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