Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8cdea7c0 BS |
2 | /* memcontrol.c - Memory Controller |
3 | * | |
4 | * Copyright IBM Corporation, 2007 | |
5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
6 | * | |
78fb7466 PE |
7 | * Copyright 2007 OpenVZ SWsoft Inc |
8 | * Author: Pavel Emelianov <xemul@openvz.org> | |
9 | * | |
2e72b634 KS |
10 | * Memory thresholds |
11 | * Copyright (C) 2009 Nokia Corporation | |
12 | * Author: Kirill A. Shutemov | |
13 | * | |
7ae1e1d0 GC |
14 | * Kernel Memory Controller |
15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
16 | * Authors: Glauber Costa and Suleiman Souhlal | |
17 | * | |
1575e68b JW |
18 | * Native page reclaim |
19 | * Charge lifetime sanitation | |
20 | * Lockless page tracking & accounting | |
21 | * Unified hierarchy configuration model | |
22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
6168d0da AS |
23 | * |
24 | * Per memcg lru locking | |
25 | * Copyright (C) 2020 Alibaba, Inc, Alex Shi | |
8cdea7c0 BS |
26 | */ |
27 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
29 | #include <linux/memcontrol.h> |
30 | #include <linux/cgroup.h> | |
a520110e | 31 | #include <linux/pagewalk.h> |
6e84f315 | 32 | #include <linux/sched/mm.h> |
3a4f8a0b | 33 | #include <linux/shmem_fs.h> |
4ffef5fe | 34 | #include <linux/hugetlb.h> |
d13d1443 | 35 | #include <linux/pagemap.h> |
4882c809 | 36 | #include <linux/pagevec.h> |
1ff9e6e1 | 37 | #include <linux/vm_event_item.h> |
d52aa412 | 38 | #include <linux/smp.h> |
8a9f3ccd | 39 | #include <linux/page-flags.h> |
66e1707b | 40 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
41 | #include <linux/bit_spinlock.h> |
42 | #include <linux/rcupdate.h> | |
e222432b | 43 | #include <linux/limits.h> |
b9e15baf | 44 | #include <linux/export.h> |
8c7c6e34 | 45 | #include <linux/mutex.h> |
bb4cc1a8 | 46 | #include <linux/rbtree.h> |
b6ac57d5 | 47 | #include <linux/slab.h> |
66e1707b | 48 | #include <linux/swap.h> |
02491447 | 49 | #include <linux/swapops.h> |
66e1707b | 50 | #include <linux/spinlock.h> |
2e72b634 | 51 | #include <linux/eventfd.h> |
79bd9814 | 52 | #include <linux/poll.h> |
2e72b634 | 53 | #include <linux/sort.h> |
66e1707b | 54 | #include <linux/fs.h> |
d2ceb9b7 | 55 | #include <linux/seq_file.h> |
70ddf637 | 56 | #include <linux/vmpressure.h> |
dc90f084 | 57 | #include <linux/memremap.h> |
b69408e8 | 58 | #include <linux/mm_inline.h> |
5d1ea48b | 59 | #include <linux/swap_cgroup.h> |
cdec2e42 | 60 | #include <linux/cpu.h> |
158e0a2d | 61 | #include <linux/oom.h> |
0056f4e6 | 62 | #include <linux/lockdep.h> |
79bd9814 | 63 | #include <linux/file.h> |
03248add | 64 | #include <linux/resume_user_mode.h> |
0e4b01df | 65 | #include <linux/psi.h> |
c8713d0b | 66 | #include <linux/seq_buf.h> |
6a792697 | 67 | #include <linux/sched/isolation.h> |
6011be59 | 68 | #include <linux/kmemleak.h> |
08e552c6 | 69 | #include "internal.h" |
d1a4c0b3 | 70 | #include <net/sock.h> |
4bd2c1ee | 71 | #include <net/ip.h> |
f35c3a8e | 72 | #include "slab.h" |
014bb1de | 73 | #include "swap.h" |
8cdea7c0 | 74 | |
7c0f6ba6 | 75 | #include <linux/uaccess.h> |
8697d331 | 76 | |
cc8e970c KM |
77 | #include <trace/events/vmscan.h> |
78 | ||
073219e9 TH |
79 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
80 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 81 | |
7d828602 JW |
82 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
83 | ||
37d5985c RG |
84 | /* Active memory cgroup to use from an interrupt context */ |
85 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); | |
c74d40e8 | 86 | EXPORT_PER_CPU_SYMBOL_GPL(int_active_memcg); |
37d5985c | 87 | |
f7e1cb6e | 88 | /* Socket memory accounting disabled? */ |
0f0cace3 | 89 | static bool cgroup_memory_nosocket __ro_after_init; |
f7e1cb6e | 90 | |
04823c83 | 91 | /* Kernel memory accounting disabled? */ |
17c17367 | 92 | static bool cgroup_memory_nokmem __ro_after_init; |
04823c83 | 93 | |
b6c1a8af YS |
94 | /* BPF memory accounting disabled? */ |
95 | static bool cgroup_memory_nobpf __ro_after_init; | |
96 | ||
97b27821 TH |
97 | #ifdef CONFIG_CGROUP_WRITEBACK |
98 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
99 | #endif | |
100 | ||
7941d214 JW |
101 | /* Whether legacy memory+swap accounting is active */ |
102 | static bool do_memsw_account(void) | |
103 | { | |
b25806dc | 104 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys); |
7941d214 JW |
105 | } |
106 | ||
a0db00fc KS |
107 | #define THRESHOLDS_EVENTS_TARGET 128 |
108 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 109 | |
bb4cc1a8 AM |
110 | /* |
111 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
112 | * their hierarchy representation | |
113 | */ | |
114 | ||
ef8f2327 | 115 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 116 | struct rb_root rb_root; |
fa90b2fd | 117 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
118 | spinlock_t lock; |
119 | }; | |
120 | ||
bb4cc1a8 AM |
121 | struct mem_cgroup_tree { |
122 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
123 | }; | |
124 | ||
125 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
126 | ||
9490ff27 KH |
127 | /* for OOM */ |
128 | struct mem_cgroup_eventfd_list { | |
129 | struct list_head list; | |
130 | struct eventfd_ctx *eventfd; | |
131 | }; | |
2e72b634 | 132 | |
79bd9814 TH |
133 | /* |
134 | * cgroup_event represents events which userspace want to receive. | |
135 | */ | |
3bc942f3 | 136 | struct mem_cgroup_event { |
79bd9814 | 137 | /* |
59b6f873 | 138 | * memcg which the event belongs to. |
79bd9814 | 139 | */ |
59b6f873 | 140 | struct mem_cgroup *memcg; |
79bd9814 TH |
141 | /* |
142 | * eventfd to signal userspace about the event. | |
143 | */ | |
144 | struct eventfd_ctx *eventfd; | |
145 | /* | |
146 | * Each of these stored in a list by the cgroup. | |
147 | */ | |
148 | struct list_head list; | |
fba94807 TH |
149 | /* |
150 | * register_event() callback will be used to add new userspace | |
151 | * waiter for changes related to this event. Use eventfd_signal() | |
152 | * on eventfd to send notification to userspace. | |
153 | */ | |
59b6f873 | 154 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 155 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
156 | /* |
157 | * unregister_event() callback will be called when userspace closes | |
158 | * the eventfd or on cgroup removing. This callback must be set, | |
159 | * if you want provide notification functionality. | |
160 | */ | |
59b6f873 | 161 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 162 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
163 | /* |
164 | * All fields below needed to unregister event when | |
165 | * userspace closes eventfd. | |
166 | */ | |
167 | poll_table pt; | |
168 | wait_queue_head_t *wqh; | |
ac6424b9 | 169 | wait_queue_entry_t wait; |
79bd9814 TH |
170 | struct work_struct remove; |
171 | }; | |
172 | ||
c0ff4b85 R |
173 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
174 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 175 | |
7dc74be0 DN |
176 | /* Stuffs for move charges at task migration. */ |
177 | /* | |
1dfab5ab | 178 | * Types of charges to be moved. |
7dc74be0 | 179 | */ |
1dfab5ab JW |
180 | #define MOVE_ANON 0x1U |
181 | #define MOVE_FILE 0x2U | |
182 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 183 | |
4ffef5fe DN |
184 | /* "mc" and its members are protected by cgroup_mutex */ |
185 | static struct move_charge_struct { | |
b1dd693e | 186 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 187 | struct mm_struct *mm; |
4ffef5fe DN |
188 | struct mem_cgroup *from; |
189 | struct mem_cgroup *to; | |
1dfab5ab | 190 | unsigned long flags; |
4ffef5fe | 191 | unsigned long precharge; |
854ffa8d | 192 | unsigned long moved_charge; |
483c30b5 | 193 | unsigned long moved_swap; |
8033b97c DN |
194 | struct task_struct *moving_task; /* a task moving charges */ |
195 | wait_queue_head_t waitq; /* a waitq for other context */ | |
196 | } mc = { | |
2bd9bb20 | 197 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
198 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
199 | }; | |
4ffef5fe | 200 | |
4e416953 | 201 | /* |
f4d005af | 202 | * Maximum loops in mem_cgroup_soft_reclaim(), used for soft |
4e416953 BS |
203 | * limit reclaim to prevent infinite loops, if they ever occur. |
204 | */ | |
a0db00fc | 205 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 206 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 207 | |
8c7c6e34 | 208 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
209 | enum res_type { |
210 | _MEM, | |
211 | _MEMSWAP, | |
510fc4e1 | 212 | _KMEM, |
d55f90bf | 213 | _TCP, |
86ae53e1 GC |
214 | }; |
215 | ||
a0db00fc KS |
216 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
217 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 KH |
218 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
219 | ||
b05706f1 KT |
220 | /* |
221 | * Iteration constructs for visiting all cgroups (under a tree). If | |
222 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
223 | * be used for reference counting. | |
224 | */ | |
225 | #define for_each_mem_cgroup_tree(iter, root) \ | |
226 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
227 | iter != NULL; \ | |
228 | iter = mem_cgroup_iter(root, iter, NULL)) | |
229 | ||
230 | #define for_each_mem_cgroup(iter) \ | |
231 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
232 | iter != NULL; \ | |
233 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
234 | ||
a4ebf1b6 | 235 | static inline bool task_is_dying(void) |
7775face TH |
236 | { |
237 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
238 | (current->flags & PF_EXITING); | |
239 | } | |
240 | ||
70ddf637 AV |
241 | /* Some nice accessors for the vmpressure. */ |
242 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
243 | { | |
244 | if (!memcg) | |
245 | memcg = root_mem_cgroup; | |
246 | return &memcg->vmpressure; | |
247 | } | |
248 | ||
9647875b | 249 | struct mem_cgroup *vmpressure_to_memcg(struct vmpressure *vmpr) |
70ddf637 | 250 | { |
9647875b | 251 | return container_of(vmpr, struct mem_cgroup, vmpressure); |
70ddf637 AV |
252 | } |
253 | ||
1aacbd35 RG |
254 | #define CURRENT_OBJCG_UPDATE_BIT 0 |
255 | #define CURRENT_OBJCG_UPDATE_FLAG (1UL << CURRENT_OBJCG_UPDATE_BIT) | |
256 | ||
84c07d11 | 257 | #ifdef CONFIG_MEMCG_KMEM |
0764db9b | 258 | static DEFINE_SPINLOCK(objcg_lock); |
bf4f0599 | 259 | |
4d5c8aed RG |
260 | bool mem_cgroup_kmem_disabled(void) |
261 | { | |
262 | return cgroup_memory_nokmem; | |
263 | } | |
264 | ||
f1286fae MS |
265 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
266 | unsigned int nr_pages); | |
c1a660de | 267 | |
bf4f0599 RG |
268 | static void obj_cgroup_release(struct percpu_ref *ref) |
269 | { | |
270 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
bf4f0599 RG |
271 | unsigned int nr_bytes; |
272 | unsigned int nr_pages; | |
273 | unsigned long flags; | |
274 | ||
275 | /* | |
276 | * At this point all allocated objects are freed, and | |
277 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
278 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
279 | * | |
280 | * The following sequence can lead to it: | |
281 | * 1) CPU0: objcg == stock->cached_objcg | |
282 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
283 | * PAGE_SIZE bytes are charged | |
284 | * 3) CPU1: a process from another memcg is allocating something, | |
285 | * the stock if flushed, | |
286 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
287 | * 5) CPU0: we do release this object, | |
288 | * 92 bytes are added to stock->nr_bytes | |
289 | * 6) CPU0: stock is flushed, | |
290 | * 92 bytes are added to objcg->nr_charged_bytes | |
291 | * | |
292 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
293 | * This page will be uncharged in obj_cgroup_release(). | |
294 | */ | |
295 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
296 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
297 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
298 | ||
bf4f0599 | 299 | if (nr_pages) |
f1286fae | 300 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
271dd6b1 | 301 | |
0764db9b | 302 | spin_lock_irqsave(&objcg_lock, flags); |
bf4f0599 | 303 | list_del(&objcg->list); |
0764db9b | 304 | spin_unlock_irqrestore(&objcg_lock, flags); |
bf4f0599 RG |
305 | |
306 | percpu_ref_exit(ref); | |
307 | kfree_rcu(objcg, rcu); | |
308 | } | |
309 | ||
310 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
311 | { | |
312 | struct obj_cgroup *objcg; | |
313 | int ret; | |
314 | ||
315 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
316 | if (!objcg) | |
317 | return NULL; | |
318 | ||
319 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
320 | GFP_KERNEL); | |
321 | if (ret) { | |
322 | kfree(objcg); | |
323 | return NULL; | |
324 | } | |
325 | INIT_LIST_HEAD(&objcg->list); | |
326 | return objcg; | |
327 | } | |
328 | ||
329 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
330 | struct mem_cgroup *parent) | |
331 | { | |
332 | struct obj_cgroup *objcg, *iter; | |
333 | ||
334 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
335 | ||
0764db9b | 336 | spin_lock_irq(&objcg_lock); |
bf4f0599 | 337 | |
9838354e MS |
338 | /* 1) Ready to reparent active objcg. */ |
339 | list_add(&objcg->list, &memcg->objcg_list); | |
340 | /* 2) Reparent active objcg and already reparented objcgs to parent. */ | |
341 | list_for_each_entry(iter, &memcg->objcg_list, list) | |
342 | WRITE_ONCE(iter->memcg, parent); | |
343 | /* 3) Move already reparented objcgs to the parent's list */ | |
bf4f0599 RG |
344 | list_splice(&memcg->objcg_list, &parent->objcg_list); |
345 | ||
0764db9b | 346 | spin_unlock_irq(&objcg_lock); |
bf4f0599 RG |
347 | |
348 | percpu_ref_kill(&objcg->refcnt); | |
349 | } | |
350 | ||
d7f25f8a GC |
351 | /* |
352 | * A lot of the calls to the cache allocation functions are expected to be | |
9f9796b4 | 353 | * inlined by the compiler. Since the calls to memcg_slab_post_alloc_hook() are |
d7f25f8a GC |
354 | * conditional to this static branch, we'll have to allow modules that does |
355 | * kmem_cache_alloc and the such to see this symbol as well | |
356 | */ | |
f7a449f7 RG |
357 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_online_key); |
358 | EXPORT_SYMBOL(memcg_kmem_online_key); | |
b6c1a8af YS |
359 | |
360 | DEFINE_STATIC_KEY_FALSE(memcg_bpf_enabled_key); | |
361 | EXPORT_SYMBOL(memcg_bpf_enabled_key); | |
0a432dcb | 362 | #endif |
17cc4dfe | 363 | |
ad7fa852 | 364 | /** |
75376c6f MWO |
365 | * mem_cgroup_css_from_folio - css of the memcg associated with a folio |
366 | * @folio: folio of interest | |
ad7fa852 TH |
367 | * |
368 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
75376c6f | 369 | * with @folio is returned. The returned css remains associated with @folio |
ad7fa852 TH |
370 | * until it is released. |
371 | * | |
372 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
373 | * is returned. | |
ad7fa852 | 374 | */ |
75376c6f | 375 | struct cgroup_subsys_state *mem_cgroup_css_from_folio(struct folio *folio) |
ad7fa852 | 376 | { |
75376c6f | 377 | struct mem_cgroup *memcg = folio_memcg(folio); |
ad7fa852 | 378 | |
9e10a130 | 379 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
380 | memcg = root_mem_cgroup; |
381 | ||
ad7fa852 TH |
382 | return &memcg->css; |
383 | } | |
384 | ||
2fc04524 VD |
385 | /** |
386 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
387 | * @page: the page | |
388 | * | |
389 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
390 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
391 | * is safe to call this function without holding a reference to @page. | |
392 | * | |
393 | * Note, this function is inherently racy, because there is nothing to prevent | |
394 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
395 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
396 | * do not care (such as procfs interfaces). | |
397 | */ | |
398 | ino_t page_cgroup_ino(struct page *page) | |
399 | { | |
400 | struct mem_cgroup *memcg; | |
401 | unsigned long ino = 0; | |
402 | ||
403 | rcu_read_lock(); | |
ec342603 YA |
404 | /* page_folio() is racy here, but the entire function is racy anyway */ |
405 | memcg = folio_memcg_check(page_folio(page)); | |
286e04b8 | 406 | |
2fc04524 VD |
407 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
408 | memcg = parent_mem_cgroup(memcg); | |
409 | if (memcg) | |
410 | ino = cgroup_ino(memcg->css.cgroup); | |
411 | rcu_read_unlock(); | |
412 | return ino; | |
413 | } | |
414 | ||
ef8f2327 MG |
415 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
416 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 417 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
418 | { |
419 | struct rb_node **p = &mctz->rb_root.rb_node; | |
420 | struct rb_node *parent = NULL; | |
ef8f2327 | 421 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 422 | bool rightmost = true; |
bb4cc1a8 AM |
423 | |
424 | if (mz->on_tree) | |
425 | return; | |
426 | ||
427 | mz->usage_in_excess = new_usage_in_excess; | |
428 | if (!mz->usage_in_excess) | |
429 | return; | |
430 | while (*p) { | |
431 | parent = *p; | |
ef8f2327 | 432 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 433 | tree_node); |
fa90b2fd | 434 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 435 | p = &(*p)->rb_left; |
fa90b2fd | 436 | rightmost = false; |
378876b0 | 437 | } else { |
bb4cc1a8 | 438 | p = &(*p)->rb_right; |
378876b0 | 439 | } |
bb4cc1a8 | 440 | } |
fa90b2fd DB |
441 | |
442 | if (rightmost) | |
443 | mctz->rb_rightmost = &mz->tree_node; | |
444 | ||
bb4cc1a8 AM |
445 | rb_link_node(&mz->tree_node, parent, p); |
446 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
447 | mz->on_tree = true; | |
448 | } | |
449 | ||
ef8f2327 MG |
450 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
451 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
452 | { |
453 | if (!mz->on_tree) | |
454 | return; | |
fa90b2fd DB |
455 | |
456 | if (&mz->tree_node == mctz->rb_rightmost) | |
457 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
458 | ||
bb4cc1a8 AM |
459 | rb_erase(&mz->tree_node, &mctz->rb_root); |
460 | mz->on_tree = false; | |
461 | } | |
462 | ||
ef8f2327 MG |
463 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
464 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 465 | { |
0a31bc97 JW |
466 | unsigned long flags; |
467 | ||
468 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 469 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 470 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
471 | } |
472 | ||
3e32cb2e JW |
473 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
474 | { | |
475 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 476 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
477 | unsigned long excess = 0; |
478 | ||
479 | if (nr_pages > soft_limit) | |
480 | excess = nr_pages - soft_limit; | |
481 | ||
482 | return excess; | |
483 | } | |
bb4cc1a8 | 484 | |
658b69c9 | 485 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, int nid) |
bb4cc1a8 | 486 | { |
3e32cb2e | 487 | unsigned long excess; |
ef8f2327 MG |
488 | struct mem_cgroup_per_node *mz; |
489 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 490 | |
e4dde56c | 491 | if (lru_gen_enabled()) { |
36c7b4db | 492 | if (soft_limit_excess(memcg)) |
5c7e7a0d | 493 | lru_gen_soft_reclaim(memcg, nid); |
e4dde56c YZ |
494 | return; |
495 | } | |
496 | ||
2ab082ba | 497 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
498 | if (!mctz) |
499 | return; | |
bb4cc1a8 AM |
500 | /* |
501 | * Necessary to update all ancestors when hierarchy is used. | |
502 | * because their event counter is not touched. | |
503 | */ | |
504 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
658b69c9 | 505 | mz = memcg->nodeinfo[nid]; |
3e32cb2e | 506 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
507 | /* |
508 | * We have to update the tree if mz is on RB-tree or | |
509 | * mem is over its softlimit. | |
510 | */ | |
511 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
512 | unsigned long flags; |
513 | ||
514 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
515 | /* if on-tree, remove it */ |
516 | if (mz->on_tree) | |
cf2c8127 | 517 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
518 | /* |
519 | * Insert again. mz->usage_in_excess will be updated. | |
520 | * If excess is 0, no tree ops. | |
521 | */ | |
cf2c8127 | 522 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 523 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
524 | } |
525 | } | |
526 | } | |
527 | ||
528 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
529 | { | |
ef8f2327 MG |
530 | struct mem_cgroup_tree_per_node *mctz; |
531 | struct mem_cgroup_per_node *mz; | |
532 | int nid; | |
bb4cc1a8 | 533 | |
e231875b | 534 | for_each_node(nid) { |
a3747b53 | 535 | mz = memcg->nodeinfo[nid]; |
2ab082ba | 536 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
537 | if (mctz) |
538 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
539 | } |
540 | } | |
541 | ||
ef8f2327 MG |
542 | static struct mem_cgroup_per_node * |
543 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 544 | { |
ef8f2327 | 545 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
546 | |
547 | retry: | |
548 | mz = NULL; | |
fa90b2fd | 549 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
550 | goto done; /* Nothing to reclaim from */ |
551 | ||
fa90b2fd DB |
552 | mz = rb_entry(mctz->rb_rightmost, |
553 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
554 | /* |
555 | * Remove the node now but someone else can add it back, | |
556 | * we will to add it back at the end of reclaim to its correct | |
557 | * position in the tree. | |
558 | */ | |
cf2c8127 | 559 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 560 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 561 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
562 | goto retry; |
563 | done: | |
564 | return mz; | |
565 | } | |
566 | ||
ef8f2327 MG |
567 | static struct mem_cgroup_per_node * |
568 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 569 | { |
ef8f2327 | 570 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 571 | |
0a31bc97 | 572 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 573 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 574 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
575 | return mz; |
576 | } | |
577 | ||
ff48c71c SB |
578 | /* Subset of node_stat_item for memcg stats */ |
579 | static const unsigned int memcg_node_stat_items[] = { | |
580 | NR_INACTIVE_ANON, | |
581 | NR_ACTIVE_ANON, | |
582 | NR_INACTIVE_FILE, | |
583 | NR_ACTIVE_FILE, | |
584 | NR_UNEVICTABLE, | |
585 | NR_SLAB_RECLAIMABLE_B, | |
586 | NR_SLAB_UNRECLAIMABLE_B, | |
587 | WORKINGSET_REFAULT_ANON, | |
588 | WORKINGSET_REFAULT_FILE, | |
589 | WORKINGSET_ACTIVATE_ANON, | |
590 | WORKINGSET_ACTIVATE_FILE, | |
591 | WORKINGSET_RESTORE_ANON, | |
592 | WORKINGSET_RESTORE_FILE, | |
593 | WORKINGSET_NODERECLAIM, | |
594 | NR_ANON_MAPPED, | |
595 | NR_FILE_MAPPED, | |
596 | NR_FILE_PAGES, | |
597 | NR_FILE_DIRTY, | |
598 | NR_WRITEBACK, | |
599 | NR_SHMEM, | |
600 | NR_SHMEM_THPS, | |
601 | NR_FILE_THPS, | |
602 | NR_ANON_THPS, | |
603 | NR_KERNEL_STACK_KB, | |
604 | NR_PAGETABLE, | |
605 | NR_SECONDARY_PAGETABLE, | |
606 | #ifdef CONFIG_SWAP | |
607 | NR_SWAPCACHE, | |
608 | #endif | |
609 | }; | |
610 | ||
611 | static const unsigned int memcg_stat_items[] = { | |
612 | MEMCG_SWAP, | |
613 | MEMCG_SOCK, | |
614 | MEMCG_PERCPU_B, | |
615 | MEMCG_VMALLOC, | |
616 | MEMCG_KMEM, | |
617 | MEMCG_ZSWAP_B, | |
618 | MEMCG_ZSWAPPED, | |
619 | }; | |
620 | ||
621 | #define NR_MEMCG_NODE_STAT_ITEMS ARRAY_SIZE(memcg_node_stat_items) | |
622 | #define MEMCG_VMSTAT_SIZE (NR_MEMCG_NODE_STAT_ITEMS + \ | |
623 | ARRAY_SIZE(memcg_stat_items)) | |
624 | static int8_t mem_cgroup_stats_index[MEMCG_NR_STAT] __read_mostly; | |
625 | ||
626 | static void init_memcg_stats(void) | |
627 | { | |
628 | int8_t i, j = 0; | |
629 | ||
630 | BUILD_BUG_ON(MEMCG_NR_STAT >= S8_MAX); | |
631 | ||
632 | for (i = 0; i < NR_MEMCG_NODE_STAT_ITEMS; ++i) | |
633 | mem_cgroup_stats_index[memcg_node_stat_items[i]] = ++j; | |
634 | ||
635 | for (i = 0; i < ARRAY_SIZE(memcg_stat_items); ++i) | |
636 | mem_cgroup_stats_index[memcg_stat_items[i]] = ++j; | |
637 | } | |
638 | ||
639 | static inline int memcg_stats_index(int idx) | |
640 | { | |
641 | return mem_cgroup_stats_index[idx] - 1; | |
642 | } | |
643 | ||
70a64b79 SB |
644 | struct lruvec_stats_percpu { |
645 | /* Local (CPU and cgroup) state */ | |
ff48c71c | 646 | long state[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
647 | |
648 | /* Delta calculation for lockless upward propagation */ | |
ff48c71c | 649 | long state_prev[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
650 | }; |
651 | ||
652 | struct lruvec_stats { | |
653 | /* Aggregated (CPU and subtree) state */ | |
ff48c71c | 654 | long state[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
655 | |
656 | /* Non-hierarchical (CPU aggregated) state */ | |
ff48c71c | 657 | long state_local[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
658 | |
659 | /* Pending child counts during tree propagation */ | |
ff48c71c | 660 | long state_pending[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
661 | }; |
662 | ||
663 | unsigned long lruvec_page_state(struct lruvec *lruvec, enum node_stat_item idx) | |
664 | { | |
665 | struct mem_cgroup_per_node *pn; | |
666 | long x; | |
ff48c71c | 667 | int i; |
70a64b79 SB |
668 | |
669 | if (mem_cgroup_disabled()) | |
670 | return node_page_state(lruvec_pgdat(lruvec), idx); | |
671 | ||
ff48c71c SB |
672 | i = memcg_stats_index(idx); |
673 | if (i < 0) | |
674 | return 0; | |
675 | ||
70a64b79 | 676 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
ff48c71c | 677 | x = READ_ONCE(pn->lruvec_stats->state[i]); |
70a64b79 SB |
678 | #ifdef CONFIG_SMP |
679 | if (x < 0) | |
680 | x = 0; | |
681 | #endif | |
682 | return x; | |
683 | } | |
684 | ||
685 | unsigned long lruvec_page_state_local(struct lruvec *lruvec, | |
686 | enum node_stat_item idx) | |
687 | { | |
688 | struct mem_cgroup_per_node *pn; | |
689 | long x = 0; | |
ff48c71c | 690 | int i; |
70a64b79 SB |
691 | |
692 | if (mem_cgroup_disabled()) | |
693 | return node_page_state(lruvec_pgdat(lruvec), idx); | |
694 | ||
ff48c71c SB |
695 | i = memcg_stats_index(idx); |
696 | if (i < 0) | |
697 | return 0; | |
698 | ||
70a64b79 | 699 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
ff48c71c | 700 | x = READ_ONCE(pn->lruvec_stats->state_local[i]); |
70a64b79 SB |
701 | #ifdef CONFIG_SMP |
702 | if (x < 0) | |
703 | x = 0; | |
704 | #endif | |
705 | return x; | |
706 | } | |
707 | ||
d396def5 SB |
708 | /* Subset of vm_event_item to report for memcg event stats */ |
709 | static const unsigned int memcg_vm_event_stat[] = { | |
8278f1c7 SB |
710 | PGPGIN, |
711 | PGPGOUT, | |
d396def5 SB |
712 | PGSCAN_KSWAPD, |
713 | PGSCAN_DIRECT, | |
57e9cc50 | 714 | PGSCAN_KHUGEPAGED, |
d396def5 SB |
715 | PGSTEAL_KSWAPD, |
716 | PGSTEAL_DIRECT, | |
57e9cc50 | 717 | PGSTEAL_KHUGEPAGED, |
d396def5 SB |
718 | PGFAULT, |
719 | PGMAJFAULT, | |
720 | PGREFILL, | |
721 | PGACTIVATE, | |
722 | PGDEACTIVATE, | |
723 | PGLAZYFREE, | |
724 | PGLAZYFREED, | |
725 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) | |
726 | ZSWPIN, | |
727 | ZSWPOUT, | |
e0bf1dc8 | 728 | ZSWPWB, |
d396def5 SB |
729 | #endif |
730 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
731 | THP_FAULT_ALLOC, | |
732 | THP_COLLAPSE_ALLOC, | |
811244a5 XH |
733 | THP_SWPOUT, |
734 | THP_SWPOUT_FALLBACK, | |
d396def5 SB |
735 | #endif |
736 | }; | |
737 | ||
8278f1c7 | 738 | #define NR_MEMCG_EVENTS ARRAY_SIZE(memcg_vm_event_stat) |
59142d87 | 739 | static int8_t mem_cgroup_events_index[NR_VM_EVENT_ITEMS] __read_mostly; |
8278f1c7 SB |
740 | |
741 | static void init_memcg_events(void) | |
742 | { | |
59142d87 SB |
743 | int8_t i; |
744 | ||
745 | BUILD_BUG_ON(NR_VM_EVENT_ITEMS >= S8_MAX); | |
8278f1c7 SB |
746 | |
747 | for (i = 0; i < NR_MEMCG_EVENTS; ++i) | |
748 | mem_cgroup_events_index[memcg_vm_event_stat[i]] = i + 1; | |
749 | } | |
750 | ||
751 | static inline int memcg_events_index(enum vm_event_item idx) | |
752 | { | |
753 | return mem_cgroup_events_index[idx] - 1; | |
754 | } | |
755 | ||
410f8e82 | 756 | struct memcg_vmstats_percpu { |
9cee7e8e YA |
757 | /* Stats updates since the last flush */ |
758 | unsigned int stats_updates; | |
759 | ||
760 | /* Cached pointers for fast iteration in memcg_rstat_updated() */ | |
761 | struct memcg_vmstats_percpu *parent; | |
762 | struct memcg_vmstats *vmstats; | |
763 | ||
764 | /* The above should fit a single cacheline for memcg_rstat_updated() */ | |
765 | ||
410f8e82 | 766 | /* Local (CPU and cgroup) page state & events */ |
ff48c71c | 767 | long state[MEMCG_VMSTAT_SIZE]; |
8278f1c7 | 768 | unsigned long events[NR_MEMCG_EVENTS]; |
410f8e82 SB |
769 | |
770 | /* Delta calculation for lockless upward propagation */ | |
ff48c71c | 771 | long state_prev[MEMCG_VMSTAT_SIZE]; |
8278f1c7 | 772 | unsigned long events_prev[NR_MEMCG_EVENTS]; |
410f8e82 SB |
773 | |
774 | /* Cgroup1: threshold notifications & softlimit tree updates */ | |
775 | unsigned long nr_page_events; | |
776 | unsigned long targets[MEM_CGROUP_NTARGETS]; | |
9cee7e8e | 777 | } ____cacheline_aligned; |
410f8e82 SB |
778 | |
779 | struct memcg_vmstats { | |
780 | /* Aggregated (CPU and subtree) page state & events */ | |
ff48c71c | 781 | long state[MEMCG_VMSTAT_SIZE]; |
8278f1c7 | 782 | unsigned long events[NR_MEMCG_EVENTS]; |
410f8e82 | 783 | |
f82e6bf9 | 784 | /* Non-hierarchical (CPU aggregated) page state & events */ |
ff48c71c | 785 | long state_local[MEMCG_VMSTAT_SIZE]; |
f82e6bf9 YA |
786 | unsigned long events_local[NR_MEMCG_EVENTS]; |
787 | ||
410f8e82 | 788 | /* Pending child counts during tree propagation */ |
ff48c71c | 789 | long state_pending[MEMCG_VMSTAT_SIZE]; |
8278f1c7 | 790 | unsigned long events_pending[NR_MEMCG_EVENTS]; |
8d59d221 YA |
791 | |
792 | /* Stats updates since the last flush */ | |
793 | atomic64_t stats_updates; | |
410f8e82 SB |
794 | }; |
795 | ||
11192d9c SB |
796 | /* |
797 | * memcg and lruvec stats flushing | |
798 | * | |
799 | * Many codepaths leading to stats update or read are performance sensitive and | |
800 | * adding stats flushing in such codepaths is not desirable. So, to optimize the | |
801 | * flushing the kernel does: | |
802 | * | |
803 | * 1) Periodically and asynchronously flush the stats every 2 seconds to not let | |
804 | * rstat update tree grow unbounded. | |
805 | * | |
806 | * 2) Flush the stats synchronously on reader side only when there are more than | |
807 | * (MEMCG_CHARGE_BATCH * nr_cpus) update events. Though this optimization | |
808 | * will let stats be out of sync by atmost (MEMCG_CHARGE_BATCH * nr_cpus) but | |
809 | * only for 2 seconds due to (1). | |
810 | */ | |
811 | static void flush_memcg_stats_dwork(struct work_struct *w); | |
812 | static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork); | |
508bed88 | 813 | static u64 flush_last_time; |
9b301615 SB |
814 | |
815 | #define FLUSH_TIME (2UL*HZ) | |
11192d9c | 816 | |
be3e67b5 SAS |
817 | /* |
818 | * Accessors to ensure that preemption is disabled on PREEMPT_RT because it can | |
819 | * not rely on this as part of an acquired spinlock_t lock. These functions are | |
820 | * never used in hardirq context on PREEMPT_RT and therefore disabling preemtion | |
821 | * is sufficient. | |
822 | */ | |
823 | static void memcg_stats_lock(void) | |
824 | { | |
e575d401 TG |
825 | preempt_disable_nested(); |
826 | VM_WARN_ON_IRQS_ENABLED(); | |
be3e67b5 SAS |
827 | } |
828 | ||
829 | static void __memcg_stats_lock(void) | |
830 | { | |
e575d401 | 831 | preempt_disable_nested(); |
be3e67b5 SAS |
832 | } |
833 | ||
834 | static void memcg_stats_unlock(void) | |
835 | { | |
e575d401 | 836 | preempt_enable_nested(); |
be3e67b5 SAS |
837 | } |
838 | ||
8d59d221 | 839 | |
9cee7e8e | 840 | static bool memcg_vmstats_needs_flush(struct memcg_vmstats *vmstats) |
8d59d221 | 841 | { |
9cee7e8e | 842 | return atomic64_read(&vmstats->stats_updates) > |
8d59d221 YA |
843 | MEMCG_CHARGE_BATCH * num_online_cpus(); |
844 | } | |
845 | ||
5b3be698 | 846 | static inline void memcg_rstat_updated(struct mem_cgroup *memcg, int val) |
11192d9c | 847 | { |
9cee7e8e | 848 | struct memcg_vmstats_percpu *statc; |
8d59d221 | 849 | int cpu = smp_processor_id(); |
78ec6f9d | 850 | unsigned int stats_updates; |
5b3be698 | 851 | |
f9d911ca YA |
852 | if (!val) |
853 | return; | |
854 | ||
8d59d221 | 855 | cgroup_rstat_updated(memcg->css.cgroup, cpu); |
9cee7e8e YA |
856 | statc = this_cpu_ptr(memcg->vmstats_percpu); |
857 | for (; statc; statc = statc->parent) { | |
78ec6f9d BL |
858 | stats_updates = READ_ONCE(statc->stats_updates) + abs(val); |
859 | WRITE_ONCE(statc->stats_updates, stats_updates); | |
860 | if (stats_updates < MEMCG_CHARGE_BATCH) | |
8d59d221 | 861 | continue; |
5b3be698 | 862 | |
873f64b7 | 863 | /* |
8d59d221 YA |
864 | * If @memcg is already flush-able, increasing stats_updates is |
865 | * redundant. Avoid the overhead of the atomic update. | |
873f64b7 | 866 | */ |
9cee7e8e | 867 | if (!memcg_vmstats_needs_flush(statc->vmstats)) |
78ec6f9d | 868 | atomic64_add(stats_updates, |
9cee7e8e | 869 | &statc->vmstats->stats_updates); |
78ec6f9d | 870 | WRITE_ONCE(statc->stats_updates, 0); |
5b3be698 | 871 | } |
11192d9c SB |
872 | } |
873 | ||
7d7ef0a4 | 874 | static void do_flush_stats(struct mem_cgroup *memcg) |
11192d9c | 875 | { |
7d7ef0a4 YA |
876 | if (mem_cgroup_is_root(memcg)) |
877 | WRITE_ONCE(flush_last_time, jiffies_64); | |
9fad9aee | 878 | |
7d7ef0a4 | 879 | cgroup_rstat_flush(memcg->css.cgroup); |
11192d9c SB |
880 | } |
881 | ||
7d7ef0a4 YA |
882 | /* |
883 | * mem_cgroup_flush_stats - flush the stats of a memory cgroup subtree | |
884 | * @memcg: root of the subtree to flush | |
885 | * | |
886 | * Flushing is serialized by the underlying global rstat lock. There is also a | |
887 | * minimum amount of work to be done even if there are no stat updates to flush. | |
888 | * Hence, we only flush the stats if the updates delta exceeds a threshold. This | |
889 | * avoids unnecessary work and contention on the underlying lock. | |
890 | */ | |
891 | void mem_cgroup_flush_stats(struct mem_cgroup *memcg) | |
11192d9c | 892 | { |
7d7ef0a4 YA |
893 | if (mem_cgroup_disabled()) |
894 | return; | |
895 | ||
896 | if (!memcg) | |
897 | memcg = root_mem_cgroup; | |
898 | ||
9cee7e8e | 899 | if (memcg_vmstats_needs_flush(memcg->vmstats)) |
7d7ef0a4 | 900 | do_flush_stats(memcg); |
9fad9aee YA |
901 | } |
902 | ||
7d7ef0a4 | 903 | void mem_cgroup_flush_stats_ratelimited(struct mem_cgroup *memcg) |
9b301615 | 904 | { |
508bed88 YA |
905 | /* Only flush if the periodic flusher is one full cycle late */ |
906 | if (time_after64(jiffies_64, READ_ONCE(flush_last_time) + 2*FLUSH_TIME)) | |
7d7ef0a4 | 907 | mem_cgroup_flush_stats(memcg); |
9b301615 SB |
908 | } |
909 | ||
11192d9c SB |
910 | static void flush_memcg_stats_dwork(struct work_struct *w) |
911 | { | |
9fad9aee | 912 | /* |
9cee7e8e | 913 | * Deliberately ignore memcg_vmstats_needs_flush() here so that flushing |
8d59d221 | 914 | * in latency-sensitive paths is as cheap as possible. |
9fad9aee | 915 | */ |
7d7ef0a4 | 916 | do_flush_stats(root_mem_cgroup); |
9b301615 | 917 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, FLUSH_TIME); |
11192d9c SB |
918 | } |
919 | ||
410f8e82 SB |
920 | unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) |
921 | { | |
ff48c71c SB |
922 | long x; |
923 | int i = memcg_stats_index(idx); | |
924 | ||
925 | if (i < 0) | |
926 | return 0; | |
927 | ||
928 | x = READ_ONCE(memcg->vmstats->state[i]); | |
410f8e82 SB |
929 | #ifdef CONFIG_SMP |
930 | if (x < 0) | |
931 | x = 0; | |
932 | #endif | |
933 | return x; | |
934 | } | |
935 | ||
7bd5bc3c YA |
936 | static int memcg_page_state_unit(int item); |
937 | ||
938 | /* | |
939 | * Normalize the value passed into memcg_rstat_updated() to be in pages. Round | |
940 | * up non-zero sub-page updates to 1 page as zero page updates are ignored. | |
941 | */ | |
942 | static int memcg_state_val_in_pages(int idx, int val) | |
943 | { | |
944 | int unit = memcg_page_state_unit(idx); | |
945 | ||
946 | if (!val || unit == PAGE_SIZE) | |
947 | return val; | |
948 | else | |
949 | return max(val * unit / PAGE_SIZE, 1UL); | |
950 | } | |
951 | ||
db9adbcb JW |
952 | /** |
953 | * __mod_memcg_state - update cgroup memory statistics | |
954 | * @memcg: the memory cgroup | |
955 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
956 | * @val: delta to add to the counter, can be negative | |
957 | */ | |
958 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
959 | { | |
ff48c71c SB |
960 | int i = memcg_stats_index(idx); |
961 | ||
962 | if (mem_cgroup_disabled() || i < 0) | |
db9adbcb JW |
963 | return; |
964 | ||
ff48c71c | 965 | __this_cpu_add(memcg->vmstats_percpu->state[i], val); |
7bd5bc3c | 966 | memcg_rstat_updated(memcg, memcg_state_val_in_pages(idx, val)); |
db9adbcb JW |
967 | } |
968 | ||
2d146aa3 | 969 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
a18e6e6e JW |
970 | static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx) |
971 | { | |
ff48c71c SB |
972 | long x; |
973 | int i = memcg_stats_index(idx); | |
974 | ||
975 | if (i < 0) | |
976 | return 0; | |
a18e6e6e | 977 | |
ff48c71c | 978 | x = READ_ONCE(memcg->vmstats->state_local[i]); |
a18e6e6e JW |
979 | #ifdef CONFIG_SMP |
980 | if (x < 0) | |
981 | x = 0; | |
982 | #endif | |
983 | return x; | |
984 | } | |
985 | ||
91882c16 SB |
986 | static void __mod_memcg_lruvec_state(struct lruvec *lruvec, |
987 | enum node_stat_item idx, | |
988 | int val) | |
db9adbcb JW |
989 | { |
990 | struct mem_cgroup_per_node *pn; | |
42a30035 | 991 | struct mem_cgroup *memcg; |
ff48c71c SB |
992 | int i = memcg_stats_index(idx); |
993 | ||
994 | if (i < 0) | |
995 | return; | |
db9adbcb | 996 | |
db9adbcb | 997 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 998 | memcg = pn->memcg; |
db9adbcb | 999 | |
be3e67b5 | 1000 | /* |
be16dd76 | 1001 | * The caller from rmap relies on disabled preemption because they never |
be3e67b5 SAS |
1002 | * update their counter from in-interrupt context. For these two |
1003 | * counters we check that the update is never performed from an | |
1004 | * interrupt context while other caller need to have disabled interrupt. | |
1005 | */ | |
1006 | __memcg_stats_lock(); | |
e575d401 | 1007 | if (IS_ENABLED(CONFIG_DEBUG_VM)) { |
be3e67b5 SAS |
1008 | switch (idx) { |
1009 | case NR_ANON_MAPPED: | |
1010 | case NR_FILE_MAPPED: | |
1011 | case NR_ANON_THPS: | |
1012 | case NR_SHMEM_PMDMAPPED: | |
1013 | case NR_FILE_PMDMAPPED: | |
1014 | WARN_ON_ONCE(!in_task()); | |
1015 | break; | |
1016 | default: | |
e575d401 | 1017 | VM_WARN_ON_IRQS_ENABLED(); |
be3e67b5 SAS |
1018 | } |
1019 | } | |
1020 | ||
db9adbcb | 1021 | /* Update memcg */ |
ff48c71c | 1022 | __this_cpu_add(memcg->vmstats_percpu->state[i], val); |
db9adbcb | 1023 | |
b4c46484 | 1024 | /* Update lruvec */ |
ff48c71c | 1025 | __this_cpu_add(pn->lruvec_stats_percpu->state[i], val); |
11192d9c | 1026 | |
7bd5bc3c | 1027 | memcg_rstat_updated(memcg, memcg_state_val_in_pages(idx, val)); |
be3e67b5 | 1028 | memcg_stats_unlock(); |
db9adbcb JW |
1029 | } |
1030 | ||
eedc4e5a RG |
1031 | /** |
1032 | * __mod_lruvec_state - update lruvec memory statistics | |
1033 | * @lruvec: the lruvec | |
1034 | * @idx: the stat item | |
1035 | * @val: delta to add to the counter, can be negative | |
1036 | * | |
1037 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
1038 | * function updates the all three counters that are affected by a | |
1039 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
1040 | */ | |
1041 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
1042 | int val) | |
1043 | { | |
1044 | /* Update node */ | |
1045 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
1046 | ||
1047 | /* Update memcg and lruvec */ | |
1048 | if (!mem_cgroup_disabled()) | |
1049 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
1050 | } | |
1051 | ||
c701123b | 1052 | void __lruvec_stat_mod_folio(struct folio *folio, enum node_stat_item idx, |
c47d5032 SB |
1053 | int val) |
1054 | { | |
b4e0b68f | 1055 | struct mem_cgroup *memcg; |
c701123b | 1056 | pg_data_t *pgdat = folio_pgdat(folio); |
c47d5032 SB |
1057 | struct lruvec *lruvec; |
1058 | ||
b4e0b68f | 1059 | rcu_read_lock(); |
c701123b | 1060 | memcg = folio_memcg(folio); |
c47d5032 | 1061 | /* Untracked pages have no memcg, no lruvec. Update only the node */ |
d635a69d | 1062 | if (!memcg) { |
b4e0b68f | 1063 | rcu_read_unlock(); |
c47d5032 SB |
1064 | __mod_node_page_state(pgdat, idx, val); |
1065 | return; | |
1066 | } | |
1067 | ||
d635a69d | 1068 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 | 1069 | __mod_lruvec_state(lruvec, idx, val); |
b4e0b68f | 1070 | rcu_read_unlock(); |
c47d5032 | 1071 | } |
c701123b | 1072 | EXPORT_SYMBOL(__lruvec_stat_mod_folio); |
c47d5032 | 1073 | |
da3ceeff | 1074 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 1075 | { |
4f103c63 | 1076 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
1077 | struct mem_cgroup *memcg; |
1078 | struct lruvec *lruvec; | |
1079 | ||
1080 | rcu_read_lock(); | |
fc4db90f | 1081 | memcg = mem_cgroup_from_slab_obj(p); |
ec9f0238 | 1082 | |
8faeb1ff MS |
1083 | /* |
1084 | * Untracked pages have no memcg, no lruvec. Update only the | |
1085 | * node. If we reparent the slab objects to the root memcg, | |
1086 | * when we free the slab object, we need to update the per-memcg | |
1087 | * vmstats to keep it correct for the root memcg. | |
1088 | */ | |
1089 | if (!memcg) { | |
ec9f0238 RG |
1090 | __mod_node_page_state(pgdat, idx, val); |
1091 | } else { | |
867e5e1d | 1092 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
1093 | __mod_lruvec_state(lruvec, idx, val); |
1094 | } | |
1095 | rcu_read_unlock(); | |
1096 | } | |
1097 | ||
db9adbcb JW |
1098 | /** |
1099 | * __count_memcg_events - account VM events in a cgroup | |
1100 | * @memcg: the memory cgroup | |
1101 | * @idx: the event item | |
f0953a1b | 1102 | * @count: the number of events that occurred |
db9adbcb JW |
1103 | */ |
1104 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
1105 | unsigned long count) | |
1106 | { | |
8278f1c7 SB |
1107 | int index = memcg_events_index(idx); |
1108 | ||
1109 | if (mem_cgroup_disabled() || index < 0) | |
db9adbcb JW |
1110 | return; |
1111 | ||
be3e67b5 | 1112 | memcg_stats_lock(); |
8278f1c7 | 1113 | __this_cpu_add(memcg->vmstats_percpu->events[index], count); |
5b3be698 | 1114 | memcg_rstat_updated(memcg, count); |
be3e67b5 | 1115 | memcg_stats_unlock(); |
db9adbcb JW |
1116 | } |
1117 | ||
42a30035 | 1118 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 1119 | { |
8278f1c7 SB |
1120 | int index = memcg_events_index(event); |
1121 | ||
1122 | if (index < 0) | |
1123 | return 0; | |
1124 | return READ_ONCE(memcg->vmstats->events[index]); | |
e9f8974f JW |
1125 | } |
1126 | ||
42a30035 JW |
1127 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
1128 | { | |
8278f1c7 SB |
1129 | int index = memcg_events_index(event); |
1130 | ||
1131 | if (index < 0) | |
1132 | return 0; | |
815744d7 | 1133 | |
f82e6bf9 | 1134 | return READ_ONCE(memcg->vmstats->events_local[index]); |
42a30035 JW |
1135 | } |
1136 | ||
c0ff4b85 | 1137 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
3fba69a5 | 1138 | int nr_pages) |
d52aa412 | 1139 | { |
e401f176 KH |
1140 | /* pagein of a big page is an event. So, ignore page size */ |
1141 | if (nr_pages > 0) | |
c9019e9b | 1142 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 1143 | else { |
c9019e9b | 1144 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
1145 | nr_pages = -nr_pages; /* for event */ |
1146 | } | |
e401f176 | 1147 | |
871789d4 | 1148 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
1149 | } |
1150 | ||
f53d7ce3 JW |
1151 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
1152 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
1153 | { |
1154 | unsigned long val, next; | |
1155 | ||
871789d4 CD |
1156 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
1157 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 1158 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 1159 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
1160 | switch (target) { |
1161 | case MEM_CGROUP_TARGET_THRESH: | |
1162 | next = val + THRESHOLDS_EVENTS_TARGET; | |
1163 | break; | |
bb4cc1a8 AM |
1164 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
1165 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
1166 | break; | |
f53d7ce3 JW |
1167 | default: |
1168 | break; | |
1169 | } | |
871789d4 | 1170 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 1171 | return true; |
7a159cc9 | 1172 | } |
f53d7ce3 | 1173 | return false; |
d2265e6f KH |
1174 | } |
1175 | ||
1176 | /* | |
1177 | * Check events in order. | |
1178 | * | |
1179 | */ | |
8e88bd2d | 1180 | static void memcg_check_events(struct mem_cgroup *memcg, int nid) |
d2265e6f | 1181 | { |
2343e88d SAS |
1182 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
1183 | return; | |
1184 | ||
d2265e6f | 1185 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1186 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1187 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1188 | bool do_softlimit; |
f53d7ce3 | 1189 | |
bb4cc1a8 AM |
1190 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1191 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 1192 | mem_cgroup_threshold(memcg); |
bb4cc1a8 | 1193 | if (unlikely(do_softlimit)) |
8e88bd2d | 1194 | mem_cgroup_update_tree(memcg, nid); |
0a31bc97 | 1195 | } |
d2265e6f KH |
1196 | } |
1197 | ||
cf475ad2 | 1198 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1199 | { |
31a78f23 BS |
1200 | /* |
1201 | * mm_update_next_owner() may clear mm->owner to NULL | |
1202 | * if it races with swapoff, page migration, etc. | |
1203 | * So this can be called with p == NULL. | |
1204 | */ | |
1205 | if (unlikely(!p)) | |
1206 | return NULL; | |
1207 | ||
073219e9 | 1208 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1209 | } |
33398cf2 | 1210 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1211 | |
04f94e3f DS |
1212 | static __always_inline struct mem_cgroup *active_memcg(void) |
1213 | { | |
55a68c82 | 1214 | if (!in_task()) |
04f94e3f DS |
1215 | return this_cpu_read(int_active_memcg); |
1216 | else | |
1217 | return current->active_memcg; | |
1218 | } | |
1219 | ||
d46eb14b SB |
1220 | /** |
1221 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1222 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1223 | * | |
04f94e3f DS |
1224 | * Obtain a reference on mm->memcg and returns it if successful. If mm |
1225 | * is NULL, then the memcg is chosen as follows: | |
1226 | * 1) The active memcg, if set. | |
1227 | * 2) current->mm->memcg, if available | |
1228 | * 3) root memcg | |
1229 | * If mem_cgroup is disabled, NULL is returned. | |
d46eb14b SB |
1230 | */ |
1231 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1232 | { |
d46eb14b SB |
1233 | struct mem_cgroup *memcg; |
1234 | ||
1235 | if (mem_cgroup_disabled()) | |
1236 | return NULL; | |
0b7f569e | 1237 | |
2884b6b7 MS |
1238 | /* |
1239 | * Page cache insertions can happen without an | |
1240 | * actual mm context, e.g. during disk probing | |
1241 | * on boot, loopback IO, acct() writes etc. | |
1242 | * | |
1243 | * No need to css_get on root memcg as the reference | |
1244 | * counting is disabled on the root level in the | |
1245 | * cgroup core. See CSS_NO_REF. | |
1246 | */ | |
04f94e3f DS |
1247 | if (unlikely(!mm)) { |
1248 | memcg = active_memcg(); | |
1249 | if (unlikely(memcg)) { | |
1250 | /* remote memcg must hold a ref */ | |
1251 | css_get(&memcg->css); | |
1252 | return memcg; | |
1253 | } | |
1254 | mm = current->mm; | |
1255 | if (unlikely(!mm)) | |
1256 | return root_mem_cgroup; | |
1257 | } | |
2884b6b7 | 1258 | |
54595fe2 KH |
1259 | rcu_read_lock(); |
1260 | do { | |
2884b6b7 MS |
1261 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1262 | if (unlikely(!memcg)) | |
df381975 | 1263 | memcg = root_mem_cgroup; |
00d484f3 | 1264 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1265 | rcu_read_unlock(); |
c0ff4b85 | 1266 | return memcg; |
54595fe2 | 1267 | } |
d46eb14b SB |
1268 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1269 | ||
4b569387 NP |
1270 | /** |
1271 | * get_mem_cgroup_from_current - Obtain a reference on current task's memcg. | |
1272 | */ | |
1273 | struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1274 | { | |
1275 | struct mem_cgroup *memcg; | |
1276 | ||
1277 | if (mem_cgroup_disabled()) | |
1278 | return NULL; | |
1279 | ||
1280 | again: | |
1281 | rcu_read_lock(); | |
1282 | memcg = mem_cgroup_from_task(current); | |
1283 | if (!css_tryget(&memcg->css)) { | |
1284 | rcu_read_unlock(); | |
1285 | goto again; | |
1286 | } | |
1287 | rcu_read_unlock(); | |
1288 | return memcg; | |
1289 | } | |
1290 | ||
5660048c JW |
1291 | /** |
1292 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1293 | * @root: hierarchy root | |
1294 | * @prev: previously returned memcg, NULL on first invocation | |
1295 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1296 | * | |
1297 | * Returns references to children of the hierarchy below @root, or | |
1298 | * @root itself, or %NULL after a full round-trip. | |
1299 | * | |
1300 | * Caller must pass the return value in @prev on subsequent | |
1301 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1302 | * to cancel a hierarchy walk before the round-trip is complete. | |
1303 | * | |
05bdc520 ML |
1304 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1305 | * in the hierarchy among all concurrent reclaimers operating on the | |
1306 | * same node. | |
5660048c | 1307 | */ |
694fbc0f | 1308 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1309 | struct mem_cgroup *prev, |
694fbc0f | 1310 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1311 | { |
3f649ab7 | 1312 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1313 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1314 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1315 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1316 | |
694fbc0f AM |
1317 | if (mem_cgroup_disabled()) |
1318 | return NULL; | |
5660048c | 1319 | |
9f3a0d09 JW |
1320 | if (!root) |
1321 | root = root_mem_cgroup; | |
7d74b06f | 1322 | |
542f85f9 | 1323 | rcu_read_lock(); |
5f578161 | 1324 | |
5ac8fb31 | 1325 | if (reclaim) { |
ef8f2327 | 1326 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1327 | |
a3747b53 | 1328 | mz = root->nodeinfo[reclaim->pgdat->node_id]; |
9da83f3f | 1329 | iter = &mz->iter; |
5ac8fb31 | 1330 | |
a9320aae WY |
1331 | /* |
1332 | * On start, join the current reclaim iteration cycle. | |
1333 | * Exit when a concurrent walker completes it. | |
1334 | */ | |
1335 | if (!prev) | |
1336 | reclaim->generation = iter->generation; | |
1337 | else if (reclaim->generation != iter->generation) | |
5ac8fb31 JW |
1338 | goto out_unlock; |
1339 | ||
6df38689 | 1340 | while (1) { |
4db0c3c2 | 1341 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1342 | if (!pos || css_tryget(&pos->css)) |
1343 | break; | |
5ac8fb31 | 1344 | /* |
6df38689 VD |
1345 | * css reference reached zero, so iter->position will |
1346 | * be cleared by ->css_released. However, we should not | |
1347 | * rely on this happening soon, because ->css_released | |
1348 | * is called from a work queue, and by busy-waiting we | |
1349 | * might block it. So we clear iter->position right | |
1350 | * away. | |
5ac8fb31 | 1351 | */ |
6df38689 VD |
1352 | (void)cmpxchg(&iter->position, pos, NULL); |
1353 | } | |
89d8330c WY |
1354 | } else if (prev) { |
1355 | pos = prev; | |
5ac8fb31 JW |
1356 | } |
1357 | ||
1358 | if (pos) | |
1359 | css = &pos->css; | |
1360 | ||
1361 | for (;;) { | |
1362 | css = css_next_descendant_pre(css, &root->css); | |
1363 | if (!css) { | |
1364 | /* | |
1365 | * Reclaimers share the hierarchy walk, and a | |
1366 | * new one might jump in right at the end of | |
1367 | * the hierarchy - make sure they see at least | |
1368 | * one group and restart from the beginning. | |
1369 | */ | |
1370 | if (!prev) | |
1371 | continue; | |
1372 | break; | |
527a5ec9 | 1373 | } |
7d74b06f | 1374 | |
5ac8fb31 JW |
1375 | /* |
1376 | * Verify the css and acquire a reference. The root | |
1377 | * is provided by the caller, so we know it's alive | |
1378 | * and kicking, and don't take an extra reference. | |
1379 | */ | |
41555dad WY |
1380 | if (css == &root->css || css_tryget(css)) { |
1381 | memcg = mem_cgroup_from_css(css); | |
0b8f73e1 | 1382 | break; |
41555dad | 1383 | } |
9f3a0d09 | 1384 | } |
5ac8fb31 JW |
1385 | |
1386 | if (reclaim) { | |
5ac8fb31 | 1387 | /* |
6df38689 VD |
1388 | * The position could have already been updated by a competing |
1389 | * thread, so check that the value hasn't changed since we read | |
1390 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1391 | */ |
6df38689 VD |
1392 | (void)cmpxchg(&iter->position, pos, memcg); |
1393 | ||
5ac8fb31 JW |
1394 | if (pos) |
1395 | css_put(&pos->css); | |
1396 | ||
1397 | if (!memcg) | |
1398 | iter->generation++; | |
9f3a0d09 | 1399 | } |
5ac8fb31 | 1400 | |
542f85f9 MH |
1401 | out_unlock: |
1402 | rcu_read_unlock(); | |
c40046f3 MH |
1403 | if (prev && prev != root) |
1404 | css_put(&prev->css); | |
1405 | ||
9f3a0d09 | 1406 | return memcg; |
14067bb3 | 1407 | } |
7d74b06f | 1408 | |
5660048c JW |
1409 | /** |
1410 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1411 | * @root: hierarchy root | |
1412 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1413 | */ | |
1414 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1415 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1416 | { |
1417 | if (!root) | |
1418 | root = root_mem_cgroup; | |
1419 | if (prev && prev != root) | |
1420 | css_put(&prev->css); | |
1421 | } | |
7d74b06f | 1422 | |
54a83d6b MC |
1423 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1424 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1425 | { |
6df38689 | 1426 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1427 | struct mem_cgroup_per_node *mz; |
1428 | int nid; | |
6df38689 | 1429 | |
54a83d6b | 1430 | for_each_node(nid) { |
a3747b53 | 1431 | mz = from->nodeinfo[nid]; |
9da83f3f YS |
1432 | iter = &mz->iter; |
1433 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1434 | } |
1435 | } | |
1436 | ||
54a83d6b MC |
1437 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1438 | { | |
1439 | struct mem_cgroup *memcg = dead_memcg; | |
1440 | struct mem_cgroup *last; | |
1441 | ||
1442 | do { | |
1443 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1444 | last = memcg; | |
1445 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1446 | ||
1447 | /* | |
b8dd3ee9 | 1448 | * When cgroup1 non-hierarchy mode is used, |
54a83d6b MC |
1449 | * parent_mem_cgroup() does not walk all the way up to the |
1450 | * cgroup root (root_mem_cgroup). So we have to handle | |
1451 | * dead_memcg from cgroup root separately. | |
1452 | */ | |
7848ed62 | 1453 | if (!mem_cgroup_is_root(last)) |
54a83d6b MC |
1454 | __invalidate_reclaim_iterators(root_mem_cgroup, |
1455 | dead_memcg); | |
1456 | } | |
1457 | ||
7c5f64f8 VD |
1458 | /** |
1459 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1460 | * @memcg: hierarchy root | |
1461 | * @fn: function to call for each task | |
1462 | * @arg: argument passed to @fn | |
1463 | * | |
1464 | * This function iterates over tasks attached to @memcg or to any of its | |
1465 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
025b7799 Z |
1466 | * value, the function breaks the iteration loop. Otherwise, it will iterate |
1467 | * over all tasks and return 0. | |
7c5f64f8 VD |
1468 | * |
1469 | * This function must not be called for the root memory cgroup. | |
1470 | */ | |
025b7799 Z |
1471 | void mem_cgroup_scan_tasks(struct mem_cgroup *memcg, |
1472 | int (*fn)(struct task_struct *, void *), void *arg) | |
7c5f64f8 VD |
1473 | { |
1474 | struct mem_cgroup *iter; | |
1475 | int ret = 0; | |
1476 | ||
7848ed62 | 1477 | BUG_ON(mem_cgroup_is_root(memcg)); |
7c5f64f8 VD |
1478 | |
1479 | for_each_mem_cgroup_tree(iter, memcg) { | |
1480 | struct css_task_iter it; | |
1481 | struct task_struct *task; | |
1482 | ||
f168a9a5 | 1483 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1484 | while (!ret && (task = css_task_iter_next(&it))) |
1485 | ret = fn(task, arg); | |
1486 | css_task_iter_end(&it); | |
1487 | if (ret) { | |
1488 | mem_cgroup_iter_break(memcg, iter); | |
1489 | break; | |
1490 | } | |
1491 | } | |
7c5f64f8 VD |
1492 | } |
1493 | ||
6168d0da | 1494 | #ifdef CONFIG_DEBUG_VM |
e809c3fe | 1495 | void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio) |
6168d0da AS |
1496 | { |
1497 | struct mem_cgroup *memcg; | |
1498 | ||
1499 | if (mem_cgroup_disabled()) | |
1500 | return; | |
1501 | ||
e809c3fe | 1502 | memcg = folio_memcg(folio); |
6168d0da AS |
1503 | |
1504 | if (!memcg) | |
7848ed62 | 1505 | VM_BUG_ON_FOLIO(!mem_cgroup_is_root(lruvec_memcg(lruvec)), folio); |
6168d0da | 1506 | else |
e809c3fe | 1507 | VM_BUG_ON_FOLIO(lruvec_memcg(lruvec) != memcg, folio); |
6168d0da AS |
1508 | } |
1509 | #endif | |
1510 | ||
6168d0da | 1511 | /** |
e809c3fe MWO |
1512 | * folio_lruvec_lock - Lock the lruvec for a folio. |
1513 | * @folio: Pointer to the folio. | |
6168d0da | 1514 | * |
d7e3aba5 | 1515 | * These functions are safe to use under any of the following conditions: |
e809c3fe MWO |
1516 | * - folio locked |
1517 | * - folio_test_lru false | |
1518 | * - folio_memcg_lock() | |
1519 | * - folio frozen (refcount of 0) | |
1520 | * | |
1521 | * Return: The lruvec this folio is on with its lock held. | |
6168d0da | 1522 | */ |
e809c3fe | 1523 | struct lruvec *folio_lruvec_lock(struct folio *folio) |
6168d0da | 1524 | { |
e809c3fe | 1525 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1526 | |
6168d0da | 1527 | spin_lock(&lruvec->lru_lock); |
e809c3fe | 1528 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1529 | |
1530 | return lruvec; | |
1531 | } | |
1532 | ||
e809c3fe MWO |
1533 | /** |
1534 | * folio_lruvec_lock_irq - Lock the lruvec for a folio. | |
1535 | * @folio: Pointer to the folio. | |
1536 | * | |
1537 | * These functions are safe to use under any of the following conditions: | |
1538 | * - folio locked | |
1539 | * - folio_test_lru false | |
1540 | * - folio_memcg_lock() | |
1541 | * - folio frozen (refcount of 0) | |
1542 | * | |
1543 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1544 | * disabled. | |
1545 | */ | |
1546 | struct lruvec *folio_lruvec_lock_irq(struct folio *folio) | |
6168d0da | 1547 | { |
e809c3fe | 1548 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1549 | |
6168d0da | 1550 | spin_lock_irq(&lruvec->lru_lock); |
e809c3fe | 1551 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1552 | |
1553 | return lruvec; | |
1554 | } | |
1555 | ||
e809c3fe MWO |
1556 | /** |
1557 | * folio_lruvec_lock_irqsave - Lock the lruvec for a folio. | |
1558 | * @folio: Pointer to the folio. | |
1559 | * @flags: Pointer to irqsave flags. | |
1560 | * | |
1561 | * These functions are safe to use under any of the following conditions: | |
1562 | * - folio locked | |
1563 | * - folio_test_lru false | |
1564 | * - folio_memcg_lock() | |
1565 | * - folio frozen (refcount of 0) | |
1566 | * | |
1567 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1568 | * disabled. | |
1569 | */ | |
1570 | struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio, | |
1571 | unsigned long *flags) | |
6168d0da | 1572 | { |
e809c3fe | 1573 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1574 | |
6168d0da | 1575 | spin_lock_irqsave(&lruvec->lru_lock, *flags); |
e809c3fe | 1576 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1577 | |
1578 | return lruvec; | |
1579 | } | |
1580 | ||
925b7673 | 1581 | /** |
fa9add64 HD |
1582 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1583 | * @lruvec: mem_cgroup per zone lru vector | |
1584 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1585 | * @zid: zone id of the accounted pages |
fa9add64 | 1586 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1587 | * |
ca707239 | 1588 | * This function must be called under lru_lock, just before a page is added |
07ca7606 | 1589 | * to or just after a page is removed from an lru list. |
3f58a829 | 1590 | */ |
fa9add64 | 1591 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1592 | int zid, int nr_pages) |
3f58a829 | 1593 | { |
ef8f2327 | 1594 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1595 | unsigned long *lru_size; |
ca707239 | 1596 | long size; |
3f58a829 MK |
1597 | |
1598 | if (mem_cgroup_disabled()) | |
1599 | return; | |
1600 | ||
ef8f2327 | 1601 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1602 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1603 | |
1604 | if (nr_pages < 0) | |
1605 | *lru_size += nr_pages; | |
1606 | ||
1607 | size = *lru_size; | |
b4536f0c MH |
1608 | if (WARN_ONCE(size < 0, |
1609 | "%s(%p, %d, %d): lru_size %ld\n", | |
1610 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1611 | VM_BUG_ON(1); |
1612 | *lru_size = 0; | |
1613 | } | |
1614 | ||
1615 | if (nr_pages > 0) | |
1616 | *lru_size += nr_pages; | |
08e552c6 | 1617 | } |
544122e5 | 1618 | |
19942822 | 1619 | /** |
9d11ea9f | 1620 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1621 | * @memcg: the memory cgroup |
19942822 | 1622 | * |
9d11ea9f | 1623 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1624 | * pages. |
19942822 | 1625 | */ |
c0ff4b85 | 1626 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1627 | { |
3e32cb2e JW |
1628 | unsigned long margin = 0; |
1629 | unsigned long count; | |
1630 | unsigned long limit; | |
9d11ea9f | 1631 | |
3e32cb2e | 1632 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1633 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1634 | if (count < limit) |
1635 | margin = limit - count; | |
1636 | ||
7941d214 | 1637 | if (do_memsw_account()) { |
3e32cb2e | 1638 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1639 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1640 | if (count < limit) |
3e32cb2e | 1641 | margin = min(margin, limit - count); |
cbedbac3 LR |
1642 | else |
1643 | margin = 0; | |
3e32cb2e JW |
1644 | } |
1645 | ||
1646 | return margin; | |
19942822 JW |
1647 | } |
1648 | ||
32047e2a | 1649 | /* |
bdcbb659 | 1650 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1651 | * |
bdcbb659 QH |
1652 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1653 | * moving cgroups. This is for waiting at high-memory pressure | |
1654 | * caused by "move". | |
32047e2a | 1655 | */ |
c0ff4b85 | 1656 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1657 | { |
2bd9bb20 KH |
1658 | struct mem_cgroup *from; |
1659 | struct mem_cgroup *to; | |
4b534334 | 1660 | bool ret = false; |
2bd9bb20 KH |
1661 | /* |
1662 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1663 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1664 | */ | |
1665 | spin_lock(&mc.lock); | |
1666 | from = mc.from; | |
1667 | to = mc.to; | |
1668 | if (!from) | |
1669 | goto unlock; | |
3e92041d | 1670 | |
2314b42d JW |
1671 | ret = mem_cgroup_is_descendant(from, memcg) || |
1672 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1673 | unlock: |
1674 | spin_unlock(&mc.lock); | |
4b534334 KH |
1675 | return ret; |
1676 | } | |
1677 | ||
c0ff4b85 | 1678 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1679 | { |
1680 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1681 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1682 | DEFINE_WAIT(wait); |
1683 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1684 | /* moving charge context might have finished. */ | |
1685 | if (mc.moving_task) | |
1686 | schedule(); | |
1687 | finish_wait(&mc.waitq, &wait); | |
1688 | return true; | |
1689 | } | |
1690 | } | |
1691 | return false; | |
1692 | } | |
1693 | ||
5f9a4f4a MS |
1694 | struct memory_stat { |
1695 | const char *name; | |
5f9a4f4a MS |
1696 | unsigned int idx; |
1697 | }; | |
1698 | ||
57b2847d | 1699 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1700 | { "anon", NR_ANON_MAPPED }, |
1701 | { "file", NR_FILE_PAGES }, | |
a8c49af3 | 1702 | { "kernel", MEMCG_KMEM }, |
fff66b79 MS |
1703 | { "kernel_stack", NR_KERNEL_STACK_KB }, |
1704 | { "pagetables", NR_PAGETABLE }, | |
ebc97a52 | 1705 | { "sec_pagetables", NR_SECONDARY_PAGETABLE }, |
fff66b79 MS |
1706 | { "percpu", MEMCG_PERCPU_B }, |
1707 | { "sock", MEMCG_SOCK }, | |
4e5aa1f4 | 1708 | { "vmalloc", MEMCG_VMALLOC }, |
fff66b79 | 1709 | { "shmem", NR_SHMEM }, |
f4840ccf JW |
1710 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
1711 | { "zswap", MEMCG_ZSWAP_B }, | |
1712 | { "zswapped", MEMCG_ZSWAPPED }, | |
1713 | #endif | |
fff66b79 MS |
1714 | { "file_mapped", NR_FILE_MAPPED }, |
1715 | { "file_dirty", NR_FILE_DIRTY }, | |
1716 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1717 | #ifdef CONFIG_SWAP |
1718 | { "swapcached", NR_SWAPCACHE }, | |
1719 | #endif | |
5f9a4f4a | 1720 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1721 | { "anon_thp", NR_ANON_THPS }, |
1722 | { "file_thp", NR_FILE_THPS }, | |
1723 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1724 | #endif |
fff66b79 MS |
1725 | { "inactive_anon", NR_INACTIVE_ANON }, |
1726 | { "active_anon", NR_ACTIVE_ANON }, | |
1727 | { "inactive_file", NR_INACTIVE_FILE }, | |
1728 | { "active_file", NR_ACTIVE_FILE }, | |
1729 | { "unevictable", NR_UNEVICTABLE }, | |
1730 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1731 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1732 | |
1733 | /* The memory events */ | |
fff66b79 MS |
1734 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1735 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1736 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1737 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1738 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1739 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1740 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1741 | }; |
1742 | ||
ff841a06 | 1743 | /* The actual unit of the state item, not the same as the output unit */ |
fff66b79 MS |
1744 | static int memcg_page_state_unit(int item) |
1745 | { | |
1746 | switch (item) { | |
1747 | case MEMCG_PERCPU_B: | |
f4840ccf | 1748 | case MEMCG_ZSWAP_B: |
fff66b79 MS |
1749 | case NR_SLAB_RECLAIMABLE_B: |
1750 | case NR_SLAB_UNRECLAIMABLE_B: | |
ff841a06 YA |
1751 | return 1; |
1752 | case NR_KERNEL_STACK_KB: | |
1753 | return SZ_1K; | |
1754 | default: | |
1755 | return PAGE_SIZE; | |
1756 | } | |
1757 | } | |
1758 | ||
1759 | /* Translate stat items to the correct unit for memory.stat output */ | |
1760 | static int memcg_page_state_output_unit(int item) | |
1761 | { | |
1762 | /* | |
1763 | * Workingset state is actually in pages, but we export it to userspace | |
1764 | * as a scalar count of events, so special case it here. | |
1765 | */ | |
1766 | switch (item) { | |
fff66b79 MS |
1767 | case WORKINGSET_REFAULT_ANON: |
1768 | case WORKINGSET_REFAULT_FILE: | |
1769 | case WORKINGSET_ACTIVATE_ANON: | |
1770 | case WORKINGSET_ACTIVATE_FILE: | |
1771 | case WORKINGSET_RESTORE_ANON: | |
1772 | case WORKINGSET_RESTORE_FILE: | |
1773 | case WORKINGSET_NODERECLAIM: | |
1774 | return 1; | |
fff66b79 | 1775 | default: |
ff841a06 | 1776 | return memcg_page_state_unit(item); |
fff66b79 MS |
1777 | } |
1778 | } | |
1779 | ||
1780 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1781 | int item) | |
1782 | { | |
ff841a06 YA |
1783 | return memcg_page_state(memcg, item) * |
1784 | memcg_page_state_output_unit(item); | |
1785 | } | |
1786 | ||
1787 | static inline unsigned long memcg_page_state_local_output( | |
1788 | struct mem_cgroup *memcg, int item) | |
1789 | { | |
1790 | return memcg_page_state_local(memcg, item) * | |
1791 | memcg_page_state_output_unit(item); | |
fff66b79 MS |
1792 | } |
1793 | ||
dddb44ff | 1794 | static void memcg_stat_format(struct mem_cgroup *memcg, struct seq_buf *s) |
c8713d0b | 1795 | { |
c8713d0b | 1796 | int i; |
71cd3113 | 1797 | |
c8713d0b JW |
1798 | /* |
1799 | * Provide statistics on the state of the memory subsystem as | |
1800 | * well as cumulative event counters that show past behavior. | |
1801 | * | |
1802 | * This list is ordered following a combination of these gradients: | |
1803 | * 1) generic big picture -> specifics and details | |
1804 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1805 | * | |
1806 | * Current memory state: | |
1807 | */ | |
7d7ef0a4 | 1808 | mem_cgroup_flush_stats(memcg); |
c8713d0b | 1809 | |
5f9a4f4a MS |
1810 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1811 | u64 size; | |
c8713d0b | 1812 | |
fff66b79 | 1813 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5b42360c | 1814 | seq_buf_printf(s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1815 | |
5f9a4f4a | 1816 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1817 | size += memcg_page_state_output(memcg, |
1818 | NR_SLAB_RECLAIMABLE_B); | |
5b42360c | 1819 | seq_buf_printf(s, "slab %llu\n", size); |
5f9a4f4a MS |
1820 | } |
1821 | } | |
c8713d0b JW |
1822 | |
1823 | /* Accumulated memory events */ | |
5b42360c | 1824 | seq_buf_printf(s, "pgscan %lu\n", |
c8713d0b | 1825 | memcg_events(memcg, PGSCAN_KSWAPD) + |
57e9cc50 JW |
1826 | memcg_events(memcg, PGSCAN_DIRECT) + |
1827 | memcg_events(memcg, PGSCAN_KHUGEPAGED)); | |
5b42360c | 1828 | seq_buf_printf(s, "pgsteal %lu\n", |
c8713d0b | 1829 | memcg_events(memcg, PGSTEAL_KSWAPD) + |
57e9cc50 JW |
1830 | memcg_events(memcg, PGSTEAL_DIRECT) + |
1831 | memcg_events(memcg, PGSTEAL_KHUGEPAGED)); | |
c8713d0b | 1832 | |
8278f1c7 SB |
1833 | for (i = 0; i < ARRAY_SIZE(memcg_vm_event_stat); i++) { |
1834 | if (memcg_vm_event_stat[i] == PGPGIN || | |
1835 | memcg_vm_event_stat[i] == PGPGOUT) | |
1836 | continue; | |
1837 | ||
5b42360c | 1838 | seq_buf_printf(s, "%s %lu\n", |
673520f8 QZ |
1839 | vm_event_name(memcg_vm_event_stat[i]), |
1840 | memcg_events(memcg, memcg_vm_event_stat[i])); | |
8278f1c7 | 1841 | } |
c8713d0b JW |
1842 | |
1843 | /* The above should easily fit into one page */ | |
5b42360c | 1844 | WARN_ON_ONCE(seq_buf_has_overflowed(s)); |
c8713d0b | 1845 | } |
71cd3113 | 1846 | |
dddb44ff YA |
1847 | static void memcg1_stat_format(struct mem_cgroup *memcg, struct seq_buf *s); |
1848 | ||
1849 | static void memory_stat_format(struct mem_cgroup *memcg, struct seq_buf *s) | |
1850 | { | |
1851 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1852 | memcg_stat_format(memcg, s); | |
1853 | else | |
1854 | memcg1_stat_format(memcg, s); | |
1855 | WARN_ON_ONCE(seq_buf_has_overflowed(s)); | |
1856 | } | |
1857 | ||
e222432b | 1858 | /** |
f0c867d9 | 1859 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1860 | * memory controller. | |
e222432b BS |
1861 | * @memcg: The memory cgroup that went over limit |
1862 | * @p: Task that is going to be killed | |
1863 | * | |
1864 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1865 | * enabled | |
1866 | */ | |
f0c867d9 | 1867 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1868 | { |
e222432b BS |
1869 | rcu_read_lock(); |
1870 | ||
f0c867d9 | 1871 | if (memcg) { |
1872 | pr_cont(",oom_memcg="); | |
1873 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1874 | } else | |
1875 | pr_cont(",global_oom"); | |
2415b9f5 | 1876 | if (p) { |
f0c867d9 | 1877 | pr_cont(",task_memcg="); |
2415b9f5 | 1878 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1879 | } |
e222432b | 1880 | rcu_read_unlock(); |
f0c867d9 | 1881 | } |
1882 | ||
1883 | /** | |
1884 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1885 | * memory controller. | |
1886 | * @memcg: The memory cgroup that went over limit | |
1887 | */ | |
1888 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1889 | { | |
68aaee14 TH |
1890 | /* Use static buffer, for the caller is holding oom_lock. */ |
1891 | static char buf[PAGE_SIZE]; | |
5b42360c | 1892 | struct seq_buf s; |
68aaee14 TH |
1893 | |
1894 | lockdep_assert_held(&oom_lock); | |
e222432b | 1895 | |
3e32cb2e JW |
1896 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1897 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1898 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1899 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1900 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1901 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1902 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1903 | else { |
1904 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1905 | K((u64)page_counter_read(&memcg->memsw)), | |
1906 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1907 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1908 | K((u64)page_counter_read(&memcg->kmem)), | |
1909 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1910 | } |
c8713d0b JW |
1911 | |
1912 | pr_info("Memory cgroup stats for "); | |
1913 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1914 | pr_cont(":"); | |
5b42360c YA |
1915 | seq_buf_init(&s, buf, sizeof(buf)); |
1916 | memory_stat_format(memcg, &s); | |
1917 | seq_buf_do_printk(&s, KERN_INFO); | |
e222432b BS |
1918 | } |
1919 | ||
a63d83f4 DR |
1920 | /* |
1921 | * Return the memory (and swap, if configured) limit for a memcg. | |
1922 | */ | |
bbec2e15 | 1923 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1924 | { |
8d387a5f WL |
1925 | unsigned long max = READ_ONCE(memcg->memory.max); |
1926 | ||
b94c4e94 | 1927 | if (do_memsw_account()) { |
8d387a5f WL |
1928 | if (mem_cgroup_swappiness(memcg)) { |
1929 | /* Calculate swap excess capacity from memsw limit */ | |
1930 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1931 | ||
1932 | max += min(swap, (unsigned long)total_swap_pages); | |
1933 | } | |
b94c4e94 JW |
1934 | } else { |
1935 | if (mem_cgroup_swappiness(memcg)) | |
1936 | max += min(READ_ONCE(memcg->swap.max), | |
1937 | (unsigned long)total_swap_pages); | |
9a5a8f19 | 1938 | } |
bbec2e15 | 1939 | return max; |
a63d83f4 DR |
1940 | } |
1941 | ||
9783aa99 CD |
1942 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1943 | { | |
1944 | return page_counter_read(&memcg->memory); | |
1945 | } | |
1946 | ||
b6e6edcf | 1947 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1948 | int order) |
9cbb78bb | 1949 | { |
6e0fc46d DR |
1950 | struct oom_control oc = { |
1951 | .zonelist = NULL, | |
1952 | .nodemask = NULL, | |
2a966b77 | 1953 | .memcg = memcg, |
6e0fc46d DR |
1954 | .gfp_mask = gfp_mask, |
1955 | .order = order, | |
6e0fc46d | 1956 | }; |
1378b37d | 1957 | bool ret = true; |
9cbb78bb | 1958 | |
7775face TH |
1959 | if (mutex_lock_killable(&oom_lock)) |
1960 | return true; | |
1378b37d YS |
1961 | |
1962 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1963 | goto unlock; | |
1964 | ||
7775face TH |
1965 | /* |
1966 | * A few threads which were not waiting at mutex_lock_killable() can | |
1967 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1968 | */ | |
a4ebf1b6 | 1969 | ret = task_is_dying() || out_of_memory(&oc); |
1378b37d YS |
1970 | |
1971 | unlock: | |
dc56401f | 1972 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1973 | return ret; |
9cbb78bb DR |
1974 | } |
1975 | ||
0608f43d | 1976 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1977 | pg_data_t *pgdat, |
0608f43d AM |
1978 | gfp_t gfp_mask, |
1979 | unsigned long *total_scanned) | |
1980 | { | |
1981 | struct mem_cgroup *victim = NULL; | |
1982 | int total = 0; | |
1983 | int loop = 0; | |
1984 | unsigned long excess; | |
1985 | unsigned long nr_scanned; | |
1986 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1987 | .pgdat = pgdat, |
0608f43d AM |
1988 | }; |
1989 | ||
3e32cb2e | 1990 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1991 | |
1992 | while (1) { | |
1993 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1994 | if (!victim) { | |
1995 | loop++; | |
1996 | if (loop >= 2) { | |
1997 | /* | |
1998 | * If we have not been able to reclaim | |
1999 | * anything, it might because there are | |
2000 | * no reclaimable pages under this hierarchy | |
2001 | */ | |
2002 | if (!total) | |
2003 | break; | |
2004 | /* | |
2005 | * We want to do more targeted reclaim. | |
2006 | * excess >> 2 is not to excessive so as to | |
2007 | * reclaim too much, nor too less that we keep | |
2008 | * coming back to reclaim from this cgroup | |
2009 | */ | |
2010 | if (total >= (excess >> 2) || | |
2011 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2012 | break; | |
2013 | } | |
2014 | continue; | |
2015 | } | |
a9dd0a83 | 2016 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 2017 | pgdat, &nr_scanned); |
0608f43d | 2018 | *total_scanned += nr_scanned; |
3e32cb2e | 2019 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 2020 | break; |
6d61ef40 | 2021 | } |
0608f43d AM |
2022 | mem_cgroup_iter_break(root_memcg, victim); |
2023 | return total; | |
6d61ef40 BS |
2024 | } |
2025 | ||
0056f4e6 JW |
2026 | #ifdef CONFIG_LOCKDEP |
2027 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2028 | .name = "memcg_oom_lock", | |
2029 | }; | |
2030 | #endif | |
2031 | ||
fb2a6fc5 JW |
2032 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2033 | ||
867578cb KH |
2034 | /* |
2035 | * Check OOM-Killer is already running under our hierarchy. | |
2036 | * If someone is running, return false. | |
2037 | */ | |
fb2a6fc5 | 2038 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2039 | { |
79dfdacc | 2040 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2041 | |
fb2a6fc5 JW |
2042 | spin_lock(&memcg_oom_lock); |
2043 | ||
9f3a0d09 | 2044 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2045 | if (iter->oom_lock) { |
79dfdacc MH |
2046 | /* |
2047 | * this subtree of our hierarchy is already locked | |
2048 | * so we cannot give a lock. | |
2049 | */ | |
79dfdacc | 2050 | failed = iter; |
9f3a0d09 JW |
2051 | mem_cgroup_iter_break(memcg, iter); |
2052 | break; | |
23751be0 JW |
2053 | } else |
2054 | iter->oom_lock = true; | |
7d74b06f | 2055 | } |
867578cb | 2056 | |
fb2a6fc5 JW |
2057 | if (failed) { |
2058 | /* | |
2059 | * OK, we failed to lock the whole subtree so we have | |
2060 | * to clean up what we set up to the failing subtree | |
2061 | */ | |
2062 | for_each_mem_cgroup_tree(iter, memcg) { | |
2063 | if (iter == failed) { | |
2064 | mem_cgroup_iter_break(memcg, iter); | |
2065 | break; | |
2066 | } | |
2067 | iter->oom_lock = false; | |
79dfdacc | 2068 | } |
0056f4e6 JW |
2069 | } else |
2070 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2071 | |
2072 | spin_unlock(&memcg_oom_lock); | |
2073 | ||
2074 | return !failed; | |
a636b327 | 2075 | } |
0b7f569e | 2076 | |
fb2a6fc5 | 2077 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2078 | { |
7d74b06f KH |
2079 | struct mem_cgroup *iter; |
2080 | ||
fb2a6fc5 | 2081 | spin_lock(&memcg_oom_lock); |
5facae4f | 2082 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 2083 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2084 | iter->oom_lock = false; |
fb2a6fc5 | 2085 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2086 | } |
2087 | ||
c0ff4b85 | 2088 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2089 | { |
2090 | struct mem_cgroup *iter; | |
2091 | ||
c2b42d3c | 2092 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2093 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
2094 | iter->under_oom++; |
2095 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
2096 | } |
2097 | ||
c0ff4b85 | 2098 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2099 | { |
2100 | struct mem_cgroup *iter; | |
2101 | ||
867578cb | 2102 | /* |
f0953a1b | 2103 | * Be careful about under_oom underflows because a child memcg |
7a52d4d8 | 2104 | * could have been added after mem_cgroup_mark_under_oom. |
867578cb | 2105 | */ |
c2b42d3c | 2106 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2107 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
2108 | if (iter->under_oom > 0) |
2109 | iter->under_oom--; | |
2110 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
2111 | } |
2112 | ||
867578cb KH |
2113 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2114 | ||
dc98df5a | 2115 | struct oom_wait_info { |
d79154bb | 2116 | struct mem_cgroup *memcg; |
ac6424b9 | 2117 | wait_queue_entry_t wait; |
dc98df5a KH |
2118 | }; |
2119 | ||
ac6424b9 | 2120 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
2121 | unsigned mode, int sync, void *arg) |
2122 | { | |
d79154bb HD |
2123 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2124 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2125 | struct oom_wait_info *oom_wait_info; |
2126 | ||
2127 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2128 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2129 | |
2314b42d JW |
2130 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
2131 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 2132 | return 0; |
dc98df5a KH |
2133 | return autoremove_wake_function(wait, mode, sync, arg); |
2134 | } | |
2135 | ||
c0ff4b85 | 2136 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2137 | { |
c2b42d3c TH |
2138 | /* |
2139 | * For the following lockless ->under_oom test, the only required | |
2140 | * guarantee is that it must see the state asserted by an OOM when | |
2141 | * this function is called as a result of userland actions | |
2142 | * triggered by the notification of the OOM. This is trivially | |
2143 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
2144 | * triggering notification. | |
2145 | */ | |
2146 | if (memcg && memcg->under_oom) | |
f4b90b70 | 2147 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
2148 | } |
2149 | ||
becdf89d SB |
2150 | /* |
2151 | * Returns true if successfully killed one or more processes. Though in some | |
2152 | * corner cases it can return true even without killing any process. | |
2153 | */ | |
2154 | static bool mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 2155 | { |
becdf89d | 2156 | bool locked, ret; |
7056d3a3 | 2157 | |
29ef680a | 2158 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
becdf89d | 2159 | return false; |
29ef680a | 2160 | |
7a1adfdd RG |
2161 | memcg_memory_event(memcg, MEMCG_OOM); |
2162 | ||
867578cb | 2163 | /* |
49426420 JW |
2164 | * We are in the middle of the charge context here, so we |
2165 | * don't want to block when potentially sitting on a callstack | |
2166 | * that holds all kinds of filesystem and mm locks. | |
2167 | * | |
29ef680a MH |
2168 | * cgroup1 allows disabling the OOM killer and waiting for outside |
2169 | * handling until the charge can succeed; remember the context and put | |
2170 | * the task to sleep at the end of the page fault when all locks are | |
2171 | * released. | |
49426420 | 2172 | * |
29ef680a MH |
2173 | * On the other hand, in-kernel OOM killer allows for an async victim |
2174 | * memory reclaim (oom_reaper) and that means that we are not solely | |
2175 | * relying on the oom victim to make a forward progress and we can | |
2176 | * invoke the oom killer here. | |
2177 | * | |
2178 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
2179 | * victim and then we have to bail out from the charge path. | |
867578cb | 2180 | */ |
17c56de6 | 2181 | if (READ_ONCE(memcg->oom_kill_disable)) { |
becdf89d SB |
2182 | if (current->in_user_fault) { |
2183 | css_get(&memcg->css); | |
2184 | current->memcg_in_oom = memcg; | |
2185 | current->memcg_oom_gfp_mask = mask; | |
2186 | current->memcg_oom_order = order; | |
2187 | } | |
2188 | return false; | |
29ef680a MH |
2189 | } |
2190 | ||
7056d3a3 MH |
2191 | mem_cgroup_mark_under_oom(memcg); |
2192 | ||
2193 | locked = mem_cgroup_oom_trylock(memcg); | |
2194 | ||
2195 | if (locked) | |
2196 | mem_cgroup_oom_notify(memcg); | |
2197 | ||
2198 | mem_cgroup_unmark_under_oom(memcg); | |
becdf89d | 2199 | ret = mem_cgroup_out_of_memory(memcg, mask, order); |
7056d3a3 MH |
2200 | |
2201 | if (locked) | |
2202 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 2203 | |
7056d3a3 | 2204 | return ret; |
3812c8c8 JW |
2205 | } |
2206 | ||
2207 | /** | |
2208 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2209 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2210 | * |
49426420 JW |
2211 | * This has to be called at the end of a page fault if the memcg OOM |
2212 | * handler was enabled. | |
3812c8c8 | 2213 | * |
49426420 | 2214 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2215 | * sleep on a waitqueue until the userspace task resolves the |
2216 | * situation. Sleeping directly in the charge context with all kinds | |
2217 | * of locks held is not a good idea, instead we remember an OOM state | |
2218 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2219 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2220 | * |
2221 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2222 | * completed, %false otherwise. |
3812c8c8 | 2223 | */ |
49426420 | 2224 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2225 | { |
626ebc41 | 2226 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 2227 | struct oom_wait_info owait; |
49426420 | 2228 | bool locked; |
3812c8c8 JW |
2229 | |
2230 | /* OOM is global, do not handle */ | |
3812c8c8 | 2231 | if (!memcg) |
49426420 | 2232 | return false; |
3812c8c8 | 2233 | |
7c5f64f8 | 2234 | if (!handle) |
49426420 | 2235 | goto cleanup; |
3812c8c8 JW |
2236 | |
2237 | owait.memcg = memcg; | |
2238 | owait.wait.flags = 0; | |
2239 | owait.wait.func = memcg_oom_wake_function; | |
2240 | owait.wait.private = current; | |
2055da97 | 2241 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 2242 | |
3812c8c8 | 2243 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2244 | mem_cgroup_mark_under_oom(memcg); |
2245 | ||
2246 | locked = mem_cgroup_oom_trylock(memcg); | |
2247 | ||
2248 | if (locked) | |
2249 | mem_cgroup_oom_notify(memcg); | |
2250 | ||
857f2139 HX |
2251 | schedule(); |
2252 | mem_cgroup_unmark_under_oom(memcg); | |
2253 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
49426420 | 2254 | |
18b1d18b | 2255 | if (locked) |
fb2a6fc5 | 2256 | mem_cgroup_oom_unlock(memcg); |
49426420 | 2257 | cleanup: |
626ebc41 | 2258 | current->memcg_in_oom = NULL; |
3812c8c8 | 2259 | css_put(&memcg->css); |
867578cb | 2260 | return true; |
0b7f569e KH |
2261 | } |
2262 | ||
3d8b38eb RG |
2263 | /** |
2264 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2265 | * @victim: task to be killed by the OOM killer | |
2266 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2267 | * | |
2268 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2269 | * by killing all belonging OOM-killable tasks. | |
2270 | * | |
2271 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2272 | */ | |
2273 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2274 | struct mem_cgroup *oom_domain) | |
2275 | { | |
2276 | struct mem_cgroup *oom_group = NULL; | |
2277 | struct mem_cgroup *memcg; | |
2278 | ||
2279 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2280 | return NULL; | |
2281 | ||
2282 | if (!oom_domain) | |
2283 | oom_domain = root_mem_cgroup; | |
2284 | ||
2285 | rcu_read_lock(); | |
2286 | ||
2287 | memcg = mem_cgroup_from_task(victim); | |
7848ed62 | 2288 | if (mem_cgroup_is_root(memcg)) |
3d8b38eb RG |
2289 | goto out; |
2290 | ||
48fe267c RG |
2291 | /* |
2292 | * If the victim task has been asynchronously moved to a different | |
2293 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2294 | * In this case it's better to ignore memory.group.oom. | |
2295 | */ | |
2296 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2297 | goto out; | |
2298 | ||
3d8b38eb RG |
2299 | /* |
2300 | * Traverse the memory cgroup hierarchy from the victim task's | |
2301 | * cgroup up to the OOMing cgroup (or root) to find the | |
2302 | * highest-level memory cgroup with oom.group set. | |
2303 | */ | |
2304 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
eaf7b66b | 2305 | if (READ_ONCE(memcg->oom_group)) |
3d8b38eb RG |
2306 | oom_group = memcg; |
2307 | ||
2308 | if (memcg == oom_domain) | |
2309 | break; | |
2310 | } | |
2311 | ||
2312 | if (oom_group) | |
2313 | css_get(&oom_group->css); | |
2314 | out: | |
2315 | rcu_read_unlock(); | |
2316 | ||
2317 | return oom_group; | |
2318 | } | |
2319 | ||
2320 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2321 | { | |
2322 | pr_info("Tasks in "); | |
2323 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2324 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2325 | } | |
2326 | ||
d7365e78 | 2327 | /** |
f70ad448 MWO |
2328 | * folio_memcg_lock - Bind a folio to its memcg. |
2329 | * @folio: The folio. | |
32047e2a | 2330 | * |
f70ad448 | 2331 | * This function prevents unlocked LRU folios from being moved to |
739f79fc JW |
2332 | * another cgroup. |
2333 | * | |
f70ad448 MWO |
2334 | * It ensures lifetime of the bound memcg. The caller is responsible |
2335 | * for the lifetime of the folio. | |
d69b042f | 2336 | */ |
f70ad448 | 2337 | void folio_memcg_lock(struct folio *folio) |
89c06bd5 KH |
2338 | { |
2339 | struct mem_cgroup *memcg; | |
6de22619 | 2340 | unsigned long flags; |
89c06bd5 | 2341 | |
6de22619 JW |
2342 | /* |
2343 | * The RCU lock is held throughout the transaction. The fast | |
2344 | * path can get away without acquiring the memcg->move_lock | |
2345 | * because page moving starts with an RCU grace period. | |
739f79fc | 2346 | */ |
d7365e78 JW |
2347 | rcu_read_lock(); |
2348 | ||
2349 | if (mem_cgroup_disabled()) | |
1c824a68 | 2350 | return; |
89c06bd5 | 2351 | again: |
f70ad448 | 2352 | memcg = folio_memcg(folio); |
29833315 | 2353 | if (unlikely(!memcg)) |
1c824a68 | 2354 | return; |
d7365e78 | 2355 | |
20ad50d6 AS |
2356 | #ifdef CONFIG_PROVE_LOCKING |
2357 | local_irq_save(flags); | |
2358 | might_lock(&memcg->move_lock); | |
2359 | local_irq_restore(flags); | |
2360 | #endif | |
2361 | ||
bdcbb659 | 2362 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 2363 | return; |
89c06bd5 | 2364 | |
6de22619 | 2365 | spin_lock_irqsave(&memcg->move_lock, flags); |
f70ad448 | 2366 | if (memcg != folio_memcg(folio)) { |
6de22619 | 2367 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2368 | goto again; |
2369 | } | |
6de22619 JW |
2370 | |
2371 | /* | |
1c824a68 JW |
2372 | * When charge migration first begins, we can have multiple |
2373 | * critical sections holding the fast-path RCU lock and one | |
2374 | * holding the slowpath move_lock. Track the task who has the | |
6c77b607 | 2375 | * move_lock for folio_memcg_unlock(). |
6de22619 JW |
2376 | */ |
2377 | memcg->move_lock_task = current; | |
2378 | memcg->move_lock_flags = flags; | |
89c06bd5 | 2379 | } |
f70ad448 | 2380 | |
f70ad448 | 2381 | static void __folio_memcg_unlock(struct mem_cgroup *memcg) |
89c06bd5 | 2382 | { |
6de22619 JW |
2383 | if (memcg && memcg->move_lock_task == current) { |
2384 | unsigned long flags = memcg->move_lock_flags; | |
2385 | ||
2386 | memcg->move_lock_task = NULL; | |
2387 | memcg->move_lock_flags = 0; | |
2388 | ||
2389 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2390 | } | |
89c06bd5 | 2391 | |
d7365e78 | 2392 | rcu_read_unlock(); |
89c06bd5 | 2393 | } |
739f79fc JW |
2394 | |
2395 | /** | |
f70ad448 MWO |
2396 | * folio_memcg_unlock - Release the binding between a folio and its memcg. |
2397 | * @folio: The folio. | |
2398 | * | |
2399 | * This releases the binding created by folio_memcg_lock(). This does | |
2400 | * not change the accounting of this folio to its memcg, but it does | |
2401 | * permit others to change it. | |
739f79fc | 2402 | */ |
f70ad448 | 2403 | void folio_memcg_unlock(struct folio *folio) |
739f79fc | 2404 | { |
f70ad448 MWO |
2405 | __folio_memcg_unlock(folio_memcg(folio)); |
2406 | } | |
9da7b521 | 2407 | |
fead2b86 | 2408 | struct memcg_stock_pcp { |
56751146 | 2409 | local_lock_t stock_lock; |
fead2b86 MH |
2410 | struct mem_cgroup *cached; /* this never be root cgroup */ |
2411 | unsigned int nr_pages; | |
2412 | ||
bf4f0599 RG |
2413 | #ifdef CONFIG_MEMCG_KMEM |
2414 | struct obj_cgroup *cached_objcg; | |
68ac5b3c | 2415 | struct pglist_data *cached_pgdat; |
bf4f0599 | 2416 | unsigned int nr_bytes; |
68ac5b3c WL |
2417 | int nr_slab_reclaimable_b; |
2418 | int nr_slab_unreclaimable_b; | |
bf4f0599 RG |
2419 | #endif |
2420 | ||
cdec2e42 | 2421 | struct work_struct work; |
26fe6168 | 2422 | unsigned long flags; |
a0db00fc | 2423 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 | 2424 | }; |
56751146 SAS |
2425 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = { |
2426 | .stock_lock = INIT_LOCAL_LOCK(stock_lock), | |
2427 | }; | |
9f50fad6 | 2428 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2429 | |
bf4f0599 | 2430 | #ifdef CONFIG_MEMCG_KMEM |
56751146 | 2431 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock); |
bf4f0599 RG |
2432 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
2433 | struct mem_cgroup *root_memcg); | |
a8c49af3 | 2434 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages); |
bf4f0599 RG |
2435 | |
2436 | #else | |
56751146 | 2437 | static inline struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 2438 | { |
56751146 | 2439 | return NULL; |
bf4f0599 RG |
2440 | } |
2441 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2442 | struct mem_cgroup *root_memcg) | |
2443 | { | |
2444 | return false; | |
2445 | } | |
a8c49af3 YA |
2446 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
2447 | { | |
2448 | } | |
bf4f0599 RG |
2449 | #endif |
2450 | ||
a0956d54 SS |
2451 | /** |
2452 | * consume_stock: Try to consume stocked charge on this cpu. | |
2453 | * @memcg: memcg to consume from. | |
2454 | * @nr_pages: how many pages to charge. | |
2455 | * | |
2456 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2457 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2458 | * service an allocation will refill the stock. | |
2459 | * | |
2460 | * returns true if successful, false otherwise. | |
cdec2e42 | 2461 | */ |
a0956d54 | 2462 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2463 | { |
2464 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2465 | unsigned long flags; |
3e32cb2e | 2466 | bool ret = false; |
cdec2e42 | 2467 | |
a983b5eb | 2468 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2469 | return ret; |
a0956d54 | 2470 | |
56751146 | 2471 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2472 | |
2473 | stock = this_cpu_ptr(&memcg_stock); | |
f785a8f2 | 2474 | if (memcg == READ_ONCE(stock->cached) && stock->nr_pages >= nr_pages) { |
a0956d54 | 2475 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2476 | ret = true; |
2477 | } | |
db2ba40c | 2478 | |
56751146 | 2479 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
db2ba40c | 2480 | |
cdec2e42 KH |
2481 | return ret; |
2482 | } | |
2483 | ||
2484 | /* | |
3e32cb2e | 2485 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2486 | */ |
2487 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2488 | { | |
f785a8f2 | 2489 | struct mem_cgroup *old = READ_ONCE(stock->cached); |
cdec2e42 | 2490 | |
1a3e1f40 JW |
2491 | if (!old) |
2492 | return; | |
2493 | ||
11c9ea4e | 2494 | if (stock->nr_pages) { |
3e32cb2e | 2495 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2496 | if (do_memsw_account()) |
3e32cb2e | 2497 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2498 | stock->nr_pages = 0; |
cdec2e42 | 2499 | } |
1a3e1f40 JW |
2500 | |
2501 | css_put(&old->css); | |
f785a8f2 | 2502 | WRITE_ONCE(stock->cached, NULL); |
cdec2e42 KH |
2503 | } |
2504 | ||
cdec2e42 KH |
2505 | static void drain_local_stock(struct work_struct *dummy) |
2506 | { | |
db2ba40c | 2507 | struct memcg_stock_pcp *stock; |
56751146 | 2508 | struct obj_cgroup *old = NULL; |
db2ba40c JW |
2509 | unsigned long flags; |
2510 | ||
72f0184c | 2511 | /* |
5c49cf9a MH |
2512 | * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs. |
2513 | * drain_stock races is that we always operate on local CPU stock | |
2514 | * here with IRQ disabled | |
72f0184c | 2515 | */ |
56751146 | 2516 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2517 | |
2518 | stock = this_cpu_ptr(&memcg_stock); | |
56751146 | 2519 | old = drain_obj_stock(stock); |
cdec2e42 | 2520 | drain_stock(stock); |
26fe6168 | 2521 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c | 2522 | |
56751146 | 2523 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 2524 | obj_cgroup_put(old); |
cdec2e42 KH |
2525 | } |
2526 | ||
2527 | /* | |
3e32cb2e | 2528 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2529 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2530 | */ |
af9a3b69 | 2531 | static void __refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2532 | { |
db2ba40c | 2533 | struct memcg_stock_pcp *stock; |
cdec2e42 | 2534 | |
db2ba40c | 2535 | stock = this_cpu_ptr(&memcg_stock); |
f785a8f2 | 2536 | if (READ_ONCE(stock->cached) != memcg) { /* reset if necessary */ |
cdec2e42 | 2537 | drain_stock(stock); |
1a3e1f40 | 2538 | css_get(&memcg->css); |
f785a8f2 | 2539 | WRITE_ONCE(stock->cached, memcg); |
cdec2e42 | 2540 | } |
11c9ea4e | 2541 | stock->nr_pages += nr_pages; |
db2ba40c | 2542 | |
a983b5eb | 2543 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 | 2544 | drain_stock(stock); |
af9a3b69 JW |
2545 | } |
2546 | ||
2547 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) | |
2548 | { | |
2549 | unsigned long flags; | |
475d0487 | 2550 | |
56751146 | 2551 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
af9a3b69 | 2552 | __refill_stock(memcg, nr_pages); |
56751146 | 2553 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
cdec2e42 KH |
2554 | } |
2555 | ||
2556 | /* | |
c0ff4b85 | 2557 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2558 | * of the hierarchy under it. |
cdec2e42 | 2559 | */ |
6d3d6aa2 | 2560 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2561 | { |
26fe6168 | 2562 | int cpu, curcpu; |
d38144b7 | 2563 | |
6d3d6aa2 JW |
2564 | /* If someone's already draining, avoid adding running more workers. */ |
2565 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2566 | return; | |
72f0184c MH |
2567 | /* |
2568 | * Notify other cpus that system-wide "drain" is running | |
2569 | * We do not care about races with the cpu hotplug because cpu down | |
2570 | * as well as workers from this path always operate on the local | |
2571 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2572 | */ | |
0790ed62 SAS |
2573 | migrate_disable(); |
2574 | curcpu = smp_processor_id(); | |
cdec2e42 KH |
2575 | for_each_online_cpu(cpu) { |
2576 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2577 | struct mem_cgroup *memcg; |
e1a366be | 2578 | bool flush = false; |
26fe6168 | 2579 | |
e1a366be | 2580 | rcu_read_lock(); |
f785a8f2 | 2581 | memcg = READ_ONCE(stock->cached); |
e1a366be RG |
2582 | if (memcg && stock->nr_pages && |
2583 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2584 | flush = true; | |
27fb0956 | 2585 | else if (obj_stock_flush_required(stock, root_memcg)) |
bf4f0599 | 2586 | flush = true; |
e1a366be RG |
2587 | rcu_read_unlock(); |
2588 | ||
2589 | if (flush && | |
2590 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2591 | if (cpu == curcpu) |
2592 | drain_local_stock(&stock->work); | |
6a792697 | 2593 | else if (!cpu_is_isolated(cpu)) |
d1a05b69 MH |
2594 | schedule_work_on(cpu, &stock->work); |
2595 | } | |
cdec2e42 | 2596 | } |
0790ed62 | 2597 | migrate_enable(); |
9f50fad6 | 2598 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2599 | } |
2600 | ||
2cd21c89 JW |
2601 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
2602 | { | |
2603 | struct memcg_stock_pcp *stock; | |
a3d4c05a | 2604 | |
2cd21c89 JW |
2605 | stock = &per_cpu(memcg_stock, cpu); |
2606 | drain_stock(stock); | |
a3d4c05a | 2607 | |
308167fc | 2608 | return 0; |
cdec2e42 KH |
2609 | } |
2610 | ||
b3ff9291 CD |
2611 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2612 | unsigned int nr_pages, | |
2613 | gfp_t gfp_mask) | |
f7e1cb6e | 2614 | { |
b3ff9291 CD |
2615 | unsigned long nr_reclaimed = 0; |
2616 | ||
f7e1cb6e | 2617 | do { |
e22c6ed9 JW |
2618 | unsigned long pflags; |
2619 | ||
d1663a90 JK |
2620 | if (page_counter_read(&memcg->memory) <= |
2621 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2622 | continue; |
e22c6ed9 | 2623 | |
e27be240 | 2624 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2625 | |
2626 | psi_memstall_enter(&pflags); | |
b3ff9291 | 2627 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
73b73bac | 2628 | gfp_mask, |
55ab834a | 2629 | MEMCG_RECLAIM_MAY_SWAP); |
e22c6ed9 | 2630 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2631 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2632 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2633 | |
2634 | return nr_reclaimed; | |
f7e1cb6e JW |
2635 | } |
2636 | ||
2637 | static void high_work_func(struct work_struct *work) | |
2638 | { | |
2639 | struct mem_cgroup *memcg; | |
2640 | ||
2641 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2642 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2643 | } |
2644 | ||
0e4b01df CD |
2645 | /* |
2646 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2647 | * enough to still cause a significant slowdown in most cases, while still | |
2648 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2649 | */ | |
2650 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2651 | ||
2652 | /* | |
2653 | * When calculating the delay, we use these either side of the exponentiation to | |
2654 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2655 | * below. | |
2656 | * | |
2657 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2658 | * overage ratio to a delay. | |
ac5ddd0f | 2659 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2660 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2661 | * to produce a reasonable delay curve. | |
2662 | * | |
2663 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2664 | * reasonable delay curve compared to precision-adjusted overage, not | |
2665 | * penalising heavily at first, but still making sure that growth beyond the | |
2666 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2667 | * example, with a high of 100 megabytes: | |
2668 | * | |
2669 | * +-------+------------------------+ | |
2670 | * | usage | time to allocate in ms | | |
2671 | * +-------+------------------------+ | |
2672 | * | 100M | 0 | | |
2673 | * | 101M | 6 | | |
2674 | * | 102M | 25 | | |
2675 | * | 103M | 57 | | |
2676 | * | 104M | 102 | | |
2677 | * | 105M | 159 | | |
2678 | * | 106M | 230 | | |
2679 | * | 107M | 313 | | |
2680 | * | 108M | 409 | | |
2681 | * | 109M | 518 | | |
2682 | * | 110M | 639 | | |
2683 | * | 111M | 774 | | |
2684 | * | 112M | 921 | | |
2685 | * | 113M | 1081 | | |
2686 | * | 114M | 1254 | | |
2687 | * | 115M | 1439 | | |
2688 | * | 116M | 1638 | | |
2689 | * | 117M | 1849 | | |
2690 | * | 118M | 2000 | | |
2691 | * | 119M | 2000 | | |
2692 | * | 120M | 2000 | | |
2693 | * +-------+------------------------+ | |
2694 | */ | |
2695 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2696 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2697 | ||
8a5dbc65 | 2698 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2699 | { |
8a5dbc65 | 2700 | u64 overage; |
b23afb93 | 2701 | |
8a5dbc65 JK |
2702 | if (usage <= high) |
2703 | return 0; | |
e26733e0 | 2704 | |
8a5dbc65 JK |
2705 | /* |
2706 | * Prevent division by 0 in overage calculation by acting as if | |
2707 | * it was a threshold of 1 page | |
2708 | */ | |
2709 | high = max(high, 1UL); | |
9b8b1754 | 2710 | |
8a5dbc65 JK |
2711 | overage = usage - high; |
2712 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2713 | return div64_u64(overage, high); | |
2714 | } | |
e26733e0 | 2715 | |
8a5dbc65 JK |
2716 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2717 | { | |
2718 | u64 overage, max_overage = 0; | |
e26733e0 | 2719 | |
8a5dbc65 JK |
2720 | do { |
2721 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2722 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2723 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2724 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2725 | !mem_cgroup_is_root(memcg)); | |
2726 | ||
8a5dbc65 JK |
2727 | return max_overage; |
2728 | } | |
2729 | ||
4b82ab4f JK |
2730 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2731 | { | |
2732 | u64 overage, max_overage = 0; | |
2733 | ||
2734 | do { | |
2735 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2736 | READ_ONCE(memcg->swap.high)); | |
2737 | if (overage) | |
2738 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2739 | max_overage = max(overage, max_overage); | |
2740 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2741 | !mem_cgroup_is_root(memcg)); | |
2742 | ||
2743 | return max_overage; | |
2744 | } | |
2745 | ||
8a5dbc65 JK |
2746 | /* |
2747 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2748 | * is exceeding its memory.high by checking both it and its ancestors. | |
2749 | */ | |
2750 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2751 | unsigned int nr_pages, | |
2752 | u64 max_overage) | |
2753 | { | |
2754 | unsigned long penalty_jiffies; | |
2755 | ||
e26733e0 CD |
2756 | if (!max_overage) |
2757 | return 0; | |
0e4b01df CD |
2758 | |
2759 | /* | |
0e4b01df CD |
2760 | * We use overage compared to memory.high to calculate the number of |
2761 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2762 | * fairly lenient on small overages, and increasingly harsh when the | |
2763 | * memcg in question makes it clear that it has no intention of stopping | |
2764 | * its crazy behaviour, so we exponentially increase the delay based on | |
2765 | * overage amount. | |
2766 | */ | |
e26733e0 CD |
2767 | penalty_jiffies = max_overage * max_overage * HZ; |
2768 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2769 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2770 | |
2771 | /* | |
2772 | * Factor in the task's own contribution to the overage, such that four | |
2773 | * N-sized allocations are throttled approximately the same as one | |
2774 | * 4N-sized allocation. | |
2775 | * | |
2776 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2777 | * larger the current charge patch is than that. | |
2778 | */ | |
ff144e69 | 2779 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2780 | } |
2781 | ||
2782 | /* | |
63fd3270 JW |
2783 | * Reclaims memory over the high limit. Called directly from |
2784 | * try_charge() (context permitting), as well as from the userland | |
2785 | * return path where reclaim is always able to block. | |
e26733e0 | 2786 | */ |
9ea9cb00 | 2787 | void mem_cgroup_handle_over_high(gfp_t gfp_mask) |
e26733e0 CD |
2788 | { |
2789 | unsigned long penalty_jiffies; | |
2790 | unsigned long pflags; | |
b3ff9291 | 2791 | unsigned long nr_reclaimed; |
e26733e0 | 2792 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2793 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2794 | struct mem_cgroup *memcg; |
b3ff9291 | 2795 | bool in_retry = false; |
e26733e0 CD |
2796 | |
2797 | if (likely(!nr_pages)) | |
2798 | return; | |
2799 | ||
2800 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2801 | current->memcg_nr_pages_over_high = 0; |
2802 | ||
b3ff9291 | 2803 | retry_reclaim: |
63fd3270 JW |
2804 | /* |
2805 | * Bail if the task is already exiting. Unlike memory.max, | |
2806 | * memory.high enforcement isn't as strict, and there is no | |
2807 | * OOM killer involved, which means the excess could already | |
2808 | * be much bigger (and still growing) than it could for | |
2809 | * memory.max; the dying task could get stuck in fruitless | |
2810 | * reclaim for a long time, which isn't desirable. | |
2811 | */ | |
2812 | if (task_is_dying()) | |
2813 | goto out; | |
2814 | ||
b3ff9291 CD |
2815 | /* |
2816 | * The allocating task should reclaim at least the batch size, but for | |
2817 | * subsequent retries we only want to do what's necessary to prevent oom | |
2818 | * or breaching resource isolation. | |
2819 | * | |
2820 | * This is distinct from memory.max or page allocator behaviour because | |
2821 | * memory.high is currently batched, whereas memory.max and the page | |
2822 | * allocator run every time an allocation is made. | |
2823 | */ | |
2824 | nr_reclaimed = reclaim_high(memcg, | |
2825 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
9ea9cb00 | 2826 | gfp_mask); |
b3ff9291 | 2827 | |
e26733e0 CD |
2828 | /* |
2829 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2830 | * allocators proactively to slow down excessive growth. | |
2831 | */ | |
8a5dbc65 JK |
2832 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2833 | mem_find_max_overage(memcg)); | |
0e4b01df | 2834 | |
4b82ab4f JK |
2835 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2836 | swap_find_max_overage(memcg)); | |
2837 | ||
ff144e69 JK |
2838 | /* |
2839 | * Clamp the max delay per usermode return so as to still keep the | |
2840 | * application moving forwards and also permit diagnostics, albeit | |
2841 | * extremely slowly. | |
2842 | */ | |
2843 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2844 | ||
0e4b01df CD |
2845 | /* |
2846 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2847 | * that it's not even worth doing, in an attempt to be nice to those who | |
2848 | * go only a small amount over their memory.high value and maybe haven't | |
2849 | * been aggressively reclaimed enough yet. | |
2850 | */ | |
2851 | if (penalty_jiffies <= HZ / 100) | |
2852 | goto out; | |
2853 | ||
b3ff9291 CD |
2854 | /* |
2855 | * If reclaim is making forward progress but we're still over | |
2856 | * memory.high, we want to encourage that rather than doing allocator | |
2857 | * throttling. | |
2858 | */ | |
2859 | if (nr_reclaimed || nr_retries--) { | |
2860 | in_retry = true; | |
2861 | goto retry_reclaim; | |
2862 | } | |
2863 | ||
0e4b01df | 2864 | /* |
63fd3270 JW |
2865 | * Reclaim didn't manage to push usage below the limit, slow |
2866 | * this allocating task down. | |
2867 | * | |
0e4b01df CD |
2868 | * If we exit early, we're guaranteed to die (since |
2869 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2870 | * need to account for any ill-begotten jiffies to pay them off later. | |
2871 | */ | |
2872 | psi_memstall_enter(&pflags); | |
2873 | schedule_timeout_killable(penalty_jiffies); | |
2874 | psi_memstall_leave(&pflags); | |
2875 | ||
2876 | out: | |
2877 | css_put(&memcg->css); | |
b23afb93 TH |
2878 | } |
2879 | ||
c5c8b16b MS |
2880 | static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2881 | unsigned int nr_pages) | |
8a9f3ccd | 2882 | { |
a983b5eb | 2883 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2884 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2885 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2886 | struct page_counter *counter; |
6539cc05 | 2887 | unsigned long nr_reclaimed; |
a4ebf1b6 | 2888 | bool passed_oom = false; |
73b73bac | 2889 | unsigned int reclaim_options = MEMCG_RECLAIM_MAY_SWAP; |
b70a2a21 | 2890 | bool drained = false; |
d6e103a7 | 2891 | bool raised_max_event = false; |
e22c6ed9 | 2892 | unsigned long pflags; |
a636b327 | 2893 | |
6539cc05 | 2894 | retry: |
b6b6cc72 | 2895 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2896 | return 0; |
8a9f3ccd | 2897 | |
7941d214 | 2898 | if (!do_memsw_account() || |
6071ca52 JW |
2899 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2900 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2901 | goto done_restock; |
7941d214 | 2902 | if (do_memsw_account()) |
3e32cb2e JW |
2903 | page_counter_uncharge(&memcg->memsw, batch); |
2904 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2905 | } else { |
3e32cb2e | 2906 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
73b73bac | 2907 | reclaim_options &= ~MEMCG_RECLAIM_MAY_SWAP; |
3fbe7244 | 2908 | } |
7a81b88c | 2909 | |
6539cc05 JW |
2910 | if (batch > nr_pages) { |
2911 | batch = nr_pages; | |
2912 | goto retry; | |
2913 | } | |
6d61ef40 | 2914 | |
89a28483 JW |
2915 | /* |
2916 | * Prevent unbounded recursion when reclaim operations need to | |
2917 | * allocate memory. This might exceed the limits temporarily, | |
2918 | * but we prefer facilitating memory reclaim and getting back | |
2919 | * under the limit over triggering OOM kills in these cases. | |
2920 | */ | |
2921 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2922 | goto force; | |
2923 | ||
06b078fc JW |
2924 | if (unlikely(task_in_memcg_oom(current))) |
2925 | goto nomem; | |
2926 | ||
d0164adc | 2927 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2928 | goto nomem; |
4b534334 | 2929 | |
e27be240 | 2930 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
d6e103a7 | 2931 | raised_max_event = true; |
241994ed | 2932 | |
e22c6ed9 | 2933 | psi_memstall_enter(&pflags); |
b70a2a21 | 2934 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
55ab834a | 2935 | gfp_mask, reclaim_options); |
e22c6ed9 | 2936 | psi_memstall_leave(&pflags); |
6539cc05 | 2937 | |
61e02c74 | 2938 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2939 | goto retry; |
28c34c29 | 2940 | |
b70a2a21 | 2941 | if (!drained) { |
6d3d6aa2 | 2942 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2943 | drained = true; |
2944 | goto retry; | |
2945 | } | |
2946 | ||
28c34c29 JW |
2947 | if (gfp_mask & __GFP_NORETRY) |
2948 | goto nomem; | |
6539cc05 JW |
2949 | /* |
2950 | * Even though the limit is exceeded at this point, reclaim | |
2951 | * may have been able to free some pages. Retry the charge | |
2952 | * before killing the task. | |
2953 | * | |
2954 | * Only for regular pages, though: huge pages are rather | |
2955 | * unlikely to succeed so close to the limit, and we fall back | |
2956 | * to regular pages anyway in case of failure. | |
2957 | */ | |
61e02c74 | 2958 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2959 | goto retry; |
2960 | /* | |
2961 | * At task move, charge accounts can be doubly counted. So, it's | |
2962 | * better to wait until the end of task_move if something is going on. | |
2963 | */ | |
2964 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2965 | goto retry; | |
2966 | ||
9b130619 JW |
2967 | if (nr_retries--) |
2968 | goto retry; | |
2969 | ||
38d38493 | 2970 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2971 | goto nomem; |
2972 | ||
a4ebf1b6 VA |
2973 | /* Avoid endless loop for tasks bypassed by the oom killer */ |
2974 | if (passed_oom && task_is_dying()) | |
2975 | goto nomem; | |
6539cc05 | 2976 | |
29ef680a MH |
2977 | /* |
2978 | * keep retrying as long as the memcg oom killer is able to make | |
2979 | * a forward progress or bypass the charge if the oom killer | |
2980 | * couldn't make any progress. | |
2981 | */ | |
becdf89d SB |
2982 | if (mem_cgroup_oom(mem_over_limit, gfp_mask, |
2983 | get_order(nr_pages * PAGE_SIZE))) { | |
a4ebf1b6 | 2984 | passed_oom = true; |
d977aa93 | 2985 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a | 2986 | goto retry; |
29ef680a | 2987 | } |
7a81b88c | 2988 | nomem: |
1461e8c2 SB |
2989 | /* |
2990 | * Memcg doesn't have a dedicated reserve for atomic | |
2991 | * allocations. But like the global atomic pool, we need to | |
2992 | * put the burden of reclaim on regular allocation requests | |
2993 | * and let these go through as privileged allocations. | |
2994 | */ | |
2995 | if (!(gfp_mask & (__GFP_NOFAIL | __GFP_HIGH))) | |
3168ecbe | 2996 | return -ENOMEM; |
10d53c74 | 2997 | force: |
d6e103a7 RG |
2998 | /* |
2999 | * If the allocation has to be enforced, don't forget to raise | |
3000 | * a MEMCG_MAX event. | |
3001 | */ | |
3002 | if (!raised_max_event) | |
3003 | memcg_memory_event(mem_over_limit, MEMCG_MAX); | |
3004 | ||
10d53c74 TH |
3005 | /* |
3006 | * The allocation either can't fail or will lead to more memory | |
3007 | * being freed very soon. Allow memory usage go over the limit | |
3008 | * temporarily by force charging it. | |
3009 | */ | |
3010 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 3011 | if (do_memsw_account()) |
10d53c74 | 3012 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
3013 | |
3014 | return 0; | |
6539cc05 JW |
3015 | |
3016 | done_restock: | |
3017 | if (batch > nr_pages) | |
3018 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 3019 | |
241994ed | 3020 | /* |
b23afb93 TH |
3021 | * If the hierarchy is above the normal consumption range, schedule |
3022 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 3023 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
3024 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
3025 | * not recorded as it most likely matches current's and won't | |
3026 | * change in the meantime. As high limit is checked again before | |
3027 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
3028 | */ |
3029 | do { | |
4b82ab4f JK |
3030 | bool mem_high, swap_high; |
3031 | ||
3032 | mem_high = page_counter_read(&memcg->memory) > | |
3033 | READ_ONCE(memcg->memory.high); | |
3034 | swap_high = page_counter_read(&memcg->swap) > | |
3035 | READ_ONCE(memcg->swap.high); | |
3036 | ||
3037 | /* Don't bother a random interrupted task */ | |
086f694a | 3038 | if (!in_task()) { |
4b82ab4f | 3039 | if (mem_high) { |
f7e1cb6e JW |
3040 | schedule_work(&memcg->high_work); |
3041 | break; | |
3042 | } | |
4b82ab4f JK |
3043 | continue; |
3044 | } | |
3045 | ||
3046 | if (mem_high || swap_high) { | |
3047 | /* | |
3048 | * The allocating tasks in this cgroup will need to do | |
3049 | * reclaim or be throttled to prevent further growth | |
3050 | * of the memory or swap footprints. | |
3051 | * | |
3052 | * Target some best-effort fairness between the tasks, | |
3053 | * and distribute reclaim work and delay penalties | |
3054 | * based on how much each task is actually allocating. | |
3055 | */ | |
9516a18a | 3056 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
3057 | set_notify_resume(current); |
3058 | break; | |
3059 | } | |
241994ed | 3060 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 | 3061 | |
63fd3270 JW |
3062 | /* |
3063 | * Reclaim is set up above to be called from the userland | |
3064 | * return path. But also attempt synchronous reclaim to avoid | |
3065 | * excessive overrun while the task is still inside the | |
3066 | * kernel. If this is successful, the return path will see it | |
3067 | * when it rechecks the overage and simply bail out. | |
3068 | */ | |
c9afe31e SB |
3069 | if (current->memcg_nr_pages_over_high > MEMCG_CHARGE_BATCH && |
3070 | !(current->flags & PF_MEMALLOC) && | |
63fd3270 | 3071 | gfpflags_allow_blocking(gfp_mask)) |
9ea9cb00 | 3072 | mem_cgroup_handle_over_high(gfp_mask); |
10d53c74 | 3073 | return 0; |
7a81b88c | 3074 | } |
8a9f3ccd | 3075 | |
c5c8b16b MS |
3076 | static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
3077 | unsigned int nr_pages) | |
3078 | { | |
3079 | if (mem_cgroup_is_root(memcg)) | |
3080 | return 0; | |
3081 | ||
3082 | return try_charge_memcg(memcg, gfp_mask, nr_pages); | |
3083 | } | |
3084 | ||
4b569387 NP |
3085 | /** |
3086 | * mem_cgroup_cancel_charge() - cancel an uncommitted try_charge() call. | |
3087 | * @memcg: memcg previously charged. | |
3088 | * @nr_pages: number of pages previously charged. | |
3089 | */ | |
3090 | void mem_cgroup_cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
a3032a2c | 3091 | { |
ce00a967 JW |
3092 | if (mem_cgroup_is_root(memcg)) |
3093 | return; | |
3094 | ||
3e32cb2e | 3095 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 3096 | if (do_memsw_account()) |
3e32cb2e | 3097 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f KH |
3098 | } |
3099 | ||
118f2875 | 3100 | static void commit_charge(struct folio *folio, struct mem_cgroup *memcg) |
0a31bc97 | 3101 | { |
118f2875 | 3102 | VM_BUG_ON_FOLIO(folio_memcg(folio), folio); |
0a31bc97 | 3103 | /* |
a5eb011a | 3104 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 3105 | * |
a0b5b414 JW |
3106 | * - the page lock |
3107 | * - LRU isolation | |
6c77b607 | 3108 | * - folio_memcg_lock() |
a0b5b414 | 3109 | * - exclusive reference |
018ee47f | 3110 | * - mem_cgroup_trylock_pages() |
0a31bc97 | 3111 | */ |
118f2875 | 3112 | folio->memcg_data = (unsigned long)memcg; |
7a81b88c | 3113 | } |
66e1707b | 3114 | |
4b569387 NP |
3115 | /** |
3116 | * mem_cgroup_commit_charge - commit a previously successful try_charge(). | |
3117 | * @folio: folio to commit the charge to. | |
3118 | * @memcg: memcg previously charged. | |
3119 | */ | |
3120 | void mem_cgroup_commit_charge(struct folio *folio, struct mem_cgroup *memcg) | |
3121 | { | |
3122 | css_get(&memcg->css); | |
3123 | commit_charge(folio, memcg); | |
3124 | ||
3125 | local_irq_disable(); | |
3126 | mem_cgroup_charge_statistics(memcg, folio_nr_pages(folio)); | |
3127 | memcg_check_events(memcg, folio_nid(folio)); | |
3128 | local_irq_enable(); | |
3129 | } | |
3130 | ||
84c07d11 | 3131 | #ifdef CONFIG_MEMCG_KMEM |
41eb5df1 | 3132 | |
91882c16 SB |
3133 | static inline void __mod_objcg_mlstate(struct obj_cgroup *objcg, |
3134 | struct pglist_data *pgdat, | |
3135 | enum node_stat_item idx, int nr) | |
a7ebf564 WL |
3136 | { |
3137 | struct mem_cgroup *memcg; | |
3138 | struct lruvec *lruvec; | |
3139 | ||
91882c16 SB |
3140 | lockdep_assert_irqs_disabled(); |
3141 | ||
a7ebf564 WL |
3142 | rcu_read_lock(); |
3143 | memcg = obj_cgroup_memcg(objcg); | |
3144 | lruvec = mem_cgroup_lruvec(memcg, pgdat); | |
91882c16 | 3145 | __mod_memcg_lruvec_state(lruvec, idx, nr); |
a7ebf564 WL |
3146 | rcu_read_unlock(); |
3147 | } | |
3148 | ||
fc4db90f RG |
3149 | static __always_inline |
3150 | struct mem_cgroup *mem_cgroup_from_obj_folio(struct folio *folio, void *p) | |
8380ce47 | 3151 | { |
8380ce47 | 3152 | /* |
9855609b RG |
3153 | * Slab objects are accounted individually, not per-page. |
3154 | * Memcg membership data for each individual object is saved in | |
21c690a3 | 3155 | * slab->obj_exts. |
8380ce47 | 3156 | */ |
4b5f8d9a | 3157 | if (folio_test_slab(folio)) { |
21c690a3 | 3158 | struct slabobj_ext *obj_exts; |
4b5f8d9a | 3159 | struct slab *slab; |
9855609b RG |
3160 | unsigned int off; |
3161 | ||
4b5f8d9a | 3162 | slab = folio_slab(folio); |
21c690a3 SB |
3163 | obj_exts = slab_obj_exts(slab); |
3164 | if (!obj_exts) | |
4b5f8d9a VB |
3165 | return NULL; |
3166 | ||
3167 | off = obj_to_index(slab->slab_cache, slab, p); | |
21c690a3 SB |
3168 | if (obj_exts[off].objcg) |
3169 | return obj_cgroup_memcg(obj_exts[off].objcg); | |
10befea9 RG |
3170 | |
3171 | return NULL; | |
9855609b | 3172 | } |
8380ce47 | 3173 | |
bcfe06bf | 3174 | /* |
becacb04 | 3175 | * folio_memcg_check() is used here, because in theory we can encounter |
4b5f8d9a | 3176 | * a folio where the slab flag has been cleared already, but |
21c690a3 | 3177 | * slab->obj_exts has not been freed yet |
becacb04 | 3178 | * folio_memcg_check() will guarantee that a proper memory |
bcfe06bf RG |
3179 | * cgroup pointer or NULL will be returned. |
3180 | */ | |
becacb04 | 3181 | return folio_memcg_check(folio); |
8380ce47 RG |
3182 | } |
3183 | ||
fc4db90f RG |
3184 | /* |
3185 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
3186 | * | |
3187 | * A passed kernel object can be a slab object, vmalloc object or a generic | |
3188 | * kernel page, so different mechanisms for getting the memory cgroup pointer | |
3189 | * should be used. | |
3190 | * | |
3191 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
3192 | * can not know for sure how the kernel object is implemented. | |
3193 | * mem_cgroup_from_obj() can be safely used in such cases. | |
3194 | * | |
3195 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
3196 | * cgroup_mutex, etc. | |
3197 | */ | |
3198 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
3199 | { | |
3200 | struct folio *folio; | |
3201 | ||
3202 | if (mem_cgroup_disabled()) | |
3203 | return NULL; | |
3204 | ||
3205 | if (unlikely(is_vmalloc_addr(p))) | |
3206 | folio = page_folio(vmalloc_to_page(p)); | |
3207 | else | |
3208 | folio = virt_to_folio(p); | |
3209 | ||
3210 | return mem_cgroup_from_obj_folio(folio, p); | |
3211 | } | |
3212 | ||
3213 | /* | |
3214 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
3215 | * Similar to mem_cgroup_from_obj(), but faster and not suitable for objects, | |
3216 | * allocated using vmalloc(). | |
3217 | * | |
3218 | * A passed kernel object must be a slab object or a generic kernel page. | |
3219 | * | |
3220 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
3221 | * cgroup_mutex, etc. | |
3222 | */ | |
3223 | struct mem_cgroup *mem_cgroup_from_slab_obj(void *p) | |
3224 | { | |
3225 | if (mem_cgroup_disabled()) | |
3226 | return NULL; | |
3227 | ||
3228 | return mem_cgroup_from_obj_folio(virt_to_folio(p), p); | |
3229 | } | |
3230 | ||
f4840ccf JW |
3231 | static struct obj_cgroup *__get_obj_cgroup_from_memcg(struct mem_cgroup *memcg) |
3232 | { | |
3233 | struct obj_cgroup *objcg = NULL; | |
3234 | ||
7848ed62 | 3235 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
f4840ccf | 3236 | objcg = rcu_dereference(memcg->objcg); |
7d0715d0 | 3237 | if (likely(objcg && obj_cgroup_tryget(objcg))) |
f4840ccf JW |
3238 | break; |
3239 | objcg = NULL; | |
3240 | } | |
3241 | return objcg; | |
3242 | } | |
3243 | ||
1aacbd35 RG |
3244 | static struct obj_cgroup *current_objcg_update(void) |
3245 | { | |
3246 | struct mem_cgroup *memcg; | |
3247 | struct obj_cgroup *old, *objcg = NULL; | |
3248 | ||
3249 | do { | |
3250 | /* Atomically drop the update bit. */ | |
3251 | old = xchg(¤t->objcg, NULL); | |
3252 | if (old) { | |
3253 | old = (struct obj_cgroup *) | |
3254 | ((unsigned long)old & ~CURRENT_OBJCG_UPDATE_FLAG); | |
91b71e78 | 3255 | obj_cgroup_put(old); |
1aacbd35 RG |
3256 | |
3257 | old = NULL; | |
3258 | } | |
3259 | ||
3260 | /* If new objcg is NULL, no reason for the second atomic update. */ | |
3261 | if (!current->mm || (current->flags & PF_KTHREAD)) | |
3262 | return NULL; | |
3263 | ||
3264 | /* | |
3265 | * Release the objcg pointer from the previous iteration, | |
3266 | * if try_cmpxcg() below fails. | |
3267 | */ | |
3268 | if (unlikely(objcg)) { | |
3269 | obj_cgroup_put(objcg); | |
3270 | objcg = NULL; | |
3271 | } | |
3272 | ||
3273 | /* | |
3274 | * Obtain the new objcg pointer. The current task can be | |
3275 | * asynchronously moved to another memcg and the previous | |
3276 | * memcg can be offlined. So let's get the memcg pointer | |
3277 | * and try get a reference to objcg under a rcu read lock. | |
3278 | */ | |
3279 | ||
3280 | rcu_read_lock(); | |
3281 | memcg = mem_cgroup_from_task(current); | |
3282 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
3283 | rcu_read_unlock(); | |
3284 | ||
3285 | /* | |
3286 | * Try set up a new objcg pointer atomically. If it | |
3287 | * fails, it means the update flag was set concurrently, so | |
3288 | * the whole procedure should be repeated. | |
3289 | */ | |
3290 | } while (!try_cmpxchg(¤t->objcg, &old, objcg)); | |
3291 | ||
3292 | return objcg; | |
3293 | } | |
3294 | ||
e86828e5 RG |
3295 | __always_inline struct obj_cgroup *current_obj_cgroup(void) |
3296 | { | |
3297 | struct mem_cgroup *memcg; | |
3298 | struct obj_cgroup *objcg; | |
3299 | ||
3300 | if (in_task()) { | |
3301 | memcg = current->active_memcg; | |
3302 | if (unlikely(memcg)) | |
3303 | goto from_memcg; | |
3304 | ||
3305 | objcg = READ_ONCE(current->objcg); | |
3306 | if (unlikely((unsigned long)objcg & CURRENT_OBJCG_UPDATE_FLAG)) | |
3307 | objcg = current_objcg_update(); | |
3308 | /* | |
3309 | * Objcg reference is kept by the task, so it's safe | |
3310 | * to use the objcg by the current task. | |
3311 | */ | |
3312 | return objcg; | |
3313 | } | |
3314 | ||
3315 | memcg = this_cpu_read(int_active_memcg); | |
3316 | if (unlikely(memcg)) | |
3317 | goto from_memcg; | |
3318 | ||
3319 | return NULL; | |
3320 | ||
3321 | from_memcg: | |
5f79489a | 3322 | objcg = NULL; |
e86828e5 RG |
3323 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
3324 | /* | |
3325 | * Memcg pointer is protected by scope (see set_active_memcg()) | |
3326 | * and is pinning the corresponding objcg, so objcg can't go | |
3327 | * away and can be used within the scope without any additional | |
3328 | * protection. | |
3329 | */ | |
3330 | objcg = rcu_dereference_check(memcg->objcg, 1); | |
3331 | if (likely(objcg)) | |
3332 | break; | |
e86828e5 RG |
3333 | } |
3334 | ||
3335 | return objcg; | |
3336 | } | |
3337 | ||
074e3e26 | 3338 | struct obj_cgroup *get_obj_cgroup_from_folio(struct folio *folio) |
f4840ccf JW |
3339 | { |
3340 | struct obj_cgroup *objcg; | |
3341 | ||
f7a449f7 | 3342 | if (!memcg_kmem_online()) |
f4840ccf JW |
3343 | return NULL; |
3344 | ||
074e3e26 MWO |
3345 | if (folio_memcg_kmem(folio)) { |
3346 | objcg = __folio_objcg(folio); | |
f4840ccf JW |
3347 | obj_cgroup_get(objcg); |
3348 | } else { | |
3349 | struct mem_cgroup *memcg; | |
bf4f0599 | 3350 | |
f4840ccf | 3351 | rcu_read_lock(); |
074e3e26 | 3352 | memcg = __folio_memcg(folio); |
f4840ccf JW |
3353 | if (memcg) |
3354 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
3355 | else | |
3356 | objcg = NULL; | |
3357 | rcu_read_unlock(); | |
3358 | } | |
bf4f0599 RG |
3359 | return objcg; |
3360 | } | |
3361 | ||
a8c49af3 YA |
3362 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
3363 | { | |
3364 | mod_memcg_state(memcg, MEMCG_KMEM, nr_pages); | |
3365 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
3366 | if (nr_pages > 0) | |
3367 | page_counter_charge(&memcg->kmem, nr_pages); | |
3368 | else | |
3369 | page_counter_uncharge(&memcg->kmem, -nr_pages); | |
3370 | } | |
3371 | } | |
3372 | ||
3373 | ||
f1286fae MS |
3374 | /* |
3375 | * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg | |
3376 | * @objcg: object cgroup to uncharge | |
3377 | * @nr_pages: number of pages to uncharge | |
3378 | */ | |
e74d2259 MS |
3379 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
3380 | unsigned int nr_pages) | |
3381 | { | |
3382 | struct mem_cgroup *memcg; | |
3383 | ||
3384 | memcg = get_mem_cgroup_from_objcg(objcg); | |
e74d2259 | 3385 | |
a8c49af3 | 3386 | memcg_account_kmem(memcg, -nr_pages); |
f1286fae | 3387 | refill_stock(memcg, nr_pages); |
e74d2259 | 3388 | |
e74d2259 | 3389 | css_put(&memcg->css); |
e74d2259 MS |
3390 | } |
3391 | ||
f1286fae MS |
3392 | /* |
3393 | * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg | |
3394 | * @objcg: object cgroup to charge | |
45264778 | 3395 | * @gfp: reclaim mode |
92d0510c | 3396 | * @nr_pages: number of pages to charge |
45264778 VD |
3397 | * |
3398 | * Returns 0 on success, an error code on failure. | |
3399 | */ | |
f1286fae MS |
3400 | static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp, |
3401 | unsigned int nr_pages) | |
7ae1e1d0 | 3402 | { |
f1286fae | 3403 | struct mem_cgroup *memcg; |
7ae1e1d0 GC |
3404 | int ret; |
3405 | ||
f1286fae MS |
3406 | memcg = get_mem_cgroup_from_objcg(objcg); |
3407 | ||
c5c8b16b | 3408 | ret = try_charge_memcg(memcg, gfp, nr_pages); |
52c29b04 | 3409 | if (ret) |
f1286fae | 3410 | goto out; |
52c29b04 | 3411 | |
a8c49af3 | 3412 | memcg_account_kmem(memcg, nr_pages); |
f1286fae MS |
3413 | out: |
3414 | css_put(&memcg->css); | |
4b13f64d | 3415 | |
f1286fae | 3416 | return ret; |
4b13f64d RG |
3417 | } |
3418 | ||
45264778 | 3419 | /** |
f4b00eab | 3420 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3421 | * @page: page to charge |
3422 | * @gfp: reclaim mode | |
3423 | * @order: allocation order | |
3424 | * | |
3425 | * Returns 0 on success, an error code on failure. | |
3426 | */ | |
f4b00eab | 3427 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3428 | { |
b4e0b68f | 3429 | struct obj_cgroup *objcg; |
fcff7d7e | 3430 | int ret = 0; |
7ae1e1d0 | 3431 | |
e86828e5 | 3432 | objcg = current_obj_cgroup(); |
b4e0b68f MS |
3433 | if (objcg) { |
3434 | ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order); | |
4d96ba35 | 3435 | if (!ret) { |
e86828e5 | 3436 | obj_cgroup_get(objcg); |
b4e0b68f | 3437 | page->memcg_data = (unsigned long)objcg | |
18b2db3b | 3438 | MEMCG_DATA_KMEM; |
1a3e1f40 | 3439 | return 0; |
4d96ba35 | 3440 | } |
c4159a75 | 3441 | } |
d05e83a6 | 3442 | return ret; |
7ae1e1d0 | 3443 | } |
49a18eae | 3444 | |
45264778 | 3445 | /** |
f4b00eab | 3446 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3447 | * @page: page to uncharge |
3448 | * @order: allocation order | |
3449 | */ | |
f4b00eab | 3450 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3451 | { |
1b7e4464 | 3452 | struct folio *folio = page_folio(page); |
b4e0b68f | 3453 | struct obj_cgroup *objcg; |
f3ccb2c4 | 3454 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3455 | |
1b7e4464 | 3456 | if (!folio_memcg_kmem(folio)) |
7ae1e1d0 GC |
3457 | return; |
3458 | ||
1b7e4464 | 3459 | objcg = __folio_objcg(folio); |
b4e0b68f | 3460 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
1b7e4464 | 3461 | folio->memcg_data = 0; |
b4e0b68f | 3462 | obj_cgroup_put(objcg); |
60d3fd32 | 3463 | } |
bf4f0599 | 3464 | |
91882c16 | 3465 | static void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
68ac5b3c WL |
3466 | enum node_stat_item idx, int nr) |
3467 | { | |
fead2b86 | 3468 | struct memcg_stock_pcp *stock; |
56751146 | 3469 | struct obj_cgroup *old = NULL; |
68ac5b3c WL |
3470 | unsigned long flags; |
3471 | int *bytes; | |
3472 | ||
56751146 | 3473 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3474 | stock = this_cpu_ptr(&memcg_stock); |
3475 | ||
68ac5b3c WL |
3476 | /* |
3477 | * Save vmstat data in stock and skip vmstat array update unless | |
3478 | * accumulating over a page of vmstat data or when pgdat or idx | |
3479 | * changes. | |
3480 | */ | |
3b8abb32 | 3481 | if (READ_ONCE(stock->cached_objcg) != objcg) { |
56751146 | 3482 | old = drain_obj_stock(stock); |
68ac5b3c WL |
3483 | obj_cgroup_get(objcg); |
3484 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) | |
3485 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3b8abb32 | 3486 | WRITE_ONCE(stock->cached_objcg, objcg); |
68ac5b3c WL |
3487 | stock->cached_pgdat = pgdat; |
3488 | } else if (stock->cached_pgdat != pgdat) { | |
3489 | /* Flush the existing cached vmstat data */ | |
7fa0dacb WL |
3490 | struct pglist_data *oldpg = stock->cached_pgdat; |
3491 | ||
68ac5b3c | 3492 | if (stock->nr_slab_reclaimable_b) { |
91882c16 | 3493 | __mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B, |
68ac5b3c WL |
3494 | stock->nr_slab_reclaimable_b); |
3495 | stock->nr_slab_reclaimable_b = 0; | |
3496 | } | |
3497 | if (stock->nr_slab_unreclaimable_b) { | |
91882c16 | 3498 | __mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B, |
68ac5b3c WL |
3499 | stock->nr_slab_unreclaimable_b); |
3500 | stock->nr_slab_unreclaimable_b = 0; | |
3501 | } | |
3502 | stock->cached_pgdat = pgdat; | |
3503 | } | |
3504 | ||
3505 | bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b | |
3506 | : &stock->nr_slab_unreclaimable_b; | |
3507 | /* | |
3508 | * Even for large object >= PAGE_SIZE, the vmstat data will still be | |
3509 | * cached locally at least once before pushing it out. | |
3510 | */ | |
3511 | if (!*bytes) { | |
3512 | *bytes = nr; | |
3513 | nr = 0; | |
3514 | } else { | |
3515 | *bytes += nr; | |
3516 | if (abs(*bytes) > PAGE_SIZE) { | |
3517 | nr = *bytes; | |
3518 | *bytes = 0; | |
3519 | } else { | |
3520 | nr = 0; | |
3521 | } | |
3522 | } | |
3523 | if (nr) | |
91882c16 | 3524 | __mod_objcg_mlstate(objcg, pgdat, idx, nr); |
68ac5b3c | 3525 | |
56751146 | 3526 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 3527 | obj_cgroup_put(old); |
68ac5b3c WL |
3528 | } |
3529 | ||
bf4f0599 RG |
3530 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
3531 | { | |
fead2b86 | 3532 | struct memcg_stock_pcp *stock; |
bf4f0599 RG |
3533 | unsigned long flags; |
3534 | bool ret = false; | |
3535 | ||
56751146 | 3536 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3537 | |
3538 | stock = this_cpu_ptr(&memcg_stock); | |
3b8abb32 | 3539 | if (objcg == READ_ONCE(stock->cached_objcg) && stock->nr_bytes >= nr_bytes) { |
bf4f0599 RG |
3540 | stock->nr_bytes -= nr_bytes; |
3541 | ret = true; | |
3542 | } | |
3543 | ||
56751146 | 3544 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
bf4f0599 RG |
3545 | |
3546 | return ret; | |
3547 | } | |
3548 | ||
56751146 | 3549 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 3550 | { |
3b8abb32 | 3551 | struct obj_cgroup *old = READ_ONCE(stock->cached_objcg); |
bf4f0599 RG |
3552 | |
3553 | if (!old) | |
56751146 | 3554 | return NULL; |
bf4f0599 RG |
3555 | |
3556 | if (stock->nr_bytes) { | |
3557 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3558 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3559 | ||
af9a3b69 JW |
3560 | if (nr_pages) { |
3561 | struct mem_cgroup *memcg; | |
3562 | ||
3563 | memcg = get_mem_cgroup_from_objcg(old); | |
3564 | ||
3565 | memcg_account_kmem(memcg, -nr_pages); | |
3566 | __refill_stock(memcg, nr_pages); | |
3567 | ||
3568 | css_put(&memcg->css); | |
3569 | } | |
bf4f0599 RG |
3570 | |
3571 | /* | |
3572 | * The leftover is flushed to the centralized per-memcg value. | |
3573 | * On the next attempt to refill obj stock it will be moved | |
3574 | * to a per-cpu stock (probably, on an other CPU), see | |
3575 | * refill_obj_stock(). | |
3576 | * | |
3577 | * How often it's flushed is a trade-off between the memory | |
3578 | * limit enforcement accuracy and potential CPU contention, | |
3579 | * so it might be changed in the future. | |
3580 | */ | |
3581 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3582 | stock->nr_bytes = 0; | |
3583 | } | |
3584 | ||
68ac5b3c WL |
3585 | /* |
3586 | * Flush the vmstat data in current stock | |
3587 | */ | |
3588 | if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) { | |
3589 | if (stock->nr_slab_reclaimable_b) { | |
91882c16 | 3590 | __mod_objcg_mlstate(old, stock->cached_pgdat, |
68ac5b3c WL |
3591 | NR_SLAB_RECLAIMABLE_B, |
3592 | stock->nr_slab_reclaimable_b); | |
3593 | stock->nr_slab_reclaimable_b = 0; | |
3594 | } | |
3595 | if (stock->nr_slab_unreclaimable_b) { | |
91882c16 | 3596 | __mod_objcg_mlstate(old, stock->cached_pgdat, |
68ac5b3c WL |
3597 | NR_SLAB_UNRECLAIMABLE_B, |
3598 | stock->nr_slab_unreclaimable_b); | |
3599 | stock->nr_slab_unreclaimable_b = 0; | |
3600 | } | |
3601 | stock->cached_pgdat = NULL; | |
3602 | } | |
3603 | ||
3b8abb32 | 3604 | WRITE_ONCE(stock->cached_objcg, NULL); |
56751146 SAS |
3605 | /* |
3606 | * The `old' objects needs to be released by the caller via | |
3607 | * obj_cgroup_put() outside of memcg_stock_pcp::stock_lock. | |
3608 | */ | |
3609 | return old; | |
bf4f0599 RG |
3610 | } |
3611 | ||
3612 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3613 | struct mem_cgroup *root_memcg) | |
3614 | { | |
3b8abb32 | 3615 | struct obj_cgroup *objcg = READ_ONCE(stock->cached_objcg); |
bf4f0599 RG |
3616 | struct mem_cgroup *memcg; |
3617 | ||
3b8abb32 RG |
3618 | if (objcg) { |
3619 | memcg = obj_cgroup_memcg(objcg); | |
bf4f0599 RG |
3620 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) |
3621 | return true; | |
3622 | } | |
3623 | ||
3624 | return false; | |
3625 | } | |
3626 | ||
5387c904 WL |
3627 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes, |
3628 | bool allow_uncharge) | |
bf4f0599 | 3629 | { |
fead2b86 | 3630 | struct memcg_stock_pcp *stock; |
56751146 | 3631 | struct obj_cgroup *old = NULL; |
bf4f0599 | 3632 | unsigned long flags; |
5387c904 | 3633 | unsigned int nr_pages = 0; |
bf4f0599 | 3634 | |
56751146 | 3635 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3636 | |
3637 | stock = this_cpu_ptr(&memcg_stock); | |
3b8abb32 | 3638 | if (READ_ONCE(stock->cached_objcg) != objcg) { /* reset if necessary */ |
56751146 | 3639 | old = drain_obj_stock(stock); |
bf4f0599 | 3640 | obj_cgroup_get(objcg); |
3b8abb32 | 3641 | WRITE_ONCE(stock->cached_objcg, objcg); |
5387c904 WL |
3642 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) |
3643 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3644 | allow_uncharge = true; /* Allow uncharge when objcg changes */ | |
bf4f0599 RG |
3645 | } |
3646 | stock->nr_bytes += nr_bytes; | |
3647 | ||
5387c904 WL |
3648 | if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) { |
3649 | nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3650 | stock->nr_bytes &= (PAGE_SIZE - 1); | |
3651 | } | |
bf4f0599 | 3652 | |
56751146 | 3653 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 3654 | obj_cgroup_put(old); |
5387c904 WL |
3655 | |
3656 | if (nr_pages) | |
3657 | obj_cgroup_uncharge_pages(objcg, nr_pages); | |
bf4f0599 RG |
3658 | } |
3659 | ||
3660 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3661 | { | |
bf4f0599 RG |
3662 | unsigned int nr_pages, nr_bytes; |
3663 | int ret; | |
3664 | ||
3665 | if (consume_obj_stock(objcg, size)) | |
3666 | return 0; | |
3667 | ||
3668 | /* | |
5387c904 | 3669 | * In theory, objcg->nr_charged_bytes can have enough |
bf4f0599 | 3670 | * pre-charged bytes to satisfy the allocation. However, |
5387c904 WL |
3671 | * flushing objcg->nr_charged_bytes requires two atomic |
3672 | * operations, and objcg->nr_charged_bytes can't be big. | |
3673 | * The shared objcg->nr_charged_bytes can also become a | |
3674 | * performance bottleneck if all tasks of the same memcg are | |
3675 | * trying to update it. So it's better to ignore it and try | |
3676 | * grab some new pages. The stock's nr_bytes will be flushed to | |
3677 | * objcg->nr_charged_bytes later on when objcg changes. | |
3678 | * | |
3679 | * The stock's nr_bytes may contain enough pre-charged bytes | |
3680 | * to allow one less page from being charged, but we can't rely | |
3681 | * on the pre-charged bytes not being changed outside of | |
3682 | * consume_obj_stock() or refill_obj_stock(). So ignore those | |
3683 | * pre-charged bytes as well when charging pages. To avoid a | |
3684 | * page uncharge right after a page charge, we set the | |
3685 | * allow_uncharge flag to false when calling refill_obj_stock() | |
3686 | * to temporarily allow the pre-charged bytes to exceed the page | |
3687 | * size limit. The maximum reachable value of the pre-charged | |
3688 | * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data | |
3689 | * race. | |
bf4f0599 | 3690 | */ |
bf4f0599 RG |
3691 | nr_pages = size >> PAGE_SHIFT; |
3692 | nr_bytes = size & (PAGE_SIZE - 1); | |
3693 | ||
3694 | if (nr_bytes) | |
3695 | nr_pages += 1; | |
3696 | ||
e74d2259 | 3697 | ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages); |
bf4f0599 | 3698 | if (!ret && nr_bytes) |
5387c904 | 3699 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false); |
bf4f0599 | 3700 | |
bf4f0599 RG |
3701 | return ret; |
3702 | } | |
3703 | ||
3704 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3705 | { | |
5387c904 | 3706 | refill_obj_stock(objcg, size, true); |
bf4f0599 RG |
3707 | } |
3708 | ||
e6100a45 VB |
3709 | static inline size_t obj_full_size(struct kmem_cache *s) |
3710 | { | |
3711 | /* | |
3712 | * For each accounted object there is an extra space which is used | |
3713 | * to store obj_cgroup membership. Charge it too. | |
3714 | */ | |
3715 | return s->size + sizeof(struct obj_cgroup *); | |
3716 | } | |
3717 | ||
3718 | bool __memcg_slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru, | |
3719 | gfp_t flags, size_t size, void **p) | |
3720 | { | |
3721 | struct obj_cgroup *objcg; | |
3722 | struct slab *slab; | |
3723 | unsigned long off; | |
3724 | size_t i; | |
3725 | ||
3726 | /* | |
3727 | * The obtained objcg pointer is safe to use within the current scope, | |
3728 | * defined by current task or set_active_memcg() pair. | |
3729 | * obj_cgroup_get() is used to get a permanent reference. | |
3730 | */ | |
3731 | objcg = current_obj_cgroup(); | |
3732 | if (!objcg) | |
3733 | return true; | |
3734 | ||
3735 | /* | |
3736 | * slab_alloc_node() avoids the NULL check, so we might be called with a | |
3737 | * single NULL object. kmem_cache_alloc_bulk() aborts if it can't fill | |
3738 | * the whole requested size. | |
3739 | * return success as there's nothing to free back | |
3740 | */ | |
3741 | if (unlikely(*p == NULL)) | |
3742 | return true; | |
3743 | ||
3744 | flags &= gfp_allowed_mask; | |
3745 | ||
3746 | if (lru) { | |
3747 | int ret; | |
3748 | struct mem_cgroup *memcg; | |
3749 | ||
3750 | memcg = get_mem_cgroup_from_objcg(objcg); | |
3751 | ret = memcg_list_lru_alloc(memcg, lru, flags); | |
3752 | css_put(&memcg->css); | |
3753 | ||
3754 | if (ret) | |
3755 | return false; | |
3756 | } | |
3757 | ||
3758 | if (obj_cgroup_charge(objcg, flags, size * obj_full_size(s))) | |
3759 | return false; | |
3760 | ||
3761 | for (i = 0; i < size; i++) { | |
3762 | slab = virt_to_slab(p[i]); | |
3763 | ||
3764 | if (!slab_obj_exts(slab) && | |
3765 | alloc_slab_obj_exts(slab, s, flags, false)) { | |
3766 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
3767 | continue; | |
3768 | } | |
3769 | ||
3770 | off = obj_to_index(s, slab, p[i]); | |
3771 | obj_cgroup_get(objcg); | |
3772 | slab_obj_exts(slab)[off].objcg = objcg; | |
3773 | mod_objcg_state(objcg, slab_pgdat(slab), | |
3774 | cache_vmstat_idx(s), obj_full_size(s)); | |
3775 | } | |
3776 | ||
3777 | return true; | |
3778 | } | |
3779 | ||
3780 | void __memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, | |
3781 | void **p, int objects, struct slabobj_ext *obj_exts) | |
3782 | { | |
3783 | for (int i = 0; i < objects; i++) { | |
3784 | struct obj_cgroup *objcg; | |
3785 | unsigned int off; | |
3786 | ||
3787 | off = obj_to_index(s, slab, p[i]); | |
3788 | objcg = obj_exts[off].objcg; | |
3789 | if (!objcg) | |
3790 | continue; | |
3791 | ||
3792 | obj_exts[off].objcg = NULL; | |
3793 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
3794 | mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s), | |
3795 | -obj_full_size(s)); | |
3796 | obj_cgroup_put(objcg); | |
3797 | } | |
3798 | } | |
84c07d11 | 3799 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3800 | |
ca3e0214 | 3801 | /* |
be6c8982 | 3802 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3803 | */ |
b8791381 | 3804 | void split_page_memcg(struct page *head, int old_order, int new_order) |
ca3e0214 | 3805 | { |
1b7e4464 MWO |
3806 | struct folio *folio = page_folio(head); |
3807 | struct mem_cgroup *memcg = folio_memcg(folio); | |
e94c8a9c | 3808 | int i; |
b8791381 ZY |
3809 | unsigned int old_nr = 1 << old_order; |
3810 | unsigned int new_nr = 1 << new_order; | |
ca3e0214 | 3811 | |
be6c8982 | 3812 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3813 | return; |
b070e65c | 3814 | |
b8791381 | 3815 | for (i = new_nr; i < old_nr; i += new_nr) |
1b7e4464 | 3816 | folio_page(folio, i)->memcg_data = folio->memcg_data; |
b4e0b68f | 3817 | |
1b7e4464 | 3818 | if (folio_memcg_kmem(folio)) |
b8791381 | 3819 | obj_cgroup_get_many(__folio_objcg(folio), old_nr / new_nr - 1); |
b4e0b68f | 3820 | else |
b8791381 | 3821 | css_get_many(&memcg->css, old_nr / new_nr - 1); |
ca3e0214 | 3822 | } |
ca3e0214 | 3823 | |
e55b9f96 | 3824 | #ifdef CONFIG_SWAP |
02491447 DN |
3825 | /** |
3826 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3827 | * @entry: swap entry to be moved | |
3828 | * @from: mem_cgroup which the entry is moved from | |
3829 | * @to: mem_cgroup which the entry is moved to | |
3830 | * | |
3831 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3832 | * as the mem_cgroup's id of @from. | |
3833 | * | |
3834 | * Returns 0 on success, -EINVAL on failure. | |
3835 | * | |
3e32cb2e | 3836 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3837 | * both res and memsw, and called css_get(). |
3838 | */ | |
3839 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3840 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3841 | { |
3842 | unsigned short old_id, new_id; | |
3843 | ||
34c00c31 LZ |
3844 | old_id = mem_cgroup_id(from); |
3845 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3846 | |
3847 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3848 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3849 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3850 | return 0; |
3851 | } | |
3852 | return -EINVAL; | |
3853 | } | |
3854 | #else | |
3855 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3856 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3857 | { |
3858 | return -EINVAL; | |
3859 | } | |
8c7c6e34 | 3860 | #endif |
d13d1443 | 3861 | |
bbec2e15 | 3862 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3863 | |
bbec2e15 RG |
3864 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3865 | unsigned long max, bool memsw) | |
628f4235 | 3866 | { |
3e32cb2e | 3867 | bool enlarge = false; |
bb4a7ea2 | 3868 | bool drained = false; |
3e32cb2e | 3869 | int ret; |
c054a78c YZ |
3870 | bool limits_invariant; |
3871 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3872 | |
3e32cb2e | 3873 | do { |
628f4235 KH |
3874 | if (signal_pending(current)) { |
3875 | ret = -EINTR; | |
3876 | break; | |
3877 | } | |
3e32cb2e | 3878 | |
bbec2e15 | 3879 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3880 | /* |
3881 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3882 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3883 | */ |
15b42562 | 3884 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3885 | max <= memcg->memsw.max; |
c054a78c | 3886 | if (!limits_invariant) { |
bbec2e15 | 3887 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3888 | ret = -EINVAL; |
8c7c6e34 KH |
3889 | break; |
3890 | } | |
bbec2e15 | 3891 | if (max > counter->max) |
3e32cb2e | 3892 | enlarge = true; |
bbec2e15 RG |
3893 | ret = page_counter_set_max(counter, max); |
3894 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3895 | |
3896 | if (!ret) | |
3897 | break; | |
3898 | ||
bb4a7ea2 SB |
3899 | if (!drained) { |
3900 | drain_all_stock(memcg); | |
3901 | drained = true; | |
3902 | continue; | |
3903 | } | |
3904 | ||
73b73bac | 3905 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
55ab834a | 3906 | memsw ? 0 : MEMCG_RECLAIM_MAY_SWAP)) { |
1ab5c056 AR |
3907 | ret = -EBUSY; |
3908 | break; | |
3909 | } | |
3910 | } while (true); | |
3e32cb2e | 3911 | |
3c11ecf4 KH |
3912 | if (!ret && enlarge) |
3913 | memcg_oom_recover(memcg); | |
3e32cb2e | 3914 | |
628f4235 KH |
3915 | return ret; |
3916 | } | |
3917 | ||
ef8f2327 | 3918 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3919 | gfp_t gfp_mask, |
3920 | unsigned long *total_scanned) | |
3921 | { | |
3922 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3923 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3924 | unsigned long reclaimed; |
3925 | int loop = 0; | |
ef8f2327 | 3926 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3927 | unsigned long excess; |
0608f43d | 3928 | |
e4dde56c YZ |
3929 | if (lru_gen_enabled()) |
3930 | return 0; | |
3931 | ||
0608f43d AM |
3932 | if (order > 0) |
3933 | return 0; | |
3934 | ||
2ab082ba | 3935 | mctz = soft_limit_tree.rb_tree_per_node[pgdat->node_id]; |
d6507ff5 MH |
3936 | |
3937 | /* | |
3938 | * Do not even bother to check the largest node if the root | |
3939 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3940 | * are acceptable as soft limit is best effort anyway. | |
3941 | */ | |
bfc7228b | 3942 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3943 | return 0; |
3944 | ||
0608f43d AM |
3945 | /* |
3946 | * This loop can run a while, specially if mem_cgroup's continuously | |
3947 | * keep exceeding their soft limit and putting the system under | |
3948 | * pressure | |
3949 | */ | |
3950 | do { | |
3951 | if (next_mz) | |
3952 | mz = next_mz; | |
3953 | else | |
3954 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3955 | if (!mz) | |
3956 | break; | |
3957 | ||
ef8f2327 | 3958 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
d8f65338 | 3959 | gfp_mask, total_scanned); |
0608f43d | 3960 | nr_reclaimed += reclaimed; |
0a31bc97 | 3961 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3962 | |
3963 | /* | |
3964 | * If we failed to reclaim anything from this memory cgroup | |
3965 | * it is time to move on to the next cgroup | |
3966 | */ | |
3967 | next_mz = NULL; | |
bc2f2e7f VD |
3968 | if (!reclaimed) |
3969 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3970 | ||
3e32cb2e | 3971 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3972 | /* |
3973 | * One school of thought says that we should not add | |
3974 | * back the node to the tree if reclaim returns 0. | |
3975 | * But our reclaim could return 0, simply because due | |
3976 | * to priority we are exposing a smaller subset of | |
3977 | * memory to reclaim from. Consider this as a longer | |
3978 | * term TODO. | |
3979 | */ | |
3980 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3981 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3982 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3983 | css_put(&mz->memcg->css); |
3984 | loop++; | |
3985 | /* | |
3986 | * Could not reclaim anything and there are no more | |
3987 | * mem cgroups to try or we seem to be looping without | |
3988 | * reclaiming anything. | |
3989 | */ | |
3990 | if (!nr_reclaimed && | |
3991 | (next_mz == NULL || | |
3992 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3993 | break; | |
3994 | } while (!nr_reclaimed); | |
3995 | if (next_mz) | |
3996 | css_put(&next_mz->memcg->css); | |
3997 | return nr_reclaimed; | |
3998 | } | |
3999 | ||
c26251f9 | 4000 | /* |
51038171 | 4001 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
4002 | * |
4003 | * Caller is responsible for holding css reference for memcg. | |
4004 | */ | |
4005 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4006 | { | |
d977aa93 | 4007 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 4008 | |
c1e862c1 KH |
4009 | /* we call try-to-free pages for make this cgroup empty */ |
4010 | lru_add_drain_all(); | |
d12c60f6 JS |
4011 | |
4012 | drain_all_stock(memcg); | |
4013 | ||
f817ed48 | 4014 | /* try to free all pages in this cgroup */ |
3e32cb2e | 4015 | while (nr_retries && page_counter_read(&memcg->memory)) { |
c26251f9 MH |
4016 | if (signal_pending(current)) |
4017 | return -EINTR; | |
4018 | ||
73b73bac | 4019 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
55ab834a | 4020 | MEMCG_RECLAIM_MAY_SWAP)) |
f817ed48 | 4021 | nr_retries--; |
f817ed48 | 4022 | } |
ab5196c2 MH |
4023 | |
4024 | return 0; | |
cc847582 KH |
4025 | } |
4026 | ||
6770c64e TH |
4027 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
4028 | char *buf, size_t nbytes, | |
4029 | loff_t off) | |
c1e862c1 | 4030 | { |
6770c64e | 4031 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 4032 | |
d8423011 MH |
4033 | if (mem_cgroup_is_root(memcg)) |
4034 | return -EINVAL; | |
6770c64e | 4035 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
4036 | } |
4037 | ||
182446d0 TH |
4038 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4039 | struct cftype *cft) | |
18f59ea7 | 4040 | { |
bef8620c | 4041 | return 1; |
18f59ea7 BS |
4042 | } |
4043 | ||
182446d0 TH |
4044 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
4045 | struct cftype *cft, u64 val) | |
18f59ea7 | 4046 | { |
bef8620c | 4047 | if (val == 1) |
0b8f73e1 | 4048 | return 0; |
567fb435 | 4049 | |
bef8620c RG |
4050 | pr_warn_once("Non-hierarchical mode is deprecated. " |
4051 | "Please report your usecase to linux-mm@kvack.org if you " | |
4052 | "depend on this functionality.\n"); | |
567fb435 | 4053 | |
bef8620c | 4054 | return -EINVAL; |
18f59ea7 BS |
4055 | } |
4056 | ||
6f646156 | 4057 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 4058 | { |
42a30035 | 4059 | unsigned long val; |
ce00a967 | 4060 | |
3e32cb2e | 4061 | if (mem_cgroup_is_root(memcg)) { |
a2174e95 | 4062 | /* |
f82a7a86 YA |
4063 | * Approximate root's usage from global state. This isn't |
4064 | * perfect, but the root usage was always an approximation. | |
a2174e95 | 4065 | */ |
f82a7a86 YA |
4066 | val = global_node_page_state(NR_FILE_PAGES) + |
4067 | global_node_page_state(NR_ANON_MAPPED); | |
42a30035 | 4068 | if (swap) |
f82a7a86 | 4069 | val += total_swap_pages - get_nr_swap_pages(); |
3e32cb2e | 4070 | } else { |
ce00a967 | 4071 | if (!swap) |
3e32cb2e | 4072 | val = page_counter_read(&memcg->memory); |
ce00a967 | 4073 | else |
3e32cb2e | 4074 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 4075 | } |
c12176d3 | 4076 | return val; |
ce00a967 JW |
4077 | } |
4078 | ||
3e32cb2e JW |
4079 | enum { |
4080 | RES_USAGE, | |
4081 | RES_LIMIT, | |
4082 | RES_MAX_USAGE, | |
4083 | RES_FAILCNT, | |
4084 | RES_SOFT_LIMIT, | |
4085 | }; | |
ce00a967 | 4086 | |
791badbd | 4087 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 4088 | struct cftype *cft) |
8cdea7c0 | 4089 | { |
182446d0 | 4090 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 4091 | struct page_counter *counter; |
af36f906 | 4092 | |
3e32cb2e | 4093 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 4094 | case _MEM: |
3e32cb2e JW |
4095 | counter = &memcg->memory; |
4096 | break; | |
8c7c6e34 | 4097 | case _MEMSWAP: |
3e32cb2e JW |
4098 | counter = &memcg->memsw; |
4099 | break; | |
510fc4e1 | 4100 | case _KMEM: |
3e32cb2e | 4101 | counter = &memcg->kmem; |
510fc4e1 | 4102 | break; |
d55f90bf | 4103 | case _TCP: |
0db15298 | 4104 | counter = &memcg->tcpmem; |
d55f90bf | 4105 | break; |
8c7c6e34 KH |
4106 | default: |
4107 | BUG(); | |
8c7c6e34 | 4108 | } |
3e32cb2e JW |
4109 | |
4110 | switch (MEMFILE_ATTR(cft->private)) { | |
4111 | case RES_USAGE: | |
4112 | if (counter == &memcg->memory) | |
c12176d3 | 4113 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 4114 | if (counter == &memcg->memsw) |
c12176d3 | 4115 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
4116 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
4117 | case RES_LIMIT: | |
bbec2e15 | 4118 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
4119 | case RES_MAX_USAGE: |
4120 | return (u64)counter->watermark * PAGE_SIZE; | |
4121 | case RES_FAILCNT: | |
4122 | return counter->failcnt; | |
4123 | case RES_SOFT_LIMIT: | |
2178e20c | 4124 | return (u64)READ_ONCE(memcg->soft_limit) * PAGE_SIZE; |
3e32cb2e JW |
4125 | default: |
4126 | BUG(); | |
4127 | } | |
8cdea7c0 | 4128 | } |
510fc4e1 | 4129 | |
6b0ba2ab FS |
4130 | /* |
4131 | * This function doesn't do anything useful. Its only job is to provide a read | |
4132 | * handler for a file so that cgroup_file_mode() will add read permissions. | |
4133 | */ | |
4134 | static int mem_cgroup_dummy_seq_show(__always_unused struct seq_file *m, | |
4135 | __always_unused void *v) | |
4136 | { | |
4137 | return -EINVAL; | |
4138 | } | |
4139 | ||
84c07d11 | 4140 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 4141 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 4142 | { |
bf4f0599 | 4143 | struct obj_cgroup *objcg; |
d6441637 | 4144 | |
9c94bef9 | 4145 | if (mem_cgroup_kmem_disabled()) |
b313aeee VD |
4146 | return 0; |
4147 | ||
da0efe30 MS |
4148 | if (unlikely(mem_cgroup_is_root(memcg))) |
4149 | return 0; | |
d6441637 | 4150 | |
bf4f0599 | 4151 | objcg = obj_cgroup_alloc(); |
f9c69d63 | 4152 | if (!objcg) |
bf4f0599 | 4153 | return -ENOMEM; |
f9c69d63 | 4154 | |
bf4f0599 RG |
4155 | objcg->memcg = memcg; |
4156 | rcu_assign_pointer(memcg->objcg, objcg); | |
675d6c9b RG |
4157 | obj_cgroup_get(objcg); |
4158 | memcg->orig_objcg = objcg; | |
bf4f0599 | 4159 | |
f7a449f7 | 4160 | static_branch_enable(&memcg_kmem_online_key); |
d648bcc7 | 4161 | |
f9c69d63 | 4162 | memcg->kmemcg_id = memcg->id.id; |
0b8f73e1 JW |
4163 | |
4164 | return 0; | |
d6441637 VD |
4165 | } |
4166 | ||
8e0a8912 JW |
4167 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
4168 | { | |
64268868 | 4169 | struct mem_cgroup *parent; |
8e0a8912 | 4170 | |
9c94bef9 | 4171 | if (mem_cgroup_kmem_disabled()) |
da0efe30 MS |
4172 | return; |
4173 | ||
4174 | if (unlikely(mem_cgroup_is_root(memcg))) | |
8e0a8912 | 4175 | return; |
9855609b | 4176 | |
8e0a8912 JW |
4177 | parent = parent_mem_cgroup(memcg); |
4178 | if (!parent) | |
4179 | parent = root_mem_cgroup; | |
4180 | ||
bf4f0599 | 4181 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a | 4182 | |
8e0a8912 | 4183 | /* |
64268868 MS |
4184 | * After we have finished memcg_reparent_objcgs(), all list_lrus |
4185 | * corresponding to this cgroup are guaranteed to remain empty. | |
4186 | * The ordering is imposed by list_lru_node->lock taken by | |
1f391eb2 | 4187 | * memcg_reparent_list_lrus(). |
8e0a8912 | 4188 | */ |
1f391eb2 | 4189 | memcg_reparent_list_lrus(memcg, parent); |
8e0a8912 | 4190 | } |
d6441637 | 4191 | #else |
0b8f73e1 | 4192 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
4193 | { |
4194 | return 0; | |
4195 | } | |
4196 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
4197 | { | |
4198 | } | |
84c07d11 | 4199 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 4200 | |
bbec2e15 | 4201 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
4202 | { |
4203 | int ret; | |
4204 | ||
bbec2e15 | 4205 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 4206 | |
bbec2e15 | 4207 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
4208 | if (ret) |
4209 | goto out; | |
4210 | ||
0db15298 | 4211 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
4212 | /* |
4213 | * The active flag needs to be written after the static_key | |
4214 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
4215 | * function is the last one to run. See mem_cgroup_sk_alloc() |
4216 | * for details, and note that we don't mark any socket as | |
4217 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
4218 | * |
4219 | * We need to do this, because static_keys will span multiple | |
4220 | * sites, but we can't control their order. If we mark a socket | |
4221 | * as accounted, but the accounting functions are not patched in | |
4222 | * yet, we'll lose accounting. | |
4223 | * | |
2d758073 | 4224 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
4225 | * because when this value change, the code to process it is not |
4226 | * patched in yet. | |
4227 | */ | |
4228 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 4229 | memcg->tcpmem_active = true; |
d55f90bf VD |
4230 | } |
4231 | out: | |
bbec2e15 | 4232 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
4233 | return ret; |
4234 | } | |
d55f90bf | 4235 | |
628f4235 KH |
4236 | /* |
4237 | * The user of this function is... | |
4238 | * RES_LIMIT. | |
4239 | */ | |
451af504 TH |
4240 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
4241 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 4242 | { |
451af504 | 4243 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4244 | unsigned long nr_pages; |
628f4235 KH |
4245 | int ret; |
4246 | ||
451af504 | 4247 | buf = strstrip(buf); |
650c5e56 | 4248 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
4249 | if (ret) |
4250 | return ret; | |
af36f906 | 4251 | |
3e32cb2e | 4252 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 4253 | case RES_LIMIT: |
4b3bde4c BS |
4254 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4255 | ret = -EINVAL; | |
4256 | break; | |
4257 | } | |
3e32cb2e JW |
4258 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4259 | case _MEM: | |
bbec2e15 | 4260 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 4261 | break; |
3e32cb2e | 4262 | case _MEMSWAP: |
bbec2e15 | 4263 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 4264 | break; |
4597648f MH |
4265 | case _KMEM: |
4266 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " | |
4267 | "Writing any value to this file has no effect. " | |
4268 | "Please report your usecase to linux-mm@kvack.org if you " | |
4269 | "depend on this functionality.\n"); | |
4270 | ret = 0; | |
4271 | break; | |
d55f90bf | 4272 | case _TCP: |
bbec2e15 | 4273 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 4274 | break; |
3e32cb2e | 4275 | } |
296c81d8 | 4276 | break; |
3e32cb2e | 4277 | case RES_SOFT_LIMIT: |
2343e88d SAS |
4278 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
4279 | ret = -EOPNOTSUPP; | |
4280 | } else { | |
2178e20c | 4281 | WRITE_ONCE(memcg->soft_limit, nr_pages); |
2343e88d SAS |
4282 | ret = 0; |
4283 | } | |
628f4235 KH |
4284 | break; |
4285 | } | |
451af504 | 4286 | return ret ?: nbytes; |
8cdea7c0 BS |
4287 | } |
4288 | ||
6770c64e TH |
4289 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
4290 | size_t nbytes, loff_t off) | |
c84872e1 | 4291 | { |
6770c64e | 4292 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4293 | struct page_counter *counter; |
c84872e1 | 4294 | |
3e32cb2e JW |
4295 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4296 | case _MEM: | |
4297 | counter = &memcg->memory; | |
4298 | break; | |
4299 | case _MEMSWAP: | |
4300 | counter = &memcg->memsw; | |
4301 | break; | |
4302 | case _KMEM: | |
4303 | counter = &memcg->kmem; | |
4304 | break; | |
d55f90bf | 4305 | case _TCP: |
0db15298 | 4306 | counter = &memcg->tcpmem; |
d55f90bf | 4307 | break; |
3e32cb2e JW |
4308 | default: |
4309 | BUG(); | |
4310 | } | |
af36f906 | 4311 | |
3e32cb2e | 4312 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 4313 | case RES_MAX_USAGE: |
3e32cb2e | 4314 | page_counter_reset_watermark(counter); |
29f2a4da PE |
4315 | break; |
4316 | case RES_FAILCNT: | |
3e32cb2e | 4317 | counter->failcnt = 0; |
29f2a4da | 4318 | break; |
3e32cb2e JW |
4319 | default: |
4320 | BUG(); | |
29f2a4da | 4321 | } |
f64c3f54 | 4322 | |
6770c64e | 4323 | return nbytes; |
c84872e1 PE |
4324 | } |
4325 | ||
182446d0 | 4326 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4327 | struct cftype *cft) |
4328 | { | |
182446d0 | 4329 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
4330 | } |
4331 | ||
02491447 | 4332 | #ifdef CONFIG_MMU |
182446d0 | 4333 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4334 | struct cftype *cft, u64 val) |
4335 | { | |
182446d0 | 4336 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 4337 | |
da34a848 JW |
4338 | pr_warn_once("Cgroup memory moving (move_charge_at_immigrate) is deprecated. " |
4339 | "Please report your usecase to linux-mm@kvack.org if you " | |
4340 | "depend on this functionality.\n"); | |
4341 | ||
1dfab5ab | 4342 | if (val & ~MOVE_MASK) |
7dc74be0 | 4343 | return -EINVAL; |
ee5e8472 | 4344 | |
7dc74be0 | 4345 | /* |
ee5e8472 GC |
4346 | * No kind of locking is needed in here, because ->can_attach() will |
4347 | * check this value once in the beginning of the process, and then carry | |
4348 | * on with stale data. This means that changes to this value will only | |
4349 | * affect task migrations starting after the change. | |
7dc74be0 | 4350 | */ |
c0ff4b85 | 4351 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4352 | return 0; |
4353 | } | |
02491447 | 4354 | #else |
182446d0 | 4355 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
4356 | struct cftype *cft, u64 val) |
4357 | { | |
4358 | return -ENOSYS; | |
4359 | } | |
4360 | #endif | |
7dc74be0 | 4361 | |
406eb0c9 | 4362 | #ifdef CONFIG_NUMA |
113b7dfd JW |
4363 | |
4364 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
4365 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
4366 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
4367 | ||
4368 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 4369 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 4370 | { |
867e5e1d | 4371 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
4372 | unsigned long nr = 0; |
4373 | enum lru_list lru; | |
4374 | ||
4375 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
4376 | ||
4377 | for_each_lru(lru) { | |
4378 | if (!(BIT(lru) & lru_mask)) | |
4379 | continue; | |
dd8657b6 SB |
4380 | if (tree) |
4381 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
4382 | else | |
4383 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
4384 | } |
4385 | return nr; | |
4386 | } | |
4387 | ||
4388 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
4389 | unsigned int lru_mask, |
4390 | bool tree) | |
113b7dfd JW |
4391 | { |
4392 | unsigned long nr = 0; | |
4393 | enum lru_list lru; | |
4394 | ||
4395 | for_each_lru(lru) { | |
4396 | if (!(BIT(lru) & lru_mask)) | |
4397 | continue; | |
dd8657b6 SB |
4398 | if (tree) |
4399 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
4400 | else | |
4401 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
4402 | } |
4403 | return nr; | |
4404 | } | |
4405 | ||
2da8ca82 | 4406 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 4407 | { |
25485de6 GT |
4408 | struct numa_stat { |
4409 | const char *name; | |
4410 | unsigned int lru_mask; | |
4411 | }; | |
4412 | ||
4413 | static const struct numa_stat stats[] = { | |
4414 | { "total", LRU_ALL }, | |
4415 | { "file", LRU_ALL_FILE }, | |
4416 | { "anon", LRU_ALL_ANON }, | |
4417 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4418 | }; | |
4419 | const struct numa_stat *stat; | |
406eb0c9 | 4420 | int nid; |
aa9694bb | 4421 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 4422 | |
7d7ef0a4 | 4423 | mem_cgroup_flush_stats(memcg); |
2d146aa3 | 4424 | |
25485de6 | 4425 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4426 | seq_printf(m, "%s=%lu", stat->name, |
4427 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4428 | false)); | |
4429 | for_each_node_state(nid, N_MEMORY) | |
4430 | seq_printf(m, " N%d=%lu", nid, | |
4431 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4432 | stat->lru_mask, false)); | |
25485de6 | 4433 | seq_putc(m, '\n'); |
406eb0c9 | 4434 | } |
406eb0c9 | 4435 | |
071aee13 | 4436 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4437 | |
4438 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
4439 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4440 | true)); | |
4441 | for_each_node_state(nid, N_MEMORY) | |
4442 | seq_printf(m, " N%d=%lu", nid, | |
4443 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4444 | stat->lru_mask, true)); | |
071aee13 | 4445 | seq_putc(m, '\n'); |
406eb0c9 | 4446 | } |
406eb0c9 | 4447 | |
406eb0c9 YH |
4448 | return 0; |
4449 | } | |
4450 | #endif /* CONFIG_NUMA */ | |
4451 | ||
c8713d0b | 4452 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4453 | NR_FILE_PAGES, |
be5d0a74 | 4454 | NR_ANON_MAPPED, |
468c3982 JW |
4455 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4456 | NR_ANON_THPS, | |
4457 | #endif | |
c8713d0b JW |
4458 | NR_SHMEM, |
4459 | NR_FILE_MAPPED, | |
4460 | NR_FILE_DIRTY, | |
4461 | NR_WRITEBACK, | |
e09b0b61 YS |
4462 | WORKINGSET_REFAULT_ANON, |
4463 | WORKINGSET_REFAULT_FILE, | |
72a14e82 | 4464 | #ifdef CONFIG_SWAP |
c8713d0b | 4465 | MEMCG_SWAP, |
72a14e82 LS |
4466 | NR_SWAPCACHE, |
4467 | #endif | |
c8713d0b JW |
4468 | }; |
4469 | ||
4470 | static const char *const memcg1_stat_names[] = { | |
4471 | "cache", | |
4472 | "rss", | |
468c3982 | 4473 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4474 | "rss_huge", |
468c3982 | 4475 | #endif |
c8713d0b JW |
4476 | "shmem", |
4477 | "mapped_file", | |
4478 | "dirty", | |
4479 | "writeback", | |
e09b0b61 YS |
4480 | "workingset_refault_anon", |
4481 | "workingset_refault_file", | |
72a14e82 | 4482 | #ifdef CONFIG_SWAP |
c8713d0b | 4483 | "swap", |
72a14e82 LS |
4484 | "swapcached", |
4485 | #endif | |
c8713d0b JW |
4486 | }; |
4487 | ||
df0e53d0 | 4488 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4489 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4490 | PGPGIN, |
4491 | PGPGOUT, | |
4492 | PGFAULT, | |
4493 | PGMAJFAULT, | |
4494 | }; | |
4495 | ||
dddb44ff | 4496 | static void memcg1_stat_format(struct mem_cgroup *memcg, struct seq_buf *s) |
d2ceb9b7 | 4497 | { |
3e32cb2e | 4498 | unsigned long memory, memsw; |
af7c4b0e JW |
4499 | struct mem_cgroup *mi; |
4500 | unsigned int i; | |
406eb0c9 | 4501 | |
71cd3113 | 4502 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4503 | |
7d7ef0a4 | 4504 | mem_cgroup_flush_stats(memcg); |
2d146aa3 | 4505 | |
71cd3113 | 4506 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4507 | unsigned long nr; |
4508 | ||
ff841a06 YA |
4509 | nr = memcg_page_state_local_output(memcg, memcg1_stats[i]); |
4510 | seq_buf_printf(s, "%s %lu\n", memcg1_stat_names[i], nr); | |
1dd3a273 | 4511 | } |
7b854121 | 4512 | |
df0e53d0 | 4513 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
dddb44ff YA |
4514 | seq_buf_printf(s, "%s %lu\n", vm_event_name(memcg1_events[i]), |
4515 | memcg_events_local(memcg, memcg1_events[i])); | |
af7c4b0e JW |
4516 | |
4517 | for (i = 0; i < NR_LRU_LISTS; i++) | |
dddb44ff YA |
4518 | seq_buf_printf(s, "%s %lu\n", lru_list_name(i), |
4519 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * | |
4520 | PAGE_SIZE); | |
af7c4b0e | 4521 | |
14067bb3 | 4522 | /* Hierarchical information */ |
3e32cb2e JW |
4523 | memory = memsw = PAGE_COUNTER_MAX; |
4524 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4525 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4526 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4527 | } |
dddb44ff YA |
4528 | seq_buf_printf(s, "hierarchical_memory_limit %llu\n", |
4529 | (u64)memory * PAGE_SIZE); | |
840ea53a LS |
4530 | seq_buf_printf(s, "hierarchical_memsw_limit %llu\n", |
4531 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4532 | |
8de7ecc6 | 4533 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4534 | unsigned long nr; |
4535 | ||
ff841a06 | 4536 | nr = memcg_page_state_output(memcg, memcg1_stats[i]); |
dddb44ff | 4537 | seq_buf_printf(s, "total_%s %llu\n", memcg1_stat_names[i], |
ff841a06 | 4538 | (u64)nr); |
af7c4b0e JW |
4539 | } |
4540 | ||
8de7ecc6 | 4541 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
dddb44ff YA |
4542 | seq_buf_printf(s, "total_%s %llu\n", |
4543 | vm_event_name(memcg1_events[i]), | |
4544 | (u64)memcg_events(memcg, memcg1_events[i])); | |
af7c4b0e | 4545 | |
8de7ecc6 | 4546 | for (i = 0; i < NR_LRU_LISTS; i++) |
dddb44ff YA |
4547 | seq_buf_printf(s, "total_%s %llu\n", lru_list_name(i), |
4548 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * | |
4549 | PAGE_SIZE); | |
14067bb3 | 4550 | |
7f016ee8 | 4551 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4552 | { |
ef8f2327 MG |
4553 | pg_data_t *pgdat; |
4554 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4555 | unsigned long anon_cost = 0; |
4556 | unsigned long file_cost = 0; | |
7f016ee8 | 4557 | |
ef8f2327 | 4558 | for_each_online_pgdat(pgdat) { |
a3747b53 | 4559 | mz = memcg->nodeinfo[pgdat->node_id]; |
7f016ee8 | 4560 | |
1431d4d1 JW |
4561 | anon_cost += mz->lruvec.anon_cost; |
4562 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4563 | } |
dddb44ff YA |
4564 | seq_buf_printf(s, "anon_cost %lu\n", anon_cost); |
4565 | seq_buf_printf(s, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4566 | } |
4567 | #endif | |
d2ceb9b7 KH |
4568 | } |
4569 | ||
182446d0 TH |
4570 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4571 | struct cftype *cft) | |
a7885eb8 | 4572 | { |
182446d0 | 4573 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4574 | |
1f4c025b | 4575 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4576 | } |
4577 | ||
182446d0 TH |
4578 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4579 | struct cftype *cft, u64 val) | |
a7885eb8 | 4580 | { |
182446d0 | 4581 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4582 | |
37bc3cb9 | 4583 | if (val > 200) |
a7885eb8 KM |
4584 | return -EINVAL; |
4585 | ||
a4792030 | 4586 | if (!mem_cgroup_is_root(memcg)) |
82b3aa26 | 4587 | WRITE_ONCE(memcg->swappiness, val); |
3dae7fec | 4588 | else |
82b3aa26 | 4589 | WRITE_ONCE(vm_swappiness, val); |
068b38c1 | 4590 | |
a7885eb8 KM |
4591 | return 0; |
4592 | } | |
4593 | ||
2e72b634 KS |
4594 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4595 | { | |
4596 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4597 | unsigned long usage; |
2e72b634 KS |
4598 | int i; |
4599 | ||
4600 | rcu_read_lock(); | |
4601 | if (!swap) | |
2c488db2 | 4602 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4603 | else |
2c488db2 | 4604 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4605 | |
4606 | if (!t) | |
4607 | goto unlock; | |
4608 | ||
ce00a967 | 4609 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4610 | |
4611 | /* | |
748dad36 | 4612 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4613 | * If it's not true, a threshold was crossed after last |
4614 | * call of __mem_cgroup_threshold(). | |
4615 | */ | |
5407a562 | 4616 | i = t->current_threshold; |
2e72b634 KS |
4617 | |
4618 | /* | |
4619 | * Iterate backward over array of thresholds starting from | |
4620 | * current_threshold and check if a threshold is crossed. | |
4621 | * If none of thresholds below usage is crossed, we read | |
4622 | * only one element of the array here. | |
4623 | */ | |
4624 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3652117f | 4625 | eventfd_signal(t->entries[i].eventfd); |
2e72b634 KS |
4626 | |
4627 | /* i = current_threshold + 1 */ | |
4628 | i++; | |
4629 | ||
4630 | /* | |
4631 | * Iterate forward over array of thresholds starting from | |
4632 | * current_threshold+1 and check if a threshold is crossed. | |
4633 | * If none of thresholds above usage is crossed, we read | |
4634 | * only one element of the array here. | |
4635 | */ | |
4636 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3652117f | 4637 | eventfd_signal(t->entries[i].eventfd); |
2e72b634 KS |
4638 | |
4639 | /* Update current_threshold */ | |
5407a562 | 4640 | t->current_threshold = i - 1; |
2e72b634 KS |
4641 | unlock: |
4642 | rcu_read_unlock(); | |
4643 | } | |
4644 | ||
4645 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4646 | { | |
ad4ca5f4 KS |
4647 | while (memcg) { |
4648 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4649 | if (do_memsw_account()) |
ad4ca5f4 KS |
4650 | __mem_cgroup_threshold(memcg, true); |
4651 | ||
4652 | memcg = parent_mem_cgroup(memcg); | |
4653 | } | |
2e72b634 KS |
4654 | } |
4655 | ||
4656 | static int compare_thresholds(const void *a, const void *b) | |
4657 | { | |
4658 | const struct mem_cgroup_threshold *_a = a; | |
4659 | const struct mem_cgroup_threshold *_b = b; | |
4660 | ||
2bff24a3 GT |
4661 | if (_a->threshold > _b->threshold) |
4662 | return 1; | |
4663 | ||
4664 | if (_a->threshold < _b->threshold) | |
4665 | return -1; | |
4666 | ||
4667 | return 0; | |
2e72b634 KS |
4668 | } |
4669 | ||
c0ff4b85 | 4670 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4671 | { |
4672 | struct mem_cgroup_eventfd_list *ev; | |
4673 | ||
2bcf2e92 MH |
4674 | spin_lock(&memcg_oom_lock); |
4675 | ||
c0ff4b85 | 4676 | list_for_each_entry(ev, &memcg->oom_notify, list) |
3652117f | 4677 | eventfd_signal(ev->eventfd); |
2bcf2e92 MH |
4678 | |
4679 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4680 | return 0; |
4681 | } | |
4682 | ||
c0ff4b85 | 4683 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4684 | { |
7d74b06f KH |
4685 | struct mem_cgroup *iter; |
4686 | ||
c0ff4b85 | 4687 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4688 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4689 | } |
4690 | ||
59b6f873 | 4691 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4692 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4693 | { |
2c488db2 KS |
4694 | struct mem_cgroup_thresholds *thresholds; |
4695 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4696 | unsigned long threshold; |
4697 | unsigned long usage; | |
2c488db2 | 4698 | int i, size, ret; |
2e72b634 | 4699 | |
650c5e56 | 4700 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4701 | if (ret) |
4702 | return ret; | |
4703 | ||
4704 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4705 | |
05b84301 | 4706 | if (type == _MEM) { |
2c488db2 | 4707 | thresholds = &memcg->thresholds; |
ce00a967 | 4708 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4709 | } else if (type == _MEMSWAP) { |
2c488db2 | 4710 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4711 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4712 | } else |
2e72b634 KS |
4713 | BUG(); |
4714 | ||
2e72b634 | 4715 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4716 | if (thresholds->primary) |
2e72b634 KS |
4717 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4718 | ||
2c488db2 | 4719 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4720 | |
4721 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4722 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4723 | if (!new) { |
2e72b634 KS |
4724 | ret = -ENOMEM; |
4725 | goto unlock; | |
4726 | } | |
2c488db2 | 4727 | new->size = size; |
2e72b634 KS |
4728 | |
4729 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4730 | if (thresholds->primary) |
4731 | memcpy(new->entries, thresholds->primary->entries, | |
4732 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4733 | |
2e72b634 | 4734 | /* Add new threshold */ |
2c488db2 KS |
4735 | new->entries[size - 1].eventfd = eventfd; |
4736 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4737 | |
4738 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4739 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4740 | compare_thresholds, NULL); |
4741 | ||
4742 | /* Find current threshold */ | |
2c488db2 | 4743 | new->current_threshold = -1; |
2e72b634 | 4744 | for (i = 0; i < size; i++) { |
748dad36 | 4745 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4746 | /* |
2c488db2 KS |
4747 | * new->current_threshold will not be used until |
4748 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4749 | * it here. |
4750 | */ | |
2c488db2 | 4751 | ++new->current_threshold; |
748dad36 SZ |
4752 | } else |
4753 | break; | |
2e72b634 KS |
4754 | } |
4755 | ||
2c488db2 KS |
4756 | /* Free old spare buffer and save old primary buffer as spare */ |
4757 | kfree(thresholds->spare); | |
4758 | thresholds->spare = thresholds->primary; | |
4759 | ||
4760 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4761 | |
907860ed | 4762 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4763 | synchronize_rcu(); |
4764 | ||
2e72b634 KS |
4765 | unlock: |
4766 | mutex_unlock(&memcg->thresholds_lock); | |
4767 | ||
4768 | return ret; | |
4769 | } | |
4770 | ||
59b6f873 | 4771 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4772 | struct eventfd_ctx *eventfd, const char *args) |
4773 | { | |
59b6f873 | 4774 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4775 | } |
4776 | ||
59b6f873 | 4777 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4778 | struct eventfd_ctx *eventfd, const char *args) |
4779 | { | |
59b6f873 | 4780 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4781 | } |
4782 | ||
59b6f873 | 4783 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4784 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4785 | { |
2c488db2 KS |
4786 | struct mem_cgroup_thresholds *thresholds; |
4787 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4788 | unsigned long usage; |
7d36665a | 4789 | int i, j, size, entries; |
2e72b634 KS |
4790 | |
4791 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4792 | |
4793 | if (type == _MEM) { | |
2c488db2 | 4794 | thresholds = &memcg->thresholds; |
ce00a967 | 4795 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4796 | } else if (type == _MEMSWAP) { |
2c488db2 | 4797 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4798 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4799 | } else |
2e72b634 KS |
4800 | BUG(); |
4801 | ||
371528ca AV |
4802 | if (!thresholds->primary) |
4803 | goto unlock; | |
4804 | ||
2e72b634 KS |
4805 | /* Check if a threshold crossed before removing */ |
4806 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4807 | ||
4808 | /* Calculate new number of threshold */ | |
7d36665a | 4809 | size = entries = 0; |
2c488db2 KS |
4810 | for (i = 0; i < thresholds->primary->size; i++) { |
4811 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4812 | size++; |
7d36665a CX |
4813 | else |
4814 | entries++; | |
2e72b634 KS |
4815 | } |
4816 | ||
2c488db2 | 4817 | new = thresholds->spare; |
907860ed | 4818 | |
7d36665a CX |
4819 | /* If no items related to eventfd have been cleared, nothing to do */ |
4820 | if (!entries) | |
4821 | goto unlock; | |
4822 | ||
2e72b634 KS |
4823 | /* Set thresholds array to NULL if we don't have thresholds */ |
4824 | if (!size) { | |
2c488db2 KS |
4825 | kfree(new); |
4826 | new = NULL; | |
907860ed | 4827 | goto swap_buffers; |
2e72b634 KS |
4828 | } |
4829 | ||
2c488db2 | 4830 | new->size = size; |
2e72b634 KS |
4831 | |
4832 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4833 | new->current_threshold = -1; |
4834 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4835 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4836 | continue; |
4837 | ||
2c488db2 | 4838 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4839 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4840 | /* |
2c488db2 | 4841 | * new->current_threshold will not be used |
2e72b634 KS |
4842 | * until rcu_assign_pointer(), so it's safe to increment |
4843 | * it here. | |
4844 | */ | |
2c488db2 | 4845 | ++new->current_threshold; |
2e72b634 KS |
4846 | } |
4847 | j++; | |
4848 | } | |
4849 | ||
907860ed | 4850 | swap_buffers: |
2c488db2 KS |
4851 | /* Swap primary and spare array */ |
4852 | thresholds->spare = thresholds->primary; | |
8c757763 | 4853 | |
2c488db2 | 4854 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4855 | |
907860ed | 4856 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4857 | synchronize_rcu(); |
6611d8d7 MC |
4858 | |
4859 | /* If all events are unregistered, free the spare array */ | |
4860 | if (!new) { | |
4861 | kfree(thresholds->spare); | |
4862 | thresholds->spare = NULL; | |
4863 | } | |
371528ca | 4864 | unlock: |
2e72b634 | 4865 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4866 | } |
c1e862c1 | 4867 | |
59b6f873 | 4868 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4869 | struct eventfd_ctx *eventfd) |
4870 | { | |
59b6f873 | 4871 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4872 | } |
4873 | ||
59b6f873 | 4874 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4875 | struct eventfd_ctx *eventfd) |
4876 | { | |
59b6f873 | 4877 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4878 | } |
4879 | ||
59b6f873 | 4880 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4881 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4882 | { |
9490ff27 | 4883 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4884 | |
9490ff27 KH |
4885 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4886 | if (!event) | |
4887 | return -ENOMEM; | |
4888 | ||
1af8efe9 | 4889 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4890 | |
4891 | event->eventfd = eventfd; | |
4892 | list_add(&event->list, &memcg->oom_notify); | |
4893 | ||
4894 | /* already in OOM ? */ | |
c2b42d3c | 4895 | if (memcg->under_oom) |
3652117f | 4896 | eventfd_signal(eventfd); |
1af8efe9 | 4897 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4898 | |
4899 | return 0; | |
4900 | } | |
4901 | ||
59b6f873 | 4902 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4903 | struct eventfd_ctx *eventfd) |
9490ff27 | 4904 | { |
9490ff27 | 4905 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4906 | |
1af8efe9 | 4907 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4908 | |
c0ff4b85 | 4909 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4910 | if (ev->eventfd == eventfd) { |
4911 | list_del(&ev->list); | |
4912 | kfree(ev); | |
4913 | } | |
4914 | } | |
4915 | ||
1af8efe9 | 4916 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4917 | } |
4918 | ||
2da8ca82 | 4919 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4920 | { |
aa9694bb | 4921 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4922 | |
17c56de6 | 4923 | seq_printf(sf, "oom_kill_disable %d\n", READ_ONCE(memcg->oom_kill_disable)); |
c2b42d3c | 4924 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4925 | seq_printf(sf, "oom_kill %lu\n", |
4926 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4927 | return 0; |
4928 | } | |
4929 | ||
182446d0 | 4930 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4931 | struct cftype *cft, u64 val) |
4932 | { | |
182446d0 | 4933 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4934 | |
4935 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4936 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4937 | return -EINVAL; |
4938 | ||
17c56de6 | 4939 | WRITE_ONCE(memcg->oom_kill_disable, val); |
4d845ebf | 4940 | if (!val) |
c0ff4b85 | 4941 | memcg_oom_recover(memcg); |
3dae7fec | 4942 | |
3c11ecf4 KH |
4943 | return 0; |
4944 | } | |
4945 | ||
52ebea74 TH |
4946 | #ifdef CONFIG_CGROUP_WRITEBACK |
4947 | ||
3a8e9ac8 TH |
4948 | #include <trace/events/writeback.h> |
4949 | ||
841710aa TH |
4950 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4951 | { | |
4952 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4953 | } | |
4954 | ||
4955 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4956 | { | |
4957 | wb_domain_exit(&memcg->cgwb_domain); | |
4958 | } | |
4959 | ||
2529bb3a TH |
4960 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4961 | { | |
4962 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4963 | } | |
4964 | ||
841710aa TH |
4965 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4966 | { | |
4967 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4968 | ||
4969 | if (!memcg->css.parent) | |
4970 | return NULL; | |
4971 | ||
4972 | return &memcg->cgwb_domain; | |
4973 | } | |
4974 | ||
c2aa723a TH |
4975 | /** |
4976 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4977 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4978 | * @pfilepages: out parameter for number of file pages |
4979 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4980 | * @pdirty: out parameter for number of dirty pages |
4981 | * @pwriteback: out parameter for number of pages under writeback | |
4982 | * | |
c5edf9cd TH |
4983 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4984 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4985 | * is a bit more involved. | |
c2aa723a | 4986 | * |
c5edf9cd TH |
4987 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4988 | * headroom is calculated as the lowest headroom of itself and the | |
4989 | * ancestors. Note that this doesn't consider the actual amount of | |
4990 | * available memory in the system. The caller should further cap | |
4991 | * *@pheadroom accordingly. | |
c2aa723a | 4992 | */ |
c5edf9cd TH |
4993 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4994 | unsigned long *pheadroom, unsigned long *pdirty, | |
4995 | unsigned long *pwriteback) | |
c2aa723a TH |
4996 | { |
4997 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4998 | struct mem_cgroup *parent; | |
c2aa723a | 4999 | |
d9b3ce87 | 5000 | mem_cgroup_flush_stats_ratelimited(memcg); |
c2aa723a | 5001 | |
2d146aa3 JW |
5002 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
5003 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); | |
5004 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + | |
5005 | memcg_page_state(memcg, NR_ACTIVE_FILE); | |
c2aa723a | 5006 | |
2d146aa3 | 5007 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 5008 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 5009 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 5010 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
5011 | unsigned long used = page_counter_read(&memcg->memory); |
5012 | ||
c5edf9cd | 5013 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
5014 | memcg = parent; |
5015 | } | |
c2aa723a TH |
5016 | } |
5017 | ||
97b27821 TH |
5018 | /* |
5019 | * Foreign dirty flushing | |
5020 | * | |
5021 | * There's an inherent mismatch between memcg and writeback. The former | |
f0953a1b | 5022 | * tracks ownership per-page while the latter per-inode. This was a |
97b27821 TH |
5023 | * deliberate design decision because honoring per-page ownership in the |
5024 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
5025 | * and deemed unnecessary given that write-sharing an inode across | |
5026 | * different cgroups isn't a common use-case. | |
5027 | * | |
5028 | * Combined with inode majority-writer ownership switching, this works well | |
5029 | * enough in most cases but there are some pathological cases. For | |
5030 | * example, let's say there are two cgroups A and B which keep writing to | |
5031 | * different but confined parts of the same inode. B owns the inode and | |
5032 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
5033 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
5034 | * triggering background writeback. A will be slowed down without a way to | |
5035 | * make writeback of the dirty pages happen. | |
5036 | * | |
f0953a1b | 5037 | * Conditions like the above can lead to a cgroup getting repeatedly and |
97b27821 | 5038 | * severely throttled after making some progress after each |
f0953a1b | 5039 | * dirty_expire_interval while the underlying IO device is almost |
97b27821 TH |
5040 | * completely idle. |
5041 | * | |
5042 | * Solving this problem completely requires matching the ownership tracking | |
5043 | * granularities between memcg and writeback in either direction. However, | |
5044 | * the more egregious behaviors can be avoided by simply remembering the | |
5045 | * most recent foreign dirtying events and initiating remote flushes on | |
5046 | * them when local writeback isn't enough to keep the memory clean enough. | |
5047 | * | |
5048 | * The following two functions implement such mechanism. When a foreign | |
5049 | * page - a page whose memcg and writeback ownerships don't match - is | |
5050 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
5051 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
5052 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
5053 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
5054 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
5055 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
5056 | * limited to MEMCG_CGWB_FRN_CNT. | |
5057 | * | |
5058 | * The mechanism only remembers IDs and doesn't hold any object references. | |
5059 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
5060 | * records are lockless and racy. | |
5061 | */ | |
9d8053fc | 5062 | void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio, |
97b27821 TH |
5063 | struct bdi_writeback *wb) |
5064 | { | |
9d8053fc | 5065 | struct mem_cgroup *memcg = folio_memcg(folio); |
97b27821 TH |
5066 | struct memcg_cgwb_frn *frn; |
5067 | u64 now = get_jiffies_64(); | |
5068 | u64 oldest_at = now; | |
5069 | int oldest = -1; | |
5070 | int i; | |
5071 | ||
9d8053fc | 5072 | trace_track_foreign_dirty(folio, wb); |
3a8e9ac8 | 5073 | |
97b27821 TH |
5074 | /* |
5075 | * Pick the slot to use. If there is already a slot for @wb, keep | |
5076 | * using it. If not replace the oldest one which isn't being | |
5077 | * written out. | |
5078 | */ | |
5079 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
5080 | frn = &memcg->cgwb_frn[i]; | |
5081 | if (frn->bdi_id == wb->bdi->id && | |
5082 | frn->memcg_id == wb->memcg_css->id) | |
5083 | break; | |
5084 | if (time_before64(frn->at, oldest_at) && | |
5085 | atomic_read(&frn->done.cnt) == 1) { | |
5086 | oldest = i; | |
5087 | oldest_at = frn->at; | |
5088 | } | |
5089 | } | |
5090 | ||
5091 | if (i < MEMCG_CGWB_FRN_CNT) { | |
5092 | /* | |
5093 | * Re-using an existing one. Update timestamp lazily to | |
5094 | * avoid making the cacheline hot. We want them to be | |
5095 | * reasonably up-to-date and significantly shorter than | |
5096 | * dirty_expire_interval as that's what expires the record. | |
5097 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
5098 | */ | |
5099 | unsigned long update_intv = | |
5100 | min_t(unsigned long, HZ, | |
5101 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
5102 | ||
5103 | if (time_before64(frn->at, now - update_intv)) | |
5104 | frn->at = now; | |
5105 | } else if (oldest >= 0) { | |
5106 | /* replace the oldest free one */ | |
5107 | frn = &memcg->cgwb_frn[oldest]; | |
5108 | frn->bdi_id = wb->bdi->id; | |
5109 | frn->memcg_id = wb->memcg_css->id; | |
5110 | frn->at = now; | |
5111 | } | |
5112 | } | |
5113 | ||
5114 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
5115 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
5116 | { | |
5117 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
5118 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
5119 | u64 now = jiffies_64; | |
5120 | int i; | |
5121 | ||
5122 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
5123 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
5124 | ||
5125 | /* | |
5126 | * If the record is older than dirty_expire_interval, | |
5127 | * writeback on it has already started. No need to kick it | |
5128 | * off again. Also, don't start a new one if there's | |
5129 | * already one in flight. | |
5130 | */ | |
5131 | if (time_after64(frn->at, now - intv) && | |
5132 | atomic_read(&frn->done.cnt) == 1) { | |
5133 | frn->at = 0; | |
3a8e9ac8 | 5134 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
7490a2d2 | 5135 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, |
97b27821 TH |
5136 | WB_REASON_FOREIGN_FLUSH, |
5137 | &frn->done); | |
5138 | } | |
5139 | } | |
5140 | } | |
5141 | ||
841710aa TH |
5142 | #else /* CONFIG_CGROUP_WRITEBACK */ |
5143 | ||
5144 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
5145 | { | |
5146 | return 0; | |
5147 | } | |
5148 | ||
5149 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
5150 | { | |
5151 | } | |
5152 | ||
2529bb3a TH |
5153 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
5154 | { | |
5155 | } | |
5156 | ||
52ebea74 TH |
5157 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
5158 | ||
3bc942f3 TH |
5159 | /* |
5160 | * DO NOT USE IN NEW FILES. | |
5161 | * | |
5162 | * "cgroup.event_control" implementation. | |
5163 | * | |
5164 | * This is way over-engineered. It tries to support fully configurable | |
5165 | * events for each user. Such level of flexibility is completely | |
5166 | * unnecessary especially in the light of the planned unified hierarchy. | |
5167 | * | |
5168 | * Please deprecate this and replace with something simpler if at all | |
5169 | * possible. | |
5170 | */ | |
5171 | ||
79bd9814 TH |
5172 | /* |
5173 | * Unregister event and free resources. | |
5174 | * | |
5175 | * Gets called from workqueue. | |
5176 | */ | |
3bc942f3 | 5177 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 5178 | { |
3bc942f3 TH |
5179 | struct mem_cgroup_event *event = |
5180 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 5181 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5182 | |
5183 | remove_wait_queue(event->wqh, &event->wait); | |
5184 | ||
59b6f873 | 5185 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
5186 | |
5187 | /* Notify userspace the event is going away. */ | |
3652117f | 5188 | eventfd_signal(event->eventfd); |
79bd9814 TH |
5189 | |
5190 | eventfd_ctx_put(event->eventfd); | |
5191 | kfree(event); | |
59b6f873 | 5192 | css_put(&memcg->css); |
79bd9814 TH |
5193 | } |
5194 | ||
5195 | /* | |
a9a08845 | 5196 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
5197 | * |
5198 | * Called with wqh->lock held and interrupts disabled. | |
5199 | */ | |
ac6424b9 | 5200 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 5201 | int sync, void *key) |
79bd9814 | 5202 | { |
3bc942f3 TH |
5203 | struct mem_cgroup_event *event = |
5204 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 5205 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 5206 | __poll_t flags = key_to_poll(key); |
79bd9814 | 5207 | |
a9a08845 | 5208 | if (flags & EPOLLHUP) { |
79bd9814 TH |
5209 | /* |
5210 | * If the event has been detached at cgroup removal, we | |
5211 | * can simply return knowing the other side will cleanup | |
5212 | * for us. | |
5213 | * | |
5214 | * We can't race against event freeing since the other | |
5215 | * side will require wqh->lock via remove_wait_queue(), | |
5216 | * which we hold. | |
5217 | */ | |
fba94807 | 5218 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
5219 | if (!list_empty(&event->list)) { |
5220 | list_del_init(&event->list); | |
5221 | /* | |
5222 | * We are in atomic context, but cgroup_event_remove() | |
5223 | * may sleep, so we have to call it in workqueue. | |
5224 | */ | |
5225 | schedule_work(&event->remove); | |
5226 | } | |
fba94807 | 5227 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
5228 | } |
5229 | ||
5230 | return 0; | |
5231 | } | |
5232 | ||
3bc942f3 | 5233 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
5234 | wait_queue_head_t *wqh, poll_table *pt) |
5235 | { | |
3bc942f3 TH |
5236 | struct mem_cgroup_event *event = |
5237 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
5238 | |
5239 | event->wqh = wqh; | |
5240 | add_wait_queue(wqh, &event->wait); | |
5241 | } | |
5242 | ||
5243 | /* | |
3bc942f3 TH |
5244 | * DO NOT USE IN NEW FILES. |
5245 | * | |
79bd9814 TH |
5246 | * Parse input and register new cgroup event handler. |
5247 | * | |
5248 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
5249 | * Interpretation of args is defined by control file implementation. | |
5250 | */ | |
451af504 TH |
5251 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
5252 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 5253 | { |
451af504 | 5254 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 5255 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5256 | struct mem_cgroup_event *event; |
79bd9814 TH |
5257 | struct cgroup_subsys_state *cfile_css; |
5258 | unsigned int efd, cfd; | |
5259 | struct fd efile; | |
5260 | struct fd cfile; | |
4a7ba45b | 5261 | struct dentry *cdentry; |
fba94807 | 5262 | const char *name; |
79bd9814 TH |
5263 | char *endp; |
5264 | int ret; | |
5265 | ||
2343e88d SAS |
5266 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
5267 | return -EOPNOTSUPP; | |
5268 | ||
451af504 TH |
5269 | buf = strstrip(buf); |
5270 | ||
5271 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
5272 | if (*endp != ' ') |
5273 | return -EINVAL; | |
451af504 | 5274 | buf = endp + 1; |
79bd9814 | 5275 | |
451af504 | 5276 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
5277 | if ((*endp != ' ') && (*endp != '\0')) |
5278 | return -EINVAL; | |
451af504 | 5279 | buf = endp + 1; |
79bd9814 TH |
5280 | |
5281 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
5282 | if (!event) | |
5283 | return -ENOMEM; | |
5284 | ||
59b6f873 | 5285 | event->memcg = memcg; |
79bd9814 | 5286 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
5287 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
5288 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
5289 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
5290 | |
5291 | efile = fdget(efd); | |
5292 | if (!efile.file) { | |
5293 | ret = -EBADF; | |
5294 | goto out_kfree; | |
5295 | } | |
5296 | ||
5297 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
5298 | if (IS_ERR(event->eventfd)) { | |
5299 | ret = PTR_ERR(event->eventfd); | |
5300 | goto out_put_efile; | |
5301 | } | |
5302 | ||
5303 | cfile = fdget(cfd); | |
5304 | if (!cfile.file) { | |
5305 | ret = -EBADF; | |
5306 | goto out_put_eventfd; | |
5307 | } | |
5308 | ||
5309 | /* the process need read permission on control file */ | |
5310 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 5311 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
5312 | if (ret < 0) |
5313 | goto out_put_cfile; | |
5314 | ||
4a7ba45b TH |
5315 | /* |
5316 | * The control file must be a regular cgroup1 file. As a regular cgroup | |
5317 | * file can't be renamed, it's safe to access its name afterwards. | |
5318 | */ | |
5319 | cdentry = cfile.file->f_path.dentry; | |
5320 | if (cdentry->d_sb->s_type != &cgroup_fs_type || !d_is_reg(cdentry)) { | |
5321 | ret = -EINVAL; | |
5322 | goto out_put_cfile; | |
5323 | } | |
5324 | ||
fba94807 TH |
5325 | /* |
5326 | * Determine the event callbacks and set them in @event. This used | |
5327 | * to be done via struct cftype but cgroup core no longer knows | |
5328 | * about these events. The following is crude but the whole thing | |
5329 | * is for compatibility anyway. | |
3bc942f3 TH |
5330 | * |
5331 | * DO NOT ADD NEW FILES. | |
fba94807 | 5332 | */ |
4a7ba45b | 5333 | name = cdentry->d_name.name; |
fba94807 TH |
5334 | |
5335 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
5336 | event->register_event = mem_cgroup_usage_register_event; | |
5337 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
5338 | } else if (!strcmp(name, "memory.oom_control")) { | |
5339 | event->register_event = mem_cgroup_oom_register_event; | |
5340 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
5341 | } else if (!strcmp(name, "memory.pressure_level")) { | |
5342 | event->register_event = vmpressure_register_event; | |
5343 | event->unregister_event = vmpressure_unregister_event; | |
5344 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
5345 | event->register_event = memsw_cgroup_usage_register_event; |
5346 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
5347 | } else { |
5348 | ret = -EINVAL; | |
5349 | goto out_put_cfile; | |
5350 | } | |
5351 | ||
79bd9814 | 5352 | /* |
b5557c4c TH |
5353 | * Verify @cfile should belong to @css. Also, remaining events are |
5354 | * automatically removed on cgroup destruction but the removal is | |
5355 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 5356 | */ |
4a7ba45b | 5357 | cfile_css = css_tryget_online_from_dir(cdentry->d_parent, |
ec903c0c | 5358 | &memory_cgrp_subsys); |
79bd9814 | 5359 | ret = -EINVAL; |
5a17f543 | 5360 | if (IS_ERR(cfile_css)) |
79bd9814 | 5361 | goto out_put_cfile; |
5a17f543 TH |
5362 | if (cfile_css != css) { |
5363 | css_put(cfile_css); | |
79bd9814 | 5364 | goto out_put_cfile; |
5a17f543 | 5365 | } |
79bd9814 | 5366 | |
451af504 | 5367 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
5368 | if (ret) |
5369 | goto out_put_css; | |
5370 | ||
9965ed17 | 5371 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 5372 | |
4ba9515d | 5373 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5374 | list_add(&event->list, &memcg->event_list); |
4ba9515d | 5375 | spin_unlock_irq(&memcg->event_list_lock); |
79bd9814 TH |
5376 | |
5377 | fdput(cfile); | |
5378 | fdput(efile); | |
5379 | ||
451af504 | 5380 | return nbytes; |
79bd9814 TH |
5381 | |
5382 | out_put_css: | |
b5557c4c | 5383 | css_put(css); |
79bd9814 TH |
5384 | out_put_cfile: |
5385 | fdput(cfile); | |
5386 | out_put_eventfd: | |
5387 | eventfd_ctx_put(event->eventfd); | |
5388 | out_put_efile: | |
5389 | fdput(efile); | |
5390 | out_kfree: | |
5391 | kfree(event); | |
5392 | ||
5393 | return ret; | |
5394 | } | |
5395 | ||
bc3dcb85 | 5396 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_SLUB_DEBUG) |
c29b5b3d MS |
5397 | static int mem_cgroup_slab_show(struct seq_file *m, void *p) |
5398 | { | |
5399 | /* | |
5400 | * Deprecated. | |
df4ae285 | 5401 | * Please, take a look at tools/cgroup/memcg_slabinfo.py . |
c29b5b3d MS |
5402 | */ |
5403 | return 0; | |
5404 | } | |
5405 | #endif | |
5406 | ||
dddb44ff YA |
5407 | static int memory_stat_show(struct seq_file *m, void *v); |
5408 | ||
241994ed | 5409 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 5410 | { |
0eea1030 | 5411 | .name = "usage_in_bytes", |
8c7c6e34 | 5412 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5413 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5414 | }, |
c84872e1 PE |
5415 | { |
5416 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5417 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5418 | .write = mem_cgroup_reset, |
791badbd | 5419 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5420 | }, |
8cdea7c0 | 5421 | { |
0eea1030 | 5422 | .name = "limit_in_bytes", |
8c7c6e34 | 5423 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5424 | .write = mem_cgroup_write, |
791badbd | 5425 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5426 | }, |
296c81d8 BS |
5427 | { |
5428 | .name = "soft_limit_in_bytes", | |
5429 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5430 | .write = mem_cgroup_write, |
791badbd | 5431 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5432 | }, |
8cdea7c0 BS |
5433 | { |
5434 | .name = "failcnt", | |
8c7c6e34 | 5435 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5436 | .write = mem_cgroup_reset, |
791badbd | 5437 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5438 | }, |
d2ceb9b7 KH |
5439 | { |
5440 | .name = "stat", | |
dddb44ff | 5441 | .seq_show = memory_stat_show, |
d2ceb9b7 | 5442 | }, |
c1e862c1 KH |
5443 | { |
5444 | .name = "force_empty", | |
6770c64e | 5445 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5446 | }, |
18f59ea7 BS |
5447 | { |
5448 | .name = "use_hierarchy", | |
5449 | .write_u64 = mem_cgroup_hierarchy_write, | |
5450 | .read_u64 = mem_cgroup_hierarchy_read, | |
5451 | }, | |
79bd9814 | 5452 | { |
3bc942f3 | 5453 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5454 | .write = memcg_write_event_control, |
7dbdb199 | 5455 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 5456 | }, |
a7885eb8 KM |
5457 | { |
5458 | .name = "swappiness", | |
5459 | .read_u64 = mem_cgroup_swappiness_read, | |
5460 | .write_u64 = mem_cgroup_swappiness_write, | |
5461 | }, | |
7dc74be0 DN |
5462 | { |
5463 | .name = "move_charge_at_immigrate", | |
5464 | .read_u64 = mem_cgroup_move_charge_read, | |
5465 | .write_u64 = mem_cgroup_move_charge_write, | |
5466 | }, | |
9490ff27 KH |
5467 | { |
5468 | .name = "oom_control", | |
2da8ca82 | 5469 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5470 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 | 5471 | }, |
70ddf637 AV |
5472 | { |
5473 | .name = "pressure_level", | |
6b0ba2ab | 5474 | .seq_show = mem_cgroup_dummy_seq_show, |
70ddf637 | 5475 | }, |
406eb0c9 YH |
5476 | #ifdef CONFIG_NUMA |
5477 | { | |
5478 | .name = "numa_stat", | |
2da8ca82 | 5479 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5480 | }, |
5481 | #endif | |
4597648f MH |
5482 | { |
5483 | .name = "kmem.limit_in_bytes", | |
5484 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
5485 | .write = mem_cgroup_write, | |
5486 | .read_u64 = mem_cgroup_read_u64, | |
5487 | }, | |
510fc4e1 GC |
5488 | { |
5489 | .name = "kmem.usage_in_bytes", | |
5490 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5491 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5492 | }, |
5493 | { | |
5494 | .name = "kmem.failcnt", | |
5495 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5496 | .write = mem_cgroup_reset, |
791badbd | 5497 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5498 | }, |
5499 | { | |
5500 | .name = "kmem.max_usage_in_bytes", | |
5501 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5502 | .write = mem_cgroup_reset, |
791badbd | 5503 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5504 | }, |
bc3dcb85 | 5505 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
5506 | { |
5507 | .name = "kmem.slabinfo", | |
c29b5b3d | 5508 | .seq_show = mem_cgroup_slab_show, |
749c5415 GC |
5509 | }, |
5510 | #endif | |
d55f90bf VD |
5511 | { |
5512 | .name = "kmem.tcp.limit_in_bytes", | |
5513 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5514 | .write = mem_cgroup_write, | |
5515 | .read_u64 = mem_cgroup_read_u64, | |
5516 | }, | |
5517 | { | |
5518 | .name = "kmem.tcp.usage_in_bytes", | |
5519 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5520 | .read_u64 = mem_cgroup_read_u64, | |
5521 | }, | |
5522 | { | |
5523 | .name = "kmem.tcp.failcnt", | |
5524 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5525 | .write = mem_cgroup_reset, | |
5526 | .read_u64 = mem_cgroup_read_u64, | |
5527 | }, | |
5528 | { | |
5529 | .name = "kmem.tcp.max_usage_in_bytes", | |
5530 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5531 | .write = mem_cgroup_reset, | |
5532 | .read_u64 = mem_cgroup_read_u64, | |
5533 | }, | |
6bc10349 | 5534 | { }, /* terminate */ |
af36f906 | 5535 | }; |
8c7c6e34 | 5536 | |
73f576c0 JW |
5537 | /* |
5538 | * Private memory cgroup IDR | |
5539 | * | |
5540 | * Swap-out records and page cache shadow entries need to store memcg | |
5541 | * references in constrained space, so we maintain an ID space that is | |
5542 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5543 | * memory-controlled cgroups to 64k. | |
5544 | * | |
b8f2935f | 5545 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5546 | * the cgroup has been destroyed, such as page cache or reclaimable |
5547 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5548 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5549 | * relatively small ID space and prevent the creation of new cgroups | |
5550 | * even when there are much fewer than 64k cgroups - possibly none. | |
5551 | * | |
5552 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5553 | * be freed and recycled when it's no longer needed, which is usually | |
5554 | * when the CSS is offlined. | |
5555 | * | |
5556 | * The only exception to that are records of swapped out tmpfs/shmem | |
5557 | * pages that need to be attributed to live ancestors on swapin. But | |
5558 | * those references are manageable from userspace. | |
5559 | */ | |
5560 | ||
60b1e24c | 5561 | #define MEM_CGROUP_ID_MAX ((1UL << MEM_CGROUP_ID_SHIFT) - 1) |
73f576c0 JW |
5562 | static DEFINE_IDR(mem_cgroup_idr); |
5563 | ||
7e97de0b KT |
5564 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5565 | { | |
5566 | if (memcg->id.id > 0) { | |
5567 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5568 | memcg->id.id = 0; | |
5569 | } | |
5570 | } | |
5571 | ||
c1514c0a VF |
5572 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5573 | unsigned int n) | |
73f576c0 | 5574 | { |
1c2d479a | 5575 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5576 | } |
5577 | ||
615d66c3 | 5578 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5579 | { |
1c2d479a | 5580 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5581 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5582 | |
5583 | /* Memcg ID pins CSS */ | |
5584 | css_put(&memcg->css); | |
5585 | } | |
5586 | } | |
5587 | ||
615d66c3 VD |
5588 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5589 | { | |
5590 | mem_cgroup_id_put_many(memcg, 1); | |
5591 | } | |
5592 | ||
73f576c0 JW |
5593 | /** |
5594 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5595 | * @id: the memcg id to look up | |
5596 | * | |
5597 | * Caller must hold rcu_read_lock(). | |
5598 | */ | |
5599 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5600 | { | |
5601 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5602 | return idr_find(&mem_cgroup_idr, id); | |
5603 | } | |
5604 | ||
c15187a4 RG |
5605 | #ifdef CONFIG_SHRINKER_DEBUG |
5606 | struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino) | |
5607 | { | |
5608 | struct cgroup *cgrp; | |
5609 | struct cgroup_subsys_state *css; | |
5610 | struct mem_cgroup *memcg; | |
5611 | ||
5612 | cgrp = cgroup_get_from_id(ino); | |
fa7e439c | 5613 | if (IS_ERR(cgrp)) |
c0f2df49 | 5614 | return ERR_CAST(cgrp); |
c15187a4 RG |
5615 | |
5616 | css = cgroup_get_e_css(cgrp, &memory_cgrp_subsys); | |
5617 | if (css) | |
5618 | memcg = container_of(css, struct mem_cgroup, css); | |
5619 | else | |
5620 | memcg = ERR_PTR(-ENOENT); | |
5621 | ||
5622 | cgroup_put(cgrp); | |
5623 | ||
5624 | return memcg; | |
5625 | } | |
5626 | #endif | |
5627 | ||
ef8f2327 | 5628 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5629 | { |
5630 | struct mem_cgroup_per_node *pn; | |
8c9bb398 WY |
5631 | |
5632 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
6d12e2d8 KH |
5633 | if (!pn) |
5634 | return 1; | |
1ecaab2b | 5635 | |
aab6103b RG |
5636 | pn->lruvec_stats = kzalloc_node(sizeof(struct lruvec_stats), |
5637 | GFP_KERNEL_ACCOUNT, node); | |
70a64b79 SB |
5638 | if (!pn->lruvec_stats) |
5639 | goto fail; | |
5640 | ||
7e1c0d6f SB |
5641 | pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu, |
5642 | GFP_KERNEL_ACCOUNT); | |
70a64b79 SB |
5643 | if (!pn->lruvec_stats_percpu) |
5644 | goto fail; | |
00f3ca2c | 5645 | |
ef8f2327 | 5646 | lruvec_init(&pn->lruvec); |
ef8f2327 MG |
5647 | pn->memcg = memcg; |
5648 | ||
54f72fe0 | 5649 | memcg->nodeinfo[node] = pn; |
6d12e2d8 | 5650 | return 0; |
70a64b79 SB |
5651 | fail: |
5652 | kfree(pn->lruvec_stats); | |
5653 | kfree(pn); | |
5654 | return 1; | |
6d12e2d8 KH |
5655 | } |
5656 | ||
ef8f2327 | 5657 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5658 | { |
00f3ca2c JW |
5659 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5660 | ||
4eaf431f MH |
5661 | if (!pn) |
5662 | return; | |
5663 | ||
7e1c0d6f | 5664 | free_percpu(pn->lruvec_stats_percpu); |
70a64b79 | 5665 | kfree(pn->lruvec_stats); |
00f3ca2c | 5666 | kfree(pn); |
1ecaab2b KH |
5667 | } |
5668 | ||
40e952f9 | 5669 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5670 | { |
c8b2a36f | 5671 | int node; |
59927fb9 | 5672 | |
91b71e78 | 5673 | obj_cgroup_put(memcg->orig_objcg); |
675d6c9b | 5674 | |
c8b2a36f | 5675 | for_each_node(node) |
ef8f2327 | 5676 | free_mem_cgroup_per_node_info(memcg, node); |
410f8e82 | 5677 | kfree(memcg->vmstats); |
871789d4 | 5678 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 5679 | kfree(memcg); |
59927fb9 | 5680 | } |
3afe36b1 | 5681 | |
40e952f9 TE |
5682 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5683 | { | |
ec1c86b2 | 5684 | lru_gen_exit_memcg(memcg); |
40e952f9 TE |
5685 | memcg_wb_domain_exit(memcg); |
5686 | __mem_cgroup_free(memcg); | |
5687 | } | |
5688 | ||
9cee7e8e | 5689 | static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent) |
8cdea7c0 | 5690 | { |
9cee7e8e | 5691 | struct memcg_vmstats_percpu *statc, *pstatc; |
d142e3e6 | 5692 | struct mem_cgroup *memcg; |
9cee7e8e | 5693 | int node, cpu; |
97b27821 | 5694 | int __maybe_unused i; |
11d67612 | 5695 | long error = -ENOMEM; |
8cdea7c0 | 5696 | |
06b2c3b0 | 5697 | memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL); |
c0ff4b85 | 5698 | if (!memcg) |
11d67612 | 5699 | return ERR_PTR(error); |
0b8f73e1 | 5700 | |
73f576c0 | 5701 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
be740503 | 5702 | 1, MEM_CGROUP_ID_MAX + 1, GFP_KERNEL); |
11d67612 YS |
5703 | if (memcg->id.id < 0) { |
5704 | error = memcg->id.id; | |
73f576c0 | 5705 | goto fail; |
11d67612 | 5706 | } |
73f576c0 | 5707 | |
aab6103b RG |
5708 | memcg->vmstats = kzalloc(sizeof(struct memcg_vmstats), |
5709 | GFP_KERNEL_ACCOUNT); | |
410f8e82 SB |
5710 | if (!memcg->vmstats) |
5711 | goto fail; | |
5712 | ||
3e38e0aa RG |
5713 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5714 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5715 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5716 | goto fail; |
78fb7466 | 5717 | |
9cee7e8e YA |
5718 | for_each_possible_cpu(cpu) { |
5719 | if (parent) | |
5720 | pstatc = per_cpu_ptr(parent->vmstats_percpu, cpu); | |
5721 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5722 | statc->parent = parent ? pstatc : NULL; | |
5723 | statc->vmstats = memcg->vmstats; | |
5724 | } | |
5725 | ||
3ed28fa1 | 5726 | for_each_node(node) |
ef8f2327 | 5727 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5728 | goto fail; |
f64c3f54 | 5729 | |
0b8f73e1 JW |
5730 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5731 | goto fail; | |
28dbc4b6 | 5732 | |
f7e1cb6e | 5733 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5734 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5735 | mutex_init(&memcg->thresholds_lock); |
5736 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5737 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5738 | INIT_LIST_HEAD(&memcg->event_list); |
5739 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5740 | memcg->socket_pressure = jiffies; |
84c07d11 | 5741 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5742 | memcg->kmemcg_id = -1; |
bf4f0599 | 5743 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5744 | #endif |
52ebea74 TH |
5745 | #ifdef CONFIG_CGROUP_WRITEBACK |
5746 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5747 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5748 | memcg->cgwb_frn[i].done = | |
5749 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5750 | #endif |
5751 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5752 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5753 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5754 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5755 | #endif |
ec1c86b2 | 5756 | lru_gen_init_memcg(memcg); |
0b8f73e1 JW |
5757 | return memcg; |
5758 | fail: | |
7e97de0b | 5759 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5760 | __mem_cgroup_free(memcg); |
11d67612 | 5761 | return ERR_PTR(error); |
d142e3e6 GC |
5762 | } |
5763 | ||
0b8f73e1 JW |
5764 | static struct cgroup_subsys_state * __ref |
5765 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5766 | { |
0b8f73e1 | 5767 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5768 | struct mem_cgroup *memcg, *old_memcg; |
d142e3e6 | 5769 | |
b87d8cef | 5770 | old_memcg = set_active_memcg(parent); |
9cee7e8e | 5771 | memcg = mem_cgroup_alloc(parent); |
b87d8cef | 5772 | set_active_memcg(old_memcg); |
11d67612 YS |
5773 | if (IS_ERR(memcg)) |
5774 | return ERR_CAST(memcg); | |
d142e3e6 | 5775 | |
d1663a90 | 5776 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
2178e20c | 5777 | WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX); |
f4840ccf JW |
5778 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
5779 | memcg->zswap_max = PAGE_COUNTER_MAX; | |
501a06fe NP |
5780 | WRITE_ONCE(memcg->zswap_writeback, |
5781 | !parent || READ_ONCE(parent->zswap_writeback)); | |
f4840ccf | 5782 | #endif |
4b82ab4f | 5783 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 | 5784 | if (parent) { |
82b3aa26 | 5785 | WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent)); |
17c56de6 | 5786 | WRITE_ONCE(memcg->oom_kill_disable, READ_ONCE(parent->oom_kill_disable)); |
bef8620c | 5787 | |
3e32cb2e | 5788 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5789 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5790 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5791 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5792 | } else { |
ff48c71c | 5793 | init_memcg_stats(); |
8278f1c7 | 5794 | init_memcg_events(); |
bef8620c RG |
5795 | page_counter_init(&memcg->memory, NULL); |
5796 | page_counter_init(&memcg->swap, NULL); | |
5797 | page_counter_init(&memcg->kmem, NULL); | |
5798 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5799 | |
0b8f73e1 JW |
5800 | root_mem_cgroup = memcg; |
5801 | return &memcg->css; | |
5802 | } | |
5803 | ||
f7e1cb6e | 5804 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5805 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5806 | |
b6c1a8af YS |
5807 | #if defined(CONFIG_MEMCG_KMEM) |
5808 | if (!cgroup_memory_nobpf) | |
5809 | static_branch_inc(&memcg_bpf_enabled_key); | |
5810 | #endif | |
5811 | ||
0b8f73e1 | 5812 | return &memcg->css; |
0b8f73e1 JW |
5813 | } |
5814 | ||
73f576c0 | 5815 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5816 | { |
58fa2a55 VD |
5817 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5818 | ||
da0efe30 MS |
5819 | if (memcg_online_kmem(memcg)) |
5820 | goto remove_id; | |
5821 | ||
0a4465d3 | 5822 | /* |
e4262c4f | 5823 | * A memcg must be visible for expand_shrinker_info() |
0a4465d3 KT |
5824 | * by the time the maps are allocated. So, we allocate maps |
5825 | * here, when for_each_mem_cgroup() can't skip it. | |
5826 | */ | |
da0efe30 MS |
5827 | if (alloc_shrinker_info(memcg)) |
5828 | goto offline_kmem; | |
0a4465d3 | 5829 | |
13ef7424 | 5830 | if (unlikely(mem_cgroup_is_root(memcg)) && !mem_cgroup_disabled()) |
aa48e47e | 5831 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, |
396faf88 | 5832 | FLUSH_TIME); |
e4dde56c | 5833 | lru_gen_online_memcg(memcg); |
6f0df8e1 JW |
5834 | |
5835 | /* Online state pins memcg ID, memcg ID pins CSS */ | |
5836 | refcount_set(&memcg->id.ref, 1); | |
5837 | css_get(css); | |
5838 | ||
5839 | /* | |
5840 | * Ensure mem_cgroup_from_id() works once we're fully online. | |
5841 | * | |
5842 | * We could do this earlier and require callers to filter with | |
5843 | * css_tryget_online(). But right now there are no users that | |
5844 | * need earlier access, and the workingset code relies on the | |
5845 | * cgroup tree linkage (mem_cgroup_get_nr_swap_pages()). So | |
5846 | * publish it here at the end of onlining. This matches the | |
5847 | * regular ID destruction during offlining. | |
5848 | */ | |
5849 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); | |
5850 | ||
2f7dd7a4 | 5851 | return 0; |
da0efe30 MS |
5852 | offline_kmem: |
5853 | memcg_offline_kmem(memcg); | |
5854 | remove_id: | |
5855 | mem_cgroup_id_remove(memcg); | |
5856 | return -ENOMEM; | |
8cdea7c0 BS |
5857 | } |
5858 | ||
eb95419b | 5859 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5860 | { |
eb95419b | 5861 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5862 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5863 | |
5864 | /* | |
5865 | * Unregister events and notify userspace. | |
5866 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5867 | * directory to avoid race between userspace and kernelspace. | |
5868 | */ | |
4ba9515d | 5869 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5870 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
79bd9814 TH |
5871 | list_del_init(&event->list); |
5872 | schedule_work(&event->remove); | |
5873 | } | |
4ba9515d | 5874 | spin_unlock_irq(&memcg->event_list_lock); |
ec64f515 | 5875 | |
bf8d5d52 | 5876 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5877 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5878 | |
a65b0e76 DC |
5879 | zswap_memcg_offline_cleanup(memcg); |
5880 | ||
567e9ab2 | 5881 | memcg_offline_kmem(memcg); |
a178015c | 5882 | reparent_shrinker_deferred(memcg); |
52ebea74 | 5883 | wb_memcg_offline(memcg); |
e4dde56c | 5884 | lru_gen_offline_memcg(memcg); |
73f576c0 | 5885 | |
591edfb1 RG |
5886 | drain_all_stock(memcg); |
5887 | ||
73f576c0 | 5888 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5889 | } |
5890 | ||
6df38689 VD |
5891 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5892 | { | |
5893 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5894 | ||
5895 | invalidate_reclaim_iterators(memcg); | |
e4dde56c | 5896 | lru_gen_release_memcg(memcg); |
6df38689 VD |
5897 | } |
5898 | ||
eb95419b | 5899 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5900 | { |
eb95419b | 5901 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5902 | int __maybe_unused i; |
c268e994 | 5903 | |
97b27821 TH |
5904 | #ifdef CONFIG_CGROUP_WRITEBACK |
5905 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5906 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5907 | #endif | |
f7e1cb6e | 5908 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5909 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5910 | |
0db15298 | 5911 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5912 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5913 | |
b6c1a8af YS |
5914 | #if defined(CONFIG_MEMCG_KMEM) |
5915 | if (!cgroup_memory_nobpf) | |
5916 | static_branch_dec(&memcg_bpf_enabled_key); | |
5917 | #endif | |
5918 | ||
0b8f73e1 JW |
5919 | vmpressure_cleanup(&memcg->vmpressure); |
5920 | cancel_work_sync(&memcg->high_work); | |
5921 | mem_cgroup_remove_from_trees(memcg); | |
e4262c4f | 5922 | free_shrinker_info(memcg); |
0b8f73e1 | 5923 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5924 | } |
5925 | ||
1ced953b TH |
5926 | /** |
5927 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5928 | * @css: the target css | |
5929 | * | |
5930 | * Reset the states of the mem_cgroup associated with @css. This is | |
5931 | * invoked when the userland requests disabling on the default hierarchy | |
5932 | * but the memcg is pinned through dependency. The memcg should stop | |
5933 | * applying policies and should revert to the vanilla state as it may be | |
5934 | * made visible again. | |
5935 | * | |
5936 | * The current implementation only resets the essential configurations. | |
5937 | * This needs to be expanded to cover all the visible parts. | |
5938 | */ | |
5939 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5940 | { | |
5941 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5942 | ||
bbec2e15 RG |
5943 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5944 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5945 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5946 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5947 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5948 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5949 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
2178e20c | 5950 | WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX); |
4b82ab4f | 5951 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5952 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5953 | } |
5954 | ||
2d146aa3 JW |
5955 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
5956 | { | |
5957 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5958 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5959 | struct memcg_vmstats_percpu *statc; | |
f82e6bf9 | 5960 | long delta, delta_cpu, v; |
7e1c0d6f | 5961 | int i, nid; |
2d146aa3 JW |
5962 | |
5963 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5964 | ||
ff48c71c | 5965 | for (i = 0; i < MEMCG_VMSTAT_SIZE; i++) { |
2d146aa3 JW |
5966 | /* |
5967 | * Collect the aggregated propagation counts of groups | |
5968 | * below us. We're in a per-cpu loop here and this is | |
5969 | * a global counter, so the first cycle will get them. | |
5970 | */ | |
410f8e82 | 5971 | delta = memcg->vmstats->state_pending[i]; |
2d146aa3 | 5972 | if (delta) |
410f8e82 | 5973 | memcg->vmstats->state_pending[i] = 0; |
2d146aa3 JW |
5974 | |
5975 | /* Add CPU changes on this level since the last flush */ | |
f82e6bf9 | 5976 | delta_cpu = 0; |
2d146aa3 JW |
5977 | v = READ_ONCE(statc->state[i]); |
5978 | if (v != statc->state_prev[i]) { | |
f82e6bf9 YA |
5979 | delta_cpu = v - statc->state_prev[i]; |
5980 | delta += delta_cpu; | |
2d146aa3 JW |
5981 | statc->state_prev[i] = v; |
5982 | } | |
5983 | ||
2d146aa3 | 5984 | /* Aggregate counts on this level and propagate upwards */ |
f82e6bf9 YA |
5985 | if (delta_cpu) |
5986 | memcg->vmstats->state_local[i] += delta_cpu; | |
5987 | ||
5988 | if (delta) { | |
5989 | memcg->vmstats->state[i] += delta; | |
5990 | if (parent) | |
5991 | parent->vmstats->state_pending[i] += delta; | |
5992 | } | |
2d146aa3 JW |
5993 | } |
5994 | ||
8278f1c7 | 5995 | for (i = 0; i < NR_MEMCG_EVENTS; i++) { |
410f8e82 | 5996 | delta = memcg->vmstats->events_pending[i]; |
2d146aa3 | 5997 | if (delta) |
410f8e82 | 5998 | memcg->vmstats->events_pending[i] = 0; |
2d146aa3 | 5999 | |
f82e6bf9 | 6000 | delta_cpu = 0; |
2d146aa3 JW |
6001 | v = READ_ONCE(statc->events[i]); |
6002 | if (v != statc->events_prev[i]) { | |
f82e6bf9 YA |
6003 | delta_cpu = v - statc->events_prev[i]; |
6004 | delta += delta_cpu; | |
2d146aa3 JW |
6005 | statc->events_prev[i] = v; |
6006 | } | |
6007 | ||
f82e6bf9 YA |
6008 | if (delta_cpu) |
6009 | memcg->vmstats->events_local[i] += delta_cpu; | |
2d146aa3 | 6010 | |
f82e6bf9 YA |
6011 | if (delta) { |
6012 | memcg->vmstats->events[i] += delta; | |
6013 | if (parent) | |
6014 | parent->vmstats->events_pending[i] += delta; | |
6015 | } | |
2d146aa3 | 6016 | } |
7e1c0d6f SB |
6017 | |
6018 | for_each_node_state(nid, N_MEMORY) { | |
6019 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; | |
70a64b79 SB |
6020 | struct lruvec_stats *lstats = pn->lruvec_stats; |
6021 | struct lruvec_stats *plstats = NULL; | |
7e1c0d6f SB |
6022 | struct lruvec_stats_percpu *lstatc; |
6023 | ||
6024 | if (parent) | |
70a64b79 | 6025 | plstats = parent->nodeinfo[nid]->lruvec_stats; |
7e1c0d6f SB |
6026 | |
6027 | lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu); | |
6028 | ||
ff48c71c | 6029 | for (i = 0; i < NR_MEMCG_NODE_STAT_ITEMS; i++) { |
70a64b79 | 6030 | delta = lstats->state_pending[i]; |
7e1c0d6f | 6031 | if (delta) |
70a64b79 | 6032 | lstats->state_pending[i] = 0; |
7e1c0d6f | 6033 | |
f82e6bf9 | 6034 | delta_cpu = 0; |
7e1c0d6f SB |
6035 | v = READ_ONCE(lstatc->state[i]); |
6036 | if (v != lstatc->state_prev[i]) { | |
f82e6bf9 YA |
6037 | delta_cpu = v - lstatc->state_prev[i]; |
6038 | delta += delta_cpu; | |
7e1c0d6f SB |
6039 | lstatc->state_prev[i] = v; |
6040 | } | |
6041 | ||
f82e6bf9 | 6042 | if (delta_cpu) |
70a64b79 | 6043 | lstats->state_local[i] += delta_cpu; |
7e1c0d6f | 6044 | |
f82e6bf9 | 6045 | if (delta) { |
70a64b79 SB |
6046 | lstats->state[i] += delta; |
6047 | if (plstats) | |
6048 | plstats->state_pending[i] += delta; | |
f82e6bf9 | 6049 | } |
7e1c0d6f SB |
6050 | } |
6051 | } | |
78ec6f9d | 6052 | WRITE_ONCE(statc->stats_updates, 0); |
8d59d221 YA |
6053 | /* We are in a per-cpu loop here, only do the atomic write once */ |
6054 | if (atomic64_read(&memcg->vmstats->stats_updates)) | |
6055 | atomic64_set(&memcg->vmstats->stats_updates, 0); | |
2d146aa3 JW |
6056 | } |
6057 | ||
02491447 | 6058 | #ifdef CONFIG_MMU |
7dc74be0 | 6059 | /* Handlers for move charge at task migration. */ |
854ffa8d | 6060 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 6061 | { |
05b84301 | 6062 | int ret; |
9476db97 | 6063 | |
d0164adc MG |
6064 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
6065 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 6066 | if (!ret) { |
854ffa8d | 6067 | mc.precharge += count; |
854ffa8d DN |
6068 | return ret; |
6069 | } | |
9476db97 | 6070 | |
3674534b | 6071 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 6072 | while (count--) { |
3674534b | 6073 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 6074 | if (ret) |
38c5d72f | 6075 | return ret; |
854ffa8d | 6076 | mc.precharge++; |
9476db97 | 6077 | cond_resched(); |
854ffa8d | 6078 | } |
9476db97 | 6079 | return 0; |
4ffef5fe DN |
6080 | } |
6081 | ||
4ffef5fe | 6082 | union mc_target { |
b46777da | 6083 | struct folio *folio; |
02491447 | 6084 | swp_entry_t ent; |
4ffef5fe DN |
6085 | }; |
6086 | ||
4ffef5fe | 6087 | enum mc_target_type { |
8d32ff84 | 6088 | MC_TARGET_NONE = 0, |
4ffef5fe | 6089 | MC_TARGET_PAGE, |
02491447 | 6090 | MC_TARGET_SWAP, |
c733a828 | 6091 | MC_TARGET_DEVICE, |
4ffef5fe DN |
6092 | }; |
6093 | ||
90254a65 DN |
6094 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6095 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6096 | { |
25b2995a | 6097 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6098 | |
58f341f7 | 6099 | if (!page) |
90254a65 DN |
6100 | return NULL; |
6101 | if (PageAnon(page)) { | |
1dfab5ab | 6102 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 6103 | return NULL; |
1dfab5ab JW |
6104 | } else { |
6105 | if (!(mc.flags & MOVE_FILE)) | |
6106 | return NULL; | |
6107 | } | |
58f341f7 | 6108 | get_page(page); |
90254a65 DN |
6109 | |
6110 | return page; | |
6111 | } | |
6112 | ||
c733a828 | 6113 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 6114 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 6115 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 6116 | { |
90254a65 DN |
6117 | struct page *page = NULL; |
6118 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6119 | ||
9a137153 | 6120 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 6121 | return NULL; |
c733a828 JG |
6122 | |
6123 | /* | |
27674ef6 CH |
6124 | * Handle device private pages that are not accessible by the CPU, but |
6125 | * stored as special swap entries in the page table. | |
c733a828 JG |
6126 | */ |
6127 | if (is_device_private_entry(ent)) { | |
af5cdaf8 | 6128 | page = pfn_swap_entry_to_page(ent); |
27674ef6 | 6129 | if (!get_page_unless_zero(page)) |
c733a828 JG |
6130 | return NULL; |
6131 | return page; | |
6132 | } | |
6133 | ||
9a137153 RC |
6134 | if (non_swap_entry(ent)) |
6135 | return NULL; | |
6136 | ||
4b91355e | 6137 | /* |
cb691e2f | 6138 | * Because swap_cache_get_folio() updates some statistics counter, |
4b91355e KH |
6139 | * we call find_get_page() with swapper_space directly. |
6140 | */ | |
f6ab1f7f | 6141 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 6142 | entry->val = ent.val; |
90254a65 DN |
6143 | |
6144 | return page; | |
6145 | } | |
4b91355e KH |
6146 | #else |
6147 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 6148 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
6149 | { |
6150 | return NULL; | |
6151 | } | |
6152 | #endif | |
90254a65 | 6153 | |
87946a72 | 6154 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
48384b0b | 6155 | unsigned long addr, pte_t ptent) |
87946a72 | 6156 | { |
524984ff MWO |
6157 | unsigned long index; |
6158 | struct folio *folio; | |
6159 | ||
87946a72 DN |
6160 | if (!vma->vm_file) /* anonymous vma */ |
6161 | return NULL; | |
1dfab5ab | 6162 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
6163 | return NULL; |
6164 | ||
524984ff | 6165 | /* folio is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 6166 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
524984ff MWO |
6167 | index = linear_page_index(vma, addr); |
6168 | folio = filemap_get_incore_folio(vma->vm_file->f_mapping, index); | |
66dabbb6 | 6169 | if (IS_ERR(folio)) |
524984ff MWO |
6170 | return NULL; |
6171 | return folio_file_page(folio, index); | |
87946a72 DN |
6172 | } |
6173 | ||
b1b0deab | 6174 | /** |
b267e1a3 MWO |
6175 | * mem_cgroup_move_account - move account of the folio |
6176 | * @folio: The folio. | |
25843c2b | 6177 | * @compound: charge the page as compound or small page |
b267e1a3 MWO |
6178 | * @from: mem_cgroup which the folio is moved from. |
6179 | * @to: mem_cgroup which the folio is moved to. @from != @to. | |
b1b0deab | 6180 | * |
b267e1a3 | 6181 | * The folio must be locked and not on the LRU. |
b1b0deab CG |
6182 | * |
6183 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
6184 | * from old cgroup. | |
6185 | */ | |
b267e1a3 | 6186 | static int mem_cgroup_move_account(struct folio *folio, |
f627c2f5 | 6187 | bool compound, |
b1b0deab CG |
6188 | struct mem_cgroup *from, |
6189 | struct mem_cgroup *to) | |
6190 | { | |
ae8af438 KK |
6191 | struct lruvec *from_vec, *to_vec; |
6192 | struct pglist_data *pgdat; | |
fcce4672 | 6193 | unsigned int nr_pages = compound ? folio_nr_pages(folio) : 1; |
8e88bd2d | 6194 | int nid, ret; |
b1b0deab CG |
6195 | |
6196 | VM_BUG_ON(from == to); | |
4e0cf05f | 6197 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
fcce4672 | 6198 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
9c325215 | 6199 | VM_BUG_ON(compound && !folio_test_large(folio)); |
b1b0deab | 6200 | |
b1b0deab | 6201 | ret = -EINVAL; |
fcce4672 | 6202 | if (folio_memcg(folio) != from) |
4e0cf05f | 6203 | goto out; |
b1b0deab | 6204 | |
fcce4672 | 6205 | pgdat = folio_pgdat(folio); |
867e5e1d JW |
6206 | from_vec = mem_cgroup_lruvec(from, pgdat); |
6207 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 6208 | |
fcce4672 | 6209 | folio_memcg_lock(folio); |
b1b0deab | 6210 | |
fcce4672 MWO |
6211 | if (folio_test_anon(folio)) { |
6212 | if (folio_mapped(folio)) { | |
be5d0a74 JW |
6213 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); |
6214 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
6199277b | 6215 | if (folio_test_pmd_mappable(folio)) { |
69473e5d MS |
6216 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
6217 | -nr_pages); | |
6218 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
6219 | nr_pages); | |
468c3982 | 6220 | } |
be5d0a74 JW |
6221 | } |
6222 | } else { | |
0d1c2072 JW |
6223 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
6224 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
6225 | ||
fcce4672 | 6226 | if (folio_test_swapbacked(folio)) { |
0d1c2072 JW |
6227 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); |
6228 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
6229 | } | |
6230 | ||
fcce4672 | 6231 | if (folio_mapped(folio)) { |
49e50d27 JW |
6232 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
6233 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
6234 | } | |
b1b0deab | 6235 | |
fcce4672 MWO |
6236 | if (folio_test_dirty(folio)) { |
6237 | struct address_space *mapping = folio_mapping(folio); | |
c4843a75 | 6238 | |
f56753ac | 6239 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
6240 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
6241 | -nr_pages); | |
6242 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
6243 | nr_pages); | |
6244 | } | |
c4843a75 GT |
6245 | } |
6246 | } | |
6247 | ||
c449deb2 HD |
6248 | #ifdef CONFIG_SWAP |
6249 | if (folio_test_swapcache(folio)) { | |
6250 | __mod_lruvec_state(from_vec, NR_SWAPCACHE, -nr_pages); | |
6251 | __mod_lruvec_state(to_vec, NR_SWAPCACHE, nr_pages); | |
6252 | } | |
6253 | #endif | |
fcce4672 | 6254 | if (folio_test_writeback(folio)) { |
ae8af438 KK |
6255 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
6256 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
6257 | } |
6258 | ||
6259 | /* | |
abb242f5 JW |
6260 | * All state has been migrated, let's switch to the new memcg. |
6261 | * | |
bcfe06bf | 6262 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
6263 | * is referenced, charged, isolated, and locked: we can't race |
6264 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 6265 | * that would rely on a stable page's memory cgroup. |
abb242f5 | 6266 | * |
6c77b607 | 6267 | * Note that folio_memcg_lock is a memcg lock, not a page lock, |
bcfe06bf | 6268 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
6269 | * new memcg that isn't locked, the above state can change |
6270 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 6271 | */ |
abb242f5 | 6272 | smp_mb(); |
b1b0deab | 6273 | |
1a3e1f40 JW |
6274 | css_get(&to->css); |
6275 | css_put(&from->css); | |
6276 | ||
fcce4672 | 6277 | folio->memcg_data = (unsigned long)to; |
87eaceb3 | 6278 | |
f70ad448 | 6279 | __folio_memcg_unlock(from); |
b1b0deab CG |
6280 | |
6281 | ret = 0; | |
fcce4672 | 6282 | nid = folio_nid(folio); |
b1b0deab CG |
6283 | |
6284 | local_irq_disable(); | |
6e0110c2 | 6285 | mem_cgroup_charge_statistics(to, nr_pages); |
8e88bd2d | 6286 | memcg_check_events(to, nid); |
6e0110c2 | 6287 | mem_cgroup_charge_statistics(from, -nr_pages); |
8e88bd2d | 6288 | memcg_check_events(from, nid); |
b1b0deab | 6289 | local_irq_enable(); |
b1b0deab CG |
6290 | out: |
6291 | return ret; | |
6292 | } | |
6293 | ||
7cf7806c LR |
6294 | /** |
6295 | * get_mctgt_type - get target type of moving charge | |
6296 | * @vma: the vma the pte to be checked belongs | |
6297 | * @addr: the address corresponding to the pte to be checked | |
6298 | * @ptent: the pte to be checked | |
6299 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
6300 | * | |
853f62a3 MWO |
6301 | * Context: Called with pte lock held. |
6302 | * Return: | |
6303 | * * MC_TARGET_NONE - If the pte is not a target for move charge. | |
6304 | * * MC_TARGET_PAGE - If the page corresponding to this pte is a target for | |
b46777da | 6305 | * move charge. If @target is not NULL, the folio is stored in target->folio |
853f62a3 MWO |
6306 | * with extra refcnt taken (Caller should release it). |
6307 | * * MC_TARGET_SWAP - If the swap entry corresponding to this pte is a | |
6308 | * target for charge migration. If @target is not NULL, the entry is | |
6309 | * stored in target->ent. | |
6310 | * * MC_TARGET_DEVICE - Like MC_TARGET_PAGE but page is device memory and | |
6311 | * thus not on the lru. For now such page is charged like a regular page | |
6312 | * would be as it is just special memory taking the place of a regular page. | |
6313 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c | 6314 | */ |
8d32ff84 | 6315 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6316 | unsigned long addr, pte_t ptent, union mc_target *target) |
6317 | { | |
6318 | struct page *page = NULL; | |
b67fa6e4 | 6319 | struct folio *folio; |
8d32ff84 | 6320 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6321 | swp_entry_t ent = { .val = 0 }; |
6322 | ||
6323 | if (pte_present(ptent)) | |
6324 | page = mc_handle_present_pte(vma, addr, ptent); | |
5c041f5d PX |
6325 | else if (pte_none_mostly(ptent)) |
6326 | /* | |
6327 | * PTE markers should be treated as a none pte here, separated | |
6328 | * from other swap handling below. | |
6329 | */ | |
6330 | page = mc_handle_file_pte(vma, addr, ptent); | |
90254a65 | 6331 | else if (is_swap_pte(ptent)) |
48406ef8 | 6332 | page = mc_handle_swap_pte(vma, ptent, &ent); |
90254a65 | 6333 | |
b67fa6e4 MWO |
6334 | if (page) |
6335 | folio = page_folio(page); | |
4e0cf05f | 6336 | if (target && page) { |
b67fa6e4 MWO |
6337 | if (!folio_trylock(folio)) { |
6338 | folio_put(folio); | |
4e0cf05f JW |
6339 | return ret; |
6340 | } | |
6341 | /* | |
6342 | * page_mapped() must be stable during the move. This | |
6343 | * pte is locked, so if it's present, the page cannot | |
6344 | * become unmapped. If it isn't, we have only partial | |
6345 | * control over the mapped state: the page lock will | |
6346 | * prevent new faults against pagecache and swapcache, | |
6347 | * so an unmapped page cannot become mapped. However, | |
6348 | * if the page is already mapped elsewhere, it can | |
6349 | * unmap, and there is nothing we can do about it. | |
6350 | * Alas, skip moving the page in this case. | |
6351 | */ | |
6352 | if (!pte_present(ptent) && page_mapped(page)) { | |
b67fa6e4 MWO |
6353 | folio_unlock(folio); |
6354 | folio_put(folio); | |
4e0cf05f JW |
6355 | return ret; |
6356 | } | |
6357 | } | |
6358 | ||
90254a65 | 6359 | if (!page && !ent.val) |
8d32ff84 | 6360 | return ret; |
02491447 | 6361 | if (page) { |
02491447 | 6362 | /* |
0a31bc97 | 6363 | * Do only loose check w/o serialization. |
1306a85a | 6364 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 6365 | * not under LRU exclusion. |
02491447 | 6366 | */ |
b67fa6e4 | 6367 | if (folio_memcg(folio) == mc.from) { |
02491447 | 6368 | ret = MC_TARGET_PAGE; |
b67fa6e4 MWO |
6369 | if (folio_is_device_private(folio) || |
6370 | folio_is_device_coherent(folio)) | |
c733a828 | 6371 | ret = MC_TARGET_DEVICE; |
02491447 | 6372 | if (target) |
b67fa6e4 | 6373 | target->folio = folio; |
02491447 | 6374 | } |
4e0cf05f JW |
6375 | if (!ret || !target) { |
6376 | if (target) | |
b67fa6e4 MWO |
6377 | folio_unlock(folio); |
6378 | folio_put(folio); | |
4e0cf05f | 6379 | } |
02491447 | 6380 | } |
3e14a57b HY |
6381 | /* |
6382 | * There is a swap entry and a page doesn't exist or isn't charged. | |
6383 | * But we cannot move a tail-page in a THP. | |
6384 | */ | |
6385 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 6386 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6387 | ret = MC_TARGET_SWAP; |
6388 | if (target) | |
6389 | target->ent = ent; | |
4ffef5fe | 6390 | } |
4ffef5fe DN |
6391 | return ret; |
6392 | } | |
6393 | ||
12724850 NH |
6394 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6395 | /* | |
d6810d73 HY |
6396 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
6397 | * not support them for now. | |
12724850 NH |
6398 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
6399 | */ | |
6400 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6401 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6402 | { | |
6403 | struct page *page = NULL; | |
f6c7590b | 6404 | struct folio *folio; |
12724850 NH |
6405 | enum mc_target_type ret = MC_TARGET_NONE; |
6406 | ||
84c3fc4e ZY |
6407 | if (unlikely(is_swap_pmd(pmd))) { |
6408 | VM_BUG_ON(thp_migration_supported() && | |
6409 | !is_pmd_migration_entry(pmd)); | |
6410 | return ret; | |
6411 | } | |
12724850 | 6412 | page = pmd_page(pmd); |
309381fe | 6413 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
f6c7590b | 6414 | folio = page_folio(page); |
1dfab5ab | 6415 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 6416 | return ret; |
f6c7590b | 6417 | if (folio_memcg(folio) == mc.from) { |
12724850 NH |
6418 | ret = MC_TARGET_PAGE; |
6419 | if (target) { | |
f6c7590b MWO |
6420 | folio_get(folio); |
6421 | if (!folio_trylock(folio)) { | |
6422 | folio_put(folio); | |
4e0cf05f JW |
6423 | return MC_TARGET_NONE; |
6424 | } | |
f6c7590b | 6425 | target->folio = folio; |
12724850 NH |
6426 | } |
6427 | } | |
6428 | return ret; | |
6429 | } | |
6430 | #else | |
6431 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6432 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6433 | { | |
6434 | return MC_TARGET_NONE; | |
6435 | } | |
6436 | #endif | |
6437 | ||
4ffef5fe DN |
6438 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6439 | unsigned long addr, unsigned long end, | |
6440 | struct mm_walk *walk) | |
6441 | { | |
26bcd64a | 6442 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6443 | pte_t *pte; |
6444 | spinlock_t *ptl; | |
6445 | ||
b6ec57f4 KS |
6446 | ptl = pmd_trans_huge_lock(pmd, vma); |
6447 | if (ptl) { | |
c733a828 JG |
6448 | /* |
6449 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
6450 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
6451 | * this might change. | |
c733a828 | 6452 | */ |
12724850 NH |
6453 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
6454 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 6455 | spin_unlock(ptl); |
1a5a9906 | 6456 | return 0; |
12724850 | 6457 | } |
03319327 | 6458 | |
4ffef5fe | 6459 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
04dee9e8 HD |
6460 | if (!pte) |
6461 | return 0; | |
4ffef5fe | 6462 | for (; addr != end; pte++, addr += PAGE_SIZE) |
c33c7948 | 6463 | if (get_mctgt_type(vma, addr, ptep_get(pte), NULL)) |
4ffef5fe DN |
6464 | mc.precharge++; /* increment precharge temporarily */ |
6465 | pte_unmap_unlock(pte - 1, ptl); | |
6466 | cond_resched(); | |
6467 | ||
7dc74be0 DN |
6468 | return 0; |
6469 | } | |
6470 | ||
7b86ac33 CH |
6471 | static const struct mm_walk_ops precharge_walk_ops = { |
6472 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
49b06385 | 6473 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 CH |
6474 | }; |
6475 | ||
4ffef5fe DN |
6476 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6477 | { | |
6478 | unsigned long precharge; | |
4ffef5fe | 6479 | |
d8ed45c5 | 6480 | mmap_read_lock(mm); |
ba0aff8e | 6481 | walk_page_range(mm, 0, ULONG_MAX, &precharge_walk_ops, NULL); |
d8ed45c5 | 6482 | mmap_read_unlock(mm); |
4ffef5fe DN |
6483 | |
6484 | precharge = mc.precharge; | |
6485 | mc.precharge = 0; | |
6486 | ||
6487 | return precharge; | |
6488 | } | |
6489 | ||
4ffef5fe DN |
6490 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6491 | { | |
dfe076b0 DN |
6492 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6493 | ||
6494 | VM_BUG_ON(mc.moving_task); | |
6495 | mc.moving_task = current; | |
6496 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6497 | } |
6498 | ||
dfe076b0 DN |
6499 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6500 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6501 | { |
2bd9bb20 KH |
6502 | struct mem_cgroup *from = mc.from; |
6503 | struct mem_cgroup *to = mc.to; | |
6504 | ||
4ffef5fe | 6505 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 6506 | if (mc.precharge) { |
4b569387 | 6507 | mem_cgroup_cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
6508 | mc.precharge = 0; |
6509 | } | |
6510 | /* | |
6511 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6512 | * we must uncharge here. | |
6513 | */ | |
6514 | if (mc.moved_charge) { | |
4b569387 | 6515 | mem_cgroup_cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 6516 | mc.moved_charge = 0; |
4ffef5fe | 6517 | } |
483c30b5 DN |
6518 | /* we must fixup refcnts and charges */ |
6519 | if (mc.moved_swap) { | |
483c30b5 | 6520 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 6521 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 6522 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 6523 | |
615d66c3 VD |
6524 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
6525 | ||
05b84301 | 6526 | /* |
3e32cb2e JW |
6527 | * we charged both to->memory and to->memsw, so we |
6528 | * should uncharge to->memory. | |
05b84301 | 6529 | */ |
ce00a967 | 6530 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
6531 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
6532 | ||
483c30b5 DN |
6533 | mc.moved_swap = 0; |
6534 | } | |
dfe076b0 DN |
6535 | memcg_oom_recover(from); |
6536 | memcg_oom_recover(to); | |
6537 | wake_up_all(&mc.waitq); | |
6538 | } | |
6539 | ||
6540 | static void mem_cgroup_clear_mc(void) | |
6541 | { | |
264a0ae1 TH |
6542 | struct mm_struct *mm = mc.mm; |
6543 | ||
dfe076b0 DN |
6544 | /* |
6545 | * we must clear moving_task before waking up waiters at the end of | |
6546 | * task migration. | |
6547 | */ | |
6548 | mc.moving_task = NULL; | |
6549 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6550 | spin_lock(&mc.lock); |
4ffef5fe DN |
6551 | mc.from = NULL; |
6552 | mc.to = NULL; | |
264a0ae1 | 6553 | mc.mm = NULL; |
2bd9bb20 | 6554 | spin_unlock(&mc.lock); |
264a0ae1 TH |
6555 | |
6556 | mmput(mm); | |
4ffef5fe DN |
6557 | } |
6558 | ||
1f7dd3e5 | 6559 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6560 | { |
1f7dd3e5 | 6561 | struct cgroup_subsys_state *css; |
eed67d75 | 6562 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 6563 | struct mem_cgroup *from; |
4530eddb | 6564 | struct task_struct *leader, *p; |
9f2115f9 | 6565 | struct mm_struct *mm; |
1dfab5ab | 6566 | unsigned long move_flags; |
9f2115f9 | 6567 | int ret = 0; |
7dc74be0 | 6568 | |
1f7dd3e5 TH |
6569 | /* charge immigration isn't supported on the default hierarchy */ |
6570 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
6571 | return 0; |
6572 | ||
4530eddb TH |
6573 | /* |
6574 | * Multi-process migrations only happen on the default hierarchy | |
6575 | * where charge immigration is not used. Perform charge | |
6576 | * immigration if @tset contains a leader and whine if there are | |
6577 | * multiple. | |
6578 | */ | |
6579 | p = NULL; | |
1f7dd3e5 | 6580 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
6581 | WARN_ON_ONCE(p); |
6582 | p = leader; | |
1f7dd3e5 | 6583 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
6584 | } |
6585 | if (!p) | |
6586 | return 0; | |
6587 | ||
1f7dd3e5 | 6588 | /* |
f0953a1b | 6589 | * We are now committed to this value whatever it is. Changes in this |
1f7dd3e5 TH |
6590 | * tunable will only affect upcoming migrations, not the current one. |
6591 | * So we need to save it, and keep it going. | |
6592 | */ | |
6593 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
6594 | if (!move_flags) | |
6595 | return 0; | |
6596 | ||
9f2115f9 TH |
6597 | from = mem_cgroup_from_task(p); |
6598 | ||
6599 | VM_BUG_ON(from == memcg); | |
6600 | ||
6601 | mm = get_task_mm(p); | |
6602 | if (!mm) | |
6603 | return 0; | |
6604 | /* We move charges only when we move a owner of the mm */ | |
6605 | if (mm->owner == p) { | |
6606 | VM_BUG_ON(mc.from); | |
6607 | VM_BUG_ON(mc.to); | |
6608 | VM_BUG_ON(mc.precharge); | |
6609 | VM_BUG_ON(mc.moved_charge); | |
6610 | VM_BUG_ON(mc.moved_swap); | |
6611 | ||
6612 | spin_lock(&mc.lock); | |
264a0ae1 | 6613 | mc.mm = mm; |
9f2115f9 TH |
6614 | mc.from = from; |
6615 | mc.to = memcg; | |
6616 | mc.flags = move_flags; | |
6617 | spin_unlock(&mc.lock); | |
6618 | /* We set mc.moving_task later */ | |
6619 | ||
6620 | ret = mem_cgroup_precharge_mc(mm); | |
6621 | if (ret) | |
6622 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6623 | } else { |
6624 | mmput(mm); | |
7dc74be0 DN |
6625 | } |
6626 | return ret; | |
6627 | } | |
6628 | ||
1f7dd3e5 | 6629 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6630 | { |
4e2f245d JW |
6631 | if (mc.to) |
6632 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6633 | } |
6634 | ||
4ffef5fe DN |
6635 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6636 | unsigned long addr, unsigned long end, | |
6637 | struct mm_walk *walk) | |
7dc74be0 | 6638 | { |
4ffef5fe | 6639 | int ret = 0; |
26bcd64a | 6640 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6641 | pte_t *pte; |
6642 | spinlock_t *ptl; | |
12724850 NH |
6643 | enum mc_target_type target_type; |
6644 | union mc_target target; | |
b267e1a3 | 6645 | struct folio *folio; |
4ffef5fe | 6646 | |
b6ec57f4 KS |
6647 | ptl = pmd_trans_huge_lock(pmd, vma); |
6648 | if (ptl) { | |
62ade86a | 6649 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6650 | spin_unlock(ptl); |
12724850 NH |
6651 | return 0; |
6652 | } | |
6653 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6654 | if (target_type == MC_TARGET_PAGE) { | |
b46777da | 6655 | folio = target.folio; |
b267e1a3 MWO |
6656 | if (folio_isolate_lru(folio)) { |
6657 | if (!mem_cgroup_move_account(folio, true, | |
1306a85a | 6658 | mc.from, mc.to)) { |
12724850 NH |
6659 | mc.precharge -= HPAGE_PMD_NR; |
6660 | mc.moved_charge += HPAGE_PMD_NR; | |
6661 | } | |
b267e1a3 | 6662 | folio_putback_lru(folio); |
12724850 | 6663 | } |
b267e1a3 MWO |
6664 | folio_unlock(folio); |
6665 | folio_put(folio); | |
c733a828 | 6666 | } else if (target_type == MC_TARGET_DEVICE) { |
b46777da | 6667 | folio = target.folio; |
b267e1a3 | 6668 | if (!mem_cgroup_move_account(folio, true, |
c733a828 JG |
6669 | mc.from, mc.to)) { |
6670 | mc.precharge -= HPAGE_PMD_NR; | |
6671 | mc.moved_charge += HPAGE_PMD_NR; | |
6672 | } | |
b267e1a3 MWO |
6673 | folio_unlock(folio); |
6674 | folio_put(folio); | |
12724850 | 6675 | } |
bf929152 | 6676 | spin_unlock(ptl); |
1a5a9906 | 6677 | return 0; |
12724850 NH |
6678 | } |
6679 | ||
4ffef5fe DN |
6680 | retry: |
6681 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
04dee9e8 HD |
6682 | if (!pte) |
6683 | return 0; | |
4ffef5fe | 6684 | for (; addr != end; addr += PAGE_SIZE) { |
c33c7948 | 6685 | pte_t ptent = ptep_get(pte++); |
c733a828 | 6686 | bool device = false; |
02491447 | 6687 | swp_entry_t ent; |
4ffef5fe DN |
6688 | |
6689 | if (!mc.precharge) | |
6690 | break; | |
6691 | ||
8d32ff84 | 6692 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6693 | case MC_TARGET_DEVICE: |
6694 | device = true; | |
e4a9bc58 | 6695 | fallthrough; |
4ffef5fe | 6696 | case MC_TARGET_PAGE: |
b46777da | 6697 | folio = target.folio; |
53f9263b KS |
6698 | /* |
6699 | * We can have a part of the split pmd here. Moving it | |
6700 | * can be done but it would be too convoluted so simply | |
6701 | * ignore such a partial THP and keep it in original | |
6702 | * memcg. There should be somebody mapping the head. | |
6703 | */ | |
b267e1a3 | 6704 | if (folio_test_large(folio)) |
53f9263b | 6705 | goto put; |
b267e1a3 | 6706 | if (!device && !folio_isolate_lru(folio)) |
4ffef5fe | 6707 | goto put; |
b267e1a3 | 6708 | if (!mem_cgroup_move_account(folio, false, |
f627c2f5 | 6709 | mc.from, mc.to)) { |
4ffef5fe | 6710 | mc.precharge--; |
854ffa8d DN |
6711 | /* we uncharge from mc.from later. */ |
6712 | mc.moved_charge++; | |
4ffef5fe | 6713 | } |
c733a828 | 6714 | if (!device) |
b267e1a3 | 6715 | folio_putback_lru(folio); |
4e0cf05f | 6716 | put: /* get_mctgt_type() gets & locks the page */ |
b267e1a3 MWO |
6717 | folio_unlock(folio); |
6718 | folio_put(folio); | |
4ffef5fe | 6719 | break; |
02491447 DN |
6720 | case MC_TARGET_SWAP: |
6721 | ent = target.ent; | |
e91cbb42 | 6722 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6723 | mc.precharge--; |
8d22a935 HD |
6724 | mem_cgroup_id_get_many(mc.to, 1); |
6725 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6726 | mc.moved_swap++; |
6727 | } | |
02491447 | 6728 | break; |
4ffef5fe DN |
6729 | default: |
6730 | break; | |
6731 | } | |
6732 | } | |
6733 | pte_unmap_unlock(pte - 1, ptl); | |
6734 | cond_resched(); | |
6735 | ||
6736 | if (addr != end) { | |
6737 | /* | |
6738 | * We have consumed all precharges we got in can_attach(). | |
6739 | * We try charge one by one, but don't do any additional | |
6740 | * charges to mc.to if we have failed in charge once in attach() | |
6741 | * phase. | |
6742 | */ | |
854ffa8d | 6743 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6744 | if (!ret) |
6745 | goto retry; | |
6746 | } | |
6747 | ||
6748 | return ret; | |
6749 | } | |
6750 | ||
7b86ac33 CH |
6751 | static const struct mm_walk_ops charge_walk_ops = { |
6752 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
49b06385 | 6753 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 CH |
6754 | }; |
6755 | ||
264a0ae1 | 6756 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6757 | { |
4ffef5fe | 6758 | lru_add_drain_all(); |
312722cb | 6759 | /* |
6c77b607 | 6760 | * Signal folio_memcg_lock() to take the memcg's move_lock |
81f8c3a4 JW |
6761 | * while we're moving its pages to another memcg. Then wait |
6762 | * for already started RCU-only updates to finish. | |
312722cb JW |
6763 | */ |
6764 | atomic_inc(&mc.from->moving_account); | |
6765 | synchronize_rcu(); | |
dfe076b0 | 6766 | retry: |
d8ed45c5 | 6767 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6768 | /* |
c1e8d7c6 | 6769 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6770 | * waitq. So we cancel all extra charges, wake up all waiters, |
6771 | * and retry. Because we cancel precharges, we might not be able | |
6772 | * to move enough charges, but moving charge is a best-effort | |
6773 | * feature anyway, so it wouldn't be a big problem. | |
6774 | */ | |
6775 | __mem_cgroup_clear_mc(); | |
6776 | cond_resched(); | |
6777 | goto retry; | |
6778 | } | |
26bcd64a NH |
6779 | /* |
6780 | * When we have consumed all precharges and failed in doing | |
6781 | * additional charge, the page walk just aborts. | |
6782 | */ | |
ba0aff8e | 6783 | walk_page_range(mc.mm, 0, ULONG_MAX, &charge_walk_ops, NULL); |
d8ed45c5 | 6784 | mmap_read_unlock(mc.mm); |
312722cb | 6785 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6786 | } |
6787 | ||
264a0ae1 | 6788 | static void mem_cgroup_move_task(void) |
67e465a7 | 6789 | { |
264a0ae1 TH |
6790 | if (mc.to) { |
6791 | mem_cgroup_move_charge(); | |
a433658c | 6792 | mem_cgroup_clear_mc(); |
264a0ae1 | 6793 | } |
67e465a7 | 6794 | } |
1aacbd35 | 6795 | |
5cfb80a7 | 6796 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6797 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6798 | { |
6799 | return 0; | |
6800 | } | |
1f7dd3e5 | 6801 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6802 | { |
6803 | } | |
264a0ae1 | 6804 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6805 | { |
6806 | } | |
6807 | #endif | |
67e465a7 | 6808 | |
1aacbd35 RG |
6809 | #ifdef CONFIG_MEMCG_KMEM |
6810 | static void mem_cgroup_fork(struct task_struct *task) | |
6811 | { | |
6812 | /* | |
6813 | * Set the update flag to cause task->objcg to be initialized lazily | |
6814 | * on the first allocation. It can be done without any synchronization | |
6815 | * because it's always performed on the current task, so does | |
6816 | * current_objcg_update(). | |
6817 | */ | |
6818 | task->objcg = (struct obj_cgroup *)CURRENT_OBJCG_UPDATE_FLAG; | |
6819 | } | |
6820 | ||
6821 | static void mem_cgroup_exit(struct task_struct *task) | |
6822 | { | |
6823 | struct obj_cgroup *objcg = task->objcg; | |
6824 | ||
6825 | objcg = (struct obj_cgroup *) | |
6826 | ((unsigned long)objcg & ~CURRENT_OBJCG_UPDATE_FLAG); | |
91b71e78 | 6827 | obj_cgroup_put(objcg); |
1aacbd35 RG |
6828 | |
6829 | /* | |
6830 | * Some kernel allocations can happen after this point, | |
6831 | * but let's ignore them. It can be done without any synchronization | |
6832 | * because it's always performed on the current task, so does | |
6833 | * current_objcg_update(). | |
6834 | */ | |
6835 | task->objcg = NULL; | |
6836 | } | |
6837 | #endif | |
6838 | ||
bd74fdae | 6839 | #ifdef CONFIG_LRU_GEN |
1aacbd35 | 6840 | static void mem_cgroup_lru_gen_attach(struct cgroup_taskset *tset) |
bd74fdae YZ |
6841 | { |
6842 | struct task_struct *task; | |
6843 | struct cgroup_subsys_state *css; | |
6844 | ||
6845 | /* find the first leader if there is any */ | |
6846 | cgroup_taskset_for_each_leader(task, css, tset) | |
6847 | break; | |
6848 | ||
6849 | if (!task) | |
6850 | return; | |
6851 | ||
6852 | task_lock(task); | |
6853 | if (task->mm && READ_ONCE(task->mm->owner) == task) | |
6854 | lru_gen_migrate_mm(task->mm); | |
6855 | task_unlock(task); | |
6856 | } | |
6857 | #else | |
1aacbd35 RG |
6858 | static void mem_cgroup_lru_gen_attach(struct cgroup_taskset *tset) {} |
6859 | #endif /* CONFIG_LRU_GEN */ | |
6860 | ||
6861 | #ifdef CONFIG_MEMCG_KMEM | |
6862 | static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset) | |
6863 | { | |
6864 | struct task_struct *task; | |
6865 | struct cgroup_subsys_state *css; | |
6866 | ||
6867 | cgroup_taskset_for_each(task, css, tset) { | |
6868 | /* atomically set the update bit */ | |
6869 | set_bit(CURRENT_OBJCG_UPDATE_BIT, (unsigned long *)&task->objcg); | |
6870 | } | |
6871 | } | |
6872 | #else | |
6873 | static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset) {} | |
6874 | #endif /* CONFIG_MEMCG_KMEM */ | |
6875 | ||
6876 | #if defined(CONFIG_LRU_GEN) || defined(CONFIG_MEMCG_KMEM) | |
bd74fdae YZ |
6877 | static void mem_cgroup_attach(struct cgroup_taskset *tset) |
6878 | { | |
1aacbd35 RG |
6879 | mem_cgroup_lru_gen_attach(tset); |
6880 | mem_cgroup_kmem_attach(tset); | |
bd74fdae | 6881 | } |
1aacbd35 | 6882 | #endif |
bd74fdae | 6883 | |
677dc973 CD |
6884 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6885 | { | |
6886 | if (value == PAGE_COUNTER_MAX) | |
6887 | seq_puts(m, "max\n"); | |
6888 | else | |
6889 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6890 | ||
6891 | return 0; | |
6892 | } | |
6893 | ||
241994ed JW |
6894 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6895 | struct cftype *cft) | |
6896 | { | |
f5fc3c5d JW |
6897 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6898 | ||
6899 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6900 | } |
6901 | ||
8e20d4b3 GR |
6902 | static u64 memory_peak_read(struct cgroup_subsys_state *css, |
6903 | struct cftype *cft) | |
6904 | { | |
6905 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6906 | ||
6907 | return (u64)memcg->memory.watermark * PAGE_SIZE; | |
6908 | } | |
6909 | ||
bf8d5d52 RG |
6910 | static int memory_min_show(struct seq_file *m, void *v) |
6911 | { | |
677dc973 CD |
6912 | return seq_puts_memcg_tunable(m, |
6913 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6914 | } |
6915 | ||
6916 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6917 | char *buf, size_t nbytes, loff_t off) | |
6918 | { | |
6919 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6920 | unsigned long min; | |
6921 | int err; | |
6922 | ||
6923 | buf = strstrip(buf); | |
6924 | err = page_counter_memparse(buf, "max", &min); | |
6925 | if (err) | |
6926 | return err; | |
6927 | ||
6928 | page_counter_set_min(&memcg->memory, min); | |
6929 | ||
6930 | return nbytes; | |
6931 | } | |
6932 | ||
241994ed JW |
6933 | static int memory_low_show(struct seq_file *m, void *v) |
6934 | { | |
677dc973 CD |
6935 | return seq_puts_memcg_tunable(m, |
6936 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6937 | } |
6938 | ||
6939 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6940 | char *buf, size_t nbytes, loff_t off) | |
6941 | { | |
6942 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6943 | unsigned long low; | |
6944 | int err; | |
6945 | ||
6946 | buf = strstrip(buf); | |
d2973697 | 6947 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6948 | if (err) |
6949 | return err; | |
6950 | ||
23067153 | 6951 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6952 | |
6953 | return nbytes; | |
6954 | } | |
6955 | ||
6956 | static int memory_high_show(struct seq_file *m, void *v) | |
6957 | { | |
d1663a90 JK |
6958 | return seq_puts_memcg_tunable(m, |
6959 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6960 | } |
6961 | ||
6962 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6963 | char *buf, size_t nbytes, loff_t off) | |
6964 | { | |
6965 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6966 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6967 | bool drained = false; |
241994ed JW |
6968 | unsigned long high; |
6969 | int err; | |
6970 | ||
6971 | buf = strstrip(buf); | |
d2973697 | 6972 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6973 | if (err) |
6974 | return err; | |
6975 | ||
e82553c1 JW |
6976 | page_counter_set_high(&memcg->memory, high); |
6977 | ||
8c8c383c JW |
6978 | for (;;) { |
6979 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6980 | unsigned long reclaimed; | |
6981 | ||
6982 | if (nr_pages <= high) | |
6983 | break; | |
6984 | ||
6985 | if (signal_pending(current)) | |
6986 | break; | |
6987 | ||
6988 | if (!drained) { | |
6989 | drain_all_stock(memcg); | |
6990 | drained = true; | |
6991 | continue; | |
6992 | } | |
6993 | ||
6994 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
55ab834a | 6995 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP); |
8c8c383c JW |
6996 | |
6997 | if (!reclaimed && !nr_retries--) | |
6998 | break; | |
6999 | } | |
588083bb | 7000 | |
19ce33ac | 7001 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
7002 | return nbytes; |
7003 | } | |
7004 | ||
7005 | static int memory_max_show(struct seq_file *m, void *v) | |
7006 | { | |
677dc973 CD |
7007 | return seq_puts_memcg_tunable(m, |
7008 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
7009 | } |
7010 | ||
7011 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
7012 | char *buf, size_t nbytes, loff_t off) | |
7013 | { | |
7014 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 7015 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 7016 | bool drained = false; |
241994ed JW |
7017 | unsigned long max; |
7018 | int err; | |
7019 | ||
7020 | buf = strstrip(buf); | |
d2973697 | 7021 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
7022 | if (err) |
7023 | return err; | |
7024 | ||
bbec2e15 | 7025 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
7026 | |
7027 | for (;;) { | |
7028 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
7029 | ||
7030 | if (nr_pages <= max) | |
7031 | break; | |
7032 | ||
7249c9f0 | 7033 | if (signal_pending(current)) |
b6e6edcf | 7034 | break; |
b6e6edcf JW |
7035 | |
7036 | if (!drained) { | |
7037 | drain_all_stock(memcg); | |
7038 | drained = true; | |
7039 | continue; | |
7040 | } | |
7041 | ||
7042 | if (nr_reclaims) { | |
7043 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
55ab834a | 7044 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP)) |
b6e6edcf JW |
7045 | nr_reclaims--; |
7046 | continue; | |
7047 | } | |
7048 | ||
e27be240 | 7049 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
7050 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
7051 | break; | |
7052 | } | |
241994ed | 7053 | |
2529bb3a | 7054 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
7055 | return nbytes; |
7056 | } | |
7057 | ||
664dc218 DR |
7058 | /* |
7059 | * Note: don't forget to update the 'samples/cgroup/memcg_event_listener' | |
7060 | * if any new events become available. | |
7061 | */ | |
1e577f97 SB |
7062 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
7063 | { | |
7064 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
7065 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
7066 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
7067 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
7068 | seq_printf(m, "oom_kill %lu\n", | |
7069 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
b6bf9abb DS |
7070 | seq_printf(m, "oom_group_kill %lu\n", |
7071 | atomic_long_read(&events[MEMCG_OOM_GROUP_KILL])); | |
1e577f97 SB |
7072 | } |
7073 | ||
241994ed JW |
7074 | static int memory_events_show(struct seq_file *m, void *v) |
7075 | { | |
aa9694bb | 7076 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 7077 | |
1e577f97 SB |
7078 | __memory_events_show(m, memcg->memory_events); |
7079 | return 0; | |
7080 | } | |
7081 | ||
7082 | static int memory_events_local_show(struct seq_file *m, void *v) | |
7083 | { | |
7084 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 7085 | |
1e577f97 | 7086 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
7087 | return 0; |
7088 | } | |
7089 | ||
587d9f72 JW |
7090 | static int memory_stat_show(struct seq_file *m, void *v) |
7091 | { | |
aa9694bb | 7092 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
68aaee14 | 7093 | char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
5b42360c | 7094 | struct seq_buf s; |
1ff9e6e1 | 7095 | |
c8713d0b JW |
7096 | if (!buf) |
7097 | return -ENOMEM; | |
5b42360c YA |
7098 | seq_buf_init(&s, buf, PAGE_SIZE); |
7099 | memory_stat_format(memcg, &s); | |
c8713d0b JW |
7100 | seq_puts(m, buf); |
7101 | kfree(buf); | |
587d9f72 JW |
7102 | return 0; |
7103 | } | |
7104 | ||
5f9a4f4a | 7105 | #ifdef CONFIG_NUMA |
fff66b79 MS |
7106 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
7107 | int item) | |
7108 | { | |
ff841a06 YA |
7109 | return lruvec_page_state(lruvec, item) * |
7110 | memcg_page_state_output_unit(item); | |
fff66b79 MS |
7111 | } |
7112 | ||
5f9a4f4a MS |
7113 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
7114 | { | |
7115 | int i; | |
7116 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
7117 | ||
7d7ef0a4 | 7118 | mem_cgroup_flush_stats(memcg); |
7e1c0d6f | 7119 | |
5f9a4f4a MS |
7120 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
7121 | int nid; | |
7122 | ||
7123 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
7124 | continue; | |
7125 | ||
7126 | seq_printf(m, "%s", memory_stats[i].name); | |
7127 | for_each_node_state(nid, N_MEMORY) { | |
7128 | u64 size; | |
7129 | struct lruvec *lruvec; | |
7130 | ||
7131 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
7132 | size = lruvec_page_state_output(lruvec, |
7133 | memory_stats[i].idx); | |
5f9a4f4a MS |
7134 | seq_printf(m, " N%d=%llu", nid, size); |
7135 | } | |
7136 | seq_putc(m, '\n'); | |
7137 | } | |
7138 | ||
7139 | return 0; | |
7140 | } | |
7141 | #endif | |
7142 | ||
3d8b38eb RG |
7143 | static int memory_oom_group_show(struct seq_file *m, void *v) |
7144 | { | |
aa9694bb | 7145 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb | 7146 | |
eaf7b66b | 7147 | seq_printf(m, "%d\n", READ_ONCE(memcg->oom_group)); |
3d8b38eb RG |
7148 | |
7149 | return 0; | |
7150 | } | |
7151 | ||
7152 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
7153 | char *buf, size_t nbytes, loff_t off) | |
7154 | { | |
7155 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7156 | int ret, oom_group; | |
7157 | ||
7158 | buf = strstrip(buf); | |
7159 | if (!buf) | |
7160 | return -EINVAL; | |
7161 | ||
7162 | ret = kstrtoint(buf, 0, &oom_group); | |
7163 | if (ret) | |
7164 | return ret; | |
7165 | ||
7166 | if (oom_group != 0 && oom_group != 1) | |
7167 | return -EINVAL; | |
7168 | ||
eaf7b66b | 7169 | WRITE_ONCE(memcg->oom_group, oom_group); |
3d8b38eb RG |
7170 | |
7171 | return nbytes; | |
7172 | } | |
7173 | ||
94968384 SB |
7174 | static ssize_t memory_reclaim(struct kernfs_open_file *of, char *buf, |
7175 | size_t nbytes, loff_t off) | |
7176 | { | |
7177 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7178 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; | |
7179 | unsigned long nr_to_reclaim, nr_reclaimed = 0; | |
55ab834a MH |
7180 | unsigned int reclaim_options; |
7181 | int err; | |
12a5d395 MA |
7182 | |
7183 | buf = strstrip(buf); | |
55ab834a MH |
7184 | err = page_counter_memparse(buf, "", &nr_to_reclaim); |
7185 | if (err) | |
7186 | return err; | |
12a5d395 | 7187 | |
55ab834a | 7188 | reclaim_options = MEMCG_RECLAIM_MAY_SWAP | MEMCG_RECLAIM_PROACTIVE; |
94968384 | 7189 | while (nr_reclaimed < nr_to_reclaim) { |
287d5fed M |
7190 | /* Will converge on zero, but reclaim enforces a minimum */ |
7191 | unsigned long batch_size = (nr_to_reclaim - nr_reclaimed) / 4; | |
94968384 SB |
7192 | unsigned long reclaimed; |
7193 | ||
7194 | if (signal_pending(current)) | |
7195 | return -EINTR; | |
7196 | ||
7197 | /* | |
7198 | * This is the final attempt, drain percpu lru caches in the | |
7199 | * hope of introducing more evictable pages for | |
7200 | * try_to_free_mem_cgroup_pages(). | |
7201 | */ | |
7202 | if (!nr_retries) | |
7203 | lru_add_drain_all(); | |
7204 | ||
7205 | reclaimed = try_to_free_mem_cgroup_pages(memcg, | |
287d5fed | 7206 | batch_size, GFP_KERNEL, reclaim_options); |
94968384 SB |
7207 | |
7208 | if (!reclaimed && !nr_retries--) | |
7209 | return -EAGAIN; | |
7210 | ||
7211 | nr_reclaimed += reclaimed; | |
7212 | } | |
7213 | ||
7214 | return nbytes; | |
7215 | } | |
7216 | ||
241994ed JW |
7217 | static struct cftype memory_files[] = { |
7218 | { | |
7219 | .name = "current", | |
f5fc3c5d | 7220 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
7221 | .read_u64 = memory_current_read, |
7222 | }, | |
8e20d4b3 GR |
7223 | { |
7224 | .name = "peak", | |
7225 | .flags = CFTYPE_NOT_ON_ROOT, | |
7226 | .read_u64 = memory_peak_read, | |
7227 | }, | |
bf8d5d52 RG |
7228 | { |
7229 | .name = "min", | |
7230 | .flags = CFTYPE_NOT_ON_ROOT, | |
7231 | .seq_show = memory_min_show, | |
7232 | .write = memory_min_write, | |
7233 | }, | |
241994ed JW |
7234 | { |
7235 | .name = "low", | |
7236 | .flags = CFTYPE_NOT_ON_ROOT, | |
7237 | .seq_show = memory_low_show, | |
7238 | .write = memory_low_write, | |
7239 | }, | |
7240 | { | |
7241 | .name = "high", | |
7242 | .flags = CFTYPE_NOT_ON_ROOT, | |
7243 | .seq_show = memory_high_show, | |
7244 | .write = memory_high_write, | |
7245 | }, | |
7246 | { | |
7247 | .name = "max", | |
7248 | .flags = CFTYPE_NOT_ON_ROOT, | |
7249 | .seq_show = memory_max_show, | |
7250 | .write = memory_max_write, | |
7251 | }, | |
7252 | { | |
7253 | .name = "events", | |
7254 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 7255 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
7256 | .seq_show = memory_events_show, |
7257 | }, | |
1e577f97 SB |
7258 | { |
7259 | .name = "events.local", | |
7260 | .flags = CFTYPE_NOT_ON_ROOT, | |
7261 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
7262 | .seq_show = memory_events_local_show, | |
7263 | }, | |
587d9f72 JW |
7264 | { |
7265 | .name = "stat", | |
587d9f72 JW |
7266 | .seq_show = memory_stat_show, |
7267 | }, | |
5f9a4f4a MS |
7268 | #ifdef CONFIG_NUMA |
7269 | { | |
7270 | .name = "numa_stat", | |
7271 | .seq_show = memory_numa_stat_show, | |
7272 | }, | |
7273 | #endif | |
3d8b38eb RG |
7274 | { |
7275 | .name = "oom.group", | |
7276 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
7277 | .seq_show = memory_oom_group_show, | |
7278 | .write = memory_oom_group_write, | |
7279 | }, | |
94968384 SB |
7280 | { |
7281 | .name = "reclaim", | |
7282 | .flags = CFTYPE_NS_DELEGATABLE, | |
7283 | .write = memory_reclaim, | |
7284 | }, | |
241994ed JW |
7285 | { } /* terminate */ |
7286 | }; | |
7287 | ||
073219e9 | 7288 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 7289 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 7290 | .css_online = mem_cgroup_css_online, |
92fb9748 | 7291 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 7292 | .css_released = mem_cgroup_css_released, |
92fb9748 | 7293 | .css_free = mem_cgroup_css_free, |
1ced953b | 7294 | .css_reset = mem_cgroup_css_reset, |
2d146aa3 | 7295 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
7dc74be0 | 7296 | .can_attach = mem_cgroup_can_attach, |
1aacbd35 | 7297 | #if defined(CONFIG_LRU_GEN) || defined(CONFIG_MEMCG_KMEM) |
bd74fdae | 7298 | .attach = mem_cgroup_attach, |
1aacbd35 | 7299 | #endif |
7dc74be0 | 7300 | .cancel_attach = mem_cgroup_cancel_attach, |
264a0ae1 | 7301 | .post_attach = mem_cgroup_move_task, |
1aacbd35 RG |
7302 | #ifdef CONFIG_MEMCG_KMEM |
7303 | .fork = mem_cgroup_fork, | |
7304 | .exit = mem_cgroup_exit, | |
7305 | #endif | |
241994ed JW |
7306 | .dfl_cftypes = memory_files, |
7307 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 7308 | .early_init = 0, |
8cdea7c0 | 7309 | }; |
c077719b | 7310 | |
bc50bcc6 JW |
7311 | /* |
7312 | * This function calculates an individual cgroup's effective | |
7313 | * protection which is derived from its own memory.min/low, its | |
7314 | * parent's and siblings' settings, as well as the actual memory | |
7315 | * distribution in the tree. | |
7316 | * | |
7317 | * The following rules apply to the effective protection values: | |
7318 | * | |
7319 | * 1. At the first level of reclaim, effective protection is equal to | |
7320 | * the declared protection in memory.min and memory.low. | |
7321 | * | |
7322 | * 2. To enable safe delegation of the protection configuration, at | |
7323 | * subsequent levels the effective protection is capped to the | |
7324 | * parent's effective protection. | |
7325 | * | |
7326 | * 3. To make complex and dynamic subtrees easier to configure, the | |
7327 | * user is allowed to overcommit the declared protection at a given | |
7328 | * level. If that is the case, the parent's effective protection is | |
7329 | * distributed to the children in proportion to how much protection | |
7330 | * they have declared and how much of it they are utilizing. | |
7331 | * | |
7332 | * This makes distribution proportional, but also work-conserving: | |
7333 | * if one cgroup claims much more protection than it uses memory, | |
7334 | * the unused remainder is available to its siblings. | |
7335 | * | |
7336 | * 4. Conversely, when the declared protection is undercommitted at a | |
7337 | * given level, the distribution of the larger parental protection | |
7338 | * budget is NOT proportional. A cgroup's protection from a sibling | |
7339 | * is capped to its own memory.min/low setting. | |
7340 | * | |
8a931f80 JW |
7341 | * 5. However, to allow protecting recursive subtrees from each other |
7342 | * without having to declare each individual cgroup's fixed share | |
7343 | * of the ancestor's claim to protection, any unutilized - | |
7344 | * "floating" - protection from up the tree is distributed in | |
7345 | * proportion to each cgroup's *usage*. This makes the protection | |
7346 | * neutral wrt sibling cgroups and lets them compete freely over | |
7347 | * the shared parental protection budget, but it protects the | |
7348 | * subtree as a whole from neighboring subtrees. | |
7349 | * | |
7350 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
7351 | * against immediate siblings whereas 5. is about protecting against | |
7352 | * neighboring subtrees. | |
bc50bcc6 JW |
7353 | */ |
7354 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 7355 | unsigned long parent_usage, |
bc50bcc6 JW |
7356 | unsigned long setting, |
7357 | unsigned long parent_effective, | |
7358 | unsigned long siblings_protected) | |
7359 | { | |
7360 | unsigned long protected; | |
8a931f80 | 7361 | unsigned long ep; |
bc50bcc6 JW |
7362 | |
7363 | protected = min(usage, setting); | |
7364 | /* | |
7365 | * If all cgroups at this level combined claim and use more | |
08e0f49e | 7366 | * protection than what the parent affords them, distribute |
bc50bcc6 JW |
7367 | * shares in proportion to utilization. |
7368 | * | |
7369 | * We are using actual utilization rather than the statically | |
7370 | * claimed protection in order to be work-conserving: claimed | |
7371 | * but unused protection is available to siblings that would | |
7372 | * otherwise get a smaller chunk than what they claimed. | |
7373 | */ | |
7374 | if (siblings_protected > parent_effective) | |
7375 | return protected * parent_effective / siblings_protected; | |
7376 | ||
7377 | /* | |
7378 | * Ok, utilized protection of all children is within what the | |
7379 | * parent affords them, so we know whatever this child claims | |
7380 | * and utilizes is effectively protected. | |
7381 | * | |
7382 | * If there is unprotected usage beyond this value, reclaim | |
7383 | * will apply pressure in proportion to that amount. | |
7384 | * | |
7385 | * If there is unutilized protection, the cgroup will be fully | |
7386 | * shielded from reclaim, but we do return a smaller value for | |
7387 | * protection than what the group could enjoy in theory. This | |
7388 | * is okay. With the overcommit distribution above, effective | |
7389 | * protection is always dependent on how memory is actually | |
7390 | * consumed among the siblings anyway. | |
7391 | */ | |
8a931f80 JW |
7392 | ep = protected; |
7393 | ||
7394 | /* | |
7395 | * If the children aren't claiming (all of) the protection | |
7396 | * afforded to them by the parent, distribute the remainder in | |
7397 | * proportion to the (unprotected) memory of each cgroup. That | |
7398 | * way, cgroups that aren't explicitly prioritized wrt each | |
7399 | * other compete freely over the allowance, but they are | |
7400 | * collectively protected from neighboring trees. | |
7401 | * | |
7402 | * We're using unprotected memory for the weight so that if | |
7403 | * some cgroups DO claim explicit protection, we don't protect | |
7404 | * the same bytes twice. | |
cd324edc JW |
7405 | * |
7406 | * Check both usage and parent_usage against the respective | |
7407 | * protected values. One should imply the other, but they | |
7408 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
7409 | */ |
7410 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
7411 | return ep; | |
cd324edc JW |
7412 | if (parent_effective > siblings_protected && |
7413 | parent_usage > siblings_protected && | |
7414 | usage > protected) { | |
8a931f80 JW |
7415 | unsigned long unclaimed; |
7416 | ||
7417 | unclaimed = parent_effective - siblings_protected; | |
7418 | unclaimed *= usage - protected; | |
7419 | unclaimed /= parent_usage - siblings_protected; | |
7420 | ||
7421 | ep += unclaimed; | |
7422 | } | |
7423 | ||
7424 | return ep; | |
bc50bcc6 JW |
7425 | } |
7426 | ||
241994ed | 7427 | /** |
05395718 | 7428 | * mem_cgroup_calculate_protection - check if memory consumption is in the normal range |
34c81057 | 7429 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
7430 | * @memcg: the memory cgroup to check |
7431 | * | |
23067153 RG |
7432 | * WARNING: This function is not stateless! It can only be used as part |
7433 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 7434 | */ |
45c7f7e1 CD |
7435 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
7436 | struct mem_cgroup *memcg) | |
241994ed | 7437 | { |
8a931f80 | 7438 | unsigned long usage, parent_usage; |
23067153 RG |
7439 | struct mem_cgroup *parent; |
7440 | ||
241994ed | 7441 | if (mem_cgroup_disabled()) |
45c7f7e1 | 7442 | return; |
241994ed | 7443 | |
34c81057 SC |
7444 | if (!root) |
7445 | root = root_mem_cgroup; | |
22f7496f YS |
7446 | |
7447 | /* | |
7448 | * Effective values of the reclaim targets are ignored so they | |
7449 | * can be stale. Have a look at mem_cgroup_protection for more | |
7450 | * details. | |
7451 | * TODO: calculation should be more robust so that we do not need | |
7452 | * that special casing. | |
7453 | */ | |
34c81057 | 7454 | if (memcg == root) |
45c7f7e1 | 7455 | return; |
241994ed | 7456 | |
23067153 | 7457 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 7458 | if (!usage) |
45c7f7e1 | 7459 | return; |
bf8d5d52 | 7460 | |
bf8d5d52 | 7461 | parent = parent_mem_cgroup(memcg); |
df2a4196 | 7462 | |
bc50bcc6 | 7463 | if (parent == root) { |
c3d53200 | 7464 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 7465 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 7466 | return; |
bf8d5d52 RG |
7467 | } |
7468 | ||
8a931f80 JW |
7469 | parent_usage = page_counter_read(&parent->memory); |
7470 | ||
b3a7822e | 7471 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
7472 | READ_ONCE(memcg->memory.min), |
7473 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 7474 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 7475 | |
b3a7822e | 7476 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
7477 | READ_ONCE(memcg->memory.low), |
7478 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 7479 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
7480 | } |
7481 | ||
8f425e4e MWO |
7482 | static int charge_memcg(struct folio *folio, struct mem_cgroup *memcg, |
7483 | gfp_t gfp) | |
0add0c77 | 7484 | { |
0add0c77 SB |
7485 | int ret; |
7486 | ||
4b569387 | 7487 | ret = try_charge(memcg, gfp, folio_nr_pages(folio)); |
0add0c77 SB |
7488 | if (ret) |
7489 | goto out; | |
7490 | ||
4b569387 | 7491 | mem_cgroup_commit_charge(folio, memcg); |
0add0c77 SB |
7492 | out: |
7493 | return ret; | |
7494 | } | |
7495 | ||
8f425e4e | 7496 | int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp) |
00501b53 | 7497 | { |
0add0c77 SB |
7498 | struct mem_cgroup *memcg; |
7499 | int ret; | |
00501b53 | 7500 | |
0add0c77 | 7501 | memcg = get_mem_cgroup_from_mm(mm); |
8f425e4e | 7502 | ret = charge_memcg(folio, memcg, gfp); |
0add0c77 | 7503 | css_put(&memcg->css); |
2d1c4980 | 7504 | |
0add0c77 SB |
7505 | return ret; |
7506 | } | |
e993d905 | 7507 | |
8cba9576 NP |
7508 | /** |
7509 | * mem_cgroup_hugetlb_try_charge - try to charge the memcg for a hugetlb folio | |
7510 | * @memcg: memcg to charge. | |
7511 | * @gfp: reclaim mode. | |
7512 | * @nr_pages: number of pages to charge. | |
7513 | * | |
7514 | * This function is called when allocating a huge page folio to determine if | |
7515 | * the memcg has the capacity for it. It does not commit the charge yet, | |
7516 | * as the hugetlb folio itself has not been obtained from the hugetlb pool. | |
7517 | * | |
7518 | * Once we have obtained the hugetlb folio, we can call | |
7519 | * mem_cgroup_commit_charge() to commit the charge. If we fail to obtain the | |
7520 | * folio, we should instead call mem_cgroup_cancel_charge() to undo the effect | |
7521 | * of try_charge(). | |
7522 | * | |
7523 | * Returns 0 on success. Otherwise, an error code is returned. | |
7524 | */ | |
7525 | int mem_cgroup_hugetlb_try_charge(struct mem_cgroup *memcg, gfp_t gfp, | |
7526 | long nr_pages) | |
7527 | { | |
7528 | /* | |
7529 | * If hugetlb memcg charging is not enabled, do not fail hugetlb allocation, | |
7530 | * but do not attempt to commit charge later (or cancel on error) either. | |
7531 | */ | |
7532 | if (mem_cgroup_disabled() || !memcg || | |
7533 | !cgroup_subsys_on_dfl(memory_cgrp_subsys) || | |
7534 | !(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING)) | |
7535 | return -EOPNOTSUPP; | |
7536 | ||
7537 | if (try_charge(memcg, gfp, nr_pages)) | |
7538 | return -ENOMEM; | |
7539 | ||
7540 | return 0; | |
7541 | } | |
7542 | ||
0add0c77 | 7543 | /** |
65995918 MWO |
7544 | * mem_cgroup_swapin_charge_folio - Charge a newly allocated folio for swapin. |
7545 | * @folio: folio to charge. | |
0add0c77 SB |
7546 | * @mm: mm context of the victim |
7547 | * @gfp: reclaim mode | |
65995918 | 7548 | * @entry: swap entry for which the folio is allocated |
0add0c77 | 7549 | * |
65995918 MWO |
7550 | * This function charges a folio allocated for swapin. Please call this before |
7551 | * adding the folio to the swapcache. | |
0add0c77 SB |
7552 | * |
7553 | * Returns 0 on success. Otherwise, an error code is returned. | |
7554 | */ | |
65995918 | 7555 | int mem_cgroup_swapin_charge_folio(struct folio *folio, struct mm_struct *mm, |
0add0c77 SB |
7556 | gfp_t gfp, swp_entry_t entry) |
7557 | { | |
7558 | struct mem_cgroup *memcg; | |
7559 | unsigned short id; | |
7560 | int ret; | |
00501b53 | 7561 | |
0add0c77 SB |
7562 | if (mem_cgroup_disabled()) |
7563 | return 0; | |
00501b53 | 7564 | |
0add0c77 SB |
7565 | id = lookup_swap_cgroup_id(entry); |
7566 | rcu_read_lock(); | |
7567 | memcg = mem_cgroup_from_id(id); | |
7568 | if (!memcg || !css_tryget_online(&memcg->css)) | |
7569 | memcg = get_mem_cgroup_from_mm(mm); | |
7570 | rcu_read_unlock(); | |
00501b53 | 7571 | |
8f425e4e | 7572 | ret = charge_memcg(folio, memcg, gfp); |
6abb5a86 | 7573 | |
0add0c77 SB |
7574 | css_put(&memcg->css); |
7575 | return ret; | |
7576 | } | |
00501b53 | 7577 | |
0add0c77 SB |
7578 | /* |
7579 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot | |
7580 | * @entry: swap entry for which the page is charged | |
7581 | * | |
7582 | * Call this function after successfully adding the charged page to swapcache. | |
7583 | * | |
7584 | * Note: This function assumes the page for which swap slot is being uncharged | |
7585 | * is order 0 page. | |
7586 | */ | |
7587 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) | |
7588 | { | |
cae3af62 MS |
7589 | /* |
7590 | * Cgroup1's unified memory+swap counter has been charged with the | |
7591 | * new swapcache page, finish the transfer by uncharging the swap | |
7592 | * slot. The swap slot would also get uncharged when it dies, but | |
7593 | * it can stick around indefinitely and we'd count the page twice | |
7594 | * the entire time. | |
7595 | * | |
7596 | * Cgroup2 has separate resource counters for memory and swap, | |
7597 | * so this is a non-issue here. Memory and swap charge lifetimes | |
7598 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
7599 | * page to memory here, and uncharge swap when the slot is freed. | |
7600 | */ | |
0add0c77 | 7601 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
00501b53 JW |
7602 | /* |
7603 | * The swap entry might not get freed for a long time, | |
7604 | * let's not wait for it. The page already received a | |
7605 | * memory+swap charge, drop the swap entry duplicate. | |
7606 | */ | |
0add0c77 | 7607 | mem_cgroup_uncharge_swap(entry, 1); |
00501b53 | 7608 | } |
3fea5a49 JW |
7609 | } |
7610 | ||
a9d5adee JG |
7611 | struct uncharge_gather { |
7612 | struct mem_cgroup *memcg; | |
b4e0b68f | 7613 | unsigned long nr_memory; |
a9d5adee | 7614 | unsigned long pgpgout; |
a9d5adee | 7615 | unsigned long nr_kmem; |
8e88bd2d | 7616 | int nid; |
a9d5adee JG |
7617 | }; |
7618 | ||
7619 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 7620 | { |
a9d5adee JG |
7621 | memset(ug, 0, sizeof(*ug)); |
7622 | } | |
7623 | ||
7624 | static void uncharge_batch(const struct uncharge_gather *ug) | |
7625 | { | |
747db954 JW |
7626 | unsigned long flags; |
7627 | ||
b4e0b68f MS |
7628 | if (ug->nr_memory) { |
7629 | page_counter_uncharge(&ug->memcg->memory, ug->nr_memory); | |
7941d214 | 7630 | if (do_memsw_account()) |
b4e0b68f | 7631 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory); |
a8c49af3 YA |
7632 | if (ug->nr_kmem) |
7633 | memcg_account_kmem(ug->memcg, -ug->nr_kmem); | |
a9d5adee | 7634 | memcg_oom_recover(ug->memcg); |
ce00a967 | 7635 | } |
747db954 JW |
7636 | |
7637 | local_irq_save(flags); | |
c9019e9b | 7638 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
b4e0b68f | 7639 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory); |
8e88bd2d | 7640 | memcg_check_events(ug->memcg, ug->nid); |
747db954 | 7641 | local_irq_restore(flags); |
f1796544 | 7642 | |
c4ed6ebf | 7643 | /* drop reference from uncharge_folio */ |
f1796544 | 7644 | css_put(&ug->memcg->css); |
a9d5adee JG |
7645 | } |
7646 | ||
c4ed6ebf | 7647 | static void uncharge_folio(struct folio *folio, struct uncharge_gather *ug) |
a9d5adee | 7648 | { |
c4ed6ebf | 7649 | long nr_pages; |
b4e0b68f MS |
7650 | struct mem_cgroup *memcg; |
7651 | struct obj_cgroup *objcg; | |
9f762dbe | 7652 | |
c4ed6ebf | 7653 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
b7b098cf MWO |
7654 | VM_BUG_ON_FOLIO(folio_order(folio) > 1 && |
7655 | !folio_test_hugetlb(folio) && | |
7656 | !list_empty(&folio->_deferred_list), folio); | |
a9d5adee | 7657 | |
a9d5adee JG |
7658 | /* |
7659 | * Nobody should be changing or seriously looking at | |
c4ed6ebf MWO |
7660 | * folio memcg or objcg at this point, we have fully |
7661 | * exclusive access to the folio. | |
a9d5adee | 7662 | */ |
fead2b86 | 7663 | if (folio_memcg_kmem(folio)) { |
1b7e4464 | 7664 | objcg = __folio_objcg(folio); |
b4e0b68f MS |
7665 | /* |
7666 | * This get matches the put at the end of the function and | |
7667 | * kmem pages do not hold memcg references anymore. | |
7668 | */ | |
7669 | memcg = get_mem_cgroup_from_objcg(objcg); | |
7670 | } else { | |
1b7e4464 | 7671 | memcg = __folio_memcg(folio); |
b4e0b68f | 7672 | } |
a9d5adee | 7673 | |
b4e0b68f MS |
7674 | if (!memcg) |
7675 | return; | |
7676 | ||
7677 | if (ug->memcg != memcg) { | |
a9d5adee JG |
7678 | if (ug->memcg) { |
7679 | uncharge_batch(ug); | |
7680 | uncharge_gather_clear(ug); | |
7681 | } | |
b4e0b68f | 7682 | ug->memcg = memcg; |
c4ed6ebf | 7683 | ug->nid = folio_nid(folio); |
f1796544 MH |
7684 | |
7685 | /* pairs with css_put in uncharge_batch */ | |
b4e0b68f | 7686 | css_get(&memcg->css); |
a9d5adee JG |
7687 | } |
7688 | ||
c4ed6ebf | 7689 | nr_pages = folio_nr_pages(folio); |
a9d5adee | 7690 | |
fead2b86 | 7691 | if (folio_memcg_kmem(folio)) { |
b4e0b68f | 7692 | ug->nr_memory += nr_pages; |
9f762dbe | 7693 | ug->nr_kmem += nr_pages; |
b4e0b68f | 7694 | |
c4ed6ebf | 7695 | folio->memcg_data = 0; |
b4e0b68f MS |
7696 | obj_cgroup_put(objcg); |
7697 | } else { | |
7698 | /* LRU pages aren't accounted at the root level */ | |
7699 | if (!mem_cgroup_is_root(memcg)) | |
7700 | ug->nr_memory += nr_pages; | |
18b2db3b | 7701 | ug->pgpgout++; |
a9d5adee | 7702 | |
c4ed6ebf | 7703 | folio->memcg_data = 0; |
b4e0b68f MS |
7704 | } |
7705 | ||
7706 | css_put(&memcg->css); | |
747db954 JW |
7707 | } |
7708 | ||
bbc6b703 | 7709 | void __mem_cgroup_uncharge(struct folio *folio) |
0a31bc97 | 7710 | { |
a9d5adee JG |
7711 | struct uncharge_gather ug; |
7712 | ||
bbc6b703 MWO |
7713 | /* Don't touch folio->lru of any random page, pre-check: */ |
7714 | if (!folio_memcg(folio)) | |
0a31bc97 JW |
7715 | return; |
7716 | ||
a9d5adee | 7717 | uncharge_gather_clear(&ug); |
bbc6b703 | 7718 | uncharge_folio(folio, &ug); |
a9d5adee | 7719 | uncharge_batch(&ug); |
747db954 | 7720 | } |
0a31bc97 | 7721 | |
4882c809 MWO |
7722 | void __mem_cgroup_uncharge_folios(struct folio_batch *folios) |
7723 | { | |
7724 | struct uncharge_gather ug; | |
7725 | unsigned int i; | |
7726 | ||
7727 | uncharge_gather_clear(&ug); | |
7728 | for (i = 0; i < folios->nr; i++) | |
7729 | uncharge_folio(folios->folios[i], &ug); | |
7730 | if (ug.memcg) | |
7731 | uncharge_batch(&ug); | |
0a31bc97 JW |
7732 | } |
7733 | ||
7734 | /** | |
85ce2c51 | 7735 | * mem_cgroup_replace_folio - Charge a folio's replacement. |
d21bba2b MWO |
7736 | * @old: Currently circulating folio. |
7737 | * @new: Replacement folio. | |
0a31bc97 | 7738 | * |
d21bba2b | 7739 | * Charge @new as a replacement folio for @old. @old will |
85ce2c51 NP |
7740 | * be uncharged upon free. This is only used by the page cache |
7741 | * (in replace_page_cache_folio()). | |
0a31bc97 | 7742 | * |
d21bba2b | 7743 | * Both folios must be locked, @new->mapping must be set up. |
0a31bc97 | 7744 | */ |
85ce2c51 | 7745 | void mem_cgroup_replace_folio(struct folio *old, struct folio *new) |
0a31bc97 | 7746 | { |
29833315 | 7747 | struct mem_cgroup *memcg; |
d21bba2b | 7748 | long nr_pages = folio_nr_pages(new); |
d93c4130 | 7749 | unsigned long flags; |
0a31bc97 | 7750 | |
d21bba2b MWO |
7751 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); |
7752 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7753 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7754 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != nr_pages, new); | |
0a31bc97 JW |
7755 | |
7756 | if (mem_cgroup_disabled()) | |
7757 | return; | |
7758 | ||
d21bba2b MWO |
7759 | /* Page cache replacement: new folio already charged? */ |
7760 | if (folio_memcg(new)) | |
0a31bc97 JW |
7761 | return; |
7762 | ||
d21bba2b MWO |
7763 | memcg = folio_memcg(old); |
7764 | VM_WARN_ON_ONCE_FOLIO(!memcg, old); | |
29833315 | 7765 | if (!memcg) |
0a31bc97 JW |
7766 | return; |
7767 | ||
44b7a8d3 | 7768 | /* Force-charge the new page. The old one will be freed soon */ |
8dc87c7d MS |
7769 | if (!mem_cgroup_is_root(memcg)) { |
7770 | page_counter_charge(&memcg->memory, nr_pages); | |
7771 | if (do_memsw_account()) | |
7772 | page_counter_charge(&memcg->memsw, nr_pages); | |
7773 | } | |
0a31bc97 | 7774 | |
1a3e1f40 | 7775 | css_get(&memcg->css); |
d21bba2b | 7776 | commit_charge(new, memcg); |
44b7a8d3 | 7777 | |
d93c4130 | 7778 | local_irq_save(flags); |
6e0110c2 | 7779 | mem_cgroup_charge_statistics(memcg, nr_pages); |
d21bba2b | 7780 | memcg_check_events(memcg, folio_nid(new)); |
d93c4130 | 7781 | local_irq_restore(flags); |
0a31bc97 JW |
7782 | } |
7783 | ||
85ce2c51 NP |
7784 | /** |
7785 | * mem_cgroup_migrate - Transfer the memcg data from the old to the new folio. | |
7786 | * @old: Currently circulating folio. | |
7787 | * @new: Replacement folio. | |
7788 | * | |
7789 | * Transfer the memcg data from the old folio to the new folio for migration. | |
7790 | * The old folio's data info will be cleared. Note that the memory counters | |
7791 | * will remain unchanged throughout the process. | |
7792 | * | |
7793 | * Both folios must be locked, @new->mapping must be set up. | |
7794 | */ | |
7795 | void mem_cgroup_migrate(struct folio *old, struct folio *new) | |
7796 | { | |
7797 | struct mem_cgroup *memcg; | |
7798 | ||
7799 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); | |
7800 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7801 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7802 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != folio_nr_pages(new), new); | |
7803 | ||
7804 | if (mem_cgroup_disabled()) | |
7805 | return; | |
7806 | ||
7807 | memcg = folio_memcg(old); | |
8cba9576 NP |
7808 | /* |
7809 | * Note that it is normal to see !memcg for a hugetlb folio. | |
7810 | * For e.g, itt could have been allocated when memory_hugetlb_accounting | |
7811 | * was not selected. | |
7812 | */ | |
7813 | VM_WARN_ON_ONCE_FOLIO(!folio_test_hugetlb(old) && !memcg, old); | |
85ce2c51 NP |
7814 | if (!memcg) |
7815 | return; | |
7816 | ||
7817 | /* Transfer the charge and the css ref */ | |
7818 | commit_charge(new, memcg); | |
9bcef597 BW |
7819 | /* |
7820 | * If the old folio is a large folio and is in the split queue, it needs | |
7821 | * to be removed from the split queue now, in case getting an incorrect | |
7822 | * split queue in destroy_large_folio() after the memcg of the old folio | |
7823 | * is cleared. | |
7824 | * | |
7825 | * In addition, the old folio is about to be freed after migration, so | |
7826 | * removing from the split queue a bit earlier seems reasonable. | |
7827 | */ | |
7828 | if (folio_test_large(old) && folio_test_large_rmappable(old)) | |
7829 | folio_undo_large_rmappable(old); | |
85ce2c51 NP |
7830 | old->memcg_data = 0; |
7831 | } | |
7832 | ||
ef12947c | 7833 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
7834 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
7835 | ||
2d758073 | 7836 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
7837 | { |
7838 | struct mem_cgroup *memcg; | |
7839 | ||
2d758073 JW |
7840 | if (!mem_cgroup_sockets_enabled) |
7841 | return; | |
7842 | ||
e876ecc6 | 7843 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
086f694a | 7844 | if (!in_task()) |
e876ecc6 SB |
7845 | return; |
7846 | ||
11092087 JW |
7847 | rcu_read_lock(); |
7848 | memcg = mem_cgroup_from_task(current); | |
7848ed62 | 7849 | if (mem_cgroup_is_root(memcg)) |
f7e1cb6e | 7850 | goto out; |
0db15298 | 7851 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 7852 | goto out; |
8965aa28 | 7853 | if (css_tryget(&memcg->css)) |
11092087 | 7854 | sk->sk_memcg = memcg; |
f7e1cb6e | 7855 | out: |
11092087 JW |
7856 | rcu_read_unlock(); |
7857 | } | |
11092087 | 7858 | |
2d758073 | 7859 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 7860 | { |
2d758073 JW |
7861 | if (sk->sk_memcg) |
7862 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7863 | } |
7864 | ||
7865 | /** | |
7866 | * mem_cgroup_charge_skmem - charge socket memory | |
7867 | * @memcg: memcg to charge | |
7868 | * @nr_pages: number of pages to charge | |
4b1327be | 7869 | * @gfp_mask: reclaim mode |
11092087 JW |
7870 | * |
7871 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
4b1327be | 7872 | * @memcg's configured limit, %false if it doesn't. |
11092087 | 7873 | */ |
4b1327be WW |
7874 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, |
7875 | gfp_t gfp_mask) | |
11092087 | 7876 | { |
f7e1cb6e | 7877 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7878 | struct page_counter *fail; |
f7e1cb6e | 7879 | |
0db15298 JW |
7880 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7881 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7882 | return true; |
7883 | } | |
0db15298 | 7884 | memcg->tcpmem_pressure = 1; |
4b1327be WW |
7885 | if (gfp_mask & __GFP_NOFAIL) { |
7886 | page_counter_charge(&memcg->tcpmem, nr_pages); | |
7887 | return true; | |
7888 | } | |
f7e1cb6e | 7889 | return false; |
11092087 | 7890 | } |
d886f4e4 | 7891 | |
4b1327be WW |
7892 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) { |
7893 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); | |
f7e1cb6e | 7894 | return true; |
4b1327be | 7895 | } |
f7e1cb6e | 7896 | |
11092087 JW |
7897 | return false; |
7898 | } | |
7899 | ||
7900 | /** | |
7901 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7902 | * @memcg: memcg to uncharge |
7903 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7904 | */ |
7905 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7906 | { | |
f7e1cb6e | 7907 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7908 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7909 | return; |
7910 | } | |
d886f4e4 | 7911 | |
c9019e9b | 7912 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7913 | |
475d0487 | 7914 | refill_stock(memcg, nr_pages); |
11092087 JW |
7915 | } |
7916 | ||
f7e1cb6e JW |
7917 | static int __init cgroup_memory(char *s) |
7918 | { | |
7919 | char *token; | |
7920 | ||
7921 | while ((token = strsep(&s, ",")) != NULL) { | |
7922 | if (!*token) | |
7923 | continue; | |
7924 | if (!strcmp(token, "nosocket")) | |
7925 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7926 | if (!strcmp(token, "nokmem")) |
7927 | cgroup_memory_nokmem = true; | |
b6c1a8af YS |
7928 | if (!strcmp(token, "nobpf")) |
7929 | cgroup_memory_nobpf = true; | |
f7e1cb6e | 7930 | } |
460a79e1 | 7931 | return 1; |
f7e1cb6e JW |
7932 | } |
7933 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7934 | |
2d11085e | 7935 | /* |
1081312f MH |
7936 | * subsys_initcall() for memory controller. |
7937 | * | |
308167fc SAS |
7938 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7939 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7940 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7941 | * should be initialized from here. | |
2d11085e MH |
7942 | */ |
7943 | static int __init mem_cgroup_init(void) | |
7944 | { | |
95a045f6 JW |
7945 | int cpu, node; |
7946 | ||
f3344adf MS |
7947 | /* |
7948 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7949 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7950 | * to work fine, we should make sure that the overfill threshold can't | |
7951 | * exceed S32_MAX / PAGE_SIZE. | |
7952 | */ | |
7953 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7954 | ||
308167fc SAS |
7955 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7956 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7957 | |
7958 | for_each_possible_cpu(cpu) | |
7959 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7960 | drain_local_stock); | |
7961 | ||
7962 | for_each_node(node) { | |
7963 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 | 7964 | |
91f0dcce | 7965 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, node); |
95a045f6 | 7966 | |
ef8f2327 | 7967 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7968 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7969 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7970 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7971 | } | |
7972 | ||
2d11085e MH |
7973 | return 0; |
7974 | } | |
7975 | subsys_initcall(mem_cgroup_init); | |
21afa38e | 7976 | |
e55b9f96 | 7977 | #ifdef CONFIG_SWAP |
358c07fc AB |
7978 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7979 | { | |
1c2d479a | 7980 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7981 | /* |
7982 | * The root cgroup cannot be destroyed, so it's refcount must | |
7983 | * always be >= 1. | |
7984 | */ | |
7848ed62 | 7985 | if (WARN_ON_ONCE(mem_cgroup_is_root(memcg))) { |
358c07fc AB |
7986 | VM_BUG_ON(1); |
7987 | break; | |
7988 | } | |
7989 | memcg = parent_mem_cgroup(memcg); | |
7990 | if (!memcg) | |
7991 | memcg = root_mem_cgroup; | |
7992 | } | |
7993 | return memcg; | |
7994 | } | |
7995 | ||
21afa38e JW |
7996 | /** |
7997 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
3ecb0087 | 7998 | * @folio: folio whose memsw charge to transfer |
21afa38e JW |
7999 | * @entry: swap entry to move the charge to |
8000 | * | |
3ecb0087 | 8001 | * Transfer the memsw charge of @folio to @entry. |
21afa38e | 8002 | */ |
3ecb0087 | 8003 | void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry) |
21afa38e | 8004 | { |
1f47b61f | 8005 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 8006 | unsigned int nr_entries; |
21afa38e JW |
8007 | unsigned short oldid; |
8008 | ||
3ecb0087 MWO |
8009 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
8010 | VM_BUG_ON_FOLIO(folio_ref_count(folio), folio); | |
21afa38e | 8011 | |
76358ab5 AS |
8012 | if (mem_cgroup_disabled()) |
8013 | return; | |
8014 | ||
b94c4e94 | 8015 | if (!do_memsw_account()) |
21afa38e JW |
8016 | return; |
8017 | ||
3ecb0087 | 8018 | memcg = folio_memcg(folio); |
21afa38e | 8019 | |
3ecb0087 | 8020 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
21afa38e JW |
8021 | if (!memcg) |
8022 | return; | |
8023 | ||
1f47b61f VD |
8024 | /* |
8025 | * In case the memcg owning these pages has been offlined and doesn't | |
8026 | * have an ID allocated to it anymore, charge the closest online | |
8027 | * ancestor for the swap instead and transfer the memory+swap charge. | |
8028 | */ | |
8029 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
3ecb0087 | 8030 | nr_entries = folio_nr_pages(folio); |
d6810d73 HY |
8031 | /* Get references for the tail pages, too */ |
8032 | if (nr_entries > 1) | |
8033 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
8034 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
8035 | nr_entries); | |
3ecb0087 | 8036 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 8037 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 8038 | |
3ecb0087 | 8039 | folio->memcg_data = 0; |
21afa38e JW |
8040 | |
8041 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 8042 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 8043 | |
b25806dc | 8044 | if (memcg != swap_memcg) { |
1f47b61f | 8045 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
8046 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
8047 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
8048 | } |
8049 | ||
ce9ce665 SAS |
8050 | /* |
8051 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 8052 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 8053 | * important here to have the interrupts disabled because it is the |
b93b0163 | 8054 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 | 8055 | */ |
be3e67b5 | 8056 | memcg_stats_lock(); |
6e0110c2 | 8057 | mem_cgroup_charge_statistics(memcg, -nr_entries); |
be3e67b5 | 8058 | memcg_stats_unlock(); |
3ecb0087 | 8059 | memcg_check_events(memcg, folio_nid(folio)); |
73f576c0 | 8060 | |
1a3e1f40 | 8061 | css_put(&memcg->css); |
21afa38e JW |
8062 | } |
8063 | ||
38d8b4e6 | 8064 | /** |
e2e3fdc7 MWO |
8065 | * __mem_cgroup_try_charge_swap - try charging swap space for a folio |
8066 | * @folio: folio being added to swap | |
37e84351 VD |
8067 | * @entry: swap entry to charge |
8068 | * | |
e2e3fdc7 | 8069 | * Try to charge @folio's memcg for the swap space at @entry. |
37e84351 VD |
8070 | * |
8071 | * Returns 0 on success, -ENOMEM on failure. | |
8072 | */ | |
e2e3fdc7 | 8073 | int __mem_cgroup_try_charge_swap(struct folio *folio, swp_entry_t entry) |
37e84351 | 8074 | { |
e2e3fdc7 | 8075 | unsigned int nr_pages = folio_nr_pages(folio); |
37e84351 | 8076 | struct page_counter *counter; |
38d8b4e6 | 8077 | struct mem_cgroup *memcg; |
37e84351 VD |
8078 | unsigned short oldid; |
8079 | ||
b94c4e94 | 8080 | if (do_memsw_account()) |
37e84351 VD |
8081 | return 0; |
8082 | ||
e2e3fdc7 | 8083 | memcg = folio_memcg(folio); |
37e84351 | 8084 | |
e2e3fdc7 | 8085 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
37e84351 VD |
8086 | if (!memcg) |
8087 | return 0; | |
8088 | ||
f3a53a3a TH |
8089 | if (!entry.val) { |
8090 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 8091 | return 0; |
f3a53a3a | 8092 | } |
bb98f2c5 | 8093 | |
1f47b61f VD |
8094 | memcg = mem_cgroup_id_get_online(memcg); |
8095 | ||
b25806dc | 8096 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 8097 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
8098 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
8099 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 8100 | mem_cgroup_id_put(memcg); |
37e84351 | 8101 | return -ENOMEM; |
1f47b61f | 8102 | } |
37e84351 | 8103 | |
38d8b4e6 HY |
8104 | /* Get references for the tail pages, too */ |
8105 | if (nr_pages > 1) | |
8106 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
8107 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
e2e3fdc7 | 8108 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 8109 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 8110 | |
37e84351 VD |
8111 | return 0; |
8112 | } | |
8113 | ||
21afa38e | 8114 | /** |
01c4b28c | 8115 | * __mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 8116 | * @entry: swap entry to uncharge |
38d8b4e6 | 8117 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 8118 | */ |
01c4b28c | 8119 | void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
8120 | { |
8121 | struct mem_cgroup *memcg; | |
8122 | unsigned short id; | |
8123 | ||
38d8b4e6 | 8124 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 8125 | rcu_read_lock(); |
adbe427b | 8126 | memcg = mem_cgroup_from_id(id); |
21afa38e | 8127 | if (memcg) { |
b25806dc | 8128 | if (!mem_cgroup_is_root(memcg)) { |
b94c4e94 | 8129 | if (do_memsw_account()) |
38d8b4e6 | 8130 | page_counter_uncharge(&memcg->memsw, nr_pages); |
b94c4e94 JW |
8131 | else |
8132 | page_counter_uncharge(&memcg->swap, nr_pages); | |
37e84351 | 8133 | } |
c9019e9b | 8134 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 8135 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
8136 | } |
8137 | rcu_read_unlock(); | |
8138 | } | |
8139 | ||
d8b38438 VD |
8140 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
8141 | { | |
8142 | long nr_swap_pages = get_nr_swap_pages(); | |
8143 | ||
b25806dc | 8144 | if (mem_cgroup_disabled() || do_memsw_account()) |
d8b38438 | 8145 | return nr_swap_pages; |
7848ed62 | 8146 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) |
d8b38438 | 8147 | nr_swap_pages = min_t(long, nr_swap_pages, |
bbec2e15 | 8148 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
8149 | page_counter_read(&memcg->swap)); |
8150 | return nr_swap_pages; | |
8151 | } | |
8152 | ||
9202d527 | 8153 | bool mem_cgroup_swap_full(struct folio *folio) |
5ccc5aba VD |
8154 | { |
8155 | struct mem_cgroup *memcg; | |
8156 | ||
9202d527 | 8157 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
5ccc5aba VD |
8158 | |
8159 | if (vm_swap_full()) | |
8160 | return true; | |
b25806dc | 8161 | if (do_memsw_account()) |
5ccc5aba VD |
8162 | return false; |
8163 | ||
9202d527 | 8164 | memcg = folio_memcg(folio); |
5ccc5aba VD |
8165 | if (!memcg) |
8166 | return false; | |
8167 | ||
7848ed62 | 8168 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
4b82ab4f JK |
8169 | unsigned long usage = page_counter_read(&memcg->swap); |
8170 | ||
8171 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
8172 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 8173 | return true; |
4b82ab4f | 8174 | } |
5ccc5aba VD |
8175 | |
8176 | return false; | |
8177 | } | |
8178 | ||
eccb52e7 | 8179 | static int __init setup_swap_account(char *s) |
21afa38e | 8180 | { |
118642d7 JW |
8181 | bool res; |
8182 | ||
8183 | if (!kstrtobool(s, &res) && !res) | |
8184 | pr_warn_once("The swapaccount=0 commandline option is deprecated " | |
8185 | "in favor of configuring swap control via cgroupfs. " | |
8186 | "Please report your usecase to linux-mm@kvack.org if you " | |
8187 | "depend on this functionality.\n"); | |
21afa38e JW |
8188 | return 1; |
8189 | } | |
eccb52e7 | 8190 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 8191 | |
37e84351 VD |
8192 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
8193 | struct cftype *cft) | |
8194 | { | |
8195 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
8196 | ||
8197 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
8198 | } | |
8199 | ||
e0e0b412 LD |
8200 | static u64 swap_peak_read(struct cgroup_subsys_state *css, |
8201 | struct cftype *cft) | |
8202 | { | |
8203 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
8204 | ||
8205 | return (u64)memcg->swap.watermark * PAGE_SIZE; | |
8206 | } | |
8207 | ||
4b82ab4f JK |
8208 | static int swap_high_show(struct seq_file *m, void *v) |
8209 | { | |
8210 | return seq_puts_memcg_tunable(m, | |
8211 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
8212 | } | |
8213 | ||
8214 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
8215 | char *buf, size_t nbytes, loff_t off) | |
8216 | { | |
8217 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8218 | unsigned long high; | |
8219 | int err; | |
8220 | ||
8221 | buf = strstrip(buf); | |
8222 | err = page_counter_memparse(buf, "max", &high); | |
8223 | if (err) | |
8224 | return err; | |
8225 | ||
8226 | page_counter_set_high(&memcg->swap, high); | |
8227 | ||
8228 | return nbytes; | |
8229 | } | |
8230 | ||
37e84351 VD |
8231 | static int swap_max_show(struct seq_file *m, void *v) |
8232 | { | |
677dc973 CD |
8233 | return seq_puts_memcg_tunable(m, |
8234 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
8235 | } |
8236 | ||
8237 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
8238 | char *buf, size_t nbytes, loff_t off) | |
8239 | { | |
8240 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8241 | unsigned long max; | |
8242 | int err; | |
8243 | ||
8244 | buf = strstrip(buf); | |
8245 | err = page_counter_memparse(buf, "max", &max); | |
8246 | if (err) | |
8247 | return err; | |
8248 | ||
be09102b | 8249 | xchg(&memcg->swap.max, max); |
37e84351 VD |
8250 | |
8251 | return nbytes; | |
8252 | } | |
8253 | ||
f3a53a3a TH |
8254 | static int swap_events_show(struct seq_file *m, void *v) |
8255 | { | |
aa9694bb | 8256 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 8257 | |
4b82ab4f JK |
8258 | seq_printf(m, "high %lu\n", |
8259 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
8260 | seq_printf(m, "max %lu\n", |
8261 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
8262 | seq_printf(m, "fail %lu\n", | |
8263 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
8264 | ||
8265 | return 0; | |
8266 | } | |
8267 | ||
37e84351 VD |
8268 | static struct cftype swap_files[] = { |
8269 | { | |
8270 | .name = "swap.current", | |
8271 | .flags = CFTYPE_NOT_ON_ROOT, | |
8272 | .read_u64 = swap_current_read, | |
8273 | }, | |
4b82ab4f JK |
8274 | { |
8275 | .name = "swap.high", | |
8276 | .flags = CFTYPE_NOT_ON_ROOT, | |
8277 | .seq_show = swap_high_show, | |
8278 | .write = swap_high_write, | |
8279 | }, | |
37e84351 VD |
8280 | { |
8281 | .name = "swap.max", | |
8282 | .flags = CFTYPE_NOT_ON_ROOT, | |
8283 | .seq_show = swap_max_show, | |
8284 | .write = swap_max_write, | |
8285 | }, | |
e0e0b412 LD |
8286 | { |
8287 | .name = "swap.peak", | |
8288 | .flags = CFTYPE_NOT_ON_ROOT, | |
8289 | .read_u64 = swap_peak_read, | |
8290 | }, | |
f3a53a3a TH |
8291 | { |
8292 | .name = "swap.events", | |
8293 | .flags = CFTYPE_NOT_ON_ROOT, | |
8294 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
8295 | .seq_show = swap_events_show, | |
8296 | }, | |
37e84351 VD |
8297 | { } /* terminate */ |
8298 | }; | |
8299 | ||
eccb52e7 | 8300 | static struct cftype memsw_files[] = { |
21afa38e JW |
8301 | { |
8302 | .name = "memsw.usage_in_bytes", | |
8303 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
8304 | .read_u64 = mem_cgroup_read_u64, | |
8305 | }, | |
8306 | { | |
8307 | .name = "memsw.max_usage_in_bytes", | |
8308 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
8309 | .write = mem_cgroup_reset, | |
8310 | .read_u64 = mem_cgroup_read_u64, | |
8311 | }, | |
8312 | { | |
8313 | .name = "memsw.limit_in_bytes", | |
8314 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
8315 | .write = mem_cgroup_write, | |
8316 | .read_u64 = mem_cgroup_read_u64, | |
8317 | }, | |
8318 | { | |
8319 | .name = "memsw.failcnt", | |
8320 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
8321 | .write = mem_cgroup_reset, | |
8322 | .read_u64 = mem_cgroup_read_u64, | |
8323 | }, | |
8324 | { }, /* terminate */ | |
8325 | }; | |
8326 | ||
f4840ccf JW |
8327 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
8328 | /** | |
8329 | * obj_cgroup_may_zswap - check if this cgroup can zswap | |
8330 | * @objcg: the object cgroup | |
8331 | * | |
8332 | * Check if the hierarchical zswap limit has been reached. | |
8333 | * | |
8334 | * This doesn't check for specific headroom, and it is not atomic | |
8335 | * either. But with zswap, the size of the allocation is only known | |
be16dd76 | 8336 | * once compression has occurred, and this optimistic pre-check avoids |
f4840ccf JW |
8337 | * spending cycles on compression when there is already no room left |
8338 | * or zswap is disabled altogether somewhere in the hierarchy. | |
8339 | */ | |
8340 | bool obj_cgroup_may_zswap(struct obj_cgroup *objcg) | |
8341 | { | |
8342 | struct mem_cgroup *memcg, *original_memcg; | |
8343 | bool ret = true; | |
8344 | ||
8345 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8346 | return true; | |
8347 | ||
8348 | original_memcg = get_mem_cgroup_from_objcg(objcg); | |
7848ed62 | 8349 | for (memcg = original_memcg; !mem_cgroup_is_root(memcg); |
f4840ccf JW |
8350 | memcg = parent_mem_cgroup(memcg)) { |
8351 | unsigned long max = READ_ONCE(memcg->zswap_max); | |
8352 | unsigned long pages; | |
8353 | ||
8354 | if (max == PAGE_COUNTER_MAX) | |
8355 | continue; | |
8356 | if (max == 0) { | |
8357 | ret = false; | |
8358 | break; | |
8359 | } | |
8360 | ||
7d7ef0a4 YA |
8361 | /* |
8362 | * mem_cgroup_flush_stats() ignores small changes. Use | |
8363 | * do_flush_stats() directly to get accurate stats for charging. | |
8364 | */ | |
8365 | do_flush_stats(memcg); | |
f4840ccf JW |
8366 | pages = memcg_page_state(memcg, MEMCG_ZSWAP_B) / PAGE_SIZE; |
8367 | if (pages < max) | |
8368 | continue; | |
8369 | ret = false; | |
8370 | break; | |
8371 | } | |
8372 | mem_cgroup_put(original_memcg); | |
8373 | return ret; | |
8374 | } | |
8375 | ||
8376 | /** | |
8377 | * obj_cgroup_charge_zswap - charge compression backend memory | |
8378 | * @objcg: the object cgroup | |
8379 | * @size: size of compressed object | |
8380 | * | |
3a1060c2 | 8381 | * This forces the charge after obj_cgroup_may_zswap() allowed |
f4840ccf JW |
8382 | * compression and storage in zwap for this cgroup to go ahead. |
8383 | */ | |
8384 | void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size) | |
8385 | { | |
8386 | struct mem_cgroup *memcg; | |
8387 | ||
8388 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8389 | return; | |
8390 | ||
8391 | VM_WARN_ON_ONCE(!(current->flags & PF_MEMALLOC)); | |
8392 | ||
8393 | /* PF_MEMALLOC context, charging must succeed */ | |
8394 | if (obj_cgroup_charge(objcg, GFP_KERNEL, size)) | |
8395 | VM_WARN_ON_ONCE(1); | |
8396 | ||
8397 | rcu_read_lock(); | |
8398 | memcg = obj_cgroup_memcg(objcg); | |
8399 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, size); | |
8400 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, 1); | |
8401 | rcu_read_unlock(); | |
8402 | } | |
8403 | ||
8404 | /** | |
8405 | * obj_cgroup_uncharge_zswap - uncharge compression backend memory | |
8406 | * @objcg: the object cgroup | |
8407 | * @size: size of compressed object | |
8408 | * | |
8409 | * Uncharges zswap memory on page in. | |
8410 | */ | |
8411 | void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size) | |
8412 | { | |
8413 | struct mem_cgroup *memcg; | |
8414 | ||
8415 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8416 | return; | |
8417 | ||
8418 | obj_cgroup_uncharge(objcg, size); | |
8419 | ||
8420 | rcu_read_lock(); | |
8421 | memcg = obj_cgroup_memcg(objcg); | |
8422 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, -size); | |
8423 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, -1); | |
8424 | rcu_read_unlock(); | |
8425 | } | |
8426 | ||
501a06fe NP |
8427 | bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg) |
8428 | { | |
8429 | /* if zswap is disabled, do not block pages going to the swapping device */ | |
8430 | return !is_zswap_enabled() || !memcg || READ_ONCE(memcg->zswap_writeback); | |
8431 | } | |
8432 | ||
f4840ccf JW |
8433 | static u64 zswap_current_read(struct cgroup_subsys_state *css, |
8434 | struct cftype *cft) | |
8435 | { | |
7d7ef0a4 YA |
8436 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
8437 | ||
8438 | mem_cgroup_flush_stats(memcg); | |
8439 | return memcg_page_state(memcg, MEMCG_ZSWAP_B); | |
f4840ccf JW |
8440 | } |
8441 | ||
8442 | static int zswap_max_show(struct seq_file *m, void *v) | |
8443 | { | |
8444 | return seq_puts_memcg_tunable(m, | |
8445 | READ_ONCE(mem_cgroup_from_seq(m)->zswap_max)); | |
8446 | } | |
8447 | ||
8448 | static ssize_t zswap_max_write(struct kernfs_open_file *of, | |
8449 | char *buf, size_t nbytes, loff_t off) | |
8450 | { | |
8451 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8452 | unsigned long max; | |
8453 | int err; | |
8454 | ||
8455 | buf = strstrip(buf); | |
8456 | err = page_counter_memparse(buf, "max", &max); | |
8457 | if (err) | |
8458 | return err; | |
8459 | ||
8460 | xchg(&memcg->zswap_max, max); | |
8461 | ||
8462 | return nbytes; | |
8463 | } | |
8464 | ||
501a06fe NP |
8465 | static int zswap_writeback_show(struct seq_file *m, void *v) |
8466 | { | |
8467 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
8468 | ||
8469 | seq_printf(m, "%d\n", READ_ONCE(memcg->zswap_writeback)); | |
8470 | return 0; | |
8471 | } | |
8472 | ||
8473 | static ssize_t zswap_writeback_write(struct kernfs_open_file *of, | |
8474 | char *buf, size_t nbytes, loff_t off) | |
8475 | { | |
8476 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8477 | int zswap_writeback; | |
8478 | ssize_t parse_ret = kstrtoint(strstrip(buf), 0, &zswap_writeback); | |
8479 | ||
8480 | if (parse_ret) | |
8481 | return parse_ret; | |
8482 | ||
8483 | if (zswap_writeback != 0 && zswap_writeback != 1) | |
8484 | return -EINVAL; | |
8485 | ||
8486 | WRITE_ONCE(memcg->zswap_writeback, zswap_writeback); | |
8487 | return nbytes; | |
8488 | } | |
8489 | ||
f4840ccf JW |
8490 | static struct cftype zswap_files[] = { |
8491 | { | |
8492 | .name = "zswap.current", | |
8493 | .flags = CFTYPE_NOT_ON_ROOT, | |
8494 | .read_u64 = zswap_current_read, | |
8495 | }, | |
8496 | { | |
8497 | .name = "zswap.max", | |
8498 | .flags = CFTYPE_NOT_ON_ROOT, | |
8499 | .seq_show = zswap_max_show, | |
8500 | .write = zswap_max_write, | |
8501 | }, | |
501a06fe NP |
8502 | { |
8503 | .name = "zswap.writeback", | |
8504 | .seq_show = zswap_writeback_show, | |
8505 | .write = zswap_writeback_write, | |
8506 | }, | |
f4840ccf JW |
8507 | { } /* terminate */ |
8508 | }; | |
8509 | #endif /* CONFIG_MEMCG_KMEM && CONFIG_ZSWAP */ | |
8510 | ||
21afa38e JW |
8511 | static int __init mem_cgroup_swap_init(void) |
8512 | { | |
2d1c4980 | 8513 | if (mem_cgroup_disabled()) |
eccb52e7 JW |
8514 | return 0; |
8515 | ||
8516 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
8517 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
f4840ccf JW |
8518 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
8519 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, zswap_files)); | |
8520 | #endif | |
21afa38e JW |
8521 | return 0; |
8522 | } | |
b25806dc | 8523 | subsys_initcall(mem_cgroup_swap_init); |
21afa38e | 8524 | |
e55b9f96 | 8525 | #endif /* CONFIG_SWAP */ |