc7 | 7413 | unsigned int nr_pages = folio_nr_pages(folio); |
37e84351 | 7414 | struct page_counter *counter; |
38d8b4e6 | 7415 | struct mem_cgroup *memcg; |
37e84351 VD |
7416 | unsigned short oldid; |
7417 | ||
2d1c4980 | 7418 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7419 | return 0; |
7420 | ||
e2e3fdc7 | 7421 | memcg = folio_memcg(folio); |
37e84351 | 7422 | |
e2e3fdc7 | 7423 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
37e84351 VD |
7424 | if (!memcg) |
7425 | return 0; | |
7426 | ||
f3a53a3a TH |
7427 | if (!entry.val) { |
7428 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7429 | return 0; |
f3a53a3a | 7430 | } |
bb98f2c5 | 7431 | |
1f47b61f VD |
7432 | memcg = mem_cgroup_id_get_online(memcg); |
7433 | ||
c1b8fdae | 7434 | if (memcg_swap_enabled() && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7435 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7436 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7437 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7438 | mem_cgroup_id_put(memcg); |
37e84351 | 7439 | return -ENOMEM; |
1f47b61f | 7440 | } |
37e84351 | 7441 | |
38d8b4e6 HY |
7442 | /* Get references for the tail pages, too */ |
7443 | if (nr_pages > 1) | |
7444 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7445 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
e2e3fdc7 | 7446 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 7447 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7448 | |
37e84351 VD |
7449 | return 0; |
7450 | } | |
7451 | ||
21afa38e | 7452 | /** |
01c4b28c | 7453 | * __mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7454 | * @entry: swap entry to uncharge |
38d8b4e6 | 7455 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7456 | */ |
01c4b28c | 7457 | void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7458 | { |
7459 | struct mem_cgroup *memcg; | |
7460 | unsigned short id; | |
7461 | ||
c91bdc93 JW |
7462 | if (mem_cgroup_disabled()) |
7463 | return; | |
7464 | ||
38d8b4e6 | 7465 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7466 | rcu_read_lock(); |
adbe427b | 7467 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7468 | if (memcg) { |
c1b8fdae | 7469 | if (memcg_swap_enabled() && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7470 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7471 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7472 | else |
38d8b4e6 | 7473 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7474 | } |
c9019e9b | 7475 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7476 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7477 | } |
7478 | rcu_read_unlock(); | |
7479 | } | |
7480 | ||
d8b38438 VD |
7481 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7482 | { | |
7483 | long nr_swap_pages = get_nr_swap_pages(); | |
7484 | ||
c1b8fdae | 7485 | if (!memcg_swap_enabled() || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7486 | return nr_swap_pages; |
7487 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7488 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7489 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7490 | page_counter_read(&memcg->swap)); |
7491 | return nr_swap_pages; | |
7492 | } | |
7493 | ||
9202d527 | 7494 | bool mem_cgroup_swap_full(struct folio *folio) |
5ccc5aba VD |
7495 | { |
7496 | struct mem_cgroup *memcg; | |
7497 | ||
9202d527 | 7498 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
5ccc5aba VD |
7499 | |
7500 | if (vm_swap_full()) | |
7501 | return true; | |
c1b8fdae | 7502 | if (!memcg_swap_enabled() || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7503 | return false; |
7504 | ||
9202d527 | 7505 | memcg = folio_memcg(folio); |
5ccc5aba VD |
7506 | if (!memcg) |
7507 | return false; | |
7508 | ||
4b82ab4f JK |
7509 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7510 | unsigned long usage = page_counter_read(&memcg->swap); | |
7511 | ||
7512 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7513 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7514 | return true; |
4b82ab4f | 7515 | } |
5ccc5aba VD |
7516 | |
7517 | return false; | |
7518 | } | |
7519 | ||
eccb52e7 | 7520 | static int __init setup_swap_account(char *s) |
21afa38e | 7521 | { |
4988fe69 LS |
7522 | bool res; |
7523 | ||
7524 | if (!kstrtobool(s, &res)) | |
7525 | cgroup_memory_noswap = !res; | |
21afa38e JW |
7526 | return 1; |
7527 | } | |
eccb52e7 | 7528 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7529 | |
37e84351 VD |
7530 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7531 | struct cftype *cft) | |
7532 | { | |
7533 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7534 | ||
7535 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7536 | } | |
7537 | ||
4b82ab4f JK |
7538 | static int swap_high_show(struct seq_file *m, void *v) |
7539 | { | |
7540 | return seq_puts_memcg_tunable(m, | |
7541 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7542 | } | |
7543 | ||
7544 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7545 | char *buf, size_t nbytes, loff_t off) | |
7546 | { | |
7547 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7548 | unsigned long high; | |
7549 | int err; | |
7550 | ||
7551 | buf = strstrip(buf); | |
7552 | err = page_counter_memparse(buf, "max", &high); | |
7553 | if (err) | |
7554 | return err; | |
7555 | ||
7556 | page_counter_set_high(&memcg->swap, high); | |
7557 | ||
7558 | return nbytes; | |
7559 | } | |
7560 | ||
37e84351 VD |
7561 | static int swap_max_show(struct seq_file *m, void *v) |
7562 | { | |
677dc973 CD |
7563 | return seq_puts_memcg_tunable(m, |
7564 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7565 | } |
7566 | ||
7567 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7568 | char *buf, size_t nbytes, loff_t off) | |
7569 | { | |
7570 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7571 | unsigned long max; | |
7572 | int err; | |
7573 | ||
7574 | buf = strstrip(buf); | |
7575 | err = page_counter_memparse(buf, "max", &max); | |
7576 | if (err) | |
7577 | return err; | |
7578 | ||
be09102b | 7579 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7580 | |
7581 | return nbytes; | |
7582 | } | |
7583 | ||
f3a53a3a TH |
7584 | static int swap_events_show(struct seq_file *m, void *v) |
7585 | { | |
aa9694bb | 7586 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7587 | |
4b82ab4f JK |
7588 | seq_printf(m, "high %lu\n", |
7589 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7590 | seq_printf(m, "max %lu\n", |
7591 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7592 | seq_printf(m, "fail %lu\n", | |
7593 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7594 | ||
7595 | return 0; | |
7596 | } | |
7597 | ||
37e84351 VD |
7598 | static struct cftype swap_files[] = { |
7599 | { | |
7600 | .name = "swap.current", | |
7601 | .flags = CFTYPE_NOT_ON_ROOT, | |
7602 | .read_u64 = swap_current_read, | |
7603 | }, | |
4b82ab4f JK |
7604 | { |
7605 | .name = "swap.high", | |
7606 | .flags = CFTYPE_NOT_ON_ROOT, | |
7607 | .seq_show = swap_high_show, | |
7608 | .write = swap_high_write, | |
7609 | }, | |
37e84351 VD |
7610 | { |
7611 | .name = "swap.max", | |
7612 | .flags = CFTYPE_NOT_ON_ROOT, | |
7613 | .seq_show = swap_max_show, | |
7614 | .write = swap_max_write, | |
7615 | }, | |
f3a53a3a TH |
7616 | { |
7617 | .name = "swap.events", | |
7618 | .flags = CFTYPE_NOT_ON_ROOT, | |
7619 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7620 | .seq_show = swap_events_show, | |
7621 | }, | |
37e84351 VD |
7622 | { } /* terminate */ |
7623 | }; | |
7624 | ||
eccb52e7 | 7625 | static struct cftype memsw_files[] = { |
21afa38e JW |
7626 | { |
7627 | .name = "memsw.usage_in_bytes", | |
7628 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7629 | .read_u64 = mem_cgroup_read_u64, | |
7630 | }, | |
7631 | { | |
7632 | .name = "memsw.max_usage_in_bytes", | |
7633 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7634 | .write = mem_cgroup_reset, | |
7635 | .read_u64 = mem_cgroup_read_u64, | |
7636 | }, | |
7637 | { | |
7638 | .name = "memsw.limit_in_bytes", | |
7639 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7640 | .write = mem_cgroup_write, | |
7641 | .read_u64 = mem_cgroup_read_u64, | |
7642 | }, | |
7643 | { | |
7644 | .name = "memsw.failcnt", | |
7645 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7646 | .write = mem_cgroup_reset, | |
7647 | .read_u64 = mem_cgroup_read_u64, | |
7648 | }, | |
7649 | { }, /* terminate */ | |
7650 | }; | |
7651 | ||
f4840ccf JW |
7652 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
7653 | /** | |
7654 | * obj_cgroup_may_zswap - check if this cgroup can zswap | |
7655 | * @objcg: the object cgroup | |
7656 | * | |
7657 | * Check if the hierarchical zswap limit has been reached. | |
7658 | * | |
7659 | * This doesn't check for specific headroom, and it is not atomic | |
7660 | * either. But with zswap, the size of the allocation is only known | |
7661 | * once compression has occured, and this optimistic pre-check avoids | |
7662 | * spending cycles on compression when there is already no room left | |
7663 | * or zswap is disabled altogether somewhere in the hierarchy. | |
7664 | */ | |
7665 | bool obj_cgroup_may_zswap(struct obj_cgroup *objcg) | |
7666 | { | |
7667 | struct mem_cgroup *memcg, *original_memcg; | |
7668 | bool ret = true; | |
7669 | ||
7670 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7671 | return true; | |
7672 | ||
7673 | original_memcg = get_mem_cgroup_from_objcg(objcg); | |
7674 | for (memcg = original_memcg; memcg != root_mem_cgroup; | |
7675 | memcg = parent_mem_cgroup(memcg)) { | |
7676 | unsigned long max = READ_ONCE(memcg->zswap_max); | |
7677 | unsigned long pages; | |
7678 | ||
7679 | if (max == PAGE_COUNTER_MAX) | |
7680 | continue; | |
7681 | if (max == 0) { | |
7682 | ret = false; | |
7683 | break; | |
7684 | } | |
7685 | ||
7686 | cgroup_rstat_flush(memcg->css.cgroup); | |
7687 | pages = memcg_page_state(memcg, MEMCG_ZSWAP_B) / PAGE_SIZE; | |
7688 | if (pages < max) | |
7689 | continue; | |
7690 | ret = false; | |
7691 | break; | |
7692 | } | |
7693 | mem_cgroup_put(original_memcg); | |
7694 | return ret; | |
7695 | } | |
7696 | ||
7697 | /** | |
7698 | * obj_cgroup_charge_zswap - charge compression backend memory | |
7699 | * @objcg: the object cgroup | |
7700 | * @size: size of compressed object | |
7701 | * | |
7702 | * This forces the charge after obj_cgroup_may_swap() allowed | |
7703 | * compression and storage in zwap for this cgroup to go ahead. | |
7704 | */ | |
7705 | void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size) | |
7706 | { | |
7707 | struct mem_cgroup *memcg; | |
7708 | ||
7709 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7710 | return; | |
7711 | ||
7712 | VM_WARN_ON_ONCE(!(current->flags & PF_MEMALLOC)); | |
7713 | ||
7714 | /* PF_MEMALLOC context, charging must succeed */ | |
7715 | if (obj_cgroup_charge(objcg, GFP_KERNEL, size)) | |
7716 | VM_WARN_ON_ONCE(1); | |
7717 | ||
7718 | rcu_read_lock(); | |
7719 | memcg = obj_cgroup_memcg(objcg); | |
7720 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, size); | |
7721 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, 1); | |
7722 | rcu_read_unlock(); | |
7723 | } | |
7724 | ||
7725 | /** | |
7726 | * obj_cgroup_uncharge_zswap - uncharge compression backend memory | |
7727 | * @objcg: the object cgroup | |
7728 | * @size: size of compressed object | |
7729 | * | |
7730 | * Uncharges zswap memory on page in. | |
7731 | */ | |
7732 | void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size) | |
7733 | { | |
7734 | struct mem_cgroup *memcg; | |
7735 | ||
7736 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7737 | return; | |
7738 | ||
7739 | obj_cgroup_uncharge(objcg, size); | |
7740 | ||
7741 | rcu_read_lock(); | |
7742 | memcg = obj_cgroup_memcg(objcg); | |
7743 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, -size); | |
7744 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, -1); | |
7745 | rcu_read_unlock(); | |
7746 | } | |
7747 | ||
7748 | static u64 zswap_current_read(struct cgroup_subsys_state *css, | |
7749 | struct cftype *cft) | |
7750 | { | |
7751 | cgroup_rstat_flush(css->cgroup); | |
7752 | return memcg_page_state(mem_cgroup_from_css(css), MEMCG_ZSWAP_B); | |
7753 | } | |
7754 | ||
7755 | static int zswap_max_show(struct seq_file *m, void *v) | |
7756 | { | |
7757 | return seq_puts_memcg_tunable(m, | |
7758 | READ_ONCE(mem_cgroup_from_seq(m)->zswap_max)); | |
7759 | } | |
7760 | ||
7761 | static ssize_t zswap_max_write(struct kernfs_open_file *of, | |
7762 | char *buf, size_t nbytes, loff_t off) | |
7763 | { | |
7764 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7765 | unsigned long max; | |
7766 | int err; | |
7767 | ||
7768 | buf = strstrip(buf); | |
7769 | err = page_counter_memparse(buf, "max", &max); | |
7770 | if (err) | |
7771 | return err; | |
7772 | ||
7773 | xchg(&memcg->zswap_max, max); | |
7774 | ||
7775 | return nbytes; | |
7776 | } | |
7777 | ||
7778 | static struct cftype zswap_files[] = { | |
7779 | { | |
7780 | .name = "zswap.current", | |
7781 | .flags = CFTYPE_NOT_ON_ROOT, | |
7782 | .read_u64 = zswap_current_read, | |
7783 | }, | |
7784 | { | |
7785 | .name = "zswap.max", | |
7786 | .flags = CFTYPE_NOT_ON_ROOT, | |
7787 | .seq_show = zswap_max_show, | |
7788 | .write = zswap_max_write, | |
7789 | }, | |
7790 | { } /* terminate */ | |
7791 | }; | |
7792 | #endif /* CONFIG_MEMCG_KMEM && CONFIG_ZSWAP */ | |
7793 | ||
82ff165c BS |
7794 | /* |
7795 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7796 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7797 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7798 | * boot parameter. This may result in premature OOPS inside | |
7799 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7800 | */ | |
21afa38e JW |
7801 | static int __init mem_cgroup_swap_init(void) |
7802 | { | |
2d1c4980 JW |
7803 | /* No memory control -> no swap control */ |
7804 | if (mem_cgroup_disabled()) | |
7805 | cgroup_memory_noswap = true; | |
7806 | ||
7807 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7808 | return 0; |
7809 | ||
c1b8fdae KS |
7810 | static_branch_enable(&memcg_swap_enabled_key); |
7811 | ||
eccb52e7 JW |
7812 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); |
7813 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
f4840ccf JW |
7814 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
7815 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, zswap_files)); | |
7816 | #endif | |
21afa38e JW |
7817 | return 0; |
7818 | } | |
82ff165c | 7819 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7820 | |
7821 | #endif /* CONFIG_MEMCG_SWAP */ |