]>
Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8cdea7c0 BS |
2 | /* memcontrol.c - Memory Controller |
3 | * | |
4 | * Copyright IBM Corporation, 2007 | |
5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
6 | * | |
78fb7466 PE |
7 | * Copyright 2007 OpenVZ SWsoft Inc |
8 | * Author: Pavel Emelianov <xemul@openvz.org> | |
9 | * | |
2e72b634 KS |
10 | * Memory thresholds |
11 | * Copyright (C) 2009 Nokia Corporation | |
12 | * Author: Kirill A. Shutemov | |
13 | * | |
7ae1e1d0 GC |
14 | * Kernel Memory Controller |
15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
16 | * Authors: Glauber Costa and Suleiman Souhlal | |
17 | * | |
1575e68b JW |
18 | * Native page reclaim |
19 | * Charge lifetime sanitation | |
20 | * Lockless page tracking & accounting | |
21 | * Unified hierarchy configuration model | |
22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
6168d0da AS |
23 | * |
24 | * Per memcg lru locking | |
25 | * Copyright (C) 2020 Alibaba, Inc, Alex Shi | |
8cdea7c0 BS |
26 | */ |
27 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
29 | #include <linux/memcontrol.h> |
30 | #include <linux/cgroup.h> | |
a520110e | 31 | #include <linux/pagewalk.h> |
6e84f315 | 32 | #include <linux/sched/mm.h> |
3a4f8a0b | 33 | #include <linux/shmem_fs.h> |
4ffef5fe | 34 | #include <linux/hugetlb.h> |
d13d1443 | 35 | #include <linux/pagemap.h> |
1ff9e6e1 | 36 | #include <linux/vm_event_item.h> |
d52aa412 | 37 | #include <linux/smp.h> |
8a9f3ccd | 38 | #include <linux/page-flags.h> |
66e1707b | 39 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
40 | #include <linux/bit_spinlock.h> |
41 | #include <linux/rcupdate.h> | |
e222432b | 42 | #include <linux/limits.h> |
b9e15baf | 43 | #include <linux/export.h> |
8c7c6e34 | 44 | #include <linux/mutex.h> |
bb4cc1a8 | 45 | #include <linux/rbtree.h> |
b6ac57d5 | 46 | #include <linux/slab.h> |
66e1707b | 47 | #include <linux/swap.h> |
02491447 | 48 | #include <linux/swapops.h> |
66e1707b | 49 | #include <linux/spinlock.h> |
2e72b634 | 50 | #include <linux/eventfd.h> |
79bd9814 | 51 | #include <linux/poll.h> |
2e72b634 | 52 | #include <linux/sort.h> |
66e1707b | 53 | #include <linux/fs.h> |
d2ceb9b7 | 54 | #include <linux/seq_file.h> |
70ddf637 | 55 | #include <linux/vmpressure.h> |
dc90f084 | 56 | #include <linux/memremap.h> |
b69408e8 | 57 | #include <linux/mm_inline.h> |
5d1ea48b | 58 | #include <linux/swap_cgroup.h> |
cdec2e42 | 59 | #include <linux/cpu.h> |
158e0a2d | 60 | #include <linux/oom.h> |
0056f4e6 | 61 | #include <linux/lockdep.h> |
79bd9814 | 62 | #include <linux/file.h> |
03248add | 63 | #include <linux/resume_user_mode.h> |
0e4b01df | 64 | #include <linux/psi.h> |
c8713d0b | 65 | #include <linux/seq_buf.h> |
08e552c6 | 66 | #include "internal.h" |
d1a4c0b3 | 67 | #include <net/sock.h> |
4bd2c1ee | 68 | #include <net/ip.h> |
f35c3a8e | 69 | #include "slab.h" |
014bb1de | 70 | #include "swap.h" |
8cdea7c0 | 71 | |
7c0f6ba6 | 72 | #include <linux/uaccess.h> |
8697d331 | 73 | |
cc8e970c KM |
74 | #include <trace/events/vmscan.h> |
75 | ||
073219e9 TH |
76 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
77 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 78 | |
7d828602 JW |
79 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
80 | ||
37d5985c RG |
81 | /* Active memory cgroup to use from an interrupt context */ |
82 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); | |
c74d40e8 | 83 | EXPORT_PER_CPU_SYMBOL_GPL(int_active_memcg); |
37d5985c | 84 | |
f7e1cb6e | 85 | /* Socket memory accounting disabled? */ |
0f0cace3 | 86 | static bool cgroup_memory_nosocket __ro_after_init; |
f7e1cb6e | 87 | |
04823c83 | 88 | /* Kernel memory accounting disabled? */ |
17c17367 | 89 | static bool cgroup_memory_nokmem __ro_after_init; |
04823c83 | 90 | |
21afa38e | 91 | /* Whether the swap controller is active */ |
c255a458 | 92 | #ifdef CONFIG_MEMCG_SWAP |
ef7a4ffc | 93 | static bool cgroup_memory_noswap __ro_after_init; |
c077719b | 94 | #else |
eccb52e7 | 95 | #define cgroup_memory_noswap 1 |
2d1c4980 | 96 | #endif |
c077719b | 97 | |
97b27821 TH |
98 | #ifdef CONFIG_CGROUP_WRITEBACK |
99 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
100 | #endif | |
101 | ||
7941d214 JW |
102 | /* Whether legacy memory+swap accounting is active */ |
103 | static bool do_memsw_account(void) | |
104 | { | |
eccb52e7 | 105 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap; |
7941d214 JW |
106 | } |
107 | ||
a0db00fc KS |
108 | #define THRESHOLDS_EVENTS_TARGET 128 |
109 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 110 | |
bb4cc1a8 AM |
111 | /* |
112 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
113 | * their hierarchy representation | |
114 | */ | |
115 | ||
ef8f2327 | 116 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 117 | struct rb_root rb_root; |
fa90b2fd | 118 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
119 | spinlock_t lock; |
120 | }; | |
121 | ||
bb4cc1a8 AM |
122 | struct mem_cgroup_tree { |
123 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
124 | }; | |
125 | ||
126 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
127 | ||
9490ff27 KH |
128 | /* for OOM */ |
129 | struct mem_cgroup_eventfd_list { | |
130 | struct list_head list; | |
131 | struct eventfd_ctx *eventfd; | |
132 | }; | |
2e72b634 | 133 | |
79bd9814 TH |
134 | /* |
135 | * cgroup_event represents events which userspace want to receive. | |
136 | */ | |
3bc942f3 | 137 | struct mem_cgroup_event { |
79bd9814 | 138 | /* |
59b6f873 | 139 | * memcg which the event belongs to. |
79bd9814 | 140 | */ |
59b6f873 | 141 | struct mem_cgroup *memcg; |
79bd9814 TH |
142 | /* |
143 | * eventfd to signal userspace about the event. | |
144 | */ | |
145 | struct eventfd_ctx *eventfd; | |
146 | /* | |
147 | * Each of these stored in a list by the cgroup. | |
148 | */ | |
149 | struct list_head list; | |
fba94807 TH |
150 | /* |
151 | * register_event() callback will be used to add new userspace | |
152 | * waiter for changes related to this event. Use eventfd_signal() | |
153 | * on eventfd to send notification to userspace. | |
154 | */ | |
59b6f873 | 155 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 156 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
157 | /* |
158 | * unregister_event() callback will be called when userspace closes | |
159 | * the eventfd or on cgroup removing. This callback must be set, | |
160 | * if you want provide notification functionality. | |
161 | */ | |
59b6f873 | 162 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 163 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
164 | /* |
165 | * All fields below needed to unregister event when | |
166 | * userspace closes eventfd. | |
167 | */ | |
168 | poll_table pt; | |
169 | wait_queue_head_t *wqh; | |
ac6424b9 | 170 | wait_queue_entry_t wait; |
79bd9814 TH |
171 | struct work_struct remove; |
172 | }; | |
173 | ||
c0ff4b85 R |
174 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
175 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 176 | |
7dc74be0 DN |
177 | /* Stuffs for move charges at task migration. */ |
178 | /* | |
1dfab5ab | 179 | * Types of charges to be moved. |
7dc74be0 | 180 | */ |
1dfab5ab JW |
181 | #define MOVE_ANON 0x1U |
182 | #define MOVE_FILE 0x2U | |
183 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 184 | |
4ffef5fe DN |
185 | /* "mc" and its members are protected by cgroup_mutex */ |
186 | static struct move_charge_struct { | |
b1dd693e | 187 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 188 | struct mm_struct *mm; |
4ffef5fe DN |
189 | struct mem_cgroup *from; |
190 | struct mem_cgroup *to; | |
1dfab5ab | 191 | unsigned long flags; |
4ffef5fe | 192 | unsigned long precharge; |
854ffa8d | 193 | unsigned long moved_charge; |
483c30b5 | 194 | unsigned long moved_swap; |
8033b97c DN |
195 | struct task_struct *moving_task; /* a task moving charges */ |
196 | wait_queue_head_t waitq; /* a waitq for other context */ | |
197 | } mc = { | |
2bd9bb20 | 198 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
199 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
200 | }; | |
4ffef5fe | 201 | |
4e416953 BS |
202 | /* |
203 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
204 | * limit reclaim to prevent infinite loops, if they ever occur. | |
205 | */ | |
a0db00fc | 206 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 207 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 208 | |
8c7c6e34 | 209 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
210 | enum res_type { |
211 | _MEM, | |
212 | _MEMSWAP, | |
510fc4e1 | 213 | _KMEM, |
d55f90bf | 214 | _TCP, |
86ae53e1 GC |
215 | }; |
216 | ||
a0db00fc KS |
217 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
218 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 KH |
219 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
220 | ||
b05706f1 KT |
221 | /* |
222 | * Iteration constructs for visiting all cgroups (under a tree). If | |
223 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
224 | * be used for reference counting. | |
225 | */ | |
226 | #define for_each_mem_cgroup_tree(iter, root) \ | |
227 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
228 | iter != NULL; \ | |
229 | iter = mem_cgroup_iter(root, iter, NULL)) | |
230 | ||
231 | #define for_each_mem_cgroup(iter) \ | |
232 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
233 | iter != NULL; \ | |
234 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
235 | ||
a4ebf1b6 | 236 | static inline bool task_is_dying(void) |
7775face TH |
237 | { |
238 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
239 | (current->flags & PF_EXITING); | |
240 | } | |
241 | ||
70ddf637 AV |
242 | /* Some nice accessors for the vmpressure. */ |
243 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
244 | { | |
245 | if (!memcg) | |
246 | memcg = root_mem_cgroup; | |
247 | return &memcg->vmpressure; | |
248 | } | |
249 | ||
9647875b | 250 | struct mem_cgroup *vmpressure_to_memcg(struct vmpressure *vmpr) |
70ddf637 | 251 | { |
9647875b | 252 | return container_of(vmpr, struct mem_cgroup, vmpressure); |
70ddf637 AV |
253 | } |
254 | ||
84c07d11 | 255 | #ifdef CONFIG_MEMCG_KMEM |
0764db9b | 256 | static DEFINE_SPINLOCK(objcg_lock); |
bf4f0599 | 257 | |
4d5c8aed RG |
258 | bool mem_cgroup_kmem_disabled(void) |
259 | { | |
260 | return cgroup_memory_nokmem; | |
261 | } | |
262 | ||
f1286fae MS |
263 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
264 | unsigned int nr_pages); | |
c1a660de | 265 | |
bf4f0599 RG |
266 | static void obj_cgroup_release(struct percpu_ref *ref) |
267 | { | |
268 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
bf4f0599 RG |
269 | unsigned int nr_bytes; |
270 | unsigned int nr_pages; | |
271 | unsigned long flags; | |
272 | ||
273 | /* | |
274 | * At this point all allocated objects are freed, and | |
275 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
276 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
277 | * | |
278 | * The following sequence can lead to it: | |
279 | * 1) CPU0: objcg == stock->cached_objcg | |
280 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
281 | * PAGE_SIZE bytes are charged | |
282 | * 3) CPU1: a process from another memcg is allocating something, | |
283 | * the stock if flushed, | |
284 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
285 | * 5) CPU0: we do release this object, | |
286 | * 92 bytes are added to stock->nr_bytes | |
287 | * 6) CPU0: stock is flushed, | |
288 | * 92 bytes are added to objcg->nr_charged_bytes | |
289 | * | |
290 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
291 | * This page will be uncharged in obj_cgroup_release(). | |
292 | */ | |
293 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
294 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
295 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
296 | ||
bf4f0599 | 297 | if (nr_pages) |
f1286fae | 298 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
271dd6b1 | 299 | |
0764db9b | 300 | spin_lock_irqsave(&objcg_lock, flags); |
bf4f0599 | 301 | list_del(&objcg->list); |
0764db9b | 302 | spin_unlock_irqrestore(&objcg_lock, flags); |
bf4f0599 RG |
303 | |
304 | percpu_ref_exit(ref); | |
305 | kfree_rcu(objcg, rcu); | |
306 | } | |
307 | ||
308 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
309 | { | |
310 | struct obj_cgroup *objcg; | |
311 | int ret; | |
312 | ||
313 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
314 | if (!objcg) | |
315 | return NULL; | |
316 | ||
317 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
318 | GFP_KERNEL); | |
319 | if (ret) { | |
320 | kfree(objcg); | |
321 | return NULL; | |
322 | } | |
323 | INIT_LIST_HEAD(&objcg->list); | |
324 | return objcg; | |
325 | } | |
326 | ||
327 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
328 | struct mem_cgroup *parent) | |
329 | { | |
330 | struct obj_cgroup *objcg, *iter; | |
331 | ||
332 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
333 | ||
0764db9b | 334 | spin_lock_irq(&objcg_lock); |
bf4f0599 | 335 | |
9838354e MS |
336 | /* 1) Ready to reparent active objcg. */ |
337 | list_add(&objcg->list, &memcg->objcg_list); | |
338 | /* 2) Reparent active objcg and already reparented objcgs to parent. */ | |
339 | list_for_each_entry(iter, &memcg->objcg_list, list) | |
340 | WRITE_ONCE(iter->memcg, parent); | |
341 | /* 3) Move already reparented objcgs to the parent's list */ | |
bf4f0599 RG |
342 | list_splice(&memcg->objcg_list, &parent->objcg_list); |
343 | ||
0764db9b | 344 | spin_unlock_irq(&objcg_lock); |
bf4f0599 RG |
345 | |
346 | percpu_ref_kill(&objcg->refcnt); | |
347 | } | |
348 | ||
d7f25f8a GC |
349 | /* |
350 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 351 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
352 | * conditional to this static branch, we'll have to allow modules that does |
353 | * kmem_cache_alloc and the such to see this symbol as well | |
354 | */ | |
ef12947c | 355 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 356 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 357 | #endif |
17cc4dfe | 358 | |
ad7fa852 TH |
359 | /** |
360 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
361 | * @page: page of interest | |
362 | * | |
363 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
364 | * with @page is returned. The returned css remains associated with @page | |
365 | * until it is released. | |
366 | * | |
367 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
368 | * is returned. | |
ad7fa852 TH |
369 | */ |
370 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
371 | { | |
372 | struct mem_cgroup *memcg; | |
373 | ||
bcfe06bf | 374 | memcg = page_memcg(page); |
ad7fa852 | 375 | |
9e10a130 | 376 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
377 | memcg = root_mem_cgroup; |
378 | ||
ad7fa852 TH |
379 | return &memcg->css; |
380 | } | |
381 | ||
2fc04524 VD |
382 | /** |
383 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
384 | * @page: the page | |
385 | * | |
386 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
387 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
388 | * is safe to call this function without holding a reference to @page. | |
389 | * | |
390 | * Note, this function is inherently racy, because there is nothing to prevent | |
391 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
392 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
393 | * do not care (such as procfs interfaces). | |
394 | */ | |
395 | ino_t page_cgroup_ino(struct page *page) | |
396 | { | |
397 | struct mem_cgroup *memcg; | |
398 | unsigned long ino = 0; | |
399 | ||
400 | rcu_read_lock(); | |
bcfe06bf | 401 | memcg = page_memcg_check(page); |
286e04b8 | 402 | |
2fc04524 VD |
403 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
404 | memcg = parent_mem_cgroup(memcg); | |
405 | if (memcg) | |
406 | ino = cgroup_ino(memcg->css.cgroup); | |
407 | rcu_read_unlock(); | |
408 | return ino; | |
409 | } | |
410 | ||
ef8f2327 MG |
411 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
412 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 413 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
414 | { |
415 | struct rb_node **p = &mctz->rb_root.rb_node; | |
416 | struct rb_node *parent = NULL; | |
ef8f2327 | 417 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 418 | bool rightmost = true; |
bb4cc1a8 AM |
419 | |
420 | if (mz->on_tree) | |
421 | return; | |
422 | ||
423 | mz->usage_in_excess = new_usage_in_excess; | |
424 | if (!mz->usage_in_excess) | |
425 | return; | |
426 | while (*p) { | |
427 | parent = *p; | |
ef8f2327 | 428 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 429 | tree_node); |
fa90b2fd | 430 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 431 | p = &(*p)->rb_left; |
fa90b2fd | 432 | rightmost = false; |
378876b0 | 433 | } else { |
bb4cc1a8 | 434 | p = &(*p)->rb_right; |
378876b0 | 435 | } |
bb4cc1a8 | 436 | } |
fa90b2fd DB |
437 | |
438 | if (rightmost) | |
439 | mctz->rb_rightmost = &mz->tree_node; | |
440 | ||
bb4cc1a8 AM |
441 | rb_link_node(&mz->tree_node, parent, p); |
442 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
443 | mz->on_tree = true; | |
444 | } | |
445 | ||
ef8f2327 MG |
446 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
447 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
448 | { |
449 | if (!mz->on_tree) | |
450 | return; | |
fa90b2fd DB |
451 | |
452 | if (&mz->tree_node == mctz->rb_rightmost) | |
453 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
454 | ||
bb4cc1a8 AM |
455 | rb_erase(&mz->tree_node, &mctz->rb_root); |
456 | mz->on_tree = false; | |
457 | } | |
458 | ||
ef8f2327 MG |
459 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
460 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 461 | { |
0a31bc97 JW |
462 | unsigned long flags; |
463 | ||
464 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 465 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 466 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
467 | } |
468 | ||
3e32cb2e JW |
469 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
470 | { | |
471 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 472 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
473 | unsigned long excess = 0; |
474 | ||
475 | if (nr_pages > soft_limit) | |
476 | excess = nr_pages - soft_limit; | |
477 | ||
478 | return excess; | |
479 | } | |
bb4cc1a8 | 480 | |
658b69c9 | 481 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, int nid) |
bb4cc1a8 | 482 | { |
3e32cb2e | 483 | unsigned long excess; |
ef8f2327 MG |
484 | struct mem_cgroup_per_node *mz; |
485 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 486 | |
2ab082ba | 487 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
488 | if (!mctz) |
489 | return; | |
bb4cc1a8 AM |
490 | /* |
491 | * Necessary to update all ancestors when hierarchy is used. | |
492 | * because their event counter is not touched. | |
493 | */ | |
494 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
658b69c9 | 495 | mz = memcg->nodeinfo[nid]; |
3e32cb2e | 496 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
497 | /* |
498 | * We have to update the tree if mz is on RB-tree or | |
499 | * mem is over its softlimit. | |
500 | */ | |
501 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
502 | unsigned long flags; |
503 | ||
504 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
505 | /* if on-tree, remove it */ |
506 | if (mz->on_tree) | |
cf2c8127 | 507 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
508 | /* |
509 | * Insert again. mz->usage_in_excess will be updated. | |
510 | * If excess is 0, no tree ops. | |
511 | */ | |
cf2c8127 | 512 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 513 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
514 | } |
515 | } | |
516 | } | |
517 | ||
518 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
519 | { | |
ef8f2327 MG |
520 | struct mem_cgroup_tree_per_node *mctz; |
521 | struct mem_cgroup_per_node *mz; | |
522 | int nid; | |
bb4cc1a8 | 523 | |
e231875b | 524 | for_each_node(nid) { |
a3747b53 | 525 | mz = memcg->nodeinfo[nid]; |
2ab082ba | 526 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
527 | if (mctz) |
528 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
529 | } |
530 | } | |
531 | ||
ef8f2327 MG |
532 | static struct mem_cgroup_per_node * |
533 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 534 | { |
ef8f2327 | 535 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
536 | |
537 | retry: | |
538 | mz = NULL; | |
fa90b2fd | 539 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
540 | goto done; /* Nothing to reclaim from */ |
541 | ||
fa90b2fd DB |
542 | mz = rb_entry(mctz->rb_rightmost, |
543 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
544 | /* |
545 | * Remove the node now but someone else can add it back, | |
546 | * we will to add it back at the end of reclaim to its correct | |
547 | * position in the tree. | |
548 | */ | |
cf2c8127 | 549 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 550 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 551 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
552 | goto retry; |
553 | done: | |
554 | return mz; | |
555 | } | |
556 | ||
ef8f2327 MG |
557 | static struct mem_cgroup_per_node * |
558 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 559 | { |
ef8f2327 | 560 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 561 | |
0a31bc97 | 562 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 563 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 564 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
565 | return mz; |
566 | } | |
567 | ||
11192d9c SB |
568 | /* |
569 | * memcg and lruvec stats flushing | |
570 | * | |
571 | * Many codepaths leading to stats update or read are performance sensitive and | |
572 | * adding stats flushing in such codepaths is not desirable. So, to optimize the | |
573 | * flushing the kernel does: | |
574 | * | |
575 | * 1) Periodically and asynchronously flush the stats every 2 seconds to not let | |
576 | * rstat update tree grow unbounded. | |
577 | * | |
578 | * 2) Flush the stats synchronously on reader side only when there are more than | |
579 | * (MEMCG_CHARGE_BATCH * nr_cpus) update events. Though this optimization | |
580 | * will let stats be out of sync by atmost (MEMCG_CHARGE_BATCH * nr_cpus) but | |
581 | * only for 2 seconds due to (1). | |
582 | */ | |
583 | static void flush_memcg_stats_dwork(struct work_struct *w); | |
584 | static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork); | |
585 | static DEFINE_SPINLOCK(stats_flush_lock); | |
586 | static DEFINE_PER_CPU(unsigned int, stats_updates); | |
587 | static atomic_t stats_flush_threshold = ATOMIC_INIT(0); | |
9b301615 SB |
588 | static u64 flush_next_time; |
589 | ||
590 | #define FLUSH_TIME (2UL*HZ) | |
11192d9c | 591 | |
be3e67b5 SAS |
592 | /* |
593 | * Accessors to ensure that preemption is disabled on PREEMPT_RT because it can | |
594 | * not rely on this as part of an acquired spinlock_t lock. These functions are | |
595 | * never used in hardirq context on PREEMPT_RT and therefore disabling preemtion | |
596 | * is sufficient. | |
597 | */ | |
598 | static void memcg_stats_lock(void) | |
599 | { | |
600 | #ifdef CONFIG_PREEMPT_RT | |
601 | preempt_disable(); | |
602 | #else | |
603 | VM_BUG_ON(!irqs_disabled()); | |
604 | #endif | |
605 | } | |
606 | ||
607 | static void __memcg_stats_lock(void) | |
608 | { | |
609 | #ifdef CONFIG_PREEMPT_RT | |
610 | preempt_disable(); | |
611 | #endif | |
612 | } | |
613 | ||
614 | static void memcg_stats_unlock(void) | |
615 | { | |
616 | #ifdef CONFIG_PREEMPT_RT | |
617 | preempt_enable(); | |
618 | #endif | |
619 | } | |
620 | ||
5b3be698 | 621 | static inline void memcg_rstat_updated(struct mem_cgroup *memcg, int val) |
11192d9c | 622 | { |
5b3be698 SB |
623 | unsigned int x; |
624 | ||
11192d9c | 625 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); |
5b3be698 SB |
626 | |
627 | x = __this_cpu_add_return(stats_updates, abs(val)); | |
628 | if (x > MEMCG_CHARGE_BATCH) { | |
873f64b7 JS |
629 | /* |
630 | * If stats_flush_threshold exceeds the threshold | |
631 | * (>num_online_cpus()), cgroup stats update will be triggered | |
632 | * in __mem_cgroup_flush_stats(). Increasing this var further | |
633 | * is redundant and simply adds overhead in atomic update. | |
634 | */ | |
635 | if (atomic_read(&stats_flush_threshold) <= num_online_cpus()) | |
636 | atomic_add(x / MEMCG_CHARGE_BATCH, &stats_flush_threshold); | |
5b3be698 SB |
637 | __this_cpu_write(stats_updates, 0); |
638 | } | |
11192d9c SB |
639 | } |
640 | ||
641 | static void __mem_cgroup_flush_stats(void) | |
642 | { | |
fd25a9e0 SB |
643 | unsigned long flag; |
644 | ||
645 | if (!spin_trylock_irqsave(&stats_flush_lock, flag)) | |
11192d9c SB |
646 | return; |
647 | ||
9b301615 | 648 | flush_next_time = jiffies_64 + 2*FLUSH_TIME; |
11192d9c SB |
649 | cgroup_rstat_flush_irqsafe(root_mem_cgroup->css.cgroup); |
650 | atomic_set(&stats_flush_threshold, 0); | |
fd25a9e0 | 651 | spin_unlock_irqrestore(&stats_flush_lock, flag); |
11192d9c SB |
652 | } |
653 | ||
654 | void mem_cgroup_flush_stats(void) | |
655 | { | |
656 | if (atomic_read(&stats_flush_threshold) > num_online_cpus()) | |
657 | __mem_cgroup_flush_stats(); | |
658 | } | |
659 | ||
9b301615 SB |
660 | void mem_cgroup_flush_stats_delayed(void) |
661 | { | |
662 | if (time_after64(jiffies_64, flush_next_time)) | |
663 | mem_cgroup_flush_stats(); | |
664 | } | |
665 | ||
11192d9c SB |
666 | static void flush_memcg_stats_dwork(struct work_struct *w) |
667 | { | |
5b3be698 | 668 | __mem_cgroup_flush_stats(); |
9b301615 | 669 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, FLUSH_TIME); |
11192d9c SB |
670 | } |
671 | ||
db9adbcb JW |
672 | /** |
673 | * __mod_memcg_state - update cgroup memory statistics | |
674 | * @memcg: the memory cgroup | |
675 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
676 | * @val: delta to add to the counter, can be negative | |
677 | */ | |
678 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
679 | { | |
db9adbcb JW |
680 | if (mem_cgroup_disabled()) |
681 | return; | |
682 | ||
2d146aa3 | 683 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
5b3be698 | 684 | memcg_rstat_updated(memcg, val); |
db9adbcb JW |
685 | } |
686 | ||
2d146aa3 | 687 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
a18e6e6e JW |
688 | static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx) |
689 | { | |
690 | long x = 0; | |
691 | int cpu; | |
692 | ||
693 | for_each_possible_cpu(cpu) | |
2d146aa3 | 694 | x += per_cpu(memcg->vmstats_percpu->state[idx], cpu); |
a18e6e6e JW |
695 | #ifdef CONFIG_SMP |
696 | if (x < 0) | |
697 | x = 0; | |
698 | #endif | |
699 | return x; | |
700 | } | |
701 | ||
eedc4e5a RG |
702 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
703 | int val) | |
db9adbcb JW |
704 | { |
705 | struct mem_cgroup_per_node *pn; | |
42a30035 | 706 | struct mem_cgroup *memcg; |
db9adbcb | 707 | |
db9adbcb | 708 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 709 | memcg = pn->memcg; |
db9adbcb | 710 | |
be3e67b5 SAS |
711 | /* |
712 | * The caller from rmap relay on disabled preemption becase they never | |
713 | * update their counter from in-interrupt context. For these two | |
714 | * counters we check that the update is never performed from an | |
715 | * interrupt context while other caller need to have disabled interrupt. | |
716 | */ | |
717 | __memcg_stats_lock(); | |
718 | if (IS_ENABLED(CONFIG_DEBUG_VM) && !IS_ENABLED(CONFIG_PREEMPT_RT)) { | |
719 | switch (idx) { | |
720 | case NR_ANON_MAPPED: | |
721 | case NR_FILE_MAPPED: | |
722 | case NR_ANON_THPS: | |
723 | case NR_SHMEM_PMDMAPPED: | |
724 | case NR_FILE_PMDMAPPED: | |
725 | WARN_ON_ONCE(!in_task()); | |
726 | break; | |
727 | default: | |
728 | WARN_ON_ONCE(!irqs_disabled()); | |
729 | } | |
730 | } | |
731 | ||
db9adbcb | 732 | /* Update memcg */ |
11192d9c | 733 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
db9adbcb | 734 | |
b4c46484 | 735 | /* Update lruvec */ |
7e1c0d6f | 736 | __this_cpu_add(pn->lruvec_stats_percpu->state[idx], val); |
11192d9c | 737 | |
5b3be698 | 738 | memcg_rstat_updated(memcg, val); |
be3e67b5 | 739 | memcg_stats_unlock(); |
db9adbcb JW |
740 | } |
741 | ||
eedc4e5a RG |
742 | /** |
743 | * __mod_lruvec_state - update lruvec memory statistics | |
744 | * @lruvec: the lruvec | |
745 | * @idx: the stat item | |
746 | * @val: delta to add to the counter, can be negative | |
747 | * | |
748 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
749 | * function updates the all three counters that are affected by a | |
750 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
751 | */ | |
752 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
753 | int val) | |
754 | { | |
755 | /* Update node */ | |
756 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
757 | ||
758 | /* Update memcg and lruvec */ | |
759 | if (!mem_cgroup_disabled()) | |
760 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
761 | } | |
762 | ||
c47d5032 SB |
763 | void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx, |
764 | int val) | |
765 | { | |
766 | struct page *head = compound_head(page); /* rmap on tail pages */ | |
b4e0b68f | 767 | struct mem_cgroup *memcg; |
c47d5032 SB |
768 | pg_data_t *pgdat = page_pgdat(page); |
769 | struct lruvec *lruvec; | |
770 | ||
b4e0b68f MS |
771 | rcu_read_lock(); |
772 | memcg = page_memcg(head); | |
c47d5032 | 773 | /* Untracked pages have no memcg, no lruvec. Update only the node */ |
d635a69d | 774 | if (!memcg) { |
b4e0b68f | 775 | rcu_read_unlock(); |
c47d5032 SB |
776 | __mod_node_page_state(pgdat, idx, val); |
777 | return; | |
778 | } | |
779 | ||
d635a69d | 780 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 | 781 | __mod_lruvec_state(lruvec, idx, val); |
b4e0b68f | 782 | rcu_read_unlock(); |
c47d5032 | 783 | } |
f0c0c115 | 784 | EXPORT_SYMBOL(__mod_lruvec_page_state); |
c47d5032 | 785 | |
da3ceeff | 786 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 787 | { |
4f103c63 | 788 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
789 | struct mem_cgroup *memcg; |
790 | struct lruvec *lruvec; | |
791 | ||
792 | rcu_read_lock(); | |
fc4db90f | 793 | memcg = mem_cgroup_from_slab_obj(p); |
ec9f0238 | 794 | |
8faeb1ff MS |
795 | /* |
796 | * Untracked pages have no memcg, no lruvec. Update only the | |
797 | * node. If we reparent the slab objects to the root memcg, | |
798 | * when we free the slab object, we need to update the per-memcg | |
799 | * vmstats to keep it correct for the root memcg. | |
800 | */ | |
801 | if (!memcg) { | |
ec9f0238 RG |
802 | __mod_node_page_state(pgdat, idx, val); |
803 | } else { | |
867e5e1d | 804 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
805 | __mod_lruvec_state(lruvec, idx, val); |
806 | } | |
807 | rcu_read_unlock(); | |
808 | } | |
809 | ||
db9adbcb JW |
810 | /** |
811 | * __count_memcg_events - account VM events in a cgroup | |
812 | * @memcg: the memory cgroup | |
813 | * @idx: the event item | |
f0953a1b | 814 | * @count: the number of events that occurred |
db9adbcb JW |
815 | */ |
816 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
817 | unsigned long count) | |
818 | { | |
db9adbcb JW |
819 | if (mem_cgroup_disabled()) |
820 | return; | |
821 | ||
be3e67b5 | 822 | memcg_stats_lock(); |
2d146aa3 | 823 | __this_cpu_add(memcg->vmstats_percpu->events[idx], count); |
5b3be698 | 824 | memcg_rstat_updated(memcg, count); |
be3e67b5 | 825 | memcg_stats_unlock(); |
db9adbcb JW |
826 | } |
827 | ||
42a30035 | 828 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 829 | { |
2d146aa3 | 830 | return READ_ONCE(memcg->vmstats.events[event]); |
e9f8974f JW |
831 | } |
832 | ||
42a30035 JW |
833 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
834 | { | |
815744d7 JW |
835 | long x = 0; |
836 | int cpu; | |
837 | ||
838 | for_each_possible_cpu(cpu) | |
2d146aa3 | 839 | x += per_cpu(memcg->vmstats_percpu->events[event], cpu); |
815744d7 | 840 | return x; |
42a30035 JW |
841 | } |
842 | ||
c0ff4b85 | 843 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
3fba69a5 | 844 | int nr_pages) |
d52aa412 | 845 | { |
e401f176 KH |
846 | /* pagein of a big page is an event. So, ignore page size */ |
847 | if (nr_pages > 0) | |
c9019e9b | 848 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 849 | else { |
c9019e9b | 850 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
851 | nr_pages = -nr_pages; /* for event */ |
852 | } | |
e401f176 | 853 | |
871789d4 | 854 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
855 | } |
856 | ||
f53d7ce3 JW |
857 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
858 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
859 | { |
860 | unsigned long val, next; | |
861 | ||
871789d4 CD |
862 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
863 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 864 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 865 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
866 | switch (target) { |
867 | case MEM_CGROUP_TARGET_THRESH: | |
868 | next = val + THRESHOLDS_EVENTS_TARGET; | |
869 | break; | |
bb4cc1a8 AM |
870 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
871 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
872 | break; | |
f53d7ce3 JW |
873 | default: |
874 | break; | |
875 | } | |
871789d4 | 876 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 877 | return true; |
7a159cc9 | 878 | } |
f53d7ce3 | 879 | return false; |
d2265e6f KH |
880 | } |
881 | ||
882 | /* | |
883 | * Check events in order. | |
884 | * | |
885 | */ | |
8e88bd2d | 886 | static void memcg_check_events(struct mem_cgroup *memcg, int nid) |
d2265e6f | 887 | { |
2343e88d SAS |
888 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
889 | return; | |
890 | ||
d2265e6f | 891 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
892 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
893 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 894 | bool do_softlimit; |
f53d7ce3 | 895 | |
bb4cc1a8 AM |
896 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
897 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 898 | mem_cgroup_threshold(memcg); |
bb4cc1a8 | 899 | if (unlikely(do_softlimit)) |
8e88bd2d | 900 | mem_cgroup_update_tree(memcg, nid); |
0a31bc97 | 901 | } |
d2265e6f KH |
902 | } |
903 | ||
cf475ad2 | 904 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 905 | { |
31a78f23 BS |
906 | /* |
907 | * mm_update_next_owner() may clear mm->owner to NULL | |
908 | * if it races with swapoff, page migration, etc. | |
909 | * So this can be called with p == NULL. | |
910 | */ | |
911 | if (unlikely(!p)) | |
912 | return NULL; | |
913 | ||
073219e9 | 914 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 915 | } |
33398cf2 | 916 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 917 | |
04f94e3f DS |
918 | static __always_inline struct mem_cgroup *active_memcg(void) |
919 | { | |
55a68c82 | 920 | if (!in_task()) |
04f94e3f DS |
921 | return this_cpu_read(int_active_memcg); |
922 | else | |
923 | return current->active_memcg; | |
924 | } | |
925 | ||
d46eb14b SB |
926 | /** |
927 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
928 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
929 | * | |
04f94e3f DS |
930 | * Obtain a reference on mm->memcg and returns it if successful. If mm |
931 | * is NULL, then the memcg is chosen as follows: | |
932 | * 1) The active memcg, if set. | |
933 | * 2) current->mm->memcg, if available | |
934 | * 3) root memcg | |
935 | * If mem_cgroup is disabled, NULL is returned. | |
d46eb14b SB |
936 | */ |
937 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 938 | { |
d46eb14b SB |
939 | struct mem_cgroup *memcg; |
940 | ||
941 | if (mem_cgroup_disabled()) | |
942 | return NULL; | |
0b7f569e | 943 | |
2884b6b7 MS |
944 | /* |
945 | * Page cache insertions can happen without an | |
946 | * actual mm context, e.g. during disk probing | |
947 | * on boot, loopback IO, acct() writes etc. | |
948 | * | |
949 | * No need to css_get on root memcg as the reference | |
950 | * counting is disabled on the root level in the | |
951 | * cgroup core. See CSS_NO_REF. | |
952 | */ | |
04f94e3f DS |
953 | if (unlikely(!mm)) { |
954 | memcg = active_memcg(); | |
955 | if (unlikely(memcg)) { | |
956 | /* remote memcg must hold a ref */ | |
957 | css_get(&memcg->css); | |
958 | return memcg; | |
959 | } | |
960 | mm = current->mm; | |
961 | if (unlikely(!mm)) | |
962 | return root_mem_cgroup; | |
963 | } | |
2884b6b7 | 964 | |
54595fe2 KH |
965 | rcu_read_lock(); |
966 | do { | |
2884b6b7 MS |
967 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
968 | if (unlikely(!memcg)) | |
df381975 | 969 | memcg = root_mem_cgroup; |
00d484f3 | 970 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 971 | rcu_read_unlock(); |
c0ff4b85 | 972 | return memcg; |
54595fe2 | 973 | } |
d46eb14b SB |
974 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
975 | ||
4127c650 RG |
976 | static __always_inline bool memcg_kmem_bypass(void) |
977 | { | |
978 | /* Allow remote memcg charging from any context. */ | |
979 | if (unlikely(active_memcg())) | |
980 | return false; | |
981 | ||
982 | /* Memcg to charge can't be determined. */ | |
6126891c | 983 | if (!in_task() || !current->mm || (current->flags & PF_KTHREAD)) |
4127c650 RG |
984 | return true; |
985 | ||
986 | return false; | |
987 | } | |
988 | ||
5660048c JW |
989 | /** |
990 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
991 | * @root: hierarchy root | |
992 | * @prev: previously returned memcg, NULL on first invocation | |
993 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
994 | * | |
995 | * Returns references to children of the hierarchy below @root, or | |
996 | * @root itself, or %NULL after a full round-trip. | |
997 | * | |
998 | * Caller must pass the return value in @prev on subsequent | |
999 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1000 | * to cancel a hierarchy walk before the round-trip is complete. | |
1001 | * | |
05bdc520 ML |
1002 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1003 | * in the hierarchy among all concurrent reclaimers operating on the | |
1004 | * same node. | |
5660048c | 1005 | */ |
694fbc0f | 1006 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1007 | struct mem_cgroup *prev, |
694fbc0f | 1008 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1009 | { |
3f649ab7 | 1010 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1011 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1012 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1013 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1014 | |
694fbc0f AM |
1015 | if (mem_cgroup_disabled()) |
1016 | return NULL; | |
5660048c | 1017 | |
9f3a0d09 JW |
1018 | if (!root) |
1019 | root = root_mem_cgroup; | |
7d74b06f | 1020 | |
542f85f9 | 1021 | rcu_read_lock(); |
5f578161 | 1022 | |
5ac8fb31 | 1023 | if (reclaim) { |
ef8f2327 | 1024 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1025 | |
a3747b53 | 1026 | mz = root->nodeinfo[reclaim->pgdat->node_id]; |
9da83f3f | 1027 | iter = &mz->iter; |
5ac8fb31 | 1028 | |
a9320aae WY |
1029 | /* |
1030 | * On start, join the current reclaim iteration cycle. | |
1031 | * Exit when a concurrent walker completes it. | |
1032 | */ | |
1033 | if (!prev) | |
1034 | reclaim->generation = iter->generation; | |
1035 | else if (reclaim->generation != iter->generation) | |
5ac8fb31 JW |
1036 | goto out_unlock; |
1037 | ||
6df38689 | 1038 | while (1) { |
4db0c3c2 | 1039 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1040 | if (!pos || css_tryget(&pos->css)) |
1041 | break; | |
5ac8fb31 | 1042 | /* |
6df38689 VD |
1043 | * css reference reached zero, so iter->position will |
1044 | * be cleared by ->css_released. However, we should not | |
1045 | * rely on this happening soon, because ->css_released | |
1046 | * is called from a work queue, and by busy-waiting we | |
1047 | * might block it. So we clear iter->position right | |
1048 | * away. | |
5ac8fb31 | 1049 | */ |
6df38689 VD |
1050 | (void)cmpxchg(&iter->position, pos, NULL); |
1051 | } | |
89d8330c WY |
1052 | } else if (prev) { |
1053 | pos = prev; | |
5ac8fb31 JW |
1054 | } |
1055 | ||
1056 | if (pos) | |
1057 | css = &pos->css; | |
1058 | ||
1059 | for (;;) { | |
1060 | css = css_next_descendant_pre(css, &root->css); | |
1061 | if (!css) { | |
1062 | /* | |
1063 | * Reclaimers share the hierarchy walk, and a | |
1064 | * new one might jump in right at the end of | |
1065 | * the hierarchy - make sure they see at least | |
1066 | * one group and restart from the beginning. | |
1067 | */ | |
1068 | if (!prev) | |
1069 | continue; | |
1070 | break; | |
527a5ec9 | 1071 | } |
7d74b06f | 1072 | |
5ac8fb31 JW |
1073 | /* |
1074 | * Verify the css and acquire a reference. The root | |
1075 | * is provided by the caller, so we know it's alive | |
1076 | * and kicking, and don't take an extra reference. | |
1077 | */ | |
41555dad WY |
1078 | if (css == &root->css || css_tryget(css)) { |
1079 | memcg = mem_cgroup_from_css(css); | |
0b8f73e1 | 1080 | break; |
41555dad | 1081 | } |
9f3a0d09 | 1082 | } |
5ac8fb31 JW |
1083 | |
1084 | if (reclaim) { | |
5ac8fb31 | 1085 | /* |
6df38689 VD |
1086 | * The position could have already been updated by a competing |
1087 | * thread, so check that the value hasn't changed since we read | |
1088 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1089 | */ |
6df38689 VD |
1090 | (void)cmpxchg(&iter->position, pos, memcg); |
1091 | ||
5ac8fb31 JW |
1092 | if (pos) |
1093 | css_put(&pos->css); | |
1094 | ||
1095 | if (!memcg) | |
1096 | iter->generation++; | |
9f3a0d09 | 1097 | } |
5ac8fb31 | 1098 | |
542f85f9 MH |
1099 | out_unlock: |
1100 | rcu_read_unlock(); | |
c40046f3 MH |
1101 | if (prev && prev != root) |
1102 | css_put(&prev->css); | |
1103 | ||
9f3a0d09 | 1104 | return memcg; |
14067bb3 | 1105 | } |
7d74b06f | 1106 | |
5660048c JW |
1107 | /** |
1108 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1109 | * @root: hierarchy root | |
1110 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1111 | */ | |
1112 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1113 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1114 | { |
1115 | if (!root) | |
1116 | root = root_mem_cgroup; | |
1117 | if (prev && prev != root) | |
1118 | css_put(&prev->css); | |
1119 | } | |
7d74b06f | 1120 | |
54a83d6b MC |
1121 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1122 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1123 | { |
6df38689 | 1124 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1125 | struct mem_cgroup_per_node *mz; |
1126 | int nid; | |
6df38689 | 1127 | |
54a83d6b | 1128 | for_each_node(nid) { |
a3747b53 | 1129 | mz = from->nodeinfo[nid]; |
9da83f3f YS |
1130 | iter = &mz->iter; |
1131 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1132 | } |
1133 | } | |
1134 | ||
54a83d6b MC |
1135 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1136 | { | |
1137 | struct mem_cgroup *memcg = dead_memcg; | |
1138 | struct mem_cgroup *last; | |
1139 | ||
1140 | do { | |
1141 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1142 | last = memcg; | |
1143 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1144 | ||
1145 | /* | |
1146 | * When cgruop1 non-hierarchy mode is used, | |
1147 | * parent_mem_cgroup() does not walk all the way up to the | |
1148 | * cgroup root (root_mem_cgroup). So we have to handle | |
1149 | * dead_memcg from cgroup root separately. | |
1150 | */ | |
1151 | if (last != root_mem_cgroup) | |
1152 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1153 | dead_memcg); | |
1154 | } | |
1155 | ||
7c5f64f8 VD |
1156 | /** |
1157 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1158 | * @memcg: hierarchy root | |
1159 | * @fn: function to call for each task | |
1160 | * @arg: argument passed to @fn | |
1161 | * | |
1162 | * This function iterates over tasks attached to @memcg or to any of its | |
1163 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1164 | * value, the function breaks the iteration loop and returns the value. | |
1165 | * Otherwise, it will iterate over all tasks and return 0. | |
1166 | * | |
1167 | * This function must not be called for the root memory cgroup. | |
1168 | */ | |
1169 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1170 | int (*fn)(struct task_struct *, void *), void *arg) | |
1171 | { | |
1172 | struct mem_cgroup *iter; | |
1173 | int ret = 0; | |
1174 | ||
1175 | BUG_ON(memcg == root_mem_cgroup); | |
1176 | ||
1177 | for_each_mem_cgroup_tree(iter, memcg) { | |
1178 | struct css_task_iter it; | |
1179 | struct task_struct *task; | |
1180 | ||
f168a9a5 | 1181 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1182 | while (!ret && (task = css_task_iter_next(&it))) |
1183 | ret = fn(task, arg); | |
1184 | css_task_iter_end(&it); | |
1185 | if (ret) { | |
1186 | mem_cgroup_iter_break(memcg, iter); | |
1187 | break; | |
1188 | } | |
1189 | } | |
1190 | return ret; | |
1191 | } | |
1192 | ||
6168d0da | 1193 | #ifdef CONFIG_DEBUG_VM |
e809c3fe | 1194 | void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio) |
6168d0da AS |
1195 | { |
1196 | struct mem_cgroup *memcg; | |
1197 | ||
1198 | if (mem_cgroup_disabled()) | |
1199 | return; | |
1200 | ||
e809c3fe | 1201 | memcg = folio_memcg(folio); |
6168d0da AS |
1202 | |
1203 | if (!memcg) | |
e809c3fe | 1204 | VM_BUG_ON_FOLIO(lruvec_memcg(lruvec) != root_mem_cgroup, folio); |
6168d0da | 1205 | else |
e809c3fe | 1206 | VM_BUG_ON_FOLIO(lruvec_memcg(lruvec) != memcg, folio); |
6168d0da AS |
1207 | } |
1208 | #endif | |
1209 | ||
6168d0da | 1210 | /** |
e809c3fe MWO |
1211 | * folio_lruvec_lock - Lock the lruvec for a folio. |
1212 | * @folio: Pointer to the folio. | |
6168d0da | 1213 | * |
d7e3aba5 | 1214 | * These functions are safe to use under any of the following conditions: |
e809c3fe MWO |
1215 | * - folio locked |
1216 | * - folio_test_lru false | |
1217 | * - folio_memcg_lock() | |
1218 | * - folio frozen (refcount of 0) | |
1219 | * | |
1220 | * Return: The lruvec this folio is on with its lock held. | |
6168d0da | 1221 | */ |
e809c3fe | 1222 | struct lruvec *folio_lruvec_lock(struct folio *folio) |
6168d0da | 1223 | { |
e809c3fe | 1224 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1225 | |
6168d0da | 1226 | spin_lock(&lruvec->lru_lock); |
e809c3fe | 1227 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1228 | |
1229 | return lruvec; | |
1230 | } | |
1231 | ||
e809c3fe MWO |
1232 | /** |
1233 | * folio_lruvec_lock_irq - Lock the lruvec for a folio. | |
1234 | * @folio: Pointer to the folio. | |
1235 | * | |
1236 | * These functions are safe to use under any of the following conditions: | |
1237 | * - folio locked | |
1238 | * - folio_test_lru false | |
1239 | * - folio_memcg_lock() | |
1240 | * - folio frozen (refcount of 0) | |
1241 | * | |
1242 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1243 | * disabled. | |
1244 | */ | |
1245 | struct lruvec *folio_lruvec_lock_irq(struct folio *folio) | |
6168d0da | 1246 | { |
e809c3fe | 1247 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1248 | |
6168d0da | 1249 | spin_lock_irq(&lruvec->lru_lock); |
e809c3fe | 1250 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1251 | |
1252 | return lruvec; | |
1253 | } | |
1254 | ||
e809c3fe MWO |
1255 | /** |
1256 | * folio_lruvec_lock_irqsave - Lock the lruvec for a folio. | |
1257 | * @folio: Pointer to the folio. | |
1258 | * @flags: Pointer to irqsave flags. | |
1259 | * | |
1260 | * These functions are safe to use under any of the following conditions: | |
1261 | * - folio locked | |
1262 | * - folio_test_lru false | |
1263 | * - folio_memcg_lock() | |
1264 | * - folio frozen (refcount of 0) | |
1265 | * | |
1266 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1267 | * disabled. | |
1268 | */ | |
1269 | struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio, | |
1270 | unsigned long *flags) | |
6168d0da | 1271 | { |
e809c3fe | 1272 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1273 | |
6168d0da | 1274 | spin_lock_irqsave(&lruvec->lru_lock, *flags); |
e809c3fe | 1275 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1276 | |
1277 | return lruvec; | |
1278 | } | |
1279 | ||
925b7673 | 1280 | /** |
fa9add64 HD |
1281 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1282 | * @lruvec: mem_cgroup per zone lru vector | |
1283 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1284 | * @zid: zone id of the accounted pages |
fa9add64 | 1285 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1286 | * |
ca707239 | 1287 | * This function must be called under lru_lock, just before a page is added |
07ca7606 | 1288 | * to or just after a page is removed from an lru list. |
3f58a829 | 1289 | */ |
fa9add64 | 1290 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1291 | int zid, int nr_pages) |
3f58a829 | 1292 | { |
ef8f2327 | 1293 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1294 | unsigned long *lru_size; |
ca707239 | 1295 | long size; |
3f58a829 MK |
1296 | |
1297 | if (mem_cgroup_disabled()) | |
1298 | return; | |
1299 | ||
ef8f2327 | 1300 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1301 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1302 | |
1303 | if (nr_pages < 0) | |
1304 | *lru_size += nr_pages; | |
1305 | ||
1306 | size = *lru_size; | |
b4536f0c MH |
1307 | if (WARN_ONCE(size < 0, |
1308 | "%s(%p, %d, %d): lru_size %ld\n", | |
1309 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1310 | VM_BUG_ON(1); |
1311 | *lru_size = 0; | |
1312 | } | |
1313 | ||
1314 | if (nr_pages > 0) | |
1315 | *lru_size += nr_pages; | |
08e552c6 | 1316 | } |
544122e5 | 1317 | |
19942822 | 1318 | /** |
9d11ea9f | 1319 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1320 | * @memcg: the memory cgroup |
19942822 | 1321 | * |
9d11ea9f | 1322 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1323 | * pages. |
19942822 | 1324 | */ |
c0ff4b85 | 1325 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1326 | { |
3e32cb2e JW |
1327 | unsigned long margin = 0; |
1328 | unsigned long count; | |
1329 | unsigned long limit; | |
9d11ea9f | 1330 | |
3e32cb2e | 1331 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1332 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1333 | if (count < limit) |
1334 | margin = limit - count; | |
1335 | ||
7941d214 | 1336 | if (do_memsw_account()) { |
3e32cb2e | 1337 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1338 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1339 | if (count < limit) |
3e32cb2e | 1340 | margin = min(margin, limit - count); |
cbedbac3 LR |
1341 | else |
1342 | margin = 0; | |
3e32cb2e JW |
1343 | } |
1344 | ||
1345 | return margin; | |
19942822 JW |
1346 | } |
1347 | ||
32047e2a | 1348 | /* |
bdcbb659 | 1349 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1350 | * |
bdcbb659 QH |
1351 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1352 | * moving cgroups. This is for waiting at high-memory pressure | |
1353 | * caused by "move". | |
32047e2a | 1354 | */ |
c0ff4b85 | 1355 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1356 | { |
2bd9bb20 KH |
1357 | struct mem_cgroup *from; |
1358 | struct mem_cgroup *to; | |
4b534334 | 1359 | bool ret = false; |
2bd9bb20 KH |
1360 | /* |
1361 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1362 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1363 | */ | |
1364 | spin_lock(&mc.lock); | |
1365 | from = mc.from; | |
1366 | to = mc.to; | |
1367 | if (!from) | |
1368 | goto unlock; | |
3e92041d | 1369 | |
2314b42d JW |
1370 | ret = mem_cgroup_is_descendant(from, memcg) || |
1371 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1372 | unlock: |
1373 | spin_unlock(&mc.lock); | |
4b534334 KH |
1374 | return ret; |
1375 | } | |
1376 | ||
c0ff4b85 | 1377 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1378 | { |
1379 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1380 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1381 | DEFINE_WAIT(wait); |
1382 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1383 | /* moving charge context might have finished. */ | |
1384 | if (mc.moving_task) | |
1385 | schedule(); | |
1386 | finish_wait(&mc.waitq, &wait); | |
1387 | return true; | |
1388 | } | |
1389 | } | |
1390 | return false; | |
1391 | } | |
1392 | ||
5f9a4f4a MS |
1393 | struct memory_stat { |
1394 | const char *name; | |
5f9a4f4a MS |
1395 | unsigned int idx; |
1396 | }; | |
1397 | ||
57b2847d | 1398 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1399 | { "anon", NR_ANON_MAPPED }, |
1400 | { "file", NR_FILE_PAGES }, | |
a8c49af3 | 1401 | { "kernel", MEMCG_KMEM }, |
fff66b79 MS |
1402 | { "kernel_stack", NR_KERNEL_STACK_KB }, |
1403 | { "pagetables", NR_PAGETABLE }, | |
1404 | { "percpu", MEMCG_PERCPU_B }, | |
1405 | { "sock", MEMCG_SOCK }, | |
4e5aa1f4 | 1406 | { "vmalloc", MEMCG_VMALLOC }, |
fff66b79 | 1407 | { "shmem", NR_SHMEM }, |
f4840ccf JW |
1408 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
1409 | { "zswap", MEMCG_ZSWAP_B }, | |
1410 | { "zswapped", MEMCG_ZSWAPPED }, | |
1411 | #endif | |
fff66b79 MS |
1412 | { "file_mapped", NR_FILE_MAPPED }, |
1413 | { "file_dirty", NR_FILE_DIRTY }, | |
1414 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1415 | #ifdef CONFIG_SWAP |
1416 | { "swapcached", NR_SWAPCACHE }, | |
1417 | #endif | |
5f9a4f4a | 1418 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1419 | { "anon_thp", NR_ANON_THPS }, |
1420 | { "file_thp", NR_FILE_THPS }, | |
1421 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1422 | #endif |
fff66b79 MS |
1423 | { "inactive_anon", NR_INACTIVE_ANON }, |
1424 | { "active_anon", NR_ACTIVE_ANON }, | |
1425 | { "inactive_file", NR_INACTIVE_FILE }, | |
1426 | { "active_file", NR_ACTIVE_FILE }, | |
1427 | { "unevictable", NR_UNEVICTABLE }, | |
1428 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1429 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1430 | |
1431 | /* The memory events */ | |
fff66b79 MS |
1432 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1433 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1434 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1435 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1436 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1437 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1438 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1439 | }; |
1440 | ||
fff66b79 MS |
1441 | /* Translate stat items to the correct unit for memory.stat output */ |
1442 | static int memcg_page_state_unit(int item) | |
1443 | { | |
1444 | switch (item) { | |
1445 | case MEMCG_PERCPU_B: | |
f4840ccf | 1446 | case MEMCG_ZSWAP_B: |
fff66b79 MS |
1447 | case NR_SLAB_RECLAIMABLE_B: |
1448 | case NR_SLAB_UNRECLAIMABLE_B: | |
1449 | case WORKINGSET_REFAULT_ANON: | |
1450 | case WORKINGSET_REFAULT_FILE: | |
1451 | case WORKINGSET_ACTIVATE_ANON: | |
1452 | case WORKINGSET_ACTIVATE_FILE: | |
1453 | case WORKINGSET_RESTORE_ANON: | |
1454 | case WORKINGSET_RESTORE_FILE: | |
1455 | case WORKINGSET_NODERECLAIM: | |
1456 | return 1; | |
1457 | case NR_KERNEL_STACK_KB: | |
1458 | return SZ_1K; | |
1459 | default: | |
1460 | return PAGE_SIZE; | |
1461 | } | |
1462 | } | |
1463 | ||
1464 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1465 | int item) | |
1466 | { | |
1467 | return memcg_page_state(memcg, item) * memcg_page_state_unit(item); | |
1468 | } | |
1469 | ||
673520f8 QZ |
1470 | /* Subset of vm_event_item to report for memcg event stats */ |
1471 | static const unsigned int memcg_vm_event_stat[] = { | |
1472 | PGSCAN_KSWAPD, | |
1473 | PGSCAN_DIRECT, | |
1474 | PGSTEAL_KSWAPD, | |
1475 | PGSTEAL_DIRECT, | |
1476 | PGFAULT, | |
1477 | PGMAJFAULT, | |
1478 | PGREFILL, | |
1479 | PGACTIVATE, | |
1480 | PGDEACTIVATE, | |
1481 | PGLAZYFREE, | |
1482 | PGLAZYFREED, | |
1483 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) | |
1484 | ZSWPIN, | |
1485 | ZSWPOUT, | |
1486 | #endif | |
1487 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1488 | THP_FAULT_ALLOC, | |
1489 | THP_COLLAPSE_ALLOC, | |
1490 | #endif | |
1491 | }; | |
1492 | ||
c8713d0b JW |
1493 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1494 | { | |
1495 | struct seq_buf s; | |
1496 | int i; | |
71cd3113 | 1497 | |
c8713d0b JW |
1498 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1499 | if (!s.buffer) | |
1500 | return NULL; | |
1501 | ||
1502 | /* | |
1503 | * Provide statistics on the state of the memory subsystem as | |
1504 | * well as cumulative event counters that show past behavior. | |
1505 | * | |
1506 | * This list is ordered following a combination of these gradients: | |
1507 | * 1) generic big picture -> specifics and details | |
1508 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1509 | * | |
1510 | * Current memory state: | |
1511 | */ | |
fd25a9e0 | 1512 | mem_cgroup_flush_stats(); |
c8713d0b | 1513 | |
5f9a4f4a MS |
1514 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1515 | u64 size; | |
c8713d0b | 1516 | |
fff66b79 | 1517 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5f9a4f4a | 1518 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1519 | |
5f9a4f4a | 1520 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1521 | size += memcg_page_state_output(memcg, |
1522 | NR_SLAB_RECLAIMABLE_B); | |
5f9a4f4a MS |
1523 | seq_buf_printf(&s, "slab %llu\n", size); |
1524 | } | |
1525 | } | |
c8713d0b JW |
1526 | |
1527 | /* Accumulated memory events */ | |
c8713d0b JW |
1528 | seq_buf_printf(&s, "pgscan %lu\n", |
1529 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1530 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1531 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1532 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1533 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
c8713d0b | 1534 | |
673520f8 QZ |
1535 | for (i = 0; i < ARRAY_SIZE(memcg_vm_event_stat); i++) |
1536 | seq_buf_printf(&s, "%s %lu\n", | |
1537 | vm_event_name(memcg_vm_event_stat[i]), | |
1538 | memcg_events(memcg, memcg_vm_event_stat[i])); | |
c8713d0b JW |
1539 | |
1540 | /* The above should easily fit into one page */ | |
1541 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1542 | ||
1543 | return s.buffer; | |
1544 | } | |
71cd3113 | 1545 | |
58cf188e | 1546 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1547 | /** |
f0c867d9 | 1548 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1549 | * memory controller. | |
e222432b BS |
1550 | * @memcg: The memory cgroup that went over limit |
1551 | * @p: Task that is going to be killed | |
1552 | * | |
1553 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1554 | * enabled | |
1555 | */ | |
f0c867d9 | 1556 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1557 | { |
e222432b BS |
1558 | rcu_read_lock(); |
1559 | ||
f0c867d9 | 1560 | if (memcg) { |
1561 | pr_cont(",oom_memcg="); | |
1562 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1563 | } else | |
1564 | pr_cont(",global_oom"); | |
2415b9f5 | 1565 | if (p) { |
f0c867d9 | 1566 | pr_cont(",task_memcg="); |
2415b9f5 | 1567 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1568 | } |
e222432b | 1569 | rcu_read_unlock(); |
f0c867d9 | 1570 | } |
1571 | ||
1572 | /** | |
1573 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1574 | * memory controller. | |
1575 | * @memcg: The memory cgroup that went over limit | |
1576 | */ | |
1577 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1578 | { | |
c8713d0b | 1579 | char *buf; |
e222432b | 1580 | |
3e32cb2e JW |
1581 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1582 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1583 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1584 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1585 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1586 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1587 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1588 | else { |
1589 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1590 | K((u64)page_counter_read(&memcg->memsw)), | |
1591 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1592 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1593 | K((u64)page_counter_read(&memcg->kmem)), | |
1594 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1595 | } |
c8713d0b JW |
1596 | |
1597 | pr_info("Memory cgroup stats for "); | |
1598 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1599 | pr_cont(":"); | |
1600 | buf = memory_stat_format(memcg); | |
1601 | if (!buf) | |
1602 | return; | |
1603 | pr_info("%s", buf); | |
1604 | kfree(buf); | |
e222432b BS |
1605 | } |
1606 | ||
a63d83f4 DR |
1607 | /* |
1608 | * Return the memory (and swap, if configured) limit for a memcg. | |
1609 | */ | |
bbec2e15 | 1610 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1611 | { |
8d387a5f WL |
1612 | unsigned long max = READ_ONCE(memcg->memory.max); |
1613 | ||
1614 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1615 | if (mem_cgroup_swappiness(memcg)) | |
1616 | max += min(READ_ONCE(memcg->swap.max), | |
1617 | (unsigned long)total_swap_pages); | |
1618 | } else { /* v1 */ | |
1619 | if (mem_cgroup_swappiness(memcg)) { | |
1620 | /* Calculate swap excess capacity from memsw limit */ | |
1621 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1622 | ||
1623 | max += min(swap, (unsigned long)total_swap_pages); | |
1624 | } | |
9a5a8f19 | 1625 | } |
bbec2e15 | 1626 | return max; |
a63d83f4 DR |
1627 | } |
1628 | ||
9783aa99 CD |
1629 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1630 | { | |
1631 | return page_counter_read(&memcg->memory); | |
1632 | } | |
1633 | ||
b6e6edcf | 1634 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1635 | int order) |
9cbb78bb | 1636 | { |
6e0fc46d DR |
1637 | struct oom_control oc = { |
1638 | .zonelist = NULL, | |
1639 | .nodemask = NULL, | |
2a966b77 | 1640 | .memcg = memcg, |
6e0fc46d DR |
1641 | .gfp_mask = gfp_mask, |
1642 | .order = order, | |
6e0fc46d | 1643 | }; |
1378b37d | 1644 | bool ret = true; |
9cbb78bb | 1645 | |
7775face TH |
1646 | if (mutex_lock_killable(&oom_lock)) |
1647 | return true; | |
1378b37d YS |
1648 | |
1649 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1650 | goto unlock; | |
1651 | ||
7775face TH |
1652 | /* |
1653 | * A few threads which were not waiting at mutex_lock_killable() can | |
1654 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1655 | */ | |
a4ebf1b6 | 1656 | ret = task_is_dying() || out_of_memory(&oc); |
1378b37d YS |
1657 | |
1658 | unlock: | |
dc56401f | 1659 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1660 | return ret; |
9cbb78bb DR |
1661 | } |
1662 | ||
0608f43d | 1663 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1664 | pg_data_t *pgdat, |
0608f43d AM |
1665 | gfp_t gfp_mask, |
1666 | unsigned long *total_scanned) | |
1667 | { | |
1668 | struct mem_cgroup *victim = NULL; | |
1669 | int total = 0; | |
1670 | int loop = 0; | |
1671 | unsigned long excess; | |
1672 | unsigned long nr_scanned; | |
1673 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1674 | .pgdat = pgdat, |
0608f43d AM |
1675 | }; |
1676 | ||
3e32cb2e | 1677 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1678 | |
1679 | while (1) { | |
1680 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1681 | if (!victim) { | |
1682 | loop++; | |
1683 | if (loop >= 2) { | |
1684 | /* | |
1685 | * If we have not been able to reclaim | |
1686 | * anything, it might because there are | |
1687 | * no reclaimable pages under this hierarchy | |
1688 | */ | |
1689 | if (!total) | |
1690 | break; | |
1691 | /* | |
1692 | * We want to do more targeted reclaim. | |
1693 | * excess >> 2 is not to excessive so as to | |
1694 | * reclaim too much, nor too less that we keep | |
1695 | * coming back to reclaim from this cgroup | |
1696 | */ | |
1697 | if (total >= (excess >> 2) || | |
1698 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1699 | break; | |
1700 | } | |
1701 | continue; | |
1702 | } | |
a9dd0a83 | 1703 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1704 | pgdat, &nr_scanned); |
0608f43d | 1705 | *total_scanned += nr_scanned; |
3e32cb2e | 1706 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1707 | break; |
6d61ef40 | 1708 | } |
0608f43d AM |
1709 | mem_cgroup_iter_break(root_memcg, victim); |
1710 | return total; | |
6d61ef40 BS |
1711 | } |
1712 | ||
0056f4e6 JW |
1713 | #ifdef CONFIG_LOCKDEP |
1714 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1715 | .name = "memcg_oom_lock", | |
1716 | }; | |
1717 | #endif | |
1718 | ||
fb2a6fc5 JW |
1719 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1720 | ||
867578cb KH |
1721 | /* |
1722 | * Check OOM-Killer is already running under our hierarchy. | |
1723 | * If someone is running, return false. | |
1724 | */ | |
fb2a6fc5 | 1725 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1726 | { |
79dfdacc | 1727 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1728 | |
fb2a6fc5 JW |
1729 | spin_lock(&memcg_oom_lock); |
1730 | ||
9f3a0d09 | 1731 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1732 | if (iter->oom_lock) { |
79dfdacc MH |
1733 | /* |
1734 | * this subtree of our hierarchy is already locked | |
1735 | * so we cannot give a lock. | |
1736 | */ | |
79dfdacc | 1737 | failed = iter; |
9f3a0d09 JW |
1738 | mem_cgroup_iter_break(memcg, iter); |
1739 | break; | |
23751be0 JW |
1740 | } else |
1741 | iter->oom_lock = true; | |
7d74b06f | 1742 | } |
867578cb | 1743 | |
fb2a6fc5 JW |
1744 | if (failed) { |
1745 | /* | |
1746 | * OK, we failed to lock the whole subtree so we have | |
1747 | * to clean up what we set up to the failing subtree | |
1748 | */ | |
1749 | for_each_mem_cgroup_tree(iter, memcg) { | |
1750 | if (iter == failed) { | |
1751 | mem_cgroup_iter_break(memcg, iter); | |
1752 | break; | |
1753 | } | |
1754 | iter->oom_lock = false; | |
79dfdacc | 1755 | } |
0056f4e6 JW |
1756 | } else |
1757 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1758 | |
1759 | spin_unlock(&memcg_oom_lock); | |
1760 | ||
1761 | return !failed; | |
a636b327 | 1762 | } |
0b7f569e | 1763 | |
fb2a6fc5 | 1764 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1765 | { |
7d74b06f KH |
1766 | struct mem_cgroup *iter; |
1767 | ||
fb2a6fc5 | 1768 | spin_lock(&memcg_oom_lock); |
5facae4f | 1769 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1770 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1771 | iter->oom_lock = false; |
fb2a6fc5 | 1772 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1773 | } |
1774 | ||
c0ff4b85 | 1775 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1776 | { |
1777 | struct mem_cgroup *iter; | |
1778 | ||
c2b42d3c | 1779 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1780 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1781 | iter->under_oom++; |
1782 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1783 | } |
1784 | ||
c0ff4b85 | 1785 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1786 | { |
1787 | struct mem_cgroup *iter; | |
1788 | ||
867578cb | 1789 | /* |
f0953a1b | 1790 | * Be careful about under_oom underflows because a child memcg |
7a52d4d8 | 1791 | * could have been added after mem_cgroup_mark_under_oom. |
867578cb | 1792 | */ |
c2b42d3c | 1793 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1794 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1795 | if (iter->under_oom > 0) |
1796 | iter->under_oom--; | |
1797 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1798 | } |
1799 | ||
867578cb KH |
1800 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1801 | ||
dc98df5a | 1802 | struct oom_wait_info { |
d79154bb | 1803 | struct mem_cgroup *memcg; |
ac6424b9 | 1804 | wait_queue_entry_t wait; |
dc98df5a KH |
1805 | }; |
1806 | ||
ac6424b9 | 1807 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1808 | unsigned mode, int sync, void *arg) |
1809 | { | |
d79154bb HD |
1810 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1811 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1812 | struct oom_wait_info *oom_wait_info; |
1813 | ||
1814 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1815 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1816 | |
2314b42d JW |
1817 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1818 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1819 | return 0; |
dc98df5a KH |
1820 | return autoremove_wake_function(wait, mode, sync, arg); |
1821 | } | |
1822 | ||
c0ff4b85 | 1823 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1824 | { |
c2b42d3c TH |
1825 | /* |
1826 | * For the following lockless ->under_oom test, the only required | |
1827 | * guarantee is that it must see the state asserted by an OOM when | |
1828 | * this function is called as a result of userland actions | |
1829 | * triggered by the notification of the OOM. This is trivially | |
1830 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1831 | * triggering notification. | |
1832 | */ | |
1833 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1834 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1835 | } |
1836 | ||
becdf89d SB |
1837 | /* |
1838 | * Returns true if successfully killed one or more processes. Though in some | |
1839 | * corner cases it can return true even without killing any process. | |
1840 | */ | |
1841 | static bool mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1842 | { |
becdf89d | 1843 | bool locked, ret; |
7056d3a3 | 1844 | |
29ef680a | 1845 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
becdf89d | 1846 | return false; |
29ef680a | 1847 | |
7a1adfdd RG |
1848 | memcg_memory_event(memcg, MEMCG_OOM); |
1849 | ||
867578cb | 1850 | /* |
49426420 JW |
1851 | * We are in the middle of the charge context here, so we |
1852 | * don't want to block when potentially sitting on a callstack | |
1853 | * that holds all kinds of filesystem and mm locks. | |
1854 | * | |
29ef680a MH |
1855 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1856 | * handling until the charge can succeed; remember the context and put | |
1857 | * the task to sleep at the end of the page fault when all locks are | |
1858 | * released. | |
49426420 | 1859 | * |
29ef680a MH |
1860 | * On the other hand, in-kernel OOM killer allows for an async victim |
1861 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1862 | * relying on the oom victim to make a forward progress and we can | |
1863 | * invoke the oom killer here. | |
1864 | * | |
1865 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1866 | * victim and then we have to bail out from the charge path. | |
867578cb | 1867 | */ |
29ef680a | 1868 | if (memcg->oom_kill_disable) { |
becdf89d SB |
1869 | if (current->in_user_fault) { |
1870 | css_get(&memcg->css); | |
1871 | current->memcg_in_oom = memcg; | |
1872 | current->memcg_oom_gfp_mask = mask; | |
1873 | current->memcg_oom_order = order; | |
1874 | } | |
1875 | return false; | |
29ef680a MH |
1876 | } |
1877 | ||
7056d3a3 MH |
1878 | mem_cgroup_mark_under_oom(memcg); |
1879 | ||
1880 | locked = mem_cgroup_oom_trylock(memcg); | |
1881 | ||
1882 | if (locked) | |
1883 | mem_cgroup_oom_notify(memcg); | |
1884 | ||
1885 | mem_cgroup_unmark_under_oom(memcg); | |
becdf89d | 1886 | ret = mem_cgroup_out_of_memory(memcg, mask, order); |
7056d3a3 MH |
1887 | |
1888 | if (locked) | |
1889 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1890 | |
7056d3a3 | 1891 | return ret; |
3812c8c8 JW |
1892 | } |
1893 | ||
1894 | /** | |
1895 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1896 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1897 | * |
49426420 JW |
1898 | * This has to be called at the end of a page fault if the memcg OOM |
1899 | * handler was enabled. | |
3812c8c8 | 1900 | * |
49426420 | 1901 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1902 | * sleep on a waitqueue until the userspace task resolves the |
1903 | * situation. Sleeping directly in the charge context with all kinds | |
1904 | * of locks held is not a good idea, instead we remember an OOM state | |
1905 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1906 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1907 | * |
1908 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1909 | * completed, %false otherwise. |
3812c8c8 | 1910 | */ |
49426420 | 1911 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1912 | { |
626ebc41 | 1913 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1914 | struct oom_wait_info owait; |
49426420 | 1915 | bool locked; |
3812c8c8 JW |
1916 | |
1917 | /* OOM is global, do not handle */ | |
3812c8c8 | 1918 | if (!memcg) |
49426420 | 1919 | return false; |
3812c8c8 | 1920 | |
7c5f64f8 | 1921 | if (!handle) |
49426420 | 1922 | goto cleanup; |
3812c8c8 JW |
1923 | |
1924 | owait.memcg = memcg; | |
1925 | owait.wait.flags = 0; | |
1926 | owait.wait.func = memcg_oom_wake_function; | |
1927 | owait.wait.private = current; | |
2055da97 | 1928 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1929 | |
3812c8c8 | 1930 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1931 | mem_cgroup_mark_under_oom(memcg); |
1932 | ||
1933 | locked = mem_cgroup_oom_trylock(memcg); | |
1934 | ||
1935 | if (locked) | |
1936 | mem_cgroup_oom_notify(memcg); | |
1937 | ||
1938 | if (locked && !memcg->oom_kill_disable) { | |
1939 | mem_cgroup_unmark_under_oom(memcg); | |
1940 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1941 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1942 | current->memcg_oom_order); | |
49426420 | 1943 | } else { |
3812c8c8 | 1944 | schedule(); |
49426420 JW |
1945 | mem_cgroup_unmark_under_oom(memcg); |
1946 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1947 | } | |
1948 | ||
1949 | if (locked) { | |
fb2a6fc5 JW |
1950 | mem_cgroup_oom_unlock(memcg); |
1951 | /* | |
1952 | * There is no guarantee that an OOM-lock contender | |
1953 | * sees the wakeups triggered by the OOM kill | |
f0953a1b | 1954 | * uncharges. Wake any sleepers explicitly. |
fb2a6fc5 JW |
1955 | */ |
1956 | memcg_oom_recover(memcg); | |
1957 | } | |
49426420 | 1958 | cleanup: |
626ebc41 | 1959 | current->memcg_in_oom = NULL; |
3812c8c8 | 1960 | css_put(&memcg->css); |
867578cb | 1961 | return true; |
0b7f569e KH |
1962 | } |
1963 | ||
3d8b38eb RG |
1964 | /** |
1965 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1966 | * @victim: task to be killed by the OOM killer | |
1967 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1968 | * | |
1969 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1970 | * by killing all belonging OOM-killable tasks. | |
1971 | * | |
1972 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1973 | */ | |
1974 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1975 | struct mem_cgroup *oom_domain) | |
1976 | { | |
1977 | struct mem_cgroup *oom_group = NULL; | |
1978 | struct mem_cgroup *memcg; | |
1979 | ||
1980 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1981 | return NULL; | |
1982 | ||
1983 | if (!oom_domain) | |
1984 | oom_domain = root_mem_cgroup; | |
1985 | ||
1986 | rcu_read_lock(); | |
1987 | ||
1988 | memcg = mem_cgroup_from_task(victim); | |
1989 | if (memcg == root_mem_cgroup) | |
1990 | goto out; | |
1991 | ||
48fe267c RG |
1992 | /* |
1993 | * If the victim task has been asynchronously moved to a different | |
1994 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
1995 | * In this case it's better to ignore memory.group.oom. | |
1996 | */ | |
1997 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
1998 | goto out; | |
1999 | ||
3d8b38eb RG |
2000 | /* |
2001 | * Traverse the memory cgroup hierarchy from the victim task's | |
2002 | * cgroup up to the OOMing cgroup (or root) to find the | |
2003 | * highest-level memory cgroup with oom.group set. | |
2004 | */ | |
2005 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2006 | if (memcg->oom_group) | |
2007 | oom_group = memcg; | |
2008 | ||
2009 | if (memcg == oom_domain) | |
2010 | break; | |
2011 | } | |
2012 | ||
2013 | if (oom_group) | |
2014 | css_get(&oom_group->css); | |
2015 | out: | |
2016 | rcu_read_unlock(); | |
2017 | ||
2018 | return oom_group; | |
2019 | } | |
2020 | ||
2021 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2022 | { | |
2023 | pr_info("Tasks in "); | |
2024 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2025 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2026 | } | |
2027 | ||
d7365e78 | 2028 | /** |
f70ad448 MWO |
2029 | * folio_memcg_lock - Bind a folio to its memcg. |
2030 | * @folio: The folio. | |
32047e2a | 2031 | * |
f70ad448 | 2032 | * This function prevents unlocked LRU folios from being moved to |
739f79fc JW |
2033 | * another cgroup. |
2034 | * | |
f70ad448 MWO |
2035 | * It ensures lifetime of the bound memcg. The caller is responsible |
2036 | * for the lifetime of the folio. | |
d69b042f | 2037 | */ |
f70ad448 | 2038 | void folio_memcg_lock(struct folio *folio) |
89c06bd5 KH |
2039 | { |
2040 | struct mem_cgroup *memcg; | |
6de22619 | 2041 | unsigned long flags; |
89c06bd5 | 2042 | |
6de22619 JW |
2043 | /* |
2044 | * The RCU lock is held throughout the transaction. The fast | |
2045 | * path can get away without acquiring the memcg->move_lock | |
2046 | * because page moving starts with an RCU grace period. | |
739f79fc | 2047 | */ |
d7365e78 JW |
2048 | rcu_read_lock(); |
2049 | ||
2050 | if (mem_cgroup_disabled()) | |
1c824a68 | 2051 | return; |
89c06bd5 | 2052 | again: |
f70ad448 | 2053 | memcg = folio_memcg(folio); |
29833315 | 2054 | if (unlikely(!memcg)) |
1c824a68 | 2055 | return; |
d7365e78 | 2056 | |
20ad50d6 AS |
2057 | #ifdef CONFIG_PROVE_LOCKING |
2058 | local_irq_save(flags); | |
2059 | might_lock(&memcg->move_lock); | |
2060 | local_irq_restore(flags); | |
2061 | #endif | |
2062 | ||
bdcbb659 | 2063 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 2064 | return; |
89c06bd5 | 2065 | |
6de22619 | 2066 | spin_lock_irqsave(&memcg->move_lock, flags); |
f70ad448 | 2067 | if (memcg != folio_memcg(folio)) { |
6de22619 | 2068 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2069 | goto again; |
2070 | } | |
6de22619 JW |
2071 | |
2072 | /* | |
1c824a68 JW |
2073 | * When charge migration first begins, we can have multiple |
2074 | * critical sections holding the fast-path RCU lock and one | |
2075 | * holding the slowpath move_lock. Track the task who has the | |
2076 | * move_lock for unlock_page_memcg(). | |
6de22619 JW |
2077 | */ |
2078 | memcg->move_lock_task = current; | |
2079 | memcg->move_lock_flags = flags; | |
89c06bd5 | 2080 | } |
f70ad448 MWO |
2081 | |
2082 | void lock_page_memcg(struct page *page) | |
2083 | { | |
2084 | folio_memcg_lock(page_folio(page)); | |
2085 | } | |
89c06bd5 | 2086 | |
f70ad448 | 2087 | static void __folio_memcg_unlock(struct mem_cgroup *memcg) |
89c06bd5 | 2088 | { |
6de22619 JW |
2089 | if (memcg && memcg->move_lock_task == current) { |
2090 | unsigned long flags = memcg->move_lock_flags; | |
2091 | ||
2092 | memcg->move_lock_task = NULL; | |
2093 | memcg->move_lock_flags = 0; | |
2094 | ||
2095 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2096 | } | |
89c06bd5 | 2097 | |
d7365e78 | 2098 | rcu_read_unlock(); |
89c06bd5 | 2099 | } |
739f79fc JW |
2100 | |
2101 | /** | |
f70ad448 MWO |
2102 | * folio_memcg_unlock - Release the binding between a folio and its memcg. |
2103 | * @folio: The folio. | |
2104 | * | |
2105 | * This releases the binding created by folio_memcg_lock(). This does | |
2106 | * not change the accounting of this folio to its memcg, but it does | |
2107 | * permit others to change it. | |
739f79fc | 2108 | */ |
f70ad448 | 2109 | void folio_memcg_unlock(struct folio *folio) |
739f79fc | 2110 | { |
f70ad448 MWO |
2111 | __folio_memcg_unlock(folio_memcg(folio)); |
2112 | } | |
9da7b521 | 2113 | |
f70ad448 MWO |
2114 | void unlock_page_memcg(struct page *page) |
2115 | { | |
2116 | folio_memcg_unlock(page_folio(page)); | |
739f79fc | 2117 | } |
89c06bd5 | 2118 | |
fead2b86 | 2119 | struct memcg_stock_pcp { |
56751146 | 2120 | local_lock_t stock_lock; |
fead2b86 MH |
2121 | struct mem_cgroup *cached; /* this never be root cgroup */ |
2122 | unsigned int nr_pages; | |
2123 | ||
bf4f0599 RG |
2124 | #ifdef CONFIG_MEMCG_KMEM |
2125 | struct obj_cgroup *cached_objcg; | |
68ac5b3c | 2126 | struct pglist_data *cached_pgdat; |
bf4f0599 | 2127 | unsigned int nr_bytes; |
68ac5b3c WL |
2128 | int nr_slab_reclaimable_b; |
2129 | int nr_slab_unreclaimable_b; | |
bf4f0599 RG |
2130 | #endif |
2131 | ||
cdec2e42 | 2132 | struct work_struct work; |
26fe6168 | 2133 | unsigned long flags; |
a0db00fc | 2134 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 | 2135 | }; |
56751146 SAS |
2136 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = { |
2137 | .stock_lock = INIT_LOCAL_LOCK(stock_lock), | |
2138 | }; | |
9f50fad6 | 2139 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2140 | |
bf4f0599 | 2141 | #ifdef CONFIG_MEMCG_KMEM |
56751146 | 2142 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock); |
bf4f0599 RG |
2143 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
2144 | struct mem_cgroup *root_memcg); | |
a8c49af3 | 2145 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages); |
bf4f0599 RG |
2146 | |
2147 | #else | |
56751146 | 2148 | static inline struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 2149 | { |
56751146 | 2150 | return NULL; |
bf4f0599 RG |
2151 | } |
2152 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2153 | struct mem_cgroup *root_memcg) | |
2154 | { | |
2155 | return false; | |
2156 | } | |
a8c49af3 YA |
2157 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
2158 | { | |
2159 | } | |
bf4f0599 RG |
2160 | #endif |
2161 | ||
a0956d54 SS |
2162 | /** |
2163 | * consume_stock: Try to consume stocked charge on this cpu. | |
2164 | * @memcg: memcg to consume from. | |
2165 | * @nr_pages: how many pages to charge. | |
2166 | * | |
2167 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2168 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2169 | * service an allocation will refill the stock. | |
2170 | * | |
2171 | * returns true if successful, false otherwise. | |
cdec2e42 | 2172 | */ |
a0956d54 | 2173 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2174 | { |
2175 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2176 | unsigned long flags; |
3e32cb2e | 2177 | bool ret = false; |
cdec2e42 | 2178 | |
a983b5eb | 2179 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2180 | return ret; |
a0956d54 | 2181 | |
56751146 | 2182 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2183 | |
2184 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2185 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2186 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2187 | ret = true; |
2188 | } | |
db2ba40c | 2189 | |
56751146 | 2190 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
db2ba40c | 2191 | |
cdec2e42 KH |
2192 | return ret; |
2193 | } | |
2194 | ||
2195 | /* | |
3e32cb2e | 2196 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2197 | */ |
2198 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2199 | { | |
2200 | struct mem_cgroup *old = stock->cached; | |
2201 | ||
1a3e1f40 JW |
2202 | if (!old) |
2203 | return; | |
2204 | ||
11c9ea4e | 2205 | if (stock->nr_pages) { |
3e32cb2e | 2206 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2207 | if (do_memsw_account()) |
3e32cb2e | 2208 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2209 | stock->nr_pages = 0; |
cdec2e42 | 2210 | } |
1a3e1f40 JW |
2211 | |
2212 | css_put(&old->css); | |
cdec2e42 | 2213 | stock->cached = NULL; |
cdec2e42 KH |
2214 | } |
2215 | ||
cdec2e42 KH |
2216 | static void drain_local_stock(struct work_struct *dummy) |
2217 | { | |
db2ba40c | 2218 | struct memcg_stock_pcp *stock; |
56751146 | 2219 | struct obj_cgroup *old = NULL; |
db2ba40c JW |
2220 | unsigned long flags; |
2221 | ||
72f0184c | 2222 | /* |
5c49cf9a MH |
2223 | * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs. |
2224 | * drain_stock races is that we always operate on local CPU stock | |
2225 | * here with IRQ disabled | |
72f0184c | 2226 | */ |
56751146 | 2227 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2228 | |
2229 | stock = this_cpu_ptr(&memcg_stock); | |
56751146 | 2230 | old = drain_obj_stock(stock); |
cdec2e42 | 2231 | drain_stock(stock); |
26fe6168 | 2232 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c | 2233 | |
56751146 SAS |
2234 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
2235 | if (old) | |
2236 | obj_cgroup_put(old); | |
cdec2e42 KH |
2237 | } |
2238 | ||
2239 | /* | |
3e32cb2e | 2240 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2241 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2242 | */ |
af9a3b69 | 2243 | static void __refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2244 | { |
db2ba40c | 2245 | struct memcg_stock_pcp *stock; |
cdec2e42 | 2246 | |
db2ba40c | 2247 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2248 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2249 | drain_stock(stock); |
1a3e1f40 | 2250 | css_get(&memcg->css); |
c0ff4b85 | 2251 | stock->cached = memcg; |
cdec2e42 | 2252 | } |
11c9ea4e | 2253 | stock->nr_pages += nr_pages; |
db2ba40c | 2254 | |
a983b5eb | 2255 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 | 2256 | drain_stock(stock); |
af9a3b69 JW |
2257 | } |
2258 | ||
2259 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) | |
2260 | { | |
2261 | unsigned long flags; | |
475d0487 | 2262 | |
56751146 | 2263 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
af9a3b69 | 2264 | __refill_stock(memcg, nr_pages); |
56751146 | 2265 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
cdec2e42 KH |
2266 | } |
2267 | ||
2268 | /* | |
c0ff4b85 | 2269 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2270 | * of the hierarchy under it. |
cdec2e42 | 2271 | */ |
6d3d6aa2 | 2272 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2273 | { |
26fe6168 | 2274 | int cpu, curcpu; |
d38144b7 | 2275 | |
6d3d6aa2 JW |
2276 | /* If someone's already draining, avoid adding running more workers. */ |
2277 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2278 | return; | |
72f0184c MH |
2279 | /* |
2280 | * Notify other cpus that system-wide "drain" is running | |
2281 | * We do not care about races with the cpu hotplug because cpu down | |
2282 | * as well as workers from this path always operate on the local | |
2283 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2284 | */ | |
0790ed62 SAS |
2285 | migrate_disable(); |
2286 | curcpu = smp_processor_id(); | |
cdec2e42 KH |
2287 | for_each_online_cpu(cpu) { |
2288 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2289 | struct mem_cgroup *memcg; |
e1a366be | 2290 | bool flush = false; |
26fe6168 | 2291 | |
e1a366be | 2292 | rcu_read_lock(); |
c0ff4b85 | 2293 | memcg = stock->cached; |
e1a366be RG |
2294 | if (memcg && stock->nr_pages && |
2295 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2296 | flush = true; | |
27fb0956 | 2297 | else if (obj_stock_flush_required(stock, root_memcg)) |
bf4f0599 | 2298 | flush = true; |
e1a366be RG |
2299 | rcu_read_unlock(); |
2300 | ||
2301 | if (flush && | |
2302 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2303 | if (cpu == curcpu) |
2304 | drain_local_stock(&stock->work); | |
2305 | else | |
2306 | schedule_work_on(cpu, &stock->work); | |
2307 | } | |
cdec2e42 | 2308 | } |
0790ed62 | 2309 | migrate_enable(); |
9f50fad6 | 2310 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2311 | } |
2312 | ||
2cd21c89 JW |
2313 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
2314 | { | |
2315 | struct memcg_stock_pcp *stock; | |
a3d4c05a | 2316 | |
2cd21c89 JW |
2317 | stock = &per_cpu(memcg_stock, cpu); |
2318 | drain_stock(stock); | |
a3d4c05a | 2319 | |
308167fc | 2320 | return 0; |
cdec2e42 KH |
2321 | } |
2322 | ||
b3ff9291 CD |
2323 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2324 | unsigned int nr_pages, | |
2325 | gfp_t gfp_mask) | |
f7e1cb6e | 2326 | { |
b3ff9291 CD |
2327 | unsigned long nr_reclaimed = 0; |
2328 | ||
f7e1cb6e | 2329 | do { |
e22c6ed9 JW |
2330 | unsigned long pflags; |
2331 | ||
d1663a90 JK |
2332 | if (page_counter_read(&memcg->memory) <= |
2333 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2334 | continue; |
e22c6ed9 | 2335 | |
e27be240 | 2336 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2337 | |
2338 | psi_memstall_enter(&pflags); | |
b3ff9291 | 2339 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
73b73bac YA |
2340 | gfp_mask, |
2341 | MEMCG_RECLAIM_MAY_SWAP); | |
e22c6ed9 | 2342 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2343 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2344 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2345 | |
2346 | return nr_reclaimed; | |
f7e1cb6e JW |
2347 | } |
2348 | ||
2349 | static void high_work_func(struct work_struct *work) | |
2350 | { | |
2351 | struct mem_cgroup *memcg; | |
2352 | ||
2353 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2354 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2355 | } |
2356 | ||
0e4b01df CD |
2357 | /* |
2358 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2359 | * enough to still cause a significant slowdown in most cases, while still | |
2360 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2361 | */ | |
2362 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2363 | ||
2364 | /* | |
2365 | * When calculating the delay, we use these either side of the exponentiation to | |
2366 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2367 | * below. | |
2368 | * | |
2369 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2370 | * overage ratio to a delay. | |
ac5ddd0f | 2371 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2372 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2373 | * to produce a reasonable delay curve. | |
2374 | * | |
2375 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2376 | * reasonable delay curve compared to precision-adjusted overage, not | |
2377 | * penalising heavily at first, but still making sure that growth beyond the | |
2378 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2379 | * example, with a high of 100 megabytes: | |
2380 | * | |
2381 | * +-------+------------------------+ | |
2382 | * | usage | time to allocate in ms | | |
2383 | * +-------+------------------------+ | |
2384 | * | 100M | 0 | | |
2385 | * | 101M | 6 | | |
2386 | * | 102M | 25 | | |
2387 | * | 103M | 57 | | |
2388 | * | 104M | 102 | | |
2389 | * | 105M | 159 | | |
2390 | * | 106M | 230 | | |
2391 | * | 107M | 313 | | |
2392 | * | 108M | 409 | | |
2393 | * | 109M | 518 | | |
2394 | * | 110M | 639 | | |
2395 | * | 111M | 774 | | |
2396 | * | 112M | 921 | | |
2397 | * | 113M | 1081 | | |
2398 | * | 114M | 1254 | | |
2399 | * | 115M | 1439 | | |
2400 | * | 116M | 1638 | | |
2401 | * | 117M | 1849 | | |
2402 | * | 118M | 2000 | | |
2403 | * | 119M | 2000 | | |
2404 | * | 120M | 2000 | | |
2405 | * +-------+------------------------+ | |
2406 | */ | |
2407 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2408 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2409 | ||
8a5dbc65 | 2410 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2411 | { |
8a5dbc65 | 2412 | u64 overage; |
b23afb93 | 2413 | |
8a5dbc65 JK |
2414 | if (usage <= high) |
2415 | return 0; | |
e26733e0 | 2416 | |
8a5dbc65 JK |
2417 | /* |
2418 | * Prevent division by 0 in overage calculation by acting as if | |
2419 | * it was a threshold of 1 page | |
2420 | */ | |
2421 | high = max(high, 1UL); | |
9b8b1754 | 2422 | |
8a5dbc65 JK |
2423 | overage = usage - high; |
2424 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2425 | return div64_u64(overage, high); | |
2426 | } | |
e26733e0 | 2427 | |
8a5dbc65 JK |
2428 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2429 | { | |
2430 | u64 overage, max_overage = 0; | |
e26733e0 | 2431 | |
8a5dbc65 JK |
2432 | do { |
2433 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2434 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2435 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2436 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2437 | !mem_cgroup_is_root(memcg)); | |
2438 | ||
8a5dbc65 JK |
2439 | return max_overage; |
2440 | } | |
2441 | ||
4b82ab4f JK |
2442 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2443 | { | |
2444 | u64 overage, max_overage = 0; | |
2445 | ||
2446 | do { | |
2447 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2448 | READ_ONCE(memcg->swap.high)); | |
2449 | if (overage) | |
2450 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2451 | max_overage = max(overage, max_overage); | |
2452 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2453 | !mem_cgroup_is_root(memcg)); | |
2454 | ||
2455 | return max_overage; | |
2456 | } | |
2457 | ||
8a5dbc65 JK |
2458 | /* |
2459 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2460 | * is exceeding its memory.high by checking both it and its ancestors. | |
2461 | */ | |
2462 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2463 | unsigned int nr_pages, | |
2464 | u64 max_overage) | |
2465 | { | |
2466 | unsigned long penalty_jiffies; | |
2467 | ||
e26733e0 CD |
2468 | if (!max_overage) |
2469 | return 0; | |
0e4b01df CD |
2470 | |
2471 | /* | |
0e4b01df CD |
2472 | * We use overage compared to memory.high to calculate the number of |
2473 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2474 | * fairly lenient on small overages, and increasingly harsh when the | |
2475 | * memcg in question makes it clear that it has no intention of stopping | |
2476 | * its crazy behaviour, so we exponentially increase the delay based on | |
2477 | * overage amount. | |
2478 | */ | |
e26733e0 CD |
2479 | penalty_jiffies = max_overage * max_overage * HZ; |
2480 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2481 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2482 | |
2483 | /* | |
2484 | * Factor in the task's own contribution to the overage, such that four | |
2485 | * N-sized allocations are throttled approximately the same as one | |
2486 | * 4N-sized allocation. | |
2487 | * | |
2488 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2489 | * larger the current charge patch is than that. | |
2490 | */ | |
ff144e69 | 2491 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2492 | } |
2493 | ||
2494 | /* | |
2495 | * Scheduled by try_charge() to be executed from the userland return path | |
2496 | * and reclaims memory over the high limit. | |
2497 | */ | |
2498 | void mem_cgroup_handle_over_high(void) | |
2499 | { | |
2500 | unsigned long penalty_jiffies; | |
2501 | unsigned long pflags; | |
b3ff9291 | 2502 | unsigned long nr_reclaimed; |
e26733e0 | 2503 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2504 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2505 | struct mem_cgroup *memcg; |
b3ff9291 | 2506 | bool in_retry = false; |
e26733e0 CD |
2507 | |
2508 | if (likely(!nr_pages)) | |
2509 | return; | |
2510 | ||
2511 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2512 | current->memcg_nr_pages_over_high = 0; |
2513 | ||
b3ff9291 CD |
2514 | retry_reclaim: |
2515 | /* | |
2516 | * The allocating task should reclaim at least the batch size, but for | |
2517 | * subsequent retries we only want to do what's necessary to prevent oom | |
2518 | * or breaching resource isolation. | |
2519 | * | |
2520 | * This is distinct from memory.max or page allocator behaviour because | |
2521 | * memory.high is currently batched, whereas memory.max and the page | |
2522 | * allocator run every time an allocation is made. | |
2523 | */ | |
2524 | nr_reclaimed = reclaim_high(memcg, | |
2525 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2526 | GFP_KERNEL); | |
2527 | ||
e26733e0 CD |
2528 | /* |
2529 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2530 | * allocators proactively to slow down excessive growth. | |
2531 | */ | |
8a5dbc65 JK |
2532 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2533 | mem_find_max_overage(memcg)); | |
0e4b01df | 2534 | |
4b82ab4f JK |
2535 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2536 | swap_find_max_overage(memcg)); | |
2537 | ||
ff144e69 JK |
2538 | /* |
2539 | * Clamp the max delay per usermode return so as to still keep the | |
2540 | * application moving forwards and also permit diagnostics, albeit | |
2541 | * extremely slowly. | |
2542 | */ | |
2543 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2544 | ||
0e4b01df CD |
2545 | /* |
2546 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2547 | * that it's not even worth doing, in an attempt to be nice to those who | |
2548 | * go only a small amount over their memory.high value and maybe haven't | |
2549 | * been aggressively reclaimed enough yet. | |
2550 | */ | |
2551 | if (penalty_jiffies <= HZ / 100) | |
2552 | goto out; | |
2553 | ||
b3ff9291 CD |
2554 | /* |
2555 | * If reclaim is making forward progress but we're still over | |
2556 | * memory.high, we want to encourage that rather than doing allocator | |
2557 | * throttling. | |
2558 | */ | |
2559 | if (nr_reclaimed || nr_retries--) { | |
2560 | in_retry = true; | |
2561 | goto retry_reclaim; | |
2562 | } | |
2563 | ||
0e4b01df CD |
2564 | /* |
2565 | * If we exit early, we're guaranteed to die (since | |
2566 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2567 | * need to account for any ill-begotten jiffies to pay them off later. | |
2568 | */ | |
2569 | psi_memstall_enter(&pflags); | |
2570 | schedule_timeout_killable(penalty_jiffies); | |
2571 | psi_memstall_leave(&pflags); | |
2572 | ||
2573 | out: | |
2574 | css_put(&memcg->css); | |
b23afb93 TH |
2575 | } |
2576 | ||
c5c8b16b MS |
2577 | static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2578 | unsigned int nr_pages) | |
8a9f3ccd | 2579 | { |
a983b5eb | 2580 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2581 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2582 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2583 | struct page_counter *counter; |
6539cc05 | 2584 | unsigned long nr_reclaimed; |
a4ebf1b6 | 2585 | bool passed_oom = false; |
73b73bac | 2586 | unsigned int reclaim_options = MEMCG_RECLAIM_MAY_SWAP; |
b70a2a21 | 2587 | bool drained = false; |
d6e103a7 | 2588 | bool raised_max_event = false; |
e22c6ed9 | 2589 | unsigned long pflags; |
a636b327 | 2590 | |
6539cc05 | 2591 | retry: |
b6b6cc72 | 2592 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2593 | return 0; |
8a9f3ccd | 2594 | |
7941d214 | 2595 | if (!do_memsw_account() || |
6071ca52 JW |
2596 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2597 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2598 | goto done_restock; |
7941d214 | 2599 | if (do_memsw_account()) |
3e32cb2e JW |
2600 | page_counter_uncharge(&memcg->memsw, batch); |
2601 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2602 | } else { |
3e32cb2e | 2603 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
73b73bac | 2604 | reclaim_options &= ~MEMCG_RECLAIM_MAY_SWAP; |
3fbe7244 | 2605 | } |
7a81b88c | 2606 | |
6539cc05 JW |
2607 | if (batch > nr_pages) { |
2608 | batch = nr_pages; | |
2609 | goto retry; | |
2610 | } | |
6d61ef40 | 2611 | |
89a28483 JW |
2612 | /* |
2613 | * Prevent unbounded recursion when reclaim operations need to | |
2614 | * allocate memory. This might exceed the limits temporarily, | |
2615 | * but we prefer facilitating memory reclaim and getting back | |
2616 | * under the limit over triggering OOM kills in these cases. | |
2617 | */ | |
2618 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2619 | goto force; | |
2620 | ||
06b078fc JW |
2621 | if (unlikely(task_in_memcg_oom(current))) |
2622 | goto nomem; | |
2623 | ||
d0164adc | 2624 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2625 | goto nomem; |
4b534334 | 2626 | |
e27be240 | 2627 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
d6e103a7 | 2628 | raised_max_event = true; |
241994ed | 2629 | |
e22c6ed9 | 2630 | psi_memstall_enter(&pflags); |
b70a2a21 | 2631 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
73b73bac | 2632 | gfp_mask, reclaim_options); |
e22c6ed9 | 2633 | psi_memstall_leave(&pflags); |
6539cc05 | 2634 | |
61e02c74 | 2635 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2636 | goto retry; |
28c34c29 | 2637 | |
b70a2a21 | 2638 | if (!drained) { |
6d3d6aa2 | 2639 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2640 | drained = true; |
2641 | goto retry; | |
2642 | } | |
2643 | ||
28c34c29 JW |
2644 | if (gfp_mask & __GFP_NORETRY) |
2645 | goto nomem; | |
6539cc05 JW |
2646 | /* |
2647 | * Even though the limit is exceeded at this point, reclaim | |
2648 | * may have been able to free some pages. Retry the charge | |
2649 | * before killing the task. | |
2650 | * | |
2651 | * Only for regular pages, though: huge pages are rather | |
2652 | * unlikely to succeed so close to the limit, and we fall back | |
2653 | * to regular pages anyway in case of failure. | |
2654 | */ | |
61e02c74 | 2655 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2656 | goto retry; |
2657 | /* | |
2658 | * At task move, charge accounts can be doubly counted. So, it's | |
2659 | * better to wait until the end of task_move if something is going on. | |
2660 | */ | |
2661 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2662 | goto retry; | |
2663 | ||
9b130619 JW |
2664 | if (nr_retries--) |
2665 | goto retry; | |
2666 | ||
38d38493 | 2667 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2668 | goto nomem; |
2669 | ||
a4ebf1b6 VA |
2670 | /* Avoid endless loop for tasks bypassed by the oom killer */ |
2671 | if (passed_oom && task_is_dying()) | |
2672 | goto nomem; | |
6539cc05 | 2673 | |
29ef680a MH |
2674 | /* |
2675 | * keep retrying as long as the memcg oom killer is able to make | |
2676 | * a forward progress or bypass the charge if the oom killer | |
2677 | * couldn't make any progress. | |
2678 | */ | |
becdf89d SB |
2679 | if (mem_cgroup_oom(mem_over_limit, gfp_mask, |
2680 | get_order(nr_pages * PAGE_SIZE))) { | |
a4ebf1b6 | 2681 | passed_oom = true; |
d977aa93 | 2682 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a | 2683 | goto retry; |
29ef680a | 2684 | } |
7a81b88c | 2685 | nomem: |
1461e8c2 SB |
2686 | /* |
2687 | * Memcg doesn't have a dedicated reserve for atomic | |
2688 | * allocations. But like the global atomic pool, we need to | |
2689 | * put the burden of reclaim on regular allocation requests | |
2690 | * and let these go through as privileged allocations. | |
2691 | */ | |
2692 | if (!(gfp_mask & (__GFP_NOFAIL | __GFP_HIGH))) | |
3168ecbe | 2693 | return -ENOMEM; |
10d53c74 | 2694 | force: |
d6e103a7 RG |
2695 | /* |
2696 | * If the allocation has to be enforced, don't forget to raise | |
2697 | * a MEMCG_MAX event. | |
2698 | */ | |
2699 | if (!raised_max_event) | |
2700 | memcg_memory_event(mem_over_limit, MEMCG_MAX); | |
2701 | ||
10d53c74 TH |
2702 | /* |
2703 | * The allocation either can't fail or will lead to more memory | |
2704 | * being freed very soon. Allow memory usage go over the limit | |
2705 | * temporarily by force charging it. | |
2706 | */ | |
2707 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2708 | if (do_memsw_account()) |
10d53c74 | 2709 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2710 | |
2711 | return 0; | |
6539cc05 JW |
2712 | |
2713 | done_restock: | |
2714 | if (batch > nr_pages) | |
2715 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2716 | |
241994ed | 2717 | /* |
b23afb93 TH |
2718 | * If the hierarchy is above the normal consumption range, schedule |
2719 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2720 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2721 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2722 | * not recorded as it most likely matches current's and won't | |
2723 | * change in the meantime. As high limit is checked again before | |
2724 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2725 | */ |
2726 | do { | |
4b82ab4f JK |
2727 | bool mem_high, swap_high; |
2728 | ||
2729 | mem_high = page_counter_read(&memcg->memory) > | |
2730 | READ_ONCE(memcg->memory.high); | |
2731 | swap_high = page_counter_read(&memcg->swap) > | |
2732 | READ_ONCE(memcg->swap.high); | |
2733 | ||
2734 | /* Don't bother a random interrupted task */ | |
086f694a | 2735 | if (!in_task()) { |
4b82ab4f | 2736 | if (mem_high) { |
f7e1cb6e JW |
2737 | schedule_work(&memcg->high_work); |
2738 | break; | |
2739 | } | |
4b82ab4f JK |
2740 | continue; |
2741 | } | |
2742 | ||
2743 | if (mem_high || swap_high) { | |
2744 | /* | |
2745 | * The allocating tasks in this cgroup will need to do | |
2746 | * reclaim or be throttled to prevent further growth | |
2747 | * of the memory or swap footprints. | |
2748 | * | |
2749 | * Target some best-effort fairness between the tasks, | |
2750 | * and distribute reclaim work and delay penalties | |
2751 | * based on how much each task is actually allocating. | |
2752 | */ | |
9516a18a | 2753 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2754 | set_notify_resume(current); |
2755 | break; | |
2756 | } | |
241994ed | 2757 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 | 2758 | |
c9afe31e SB |
2759 | if (current->memcg_nr_pages_over_high > MEMCG_CHARGE_BATCH && |
2760 | !(current->flags & PF_MEMALLOC) && | |
2761 | gfpflags_allow_blocking(gfp_mask)) { | |
2762 | mem_cgroup_handle_over_high(); | |
2763 | } | |
10d53c74 | 2764 | return 0; |
7a81b88c | 2765 | } |
8a9f3ccd | 2766 | |
c5c8b16b MS |
2767 | static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2768 | unsigned int nr_pages) | |
2769 | { | |
2770 | if (mem_cgroup_is_root(memcg)) | |
2771 | return 0; | |
2772 | ||
2773 | return try_charge_memcg(memcg, gfp_mask, nr_pages); | |
2774 | } | |
2775 | ||
58056f77 | 2776 | static inline void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2777 | { |
ce00a967 JW |
2778 | if (mem_cgroup_is_root(memcg)) |
2779 | return; | |
2780 | ||
3e32cb2e | 2781 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2782 | if (do_memsw_account()) |
3e32cb2e | 2783 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f KH |
2784 | } |
2785 | ||
118f2875 | 2786 | static void commit_charge(struct folio *folio, struct mem_cgroup *memcg) |
0a31bc97 | 2787 | { |
118f2875 | 2788 | VM_BUG_ON_FOLIO(folio_memcg(folio), folio); |
0a31bc97 | 2789 | /* |
a5eb011a | 2790 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2791 | * |
a0b5b414 JW |
2792 | * - the page lock |
2793 | * - LRU isolation | |
2794 | * - lock_page_memcg() | |
2795 | * - exclusive reference | |
0a31bc97 | 2796 | */ |
118f2875 | 2797 | folio->memcg_data = (unsigned long)memcg; |
7a81b88c | 2798 | } |
66e1707b | 2799 | |
84c07d11 | 2800 | #ifdef CONFIG_MEMCG_KMEM |
41eb5df1 WL |
2801 | /* |
2802 | * The allocated objcg pointers array is not accounted directly. | |
2803 | * Moreover, it should not come from DMA buffer and is not readily | |
2804 | * reclaimable. So those GFP bits should be masked off. | |
2805 | */ | |
2806 | #define OBJCGS_CLEAR_MASK (__GFP_DMA | __GFP_RECLAIMABLE | __GFP_ACCOUNT) | |
2807 | ||
a7ebf564 WL |
2808 | /* |
2809 | * mod_objcg_mlstate() may be called with irq enabled, so | |
2810 | * mod_memcg_lruvec_state() should be used. | |
2811 | */ | |
2812 | static inline void mod_objcg_mlstate(struct obj_cgroup *objcg, | |
2813 | struct pglist_data *pgdat, | |
2814 | enum node_stat_item idx, int nr) | |
2815 | { | |
2816 | struct mem_cgroup *memcg; | |
2817 | struct lruvec *lruvec; | |
2818 | ||
2819 | rcu_read_lock(); | |
2820 | memcg = obj_cgroup_memcg(objcg); | |
2821 | lruvec = mem_cgroup_lruvec(memcg, pgdat); | |
2822 | mod_memcg_lruvec_state(lruvec, idx, nr); | |
2823 | rcu_read_unlock(); | |
2824 | } | |
2825 | ||
4b5f8d9a VB |
2826 | int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s, |
2827 | gfp_t gfp, bool new_slab) | |
10befea9 | 2828 | { |
4b5f8d9a | 2829 | unsigned int objects = objs_per_slab(s, slab); |
2e9bd483 | 2830 | unsigned long memcg_data; |
10befea9 RG |
2831 | void *vec; |
2832 | ||
41eb5df1 | 2833 | gfp &= ~OBJCGS_CLEAR_MASK; |
10befea9 | 2834 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, |
4b5f8d9a | 2835 | slab_nid(slab)); |
10befea9 RG |
2836 | if (!vec) |
2837 | return -ENOMEM; | |
2838 | ||
2e9bd483 | 2839 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
4b5f8d9a | 2840 | if (new_slab) { |
2e9bd483 | 2841 | /* |
4b5f8d9a VB |
2842 | * If the slab is brand new and nobody can yet access its |
2843 | * memcg_data, no synchronization is required and memcg_data can | |
2844 | * be simply assigned. | |
2e9bd483 | 2845 | */ |
4b5f8d9a VB |
2846 | slab->memcg_data = memcg_data; |
2847 | } else if (cmpxchg(&slab->memcg_data, 0, memcg_data)) { | |
2e9bd483 | 2848 | /* |
4b5f8d9a VB |
2849 | * If the slab is already in use, somebody can allocate and |
2850 | * assign obj_cgroups in parallel. In this case the existing | |
2e9bd483 RG |
2851 | * objcg vector should be reused. |
2852 | */ | |
10befea9 | 2853 | kfree(vec); |
2e9bd483 RG |
2854 | return 0; |
2855 | } | |
10befea9 | 2856 | |
2e9bd483 | 2857 | kmemleak_not_leak(vec); |
10befea9 RG |
2858 | return 0; |
2859 | } | |
2860 | ||
fc4db90f RG |
2861 | static __always_inline |
2862 | struct mem_cgroup *mem_cgroup_from_obj_folio(struct folio *folio, void *p) | |
8380ce47 | 2863 | { |
8380ce47 | 2864 | /* |
9855609b RG |
2865 | * Slab objects are accounted individually, not per-page. |
2866 | * Memcg membership data for each individual object is saved in | |
4b5f8d9a | 2867 | * slab->memcg_data. |
8380ce47 | 2868 | */ |
4b5f8d9a VB |
2869 | if (folio_test_slab(folio)) { |
2870 | struct obj_cgroup **objcgs; | |
2871 | struct slab *slab; | |
9855609b RG |
2872 | unsigned int off; |
2873 | ||
4b5f8d9a VB |
2874 | slab = folio_slab(folio); |
2875 | objcgs = slab_objcgs(slab); | |
2876 | if (!objcgs) | |
2877 | return NULL; | |
2878 | ||
2879 | off = obj_to_index(slab->slab_cache, slab, p); | |
2880 | if (objcgs[off]) | |
2881 | return obj_cgroup_memcg(objcgs[off]); | |
10befea9 RG |
2882 | |
2883 | return NULL; | |
9855609b | 2884 | } |
8380ce47 | 2885 | |
bcfe06bf | 2886 | /* |
4b5f8d9a VB |
2887 | * page_memcg_check() is used here, because in theory we can encounter |
2888 | * a folio where the slab flag has been cleared already, but | |
2889 | * slab->memcg_data has not been freed yet | |
bcfe06bf RG |
2890 | * page_memcg_check(page) will guarantee that a proper memory |
2891 | * cgroup pointer or NULL will be returned. | |
2892 | */ | |
4b5f8d9a | 2893 | return page_memcg_check(folio_page(folio, 0)); |
8380ce47 RG |
2894 | } |
2895 | ||
fc4db90f RG |
2896 | /* |
2897 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2898 | * | |
2899 | * A passed kernel object can be a slab object, vmalloc object or a generic | |
2900 | * kernel page, so different mechanisms for getting the memory cgroup pointer | |
2901 | * should be used. | |
2902 | * | |
2903 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
2904 | * can not know for sure how the kernel object is implemented. | |
2905 | * mem_cgroup_from_obj() can be safely used in such cases. | |
2906 | * | |
2907 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2908 | * cgroup_mutex, etc. | |
2909 | */ | |
2910 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2911 | { | |
2912 | struct folio *folio; | |
2913 | ||
2914 | if (mem_cgroup_disabled()) | |
2915 | return NULL; | |
2916 | ||
2917 | if (unlikely(is_vmalloc_addr(p))) | |
2918 | folio = page_folio(vmalloc_to_page(p)); | |
2919 | else | |
2920 | folio = virt_to_folio(p); | |
2921 | ||
2922 | return mem_cgroup_from_obj_folio(folio, p); | |
2923 | } | |
2924 | ||
2925 | /* | |
2926 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2927 | * Similar to mem_cgroup_from_obj(), but faster and not suitable for objects, | |
2928 | * allocated using vmalloc(). | |
2929 | * | |
2930 | * A passed kernel object must be a slab object or a generic kernel page. | |
2931 | * | |
2932 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2933 | * cgroup_mutex, etc. | |
2934 | */ | |
2935 | struct mem_cgroup *mem_cgroup_from_slab_obj(void *p) | |
2936 | { | |
2937 | if (mem_cgroup_disabled()) | |
2938 | return NULL; | |
2939 | ||
2940 | return mem_cgroup_from_obj_folio(virt_to_folio(p), p); | |
2941 | } | |
2942 | ||
f4840ccf JW |
2943 | static struct obj_cgroup *__get_obj_cgroup_from_memcg(struct mem_cgroup *memcg) |
2944 | { | |
2945 | struct obj_cgroup *objcg = NULL; | |
2946 | ||
2947 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2948 | objcg = rcu_dereference(memcg->objcg); | |
2949 | if (objcg && obj_cgroup_tryget(objcg)) | |
2950 | break; | |
2951 | objcg = NULL; | |
2952 | } | |
2953 | return objcg; | |
2954 | } | |
2955 | ||
bf4f0599 RG |
2956 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2957 | { | |
2958 | struct obj_cgroup *objcg = NULL; | |
2959 | struct mem_cgroup *memcg; | |
2960 | ||
279c3393 RG |
2961 | if (memcg_kmem_bypass()) |
2962 | return NULL; | |
2963 | ||
bf4f0599 | 2964 | rcu_read_lock(); |
37d5985c RG |
2965 | if (unlikely(active_memcg())) |
2966 | memcg = active_memcg(); | |
bf4f0599 RG |
2967 | else |
2968 | memcg = mem_cgroup_from_task(current); | |
f4840ccf | 2969 | objcg = __get_obj_cgroup_from_memcg(memcg); |
bf4f0599 | 2970 | rcu_read_unlock(); |
f4840ccf JW |
2971 | return objcg; |
2972 | } | |
2973 | ||
2974 | struct obj_cgroup *get_obj_cgroup_from_page(struct page *page) | |
2975 | { | |
2976 | struct obj_cgroup *objcg; | |
2977 | ||
2978 | if (!memcg_kmem_enabled() || memcg_kmem_bypass()) | |
2979 | return NULL; | |
2980 | ||
2981 | if (PageMemcgKmem(page)) { | |
2982 | objcg = __folio_objcg(page_folio(page)); | |
2983 | obj_cgroup_get(objcg); | |
2984 | } else { | |
2985 | struct mem_cgroup *memcg; | |
bf4f0599 | 2986 | |
f4840ccf JW |
2987 | rcu_read_lock(); |
2988 | memcg = __folio_memcg(page_folio(page)); | |
2989 | if (memcg) | |
2990 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
2991 | else | |
2992 | objcg = NULL; | |
2993 | rcu_read_unlock(); | |
2994 | } | |
bf4f0599 RG |
2995 | return objcg; |
2996 | } | |
2997 | ||
a8c49af3 YA |
2998 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
2999 | { | |
3000 | mod_memcg_state(memcg, MEMCG_KMEM, nr_pages); | |
3001 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
3002 | if (nr_pages > 0) | |
3003 | page_counter_charge(&memcg->kmem, nr_pages); | |
3004 | else | |
3005 | page_counter_uncharge(&memcg->kmem, -nr_pages); | |
3006 | } | |
3007 | } | |
3008 | ||
3009 | ||
f1286fae MS |
3010 | /* |
3011 | * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg | |
3012 | * @objcg: object cgroup to uncharge | |
3013 | * @nr_pages: number of pages to uncharge | |
3014 | */ | |
e74d2259 MS |
3015 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
3016 | unsigned int nr_pages) | |
3017 | { | |
3018 | struct mem_cgroup *memcg; | |
3019 | ||
3020 | memcg = get_mem_cgroup_from_objcg(objcg); | |
e74d2259 | 3021 | |
a8c49af3 | 3022 | memcg_account_kmem(memcg, -nr_pages); |
f1286fae | 3023 | refill_stock(memcg, nr_pages); |
e74d2259 | 3024 | |
e74d2259 | 3025 | css_put(&memcg->css); |
e74d2259 MS |
3026 | } |
3027 | ||
f1286fae MS |
3028 | /* |
3029 | * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg | |
3030 | * @objcg: object cgroup to charge | |
45264778 | 3031 | * @gfp: reclaim mode |
92d0510c | 3032 | * @nr_pages: number of pages to charge |
45264778 VD |
3033 | * |
3034 | * Returns 0 on success, an error code on failure. | |
3035 | */ | |
f1286fae MS |
3036 | static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp, |
3037 | unsigned int nr_pages) | |
7ae1e1d0 | 3038 | { |
f1286fae | 3039 | struct mem_cgroup *memcg; |
7ae1e1d0 GC |
3040 | int ret; |
3041 | ||
f1286fae MS |
3042 | memcg = get_mem_cgroup_from_objcg(objcg); |
3043 | ||
c5c8b16b | 3044 | ret = try_charge_memcg(memcg, gfp, nr_pages); |
52c29b04 | 3045 | if (ret) |
f1286fae | 3046 | goto out; |
52c29b04 | 3047 | |
a8c49af3 | 3048 | memcg_account_kmem(memcg, nr_pages); |
f1286fae MS |
3049 | out: |
3050 | css_put(&memcg->css); | |
4b13f64d | 3051 | |
f1286fae | 3052 | return ret; |
4b13f64d RG |
3053 | } |
3054 | ||
45264778 | 3055 | /** |
f4b00eab | 3056 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3057 | * @page: page to charge |
3058 | * @gfp: reclaim mode | |
3059 | * @order: allocation order | |
3060 | * | |
3061 | * Returns 0 on success, an error code on failure. | |
3062 | */ | |
f4b00eab | 3063 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3064 | { |
b4e0b68f | 3065 | struct obj_cgroup *objcg; |
fcff7d7e | 3066 | int ret = 0; |
7ae1e1d0 | 3067 | |
b4e0b68f MS |
3068 | objcg = get_obj_cgroup_from_current(); |
3069 | if (objcg) { | |
3070 | ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order); | |
4d96ba35 | 3071 | if (!ret) { |
b4e0b68f | 3072 | page->memcg_data = (unsigned long)objcg | |
18b2db3b | 3073 | MEMCG_DATA_KMEM; |
1a3e1f40 | 3074 | return 0; |
4d96ba35 | 3075 | } |
b4e0b68f | 3076 | obj_cgroup_put(objcg); |
c4159a75 | 3077 | } |
d05e83a6 | 3078 | return ret; |
7ae1e1d0 | 3079 | } |
49a18eae | 3080 | |
45264778 | 3081 | /** |
f4b00eab | 3082 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3083 | * @page: page to uncharge |
3084 | * @order: allocation order | |
3085 | */ | |
f4b00eab | 3086 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3087 | { |
1b7e4464 | 3088 | struct folio *folio = page_folio(page); |
b4e0b68f | 3089 | struct obj_cgroup *objcg; |
f3ccb2c4 | 3090 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3091 | |
1b7e4464 | 3092 | if (!folio_memcg_kmem(folio)) |
7ae1e1d0 GC |
3093 | return; |
3094 | ||
1b7e4464 | 3095 | objcg = __folio_objcg(folio); |
b4e0b68f | 3096 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
1b7e4464 | 3097 | folio->memcg_data = 0; |
b4e0b68f | 3098 | obj_cgroup_put(objcg); |
60d3fd32 | 3099 | } |
bf4f0599 | 3100 | |
68ac5b3c WL |
3101 | void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
3102 | enum node_stat_item idx, int nr) | |
3103 | { | |
fead2b86 | 3104 | struct memcg_stock_pcp *stock; |
56751146 | 3105 | struct obj_cgroup *old = NULL; |
68ac5b3c WL |
3106 | unsigned long flags; |
3107 | int *bytes; | |
3108 | ||
56751146 | 3109 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3110 | stock = this_cpu_ptr(&memcg_stock); |
3111 | ||
68ac5b3c WL |
3112 | /* |
3113 | * Save vmstat data in stock and skip vmstat array update unless | |
3114 | * accumulating over a page of vmstat data or when pgdat or idx | |
3115 | * changes. | |
3116 | */ | |
3117 | if (stock->cached_objcg != objcg) { | |
56751146 | 3118 | old = drain_obj_stock(stock); |
68ac5b3c WL |
3119 | obj_cgroup_get(objcg); |
3120 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) | |
3121 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3122 | stock->cached_objcg = objcg; | |
3123 | stock->cached_pgdat = pgdat; | |
3124 | } else if (stock->cached_pgdat != pgdat) { | |
3125 | /* Flush the existing cached vmstat data */ | |
7fa0dacb WL |
3126 | struct pglist_data *oldpg = stock->cached_pgdat; |
3127 | ||
68ac5b3c | 3128 | if (stock->nr_slab_reclaimable_b) { |
7fa0dacb | 3129 | mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B, |
68ac5b3c WL |
3130 | stock->nr_slab_reclaimable_b); |
3131 | stock->nr_slab_reclaimable_b = 0; | |
3132 | } | |
3133 | if (stock->nr_slab_unreclaimable_b) { | |
7fa0dacb | 3134 | mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B, |
68ac5b3c WL |
3135 | stock->nr_slab_unreclaimable_b); |
3136 | stock->nr_slab_unreclaimable_b = 0; | |
3137 | } | |
3138 | stock->cached_pgdat = pgdat; | |
3139 | } | |
3140 | ||
3141 | bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b | |
3142 | : &stock->nr_slab_unreclaimable_b; | |
3143 | /* | |
3144 | * Even for large object >= PAGE_SIZE, the vmstat data will still be | |
3145 | * cached locally at least once before pushing it out. | |
3146 | */ | |
3147 | if (!*bytes) { | |
3148 | *bytes = nr; | |
3149 | nr = 0; | |
3150 | } else { | |
3151 | *bytes += nr; | |
3152 | if (abs(*bytes) > PAGE_SIZE) { | |
3153 | nr = *bytes; | |
3154 | *bytes = 0; | |
3155 | } else { | |
3156 | nr = 0; | |
3157 | } | |
3158 | } | |
3159 | if (nr) | |
3160 | mod_objcg_mlstate(objcg, pgdat, idx, nr); | |
3161 | ||
56751146 SAS |
3162 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
3163 | if (old) | |
3164 | obj_cgroup_put(old); | |
68ac5b3c WL |
3165 | } |
3166 | ||
bf4f0599 RG |
3167 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
3168 | { | |
fead2b86 | 3169 | struct memcg_stock_pcp *stock; |
bf4f0599 RG |
3170 | unsigned long flags; |
3171 | bool ret = false; | |
3172 | ||
56751146 | 3173 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3174 | |
3175 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 RG |
3176 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { |
3177 | stock->nr_bytes -= nr_bytes; | |
3178 | ret = true; | |
3179 | } | |
3180 | ||
56751146 | 3181 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
bf4f0599 RG |
3182 | |
3183 | return ret; | |
3184 | } | |
3185 | ||
56751146 | 3186 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 RG |
3187 | { |
3188 | struct obj_cgroup *old = stock->cached_objcg; | |
3189 | ||
3190 | if (!old) | |
56751146 | 3191 | return NULL; |
bf4f0599 RG |
3192 | |
3193 | if (stock->nr_bytes) { | |
3194 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3195 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3196 | ||
af9a3b69 JW |
3197 | if (nr_pages) { |
3198 | struct mem_cgroup *memcg; | |
3199 | ||
3200 | memcg = get_mem_cgroup_from_objcg(old); | |
3201 | ||
3202 | memcg_account_kmem(memcg, -nr_pages); | |
3203 | __refill_stock(memcg, nr_pages); | |
3204 | ||
3205 | css_put(&memcg->css); | |
3206 | } | |
bf4f0599 RG |
3207 | |
3208 | /* | |
3209 | * The leftover is flushed to the centralized per-memcg value. | |
3210 | * On the next attempt to refill obj stock it will be moved | |
3211 | * to a per-cpu stock (probably, on an other CPU), see | |
3212 | * refill_obj_stock(). | |
3213 | * | |
3214 | * How often it's flushed is a trade-off between the memory | |
3215 | * limit enforcement accuracy and potential CPU contention, | |
3216 | * so it might be changed in the future. | |
3217 | */ | |
3218 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3219 | stock->nr_bytes = 0; | |
3220 | } | |
3221 | ||
68ac5b3c WL |
3222 | /* |
3223 | * Flush the vmstat data in current stock | |
3224 | */ | |
3225 | if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) { | |
3226 | if (stock->nr_slab_reclaimable_b) { | |
3227 | mod_objcg_mlstate(old, stock->cached_pgdat, | |
3228 | NR_SLAB_RECLAIMABLE_B, | |
3229 | stock->nr_slab_reclaimable_b); | |
3230 | stock->nr_slab_reclaimable_b = 0; | |
3231 | } | |
3232 | if (stock->nr_slab_unreclaimable_b) { | |
3233 | mod_objcg_mlstate(old, stock->cached_pgdat, | |
3234 | NR_SLAB_UNRECLAIMABLE_B, | |
3235 | stock->nr_slab_unreclaimable_b); | |
3236 | stock->nr_slab_unreclaimable_b = 0; | |
3237 | } | |
3238 | stock->cached_pgdat = NULL; | |
3239 | } | |
3240 | ||
bf4f0599 | 3241 | stock->cached_objcg = NULL; |
56751146 SAS |
3242 | /* |
3243 | * The `old' objects needs to be released by the caller via | |
3244 | * obj_cgroup_put() outside of memcg_stock_pcp::stock_lock. | |
3245 | */ | |
3246 | return old; | |
bf4f0599 RG |
3247 | } |
3248 | ||
3249 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3250 | struct mem_cgroup *root_memcg) | |
3251 | { | |
3252 | struct mem_cgroup *memcg; | |
3253 | ||
fead2b86 MH |
3254 | if (stock->cached_objcg) { |
3255 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
bf4f0599 RG |
3256 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) |
3257 | return true; | |
3258 | } | |
3259 | ||
3260 | return false; | |
3261 | } | |
3262 | ||
5387c904 WL |
3263 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes, |
3264 | bool allow_uncharge) | |
bf4f0599 | 3265 | { |
fead2b86 | 3266 | struct memcg_stock_pcp *stock; |
56751146 | 3267 | struct obj_cgroup *old = NULL; |
bf4f0599 | 3268 | unsigned long flags; |
5387c904 | 3269 | unsigned int nr_pages = 0; |
bf4f0599 | 3270 | |
56751146 | 3271 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3272 | |
3273 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 3274 | if (stock->cached_objcg != objcg) { /* reset if necessary */ |
56751146 | 3275 | old = drain_obj_stock(stock); |
bf4f0599 RG |
3276 | obj_cgroup_get(objcg); |
3277 | stock->cached_objcg = objcg; | |
5387c904 WL |
3278 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) |
3279 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3280 | allow_uncharge = true; /* Allow uncharge when objcg changes */ | |
bf4f0599 RG |
3281 | } |
3282 | stock->nr_bytes += nr_bytes; | |
3283 | ||
5387c904 WL |
3284 | if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) { |
3285 | nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3286 | stock->nr_bytes &= (PAGE_SIZE - 1); | |
3287 | } | |
bf4f0599 | 3288 | |
56751146 SAS |
3289 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
3290 | if (old) | |
3291 | obj_cgroup_put(old); | |
5387c904 WL |
3292 | |
3293 | if (nr_pages) | |
3294 | obj_cgroup_uncharge_pages(objcg, nr_pages); | |
bf4f0599 RG |
3295 | } |
3296 | ||
3297 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3298 | { | |
bf4f0599 RG |
3299 | unsigned int nr_pages, nr_bytes; |
3300 | int ret; | |
3301 | ||
3302 | if (consume_obj_stock(objcg, size)) | |
3303 | return 0; | |
3304 | ||
3305 | /* | |
5387c904 | 3306 | * In theory, objcg->nr_charged_bytes can have enough |
bf4f0599 | 3307 | * pre-charged bytes to satisfy the allocation. However, |
5387c904 WL |
3308 | * flushing objcg->nr_charged_bytes requires two atomic |
3309 | * operations, and objcg->nr_charged_bytes can't be big. | |
3310 | * The shared objcg->nr_charged_bytes can also become a | |
3311 | * performance bottleneck if all tasks of the same memcg are | |
3312 | * trying to update it. So it's better to ignore it and try | |
3313 | * grab some new pages. The stock's nr_bytes will be flushed to | |
3314 | * objcg->nr_charged_bytes later on when objcg changes. | |
3315 | * | |
3316 | * The stock's nr_bytes may contain enough pre-charged bytes | |
3317 | * to allow one less page from being charged, but we can't rely | |
3318 | * on the pre-charged bytes not being changed outside of | |
3319 | * consume_obj_stock() or refill_obj_stock(). So ignore those | |
3320 | * pre-charged bytes as well when charging pages. To avoid a | |
3321 | * page uncharge right after a page charge, we set the | |
3322 | * allow_uncharge flag to false when calling refill_obj_stock() | |
3323 | * to temporarily allow the pre-charged bytes to exceed the page | |
3324 | * size limit. The maximum reachable value of the pre-charged | |
3325 | * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data | |
3326 | * race. | |
bf4f0599 | 3327 | */ |
bf4f0599 RG |
3328 | nr_pages = size >> PAGE_SHIFT; |
3329 | nr_bytes = size & (PAGE_SIZE - 1); | |
3330 | ||
3331 | if (nr_bytes) | |
3332 | nr_pages += 1; | |
3333 | ||
e74d2259 | 3334 | ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages); |
bf4f0599 | 3335 | if (!ret && nr_bytes) |
5387c904 | 3336 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false); |
bf4f0599 | 3337 | |
bf4f0599 RG |
3338 | return ret; |
3339 | } | |
3340 | ||
3341 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3342 | { | |
5387c904 | 3343 | refill_obj_stock(objcg, size, true); |
bf4f0599 RG |
3344 | } |
3345 | ||
84c07d11 | 3346 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3347 | |
ca3e0214 | 3348 | /* |
be6c8982 | 3349 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3350 | */ |
be6c8982 | 3351 | void split_page_memcg(struct page *head, unsigned int nr) |
ca3e0214 | 3352 | { |
1b7e4464 MWO |
3353 | struct folio *folio = page_folio(head); |
3354 | struct mem_cgroup *memcg = folio_memcg(folio); | |
e94c8a9c | 3355 | int i; |
ca3e0214 | 3356 | |
be6c8982 | 3357 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3358 | return; |
b070e65c | 3359 | |
be6c8982 | 3360 | for (i = 1; i < nr; i++) |
1b7e4464 | 3361 | folio_page(folio, i)->memcg_data = folio->memcg_data; |
b4e0b68f | 3362 | |
1b7e4464 MWO |
3363 | if (folio_memcg_kmem(folio)) |
3364 | obj_cgroup_get_many(__folio_objcg(folio), nr - 1); | |
b4e0b68f MS |
3365 | else |
3366 | css_get_many(&memcg->css, nr - 1); | |
ca3e0214 | 3367 | } |
ca3e0214 | 3368 | |
c255a458 | 3369 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3370 | /** |
3371 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3372 | * @entry: swap entry to be moved | |
3373 | * @from: mem_cgroup which the entry is moved from | |
3374 | * @to: mem_cgroup which the entry is moved to | |
3375 | * | |
3376 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3377 | * as the mem_cgroup's id of @from. | |
3378 | * | |
3379 | * Returns 0 on success, -EINVAL on failure. | |
3380 | * | |
3e32cb2e | 3381 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3382 | * both res and memsw, and called css_get(). |
3383 | */ | |
3384 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3385 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3386 | { |
3387 | unsigned short old_id, new_id; | |
3388 | ||
34c00c31 LZ |
3389 | old_id = mem_cgroup_id(from); |
3390 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3391 | |
3392 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3393 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3394 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3395 | return 0; |
3396 | } | |
3397 | return -EINVAL; | |
3398 | } | |
3399 | #else | |
3400 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3401 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3402 | { |
3403 | return -EINVAL; | |
3404 | } | |
8c7c6e34 | 3405 | #endif |
d13d1443 | 3406 | |
bbec2e15 | 3407 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3408 | |
bbec2e15 RG |
3409 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3410 | unsigned long max, bool memsw) | |
628f4235 | 3411 | { |
3e32cb2e | 3412 | bool enlarge = false; |
bb4a7ea2 | 3413 | bool drained = false; |
3e32cb2e | 3414 | int ret; |
c054a78c YZ |
3415 | bool limits_invariant; |
3416 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3417 | |
3e32cb2e | 3418 | do { |
628f4235 KH |
3419 | if (signal_pending(current)) { |
3420 | ret = -EINTR; | |
3421 | break; | |
3422 | } | |
3e32cb2e | 3423 | |
bbec2e15 | 3424 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3425 | /* |
3426 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3427 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3428 | */ |
15b42562 | 3429 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3430 | max <= memcg->memsw.max; |
c054a78c | 3431 | if (!limits_invariant) { |
bbec2e15 | 3432 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3433 | ret = -EINVAL; |
8c7c6e34 KH |
3434 | break; |
3435 | } | |
bbec2e15 | 3436 | if (max > counter->max) |
3e32cb2e | 3437 | enlarge = true; |
bbec2e15 RG |
3438 | ret = page_counter_set_max(counter, max); |
3439 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3440 | |
3441 | if (!ret) | |
3442 | break; | |
3443 | ||
bb4a7ea2 SB |
3444 | if (!drained) { |
3445 | drain_all_stock(memcg); | |
3446 | drained = true; | |
3447 | continue; | |
3448 | } | |
3449 | ||
73b73bac YA |
3450 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
3451 | memsw ? 0 : MEMCG_RECLAIM_MAY_SWAP)) { | |
1ab5c056 AR |
3452 | ret = -EBUSY; |
3453 | break; | |
3454 | } | |
3455 | } while (true); | |
3e32cb2e | 3456 | |
3c11ecf4 KH |
3457 | if (!ret && enlarge) |
3458 | memcg_oom_recover(memcg); | |
3e32cb2e | 3459 | |
628f4235 KH |
3460 | return ret; |
3461 | } | |
3462 | ||
ef8f2327 | 3463 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3464 | gfp_t gfp_mask, |
3465 | unsigned long *total_scanned) | |
3466 | { | |
3467 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3468 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3469 | unsigned long reclaimed; |
3470 | int loop = 0; | |
ef8f2327 | 3471 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3472 | unsigned long excess; |
0608f43d AM |
3473 | |
3474 | if (order > 0) | |
3475 | return 0; | |
3476 | ||
2ab082ba | 3477 | mctz = soft_limit_tree.rb_tree_per_node[pgdat->node_id]; |
d6507ff5 MH |
3478 | |
3479 | /* | |
3480 | * Do not even bother to check the largest node if the root | |
3481 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3482 | * are acceptable as soft limit is best effort anyway. | |
3483 | */ | |
bfc7228b | 3484 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3485 | return 0; |
3486 | ||
0608f43d AM |
3487 | /* |
3488 | * This loop can run a while, specially if mem_cgroup's continuously | |
3489 | * keep exceeding their soft limit and putting the system under | |
3490 | * pressure | |
3491 | */ | |
3492 | do { | |
3493 | if (next_mz) | |
3494 | mz = next_mz; | |
3495 | else | |
3496 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3497 | if (!mz) | |
3498 | break; | |
3499 | ||
ef8f2327 | 3500 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
d8f65338 | 3501 | gfp_mask, total_scanned); |
0608f43d | 3502 | nr_reclaimed += reclaimed; |
0a31bc97 | 3503 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3504 | |
3505 | /* | |
3506 | * If we failed to reclaim anything from this memory cgroup | |
3507 | * it is time to move on to the next cgroup | |
3508 | */ | |
3509 | next_mz = NULL; | |
bc2f2e7f VD |
3510 | if (!reclaimed) |
3511 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3512 | ||
3e32cb2e | 3513 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3514 | /* |
3515 | * One school of thought says that we should not add | |
3516 | * back the node to the tree if reclaim returns 0. | |
3517 | * But our reclaim could return 0, simply because due | |
3518 | * to priority we are exposing a smaller subset of | |
3519 | * memory to reclaim from. Consider this as a longer | |
3520 | * term TODO. | |
3521 | */ | |
3522 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3523 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3524 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3525 | css_put(&mz->memcg->css); |
3526 | loop++; | |
3527 | /* | |
3528 | * Could not reclaim anything and there are no more | |
3529 | * mem cgroups to try or we seem to be looping without | |
3530 | * reclaiming anything. | |
3531 | */ | |
3532 | if (!nr_reclaimed && | |
3533 | (next_mz == NULL || | |
3534 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3535 | break; | |
3536 | } while (!nr_reclaimed); | |
3537 | if (next_mz) | |
3538 | css_put(&next_mz->memcg->css); | |
3539 | return nr_reclaimed; | |
3540 | } | |
3541 | ||
c26251f9 | 3542 | /* |
51038171 | 3543 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3544 | * |
3545 | * Caller is responsible for holding css reference for memcg. | |
3546 | */ | |
3547 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3548 | { | |
d977aa93 | 3549 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3550 | |
c1e862c1 KH |
3551 | /* we call try-to-free pages for make this cgroup empty */ |
3552 | lru_add_drain_all(); | |
d12c60f6 JS |
3553 | |
3554 | drain_all_stock(memcg); | |
3555 | ||
f817ed48 | 3556 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3557 | while (nr_retries && page_counter_read(&memcg->memory)) { |
c26251f9 MH |
3558 | if (signal_pending(current)) |
3559 | return -EINTR; | |
3560 | ||
73b73bac YA |
3561 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
3562 | MEMCG_RECLAIM_MAY_SWAP)) | |
f817ed48 | 3563 | nr_retries--; |
f817ed48 | 3564 | } |
ab5196c2 MH |
3565 | |
3566 | return 0; | |
cc847582 KH |
3567 | } |
3568 | ||
6770c64e TH |
3569 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3570 | char *buf, size_t nbytes, | |
3571 | loff_t off) | |
c1e862c1 | 3572 | { |
6770c64e | 3573 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3574 | |
d8423011 MH |
3575 | if (mem_cgroup_is_root(memcg)) |
3576 | return -EINVAL; | |
6770c64e | 3577 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3578 | } |
3579 | ||
182446d0 TH |
3580 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3581 | struct cftype *cft) | |
18f59ea7 | 3582 | { |
bef8620c | 3583 | return 1; |
18f59ea7 BS |
3584 | } |
3585 | ||
182446d0 TH |
3586 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3587 | struct cftype *cft, u64 val) | |
18f59ea7 | 3588 | { |
bef8620c | 3589 | if (val == 1) |
0b8f73e1 | 3590 | return 0; |
567fb435 | 3591 | |
bef8620c RG |
3592 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3593 | "Please report your usecase to linux-mm@kvack.org if you " | |
3594 | "depend on this functionality.\n"); | |
567fb435 | 3595 | |
bef8620c | 3596 | return -EINVAL; |
18f59ea7 BS |
3597 | } |
3598 | ||
6f646156 | 3599 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3600 | { |
42a30035 | 3601 | unsigned long val; |
ce00a967 | 3602 | |
3e32cb2e | 3603 | if (mem_cgroup_is_root(memcg)) { |
fd25a9e0 | 3604 | mem_cgroup_flush_stats(); |
0d1c2072 | 3605 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3606 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3607 | if (swap) |
3608 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3609 | } else { |
ce00a967 | 3610 | if (!swap) |
3e32cb2e | 3611 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3612 | else |
3e32cb2e | 3613 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3614 | } |
c12176d3 | 3615 | return val; |
ce00a967 JW |
3616 | } |
3617 | ||
3e32cb2e JW |
3618 | enum { |
3619 | RES_USAGE, | |
3620 | RES_LIMIT, | |
3621 | RES_MAX_USAGE, | |
3622 | RES_FAILCNT, | |
3623 | RES_SOFT_LIMIT, | |
3624 | }; | |
ce00a967 | 3625 | |
791badbd | 3626 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3627 | struct cftype *cft) |
8cdea7c0 | 3628 | { |
182446d0 | 3629 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3630 | struct page_counter *counter; |
af36f906 | 3631 | |
3e32cb2e | 3632 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3633 | case _MEM: |
3e32cb2e JW |
3634 | counter = &memcg->memory; |
3635 | break; | |
8c7c6e34 | 3636 | case _MEMSWAP: |
3e32cb2e JW |
3637 | counter = &memcg->memsw; |
3638 | break; | |
510fc4e1 | 3639 | case _KMEM: |
3e32cb2e | 3640 | counter = &memcg->kmem; |
510fc4e1 | 3641 | break; |
d55f90bf | 3642 | case _TCP: |
0db15298 | 3643 | counter = &memcg->tcpmem; |
d55f90bf | 3644 | break; |
8c7c6e34 KH |
3645 | default: |
3646 | BUG(); | |
8c7c6e34 | 3647 | } |
3e32cb2e JW |
3648 | |
3649 | switch (MEMFILE_ATTR(cft->private)) { | |
3650 | case RES_USAGE: | |
3651 | if (counter == &memcg->memory) | |
c12176d3 | 3652 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3653 | if (counter == &memcg->memsw) |
c12176d3 | 3654 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3655 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3656 | case RES_LIMIT: | |
bbec2e15 | 3657 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3658 | case RES_MAX_USAGE: |
3659 | return (u64)counter->watermark * PAGE_SIZE; | |
3660 | case RES_FAILCNT: | |
3661 | return counter->failcnt; | |
3662 | case RES_SOFT_LIMIT: | |
3663 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3664 | default: | |
3665 | BUG(); | |
3666 | } | |
8cdea7c0 | 3667 | } |
510fc4e1 | 3668 | |
84c07d11 | 3669 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3670 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3671 | { |
bf4f0599 | 3672 | struct obj_cgroup *objcg; |
d6441637 | 3673 | |
9c94bef9 | 3674 | if (mem_cgroup_kmem_disabled()) |
b313aeee VD |
3675 | return 0; |
3676 | ||
da0efe30 MS |
3677 | if (unlikely(mem_cgroup_is_root(memcg))) |
3678 | return 0; | |
d6441637 | 3679 | |
bf4f0599 | 3680 | objcg = obj_cgroup_alloc(); |
f9c69d63 | 3681 | if (!objcg) |
bf4f0599 | 3682 | return -ENOMEM; |
f9c69d63 | 3683 | |
bf4f0599 RG |
3684 | objcg->memcg = memcg; |
3685 | rcu_assign_pointer(memcg->objcg, objcg); | |
3686 | ||
d648bcc7 RG |
3687 | static_branch_enable(&memcg_kmem_enabled_key); |
3688 | ||
f9c69d63 | 3689 | memcg->kmemcg_id = memcg->id.id; |
0b8f73e1 JW |
3690 | |
3691 | return 0; | |
d6441637 VD |
3692 | } |
3693 | ||
8e0a8912 JW |
3694 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3695 | { | |
64268868 | 3696 | struct mem_cgroup *parent; |
8e0a8912 | 3697 | |
9c94bef9 | 3698 | if (mem_cgroup_kmem_disabled()) |
da0efe30 MS |
3699 | return; |
3700 | ||
3701 | if (unlikely(mem_cgroup_is_root(memcg))) | |
8e0a8912 | 3702 | return; |
9855609b | 3703 | |
8e0a8912 JW |
3704 | parent = parent_mem_cgroup(memcg); |
3705 | if (!parent) | |
3706 | parent = root_mem_cgroup; | |
3707 | ||
bf4f0599 | 3708 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a | 3709 | |
8e0a8912 | 3710 | /* |
64268868 MS |
3711 | * After we have finished memcg_reparent_objcgs(), all list_lrus |
3712 | * corresponding to this cgroup are guaranteed to remain empty. | |
3713 | * The ordering is imposed by list_lru_node->lock taken by | |
1f391eb2 | 3714 | * memcg_reparent_list_lrus(). |
8e0a8912 | 3715 | */ |
1f391eb2 | 3716 | memcg_reparent_list_lrus(memcg, parent); |
8e0a8912 | 3717 | } |
d6441637 | 3718 | #else |
0b8f73e1 | 3719 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3720 | { |
3721 | return 0; | |
3722 | } | |
3723 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3724 | { | |
3725 | } | |
84c07d11 | 3726 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3727 | |
bbec2e15 | 3728 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3729 | { |
3730 | int ret; | |
3731 | ||
bbec2e15 | 3732 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3733 | |
bbec2e15 | 3734 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3735 | if (ret) |
3736 | goto out; | |
3737 | ||
0db15298 | 3738 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3739 | /* |
3740 | * The active flag needs to be written after the static_key | |
3741 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3742 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3743 | * for details, and note that we don't mark any socket as | |
3744 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3745 | * |
3746 | * We need to do this, because static_keys will span multiple | |
3747 | * sites, but we can't control their order. If we mark a socket | |
3748 | * as accounted, but the accounting functions are not patched in | |
3749 | * yet, we'll lose accounting. | |
3750 | * | |
2d758073 | 3751 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3752 | * because when this value change, the code to process it is not |
3753 | * patched in yet. | |
3754 | */ | |
3755 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3756 | memcg->tcpmem_active = true; |
d55f90bf VD |
3757 | } |
3758 | out: | |
bbec2e15 | 3759 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3760 | return ret; |
3761 | } | |
d55f90bf | 3762 | |
628f4235 KH |
3763 | /* |
3764 | * The user of this function is... | |
3765 | * RES_LIMIT. | |
3766 | */ | |
451af504 TH |
3767 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3768 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3769 | { |
451af504 | 3770 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3771 | unsigned long nr_pages; |
628f4235 KH |
3772 | int ret; |
3773 | ||
451af504 | 3774 | buf = strstrip(buf); |
650c5e56 | 3775 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3776 | if (ret) |
3777 | return ret; | |
af36f906 | 3778 | |
3e32cb2e | 3779 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3780 | case RES_LIMIT: |
4b3bde4c BS |
3781 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3782 | ret = -EINVAL; | |
3783 | break; | |
3784 | } | |
3e32cb2e JW |
3785 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3786 | case _MEM: | |
bbec2e15 | 3787 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3788 | break; |
3e32cb2e | 3789 | case _MEMSWAP: |
bbec2e15 | 3790 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3791 | break; |
3e32cb2e | 3792 | case _KMEM: |
58056f77 SB |
3793 | /* kmem.limit_in_bytes is deprecated. */ |
3794 | ret = -EOPNOTSUPP; | |
3e32cb2e | 3795 | break; |
d55f90bf | 3796 | case _TCP: |
bbec2e15 | 3797 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3798 | break; |
3e32cb2e | 3799 | } |
296c81d8 | 3800 | break; |
3e32cb2e | 3801 | case RES_SOFT_LIMIT: |
2343e88d SAS |
3802 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
3803 | ret = -EOPNOTSUPP; | |
3804 | } else { | |
3805 | memcg->soft_limit = nr_pages; | |
3806 | ret = 0; | |
3807 | } | |
628f4235 KH |
3808 | break; |
3809 | } | |
451af504 | 3810 | return ret ?: nbytes; |
8cdea7c0 BS |
3811 | } |
3812 | ||
6770c64e TH |
3813 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3814 | size_t nbytes, loff_t off) | |
c84872e1 | 3815 | { |
6770c64e | 3816 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3817 | struct page_counter *counter; |
c84872e1 | 3818 | |
3e32cb2e JW |
3819 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3820 | case _MEM: | |
3821 | counter = &memcg->memory; | |
3822 | break; | |
3823 | case _MEMSWAP: | |
3824 | counter = &memcg->memsw; | |
3825 | break; | |
3826 | case _KMEM: | |
3827 | counter = &memcg->kmem; | |
3828 | break; | |
d55f90bf | 3829 | case _TCP: |
0db15298 | 3830 | counter = &memcg->tcpmem; |
d55f90bf | 3831 | break; |
3e32cb2e JW |
3832 | default: |
3833 | BUG(); | |
3834 | } | |
af36f906 | 3835 | |
3e32cb2e | 3836 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3837 | case RES_MAX_USAGE: |
3e32cb2e | 3838 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3839 | break; |
3840 | case RES_FAILCNT: | |
3e32cb2e | 3841 | counter->failcnt = 0; |
29f2a4da | 3842 | break; |
3e32cb2e JW |
3843 | default: |
3844 | BUG(); | |
29f2a4da | 3845 | } |
f64c3f54 | 3846 | |
6770c64e | 3847 | return nbytes; |
c84872e1 PE |
3848 | } |
3849 | ||
182446d0 | 3850 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3851 | struct cftype *cft) |
3852 | { | |
182446d0 | 3853 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3854 | } |
3855 | ||
02491447 | 3856 | #ifdef CONFIG_MMU |
182446d0 | 3857 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3858 | struct cftype *cft, u64 val) |
3859 | { | |
182446d0 | 3860 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3861 | |
1dfab5ab | 3862 | if (val & ~MOVE_MASK) |
7dc74be0 | 3863 | return -EINVAL; |
ee5e8472 | 3864 | |
7dc74be0 | 3865 | /* |
ee5e8472 GC |
3866 | * No kind of locking is needed in here, because ->can_attach() will |
3867 | * check this value once in the beginning of the process, and then carry | |
3868 | * on with stale data. This means that changes to this value will only | |
3869 | * affect task migrations starting after the change. | |
7dc74be0 | 3870 | */ |
c0ff4b85 | 3871 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3872 | return 0; |
3873 | } | |
02491447 | 3874 | #else |
182446d0 | 3875 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3876 | struct cftype *cft, u64 val) |
3877 | { | |
3878 | return -ENOSYS; | |
3879 | } | |
3880 | #endif | |
7dc74be0 | 3881 | |
406eb0c9 | 3882 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3883 | |
3884 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3885 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3886 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3887 | ||
3888 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3889 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3890 | { |
867e5e1d | 3891 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3892 | unsigned long nr = 0; |
3893 | enum lru_list lru; | |
3894 | ||
3895 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3896 | ||
3897 | for_each_lru(lru) { | |
3898 | if (!(BIT(lru) & lru_mask)) | |
3899 | continue; | |
dd8657b6 SB |
3900 | if (tree) |
3901 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3902 | else | |
3903 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3904 | } |
3905 | return nr; | |
3906 | } | |
3907 | ||
3908 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3909 | unsigned int lru_mask, |
3910 | bool tree) | |
113b7dfd JW |
3911 | { |
3912 | unsigned long nr = 0; | |
3913 | enum lru_list lru; | |
3914 | ||
3915 | for_each_lru(lru) { | |
3916 | if (!(BIT(lru) & lru_mask)) | |
3917 | continue; | |
dd8657b6 SB |
3918 | if (tree) |
3919 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3920 | else | |
3921 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3922 | } |
3923 | return nr; | |
3924 | } | |
3925 | ||
2da8ca82 | 3926 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3927 | { |
25485de6 GT |
3928 | struct numa_stat { |
3929 | const char *name; | |
3930 | unsigned int lru_mask; | |
3931 | }; | |
3932 | ||
3933 | static const struct numa_stat stats[] = { | |
3934 | { "total", LRU_ALL }, | |
3935 | { "file", LRU_ALL_FILE }, | |
3936 | { "anon", LRU_ALL_ANON }, | |
3937 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3938 | }; | |
3939 | const struct numa_stat *stat; | |
406eb0c9 | 3940 | int nid; |
aa9694bb | 3941 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3942 | |
fd25a9e0 | 3943 | mem_cgroup_flush_stats(); |
2d146aa3 | 3944 | |
25485de6 | 3945 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3946 | seq_printf(m, "%s=%lu", stat->name, |
3947 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3948 | false)); | |
3949 | for_each_node_state(nid, N_MEMORY) | |
3950 | seq_printf(m, " N%d=%lu", nid, | |
3951 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3952 | stat->lru_mask, false)); | |
25485de6 | 3953 | seq_putc(m, '\n'); |
406eb0c9 | 3954 | } |
406eb0c9 | 3955 | |
071aee13 | 3956 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3957 | |
3958 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3959 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3960 | true)); | |
3961 | for_each_node_state(nid, N_MEMORY) | |
3962 | seq_printf(m, " N%d=%lu", nid, | |
3963 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3964 | stat->lru_mask, true)); | |
071aee13 | 3965 | seq_putc(m, '\n'); |
406eb0c9 | 3966 | } |
406eb0c9 | 3967 | |
406eb0c9 YH |
3968 | return 0; |
3969 | } | |
3970 | #endif /* CONFIG_NUMA */ | |
3971 | ||
c8713d0b | 3972 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 3973 | NR_FILE_PAGES, |
be5d0a74 | 3974 | NR_ANON_MAPPED, |
468c3982 JW |
3975 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3976 | NR_ANON_THPS, | |
3977 | #endif | |
c8713d0b JW |
3978 | NR_SHMEM, |
3979 | NR_FILE_MAPPED, | |
3980 | NR_FILE_DIRTY, | |
3981 | NR_WRITEBACK, | |
3982 | MEMCG_SWAP, | |
3983 | }; | |
3984 | ||
3985 | static const char *const memcg1_stat_names[] = { | |
3986 | "cache", | |
3987 | "rss", | |
468c3982 | 3988 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 3989 | "rss_huge", |
468c3982 | 3990 | #endif |
c8713d0b JW |
3991 | "shmem", |
3992 | "mapped_file", | |
3993 | "dirty", | |
3994 | "writeback", | |
3995 | "swap", | |
3996 | }; | |
3997 | ||
df0e53d0 | 3998 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3999 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4000 | PGPGIN, |
4001 | PGPGOUT, | |
4002 | PGFAULT, | |
4003 | PGMAJFAULT, | |
4004 | }; | |
4005 | ||
2da8ca82 | 4006 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4007 | { |
aa9694bb | 4008 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4009 | unsigned long memory, memsw; |
af7c4b0e JW |
4010 | struct mem_cgroup *mi; |
4011 | unsigned int i; | |
406eb0c9 | 4012 | |
71cd3113 | 4013 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4014 | |
fd25a9e0 | 4015 | mem_cgroup_flush_stats(); |
2d146aa3 | 4016 | |
71cd3113 | 4017 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4018 | unsigned long nr; |
4019 | ||
71cd3113 | 4020 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4021 | continue; |
468c3982 | 4022 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
468c3982 | 4023 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); |
1dd3a273 | 4024 | } |
7b854121 | 4025 | |
df0e53d0 | 4026 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4027 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4028 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4029 | |
4030 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4031 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4032 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4033 | PAGE_SIZE); |
af7c4b0e | 4034 | |
14067bb3 | 4035 | /* Hierarchical information */ |
3e32cb2e JW |
4036 | memory = memsw = PAGE_COUNTER_MAX; |
4037 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4038 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4039 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4040 | } |
3e32cb2e JW |
4041 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4042 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4043 | if (do_memsw_account()) |
3e32cb2e JW |
4044 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4045 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4046 | |
8de7ecc6 | 4047 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4048 | unsigned long nr; |
4049 | ||
71cd3113 | 4050 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4051 | continue; |
7de2e9f1 | 4052 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
8de7ecc6 | 4053 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
7de2e9f1 | 4054 | (u64)nr * PAGE_SIZE); |
af7c4b0e JW |
4055 | } |
4056 | ||
8de7ecc6 | 4057 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4058 | seq_printf(m, "total_%s %llu\n", |
4059 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4060 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4061 | |
8de7ecc6 | 4062 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4063 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4064 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4065 | PAGE_SIZE); | |
14067bb3 | 4066 | |
7f016ee8 | 4067 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4068 | { |
ef8f2327 MG |
4069 | pg_data_t *pgdat; |
4070 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4071 | unsigned long anon_cost = 0; |
4072 | unsigned long file_cost = 0; | |
7f016ee8 | 4073 | |
ef8f2327 | 4074 | for_each_online_pgdat(pgdat) { |
a3747b53 | 4075 | mz = memcg->nodeinfo[pgdat->node_id]; |
7f016ee8 | 4076 | |
1431d4d1 JW |
4077 | anon_cost += mz->lruvec.anon_cost; |
4078 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4079 | } |
1431d4d1 JW |
4080 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4081 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4082 | } |
4083 | #endif | |
4084 | ||
d2ceb9b7 KH |
4085 | return 0; |
4086 | } | |
4087 | ||
182446d0 TH |
4088 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4089 | struct cftype *cft) | |
a7885eb8 | 4090 | { |
182446d0 | 4091 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4092 | |
1f4c025b | 4093 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4094 | } |
4095 | ||
182446d0 TH |
4096 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4097 | struct cftype *cft, u64 val) | |
a7885eb8 | 4098 | { |
182446d0 | 4099 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4100 | |
37bc3cb9 | 4101 | if (val > 200) |
a7885eb8 KM |
4102 | return -EINVAL; |
4103 | ||
a4792030 | 4104 | if (!mem_cgroup_is_root(memcg)) |
3dae7fec JW |
4105 | memcg->swappiness = val; |
4106 | else | |
4107 | vm_swappiness = val; | |
068b38c1 | 4108 | |
a7885eb8 KM |
4109 | return 0; |
4110 | } | |
4111 | ||
2e72b634 KS |
4112 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4113 | { | |
4114 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4115 | unsigned long usage; |
2e72b634 KS |
4116 | int i; |
4117 | ||
4118 | rcu_read_lock(); | |
4119 | if (!swap) | |
2c488db2 | 4120 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4121 | else |
2c488db2 | 4122 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4123 | |
4124 | if (!t) | |
4125 | goto unlock; | |
4126 | ||
ce00a967 | 4127 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4128 | |
4129 | /* | |
748dad36 | 4130 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4131 | * If it's not true, a threshold was crossed after last |
4132 | * call of __mem_cgroup_threshold(). | |
4133 | */ | |
5407a562 | 4134 | i = t->current_threshold; |
2e72b634 KS |
4135 | |
4136 | /* | |
4137 | * Iterate backward over array of thresholds starting from | |
4138 | * current_threshold and check if a threshold is crossed. | |
4139 | * If none of thresholds below usage is crossed, we read | |
4140 | * only one element of the array here. | |
4141 | */ | |
4142 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4143 | eventfd_signal(t->entries[i].eventfd, 1); | |
4144 | ||
4145 | /* i = current_threshold + 1 */ | |
4146 | i++; | |
4147 | ||
4148 | /* | |
4149 | * Iterate forward over array of thresholds starting from | |
4150 | * current_threshold+1 and check if a threshold is crossed. | |
4151 | * If none of thresholds above usage is crossed, we read | |
4152 | * only one element of the array here. | |
4153 | */ | |
4154 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4155 | eventfd_signal(t->entries[i].eventfd, 1); | |
4156 | ||
4157 | /* Update current_threshold */ | |
5407a562 | 4158 | t->current_threshold = i - 1; |
2e72b634 KS |
4159 | unlock: |
4160 | rcu_read_unlock(); | |
4161 | } | |
4162 | ||
4163 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4164 | { | |
ad4ca5f4 KS |
4165 | while (memcg) { |
4166 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4167 | if (do_memsw_account()) |
ad4ca5f4 KS |
4168 | __mem_cgroup_threshold(memcg, true); |
4169 | ||
4170 | memcg = parent_mem_cgroup(memcg); | |
4171 | } | |
2e72b634 KS |
4172 | } |
4173 | ||
4174 | static int compare_thresholds(const void *a, const void *b) | |
4175 | { | |
4176 | const struct mem_cgroup_threshold *_a = a; | |
4177 | const struct mem_cgroup_threshold *_b = b; | |
4178 | ||
2bff24a3 GT |
4179 | if (_a->threshold > _b->threshold) |
4180 | return 1; | |
4181 | ||
4182 | if (_a->threshold < _b->threshold) | |
4183 | return -1; | |
4184 | ||
4185 | return 0; | |
2e72b634 KS |
4186 | } |
4187 | ||
c0ff4b85 | 4188 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4189 | { |
4190 | struct mem_cgroup_eventfd_list *ev; | |
4191 | ||
2bcf2e92 MH |
4192 | spin_lock(&memcg_oom_lock); |
4193 | ||
c0ff4b85 | 4194 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4195 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4196 | |
4197 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4198 | return 0; |
4199 | } | |
4200 | ||
c0ff4b85 | 4201 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4202 | { |
7d74b06f KH |
4203 | struct mem_cgroup *iter; |
4204 | ||
c0ff4b85 | 4205 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4206 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4207 | } |
4208 | ||
59b6f873 | 4209 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4210 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4211 | { |
2c488db2 KS |
4212 | struct mem_cgroup_thresholds *thresholds; |
4213 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4214 | unsigned long threshold; |
4215 | unsigned long usage; | |
2c488db2 | 4216 | int i, size, ret; |
2e72b634 | 4217 | |
650c5e56 | 4218 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4219 | if (ret) |
4220 | return ret; | |
4221 | ||
4222 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4223 | |
05b84301 | 4224 | if (type == _MEM) { |
2c488db2 | 4225 | thresholds = &memcg->thresholds; |
ce00a967 | 4226 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4227 | } else if (type == _MEMSWAP) { |
2c488db2 | 4228 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4229 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4230 | } else |
2e72b634 KS |
4231 | BUG(); |
4232 | ||
2e72b634 | 4233 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4234 | if (thresholds->primary) |
2e72b634 KS |
4235 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4236 | ||
2c488db2 | 4237 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4238 | |
4239 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4240 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4241 | if (!new) { |
2e72b634 KS |
4242 | ret = -ENOMEM; |
4243 | goto unlock; | |
4244 | } | |
2c488db2 | 4245 | new->size = size; |
2e72b634 KS |
4246 | |
4247 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4248 | if (thresholds->primary) |
4249 | memcpy(new->entries, thresholds->primary->entries, | |
4250 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4251 | |
2e72b634 | 4252 | /* Add new threshold */ |
2c488db2 KS |
4253 | new->entries[size - 1].eventfd = eventfd; |
4254 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4255 | |
4256 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4257 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4258 | compare_thresholds, NULL); |
4259 | ||
4260 | /* Find current threshold */ | |
2c488db2 | 4261 | new->current_threshold = -1; |
2e72b634 | 4262 | for (i = 0; i < size; i++) { |
748dad36 | 4263 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4264 | /* |
2c488db2 KS |
4265 | * new->current_threshold will not be used until |
4266 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4267 | * it here. |
4268 | */ | |
2c488db2 | 4269 | ++new->current_threshold; |
748dad36 SZ |
4270 | } else |
4271 | break; | |
2e72b634 KS |
4272 | } |
4273 | ||
2c488db2 KS |
4274 | /* Free old spare buffer and save old primary buffer as spare */ |
4275 | kfree(thresholds->spare); | |
4276 | thresholds->spare = thresholds->primary; | |
4277 | ||
4278 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4279 | |
907860ed | 4280 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4281 | synchronize_rcu(); |
4282 | ||
2e72b634 KS |
4283 | unlock: |
4284 | mutex_unlock(&memcg->thresholds_lock); | |
4285 | ||
4286 | return ret; | |
4287 | } | |
4288 | ||
59b6f873 | 4289 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4290 | struct eventfd_ctx *eventfd, const char *args) |
4291 | { | |
59b6f873 | 4292 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4293 | } |
4294 | ||
59b6f873 | 4295 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4296 | struct eventfd_ctx *eventfd, const char *args) |
4297 | { | |
59b6f873 | 4298 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4299 | } |
4300 | ||
59b6f873 | 4301 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4302 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4303 | { |
2c488db2 KS |
4304 | struct mem_cgroup_thresholds *thresholds; |
4305 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4306 | unsigned long usage; |
7d36665a | 4307 | int i, j, size, entries; |
2e72b634 KS |
4308 | |
4309 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4310 | |
4311 | if (type == _MEM) { | |
2c488db2 | 4312 | thresholds = &memcg->thresholds; |
ce00a967 | 4313 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4314 | } else if (type == _MEMSWAP) { |
2c488db2 | 4315 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4316 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4317 | } else |
2e72b634 KS |
4318 | BUG(); |
4319 | ||
371528ca AV |
4320 | if (!thresholds->primary) |
4321 | goto unlock; | |
4322 | ||
2e72b634 KS |
4323 | /* Check if a threshold crossed before removing */ |
4324 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4325 | ||
4326 | /* Calculate new number of threshold */ | |
7d36665a | 4327 | size = entries = 0; |
2c488db2 KS |
4328 | for (i = 0; i < thresholds->primary->size; i++) { |
4329 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4330 | size++; |
7d36665a CX |
4331 | else |
4332 | entries++; | |
2e72b634 KS |
4333 | } |
4334 | ||
2c488db2 | 4335 | new = thresholds->spare; |
907860ed | 4336 | |
7d36665a CX |
4337 | /* If no items related to eventfd have been cleared, nothing to do */ |
4338 | if (!entries) | |
4339 | goto unlock; | |
4340 | ||
2e72b634 KS |
4341 | /* Set thresholds array to NULL if we don't have thresholds */ |
4342 | if (!size) { | |
2c488db2 KS |
4343 | kfree(new); |
4344 | new = NULL; | |
907860ed | 4345 | goto swap_buffers; |
2e72b634 KS |
4346 | } |
4347 | ||
2c488db2 | 4348 | new->size = size; |
2e72b634 KS |
4349 | |
4350 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4351 | new->current_threshold = -1; |
4352 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4353 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4354 | continue; |
4355 | ||
2c488db2 | 4356 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4357 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4358 | /* |
2c488db2 | 4359 | * new->current_threshold will not be used |
2e72b634 KS |
4360 | * until rcu_assign_pointer(), so it's safe to increment |
4361 | * it here. | |
4362 | */ | |
2c488db2 | 4363 | ++new->current_threshold; |
2e72b634 KS |
4364 | } |
4365 | j++; | |
4366 | } | |
4367 | ||
907860ed | 4368 | swap_buffers: |
2c488db2 KS |
4369 | /* Swap primary and spare array */ |
4370 | thresholds->spare = thresholds->primary; | |
8c757763 | 4371 | |
2c488db2 | 4372 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4373 | |
907860ed | 4374 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4375 | synchronize_rcu(); |
6611d8d7 MC |
4376 | |
4377 | /* If all events are unregistered, free the spare array */ | |
4378 | if (!new) { | |
4379 | kfree(thresholds->spare); | |
4380 | thresholds->spare = NULL; | |
4381 | } | |
371528ca | 4382 | unlock: |
2e72b634 | 4383 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4384 | } |
c1e862c1 | 4385 | |
59b6f873 | 4386 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4387 | struct eventfd_ctx *eventfd) |
4388 | { | |
59b6f873 | 4389 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4390 | } |
4391 | ||
59b6f873 | 4392 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4393 | struct eventfd_ctx *eventfd) |
4394 | { | |
59b6f873 | 4395 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4396 | } |
4397 | ||
59b6f873 | 4398 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4399 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4400 | { |
9490ff27 | 4401 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4402 | |
9490ff27 KH |
4403 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4404 | if (!event) | |
4405 | return -ENOMEM; | |
4406 | ||
1af8efe9 | 4407 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4408 | |
4409 | event->eventfd = eventfd; | |
4410 | list_add(&event->list, &memcg->oom_notify); | |
4411 | ||
4412 | /* already in OOM ? */ | |
c2b42d3c | 4413 | if (memcg->under_oom) |
9490ff27 | 4414 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4415 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4416 | |
4417 | return 0; | |
4418 | } | |
4419 | ||
59b6f873 | 4420 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4421 | struct eventfd_ctx *eventfd) |
9490ff27 | 4422 | { |
9490ff27 | 4423 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4424 | |
1af8efe9 | 4425 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4426 | |
c0ff4b85 | 4427 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4428 | if (ev->eventfd == eventfd) { |
4429 | list_del(&ev->list); | |
4430 | kfree(ev); | |
4431 | } | |
4432 | } | |
4433 | ||
1af8efe9 | 4434 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4435 | } |
4436 | ||
2da8ca82 | 4437 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4438 | { |
aa9694bb | 4439 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4440 | |
791badbd | 4441 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4442 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4443 | seq_printf(sf, "oom_kill %lu\n", |
4444 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4445 | return 0; |
4446 | } | |
4447 | ||
182446d0 | 4448 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4449 | struct cftype *cft, u64 val) |
4450 | { | |
182446d0 | 4451 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4452 | |
4453 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4454 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4455 | return -EINVAL; |
4456 | ||
c0ff4b85 | 4457 | memcg->oom_kill_disable = val; |
4d845ebf | 4458 | if (!val) |
c0ff4b85 | 4459 | memcg_oom_recover(memcg); |
3dae7fec | 4460 | |
3c11ecf4 KH |
4461 | return 0; |
4462 | } | |
4463 | ||
52ebea74 TH |
4464 | #ifdef CONFIG_CGROUP_WRITEBACK |
4465 | ||
3a8e9ac8 TH |
4466 | #include <trace/events/writeback.h> |
4467 | ||
841710aa TH |
4468 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4469 | { | |
4470 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4471 | } | |
4472 | ||
4473 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4474 | { | |
4475 | wb_domain_exit(&memcg->cgwb_domain); | |
4476 | } | |
4477 | ||
2529bb3a TH |
4478 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4479 | { | |
4480 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4481 | } | |
4482 | ||
841710aa TH |
4483 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4484 | { | |
4485 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4486 | ||
4487 | if (!memcg->css.parent) | |
4488 | return NULL; | |
4489 | ||
4490 | return &memcg->cgwb_domain; | |
4491 | } | |
4492 | ||
c2aa723a TH |
4493 | /** |
4494 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4495 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4496 | * @pfilepages: out parameter for number of file pages |
4497 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4498 | * @pdirty: out parameter for number of dirty pages |
4499 | * @pwriteback: out parameter for number of pages under writeback | |
4500 | * | |
c5edf9cd TH |
4501 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4502 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4503 | * is a bit more involved. | |
c2aa723a | 4504 | * |
c5edf9cd TH |
4505 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4506 | * headroom is calculated as the lowest headroom of itself and the | |
4507 | * ancestors. Note that this doesn't consider the actual amount of | |
4508 | * available memory in the system. The caller should further cap | |
4509 | * *@pheadroom accordingly. | |
c2aa723a | 4510 | */ |
c5edf9cd TH |
4511 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4512 | unsigned long *pheadroom, unsigned long *pdirty, | |
4513 | unsigned long *pwriteback) | |
c2aa723a TH |
4514 | { |
4515 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4516 | struct mem_cgroup *parent; | |
c2aa723a | 4517 | |
fd25a9e0 | 4518 | mem_cgroup_flush_stats(); |
c2aa723a | 4519 | |
2d146aa3 JW |
4520 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
4521 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); | |
4522 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + | |
4523 | memcg_page_state(memcg, NR_ACTIVE_FILE); | |
c2aa723a | 4524 | |
2d146aa3 | 4525 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4526 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4527 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4528 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4529 | unsigned long used = page_counter_read(&memcg->memory); |
4530 | ||
c5edf9cd | 4531 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4532 | memcg = parent; |
4533 | } | |
c2aa723a TH |
4534 | } |
4535 | ||
97b27821 TH |
4536 | /* |
4537 | * Foreign dirty flushing | |
4538 | * | |
4539 | * There's an inherent mismatch between memcg and writeback. The former | |
f0953a1b | 4540 | * tracks ownership per-page while the latter per-inode. This was a |
97b27821 TH |
4541 | * deliberate design decision because honoring per-page ownership in the |
4542 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4543 | * and deemed unnecessary given that write-sharing an inode across | |
4544 | * different cgroups isn't a common use-case. | |
4545 | * | |
4546 | * Combined with inode majority-writer ownership switching, this works well | |
4547 | * enough in most cases but there are some pathological cases. For | |
4548 | * example, let's say there are two cgroups A and B which keep writing to | |
4549 | * different but confined parts of the same inode. B owns the inode and | |
4550 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4551 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4552 | * triggering background writeback. A will be slowed down without a way to | |
4553 | * make writeback of the dirty pages happen. | |
4554 | * | |
f0953a1b | 4555 | * Conditions like the above can lead to a cgroup getting repeatedly and |
97b27821 | 4556 | * severely throttled after making some progress after each |
f0953a1b | 4557 | * dirty_expire_interval while the underlying IO device is almost |
97b27821 TH |
4558 | * completely idle. |
4559 | * | |
4560 | * Solving this problem completely requires matching the ownership tracking | |
4561 | * granularities between memcg and writeback in either direction. However, | |
4562 | * the more egregious behaviors can be avoided by simply remembering the | |
4563 | * most recent foreign dirtying events and initiating remote flushes on | |
4564 | * them when local writeback isn't enough to keep the memory clean enough. | |
4565 | * | |
4566 | * The following two functions implement such mechanism. When a foreign | |
4567 | * page - a page whose memcg and writeback ownerships don't match - is | |
4568 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4569 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4570 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4571 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4572 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4573 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4574 | * limited to MEMCG_CGWB_FRN_CNT. | |
4575 | * | |
4576 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4577 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4578 | * records are lockless and racy. | |
4579 | */ | |
9d8053fc | 4580 | void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio, |
97b27821 TH |
4581 | struct bdi_writeback *wb) |
4582 | { | |
9d8053fc | 4583 | struct mem_cgroup *memcg = folio_memcg(folio); |
97b27821 TH |
4584 | struct memcg_cgwb_frn *frn; |
4585 | u64 now = get_jiffies_64(); | |
4586 | u64 oldest_at = now; | |
4587 | int oldest = -1; | |
4588 | int i; | |
4589 | ||
9d8053fc | 4590 | trace_track_foreign_dirty(folio, wb); |
3a8e9ac8 | 4591 | |
97b27821 TH |
4592 | /* |
4593 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4594 | * using it. If not replace the oldest one which isn't being | |
4595 | * written out. | |
4596 | */ | |
4597 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4598 | frn = &memcg->cgwb_frn[i]; | |
4599 | if (frn->bdi_id == wb->bdi->id && | |
4600 | frn->memcg_id == wb->memcg_css->id) | |
4601 | break; | |
4602 | if (time_before64(frn->at, oldest_at) && | |
4603 | atomic_read(&frn->done.cnt) == 1) { | |
4604 | oldest = i; | |
4605 | oldest_at = frn->at; | |
4606 | } | |
4607 | } | |
4608 | ||
4609 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4610 | /* | |
4611 | * Re-using an existing one. Update timestamp lazily to | |
4612 | * avoid making the cacheline hot. We want them to be | |
4613 | * reasonably up-to-date and significantly shorter than | |
4614 | * dirty_expire_interval as that's what expires the record. | |
4615 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4616 | */ | |
4617 | unsigned long update_intv = | |
4618 | min_t(unsigned long, HZ, | |
4619 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4620 | ||
4621 | if (time_before64(frn->at, now - update_intv)) | |
4622 | frn->at = now; | |
4623 | } else if (oldest >= 0) { | |
4624 | /* replace the oldest free one */ | |
4625 | frn = &memcg->cgwb_frn[oldest]; | |
4626 | frn->bdi_id = wb->bdi->id; | |
4627 | frn->memcg_id = wb->memcg_css->id; | |
4628 | frn->at = now; | |
4629 | } | |
4630 | } | |
4631 | ||
4632 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4633 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4634 | { | |
4635 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4636 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4637 | u64 now = jiffies_64; | |
4638 | int i; | |
4639 | ||
4640 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4641 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4642 | ||
4643 | /* | |
4644 | * If the record is older than dirty_expire_interval, | |
4645 | * writeback on it has already started. No need to kick it | |
4646 | * off again. Also, don't start a new one if there's | |
4647 | * already one in flight. | |
4648 | */ | |
4649 | if (time_after64(frn->at, now - intv) && | |
4650 | atomic_read(&frn->done.cnt) == 1) { | |
4651 | frn->at = 0; | |
3a8e9ac8 | 4652 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
7490a2d2 | 4653 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, |
97b27821 TH |
4654 | WB_REASON_FOREIGN_FLUSH, |
4655 | &frn->done); | |
4656 | } | |
4657 | } | |
4658 | } | |
4659 | ||
841710aa TH |
4660 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4661 | ||
4662 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4663 | { | |
4664 | return 0; | |
4665 | } | |
4666 | ||
4667 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4668 | { | |
4669 | } | |
4670 | ||
2529bb3a TH |
4671 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4672 | { | |
4673 | } | |
4674 | ||
52ebea74 TH |
4675 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4676 | ||
3bc942f3 TH |
4677 | /* |
4678 | * DO NOT USE IN NEW FILES. | |
4679 | * | |
4680 | * "cgroup.event_control" implementation. | |
4681 | * | |
4682 | * This is way over-engineered. It tries to support fully configurable | |
4683 | * events for each user. Such level of flexibility is completely | |
4684 | * unnecessary especially in the light of the planned unified hierarchy. | |
4685 | * | |
4686 | * Please deprecate this and replace with something simpler if at all | |
4687 | * possible. | |
4688 | */ | |
4689 | ||
79bd9814 TH |
4690 | /* |
4691 | * Unregister event and free resources. | |
4692 | * | |
4693 | * Gets called from workqueue. | |
4694 | */ | |
3bc942f3 | 4695 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4696 | { |
3bc942f3 TH |
4697 | struct mem_cgroup_event *event = |
4698 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4699 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4700 | |
4701 | remove_wait_queue(event->wqh, &event->wait); | |
4702 | ||
59b6f873 | 4703 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4704 | |
4705 | /* Notify userspace the event is going away. */ | |
4706 | eventfd_signal(event->eventfd, 1); | |
4707 | ||
4708 | eventfd_ctx_put(event->eventfd); | |
4709 | kfree(event); | |
59b6f873 | 4710 | css_put(&memcg->css); |
79bd9814 TH |
4711 | } |
4712 | ||
4713 | /* | |
a9a08845 | 4714 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4715 | * |
4716 | * Called with wqh->lock held and interrupts disabled. | |
4717 | */ | |
ac6424b9 | 4718 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4719 | int sync, void *key) |
79bd9814 | 4720 | { |
3bc942f3 TH |
4721 | struct mem_cgroup_event *event = |
4722 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4723 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4724 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4725 | |
a9a08845 | 4726 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4727 | /* |
4728 | * If the event has been detached at cgroup removal, we | |
4729 | * can simply return knowing the other side will cleanup | |
4730 | * for us. | |
4731 | * | |
4732 | * We can't race against event freeing since the other | |
4733 | * side will require wqh->lock via remove_wait_queue(), | |
4734 | * which we hold. | |
4735 | */ | |
fba94807 | 4736 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4737 | if (!list_empty(&event->list)) { |
4738 | list_del_init(&event->list); | |
4739 | /* | |
4740 | * We are in atomic context, but cgroup_event_remove() | |
4741 | * may sleep, so we have to call it in workqueue. | |
4742 | */ | |
4743 | schedule_work(&event->remove); | |
4744 | } | |
fba94807 | 4745 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4746 | } |
4747 | ||
4748 | return 0; | |
4749 | } | |
4750 | ||
3bc942f3 | 4751 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4752 | wait_queue_head_t *wqh, poll_table *pt) |
4753 | { | |
3bc942f3 TH |
4754 | struct mem_cgroup_event *event = |
4755 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4756 | |
4757 | event->wqh = wqh; | |
4758 | add_wait_queue(wqh, &event->wait); | |
4759 | } | |
4760 | ||
4761 | /* | |
3bc942f3 TH |
4762 | * DO NOT USE IN NEW FILES. |
4763 | * | |
79bd9814 TH |
4764 | * Parse input and register new cgroup event handler. |
4765 | * | |
4766 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4767 | * Interpretation of args is defined by control file implementation. | |
4768 | */ | |
451af504 TH |
4769 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4770 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4771 | { |
451af504 | 4772 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4773 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4774 | struct mem_cgroup_event *event; |
79bd9814 TH |
4775 | struct cgroup_subsys_state *cfile_css; |
4776 | unsigned int efd, cfd; | |
4777 | struct fd efile; | |
4778 | struct fd cfile; | |
fba94807 | 4779 | const char *name; |
79bd9814 TH |
4780 | char *endp; |
4781 | int ret; | |
4782 | ||
2343e88d SAS |
4783 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
4784 | return -EOPNOTSUPP; | |
4785 | ||
451af504 TH |
4786 | buf = strstrip(buf); |
4787 | ||
4788 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4789 | if (*endp != ' ') |
4790 | return -EINVAL; | |
451af504 | 4791 | buf = endp + 1; |
79bd9814 | 4792 | |
451af504 | 4793 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4794 | if ((*endp != ' ') && (*endp != '\0')) |
4795 | return -EINVAL; | |
451af504 | 4796 | buf = endp + 1; |
79bd9814 TH |
4797 | |
4798 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4799 | if (!event) | |
4800 | return -ENOMEM; | |
4801 | ||
59b6f873 | 4802 | event->memcg = memcg; |
79bd9814 | 4803 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4804 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4805 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4806 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4807 | |
4808 | efile = fdget(efd); | |
4809 | if (!efile.file) { | |
4810 | ret = -EBADF; | |
4811 | goto out_kfree; | |
4812 | } | |
4813 | ||
4814 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4815 | if (IS_ERR(event->eventfd)) { | |
4816 | ret = PTR_ERR(event->eventfd); | |
4817 | goto out_put_efile; | |
4818 | } | |
4819 | ||
4820 | cfile = fdget(cfd); | |
4821 | if (!cfile.file) { | |
4822 | ret = -EBADF; | |
4823 | goto out_put_eventfd; | |
4824 | } | |
4825 | ||
4826 | /* the process need read permission on control file */ | |
4827 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 4828 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
4829 | if (ret < 0) |
4830 | goto out_put_cfile; | |
4831 | ||
fba94807 TH |
4832 | /* |
4833 | * Determine the event callbacks and set them in @event. This used | |
4834 | * to be done via struct cftype but cgroup core no longer knows | |
4835 | * about these events. The following is crude but the whole thing | |
4836 | * is for compatibility anyway. | |
3bc942f3 TH |
4837 | * |
4838 | * DO NOT ADD NEW FILES. | |
fba94807 | 4839 | */ |
b583043e | 4840 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4841 | |
4842 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4843 | event->register_event = mem_cgroup_usage_register_event; | |
4844 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4845 | } else if (!strcmp(name, "memory.oom_control")) { | |
4846 | event->register_event = mem_cgroup_oom_register_event; | |
4847 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4848 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4849 | event->register_event = vmpressure_register_event; | |
4850 | event->unregister_event = vmpressure_unregister_event; | |
4851 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4852 | event->register_event = memsw_cgroup_usage_register_event; |
4853 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4854 | } else { |
4855 | ret = -EINVAL; | |
4856 | goto out_put_cfile; | |
4857 | } | |
4858 | ||
79bd9814 | 4859 | /* |
b5557c4c TH |
4860 | * Verify @cfile should belong to @css. Also, remaining events are |
4861 | * automatically removed on cgroup destruction but the removal is | |
4862 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4863 | */ |
b583043e | 4864 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4865 | &memory_cgrp_subsys); |
79bd9814 | 4866 | ret = -EINVAL; |
5a17f543 | 4867 | if (IS_ERR(cfile_css)) |
79bd9814 | 4868 | goto out_put_cfile; |
5a17f543 TH |
4869 | if (cfile_css != css) { |
4870 | css_put(cfile_css); | |
79bd9814 | 4871 | goto out_put_cfile; |
5a17f543 | 4872 | } |
79bd9814 | 4873 | |
451af504 | 4874 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4875 | if (ret) |
4876 | goto out_put_css; | |
4877 | ||
9965ed17 | 4878 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4879 | |
4ba9515d | 4880 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 4881 | list_add(&event->list, &memcg->event_list); |
4ba9515d | 4882 | spin_unlock_irq(&memcg->event_list_lock); |
79bd9814 TH |
4883 | |
4884 | fdput(cfile); | |
4885 | fdput(efile); | |
4886 | ||
451af504 | 4887 | return nbytes; |
79bd9814 TH |
4888 | |
4889 | out_put_css: | |
b5557c4c | 4890 | css_put(css); |
79bd9814 TH |
4891 | out_put_cfile: |
4892 | fdput(cfile); | |
4893 | out_put_eventfd: | |
4894 | eventfd_ctx_put(event->eventfd); | |
4895 | out_put_efile: | |
4896 | fdput(efile); | |
4897 | out_kfree: | |
4898 | kfree(event); | |
4899 | ||
4900 | return ret; | |
4901 | } | |
4902 | ||
c29b5b3d MS |
4903 | #if defined(CONFIG_MEMCG_KMEM) && (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) |
4904 | static int mem_cgroup_slab_show(struct seq_file *m, void *p) | |
4905 | { | |
4906 | /* | |
4907 | * Deprecated. | |
df4ae285 | 4908 | * Please, take a look at tools/cgroup/memcg_slabinfo.py . |
c29b5b3d MS |
4909 | */ |
4910 | return 0; | |
4911 | } | |
4912 | #endif | |
4913 | ||
241994ed | 4914 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4915 | { |
0eea1030 | 4916 | .name = "usage_in_bytes", |
8c7c6e34 | 4917 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4918 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4919 | }, |
c84872e1 PE |
4920 | { |
4921 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4922 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4923 | .write = mem_cgroup_reset, |
791badbd | 4924 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4925 | }, |
8cdea7c0 | 4926 | { |
0eea1030 | 4927 | .name = "limit_in_bytes", |
8c7c6e34 | 4928 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4929 | .write = mem_cgroup_write, |
791badbd | 4930 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4931 | }, |
296c81d8 BS |
4932 | { |
4933 | .name = "soft_limit_in_bytes", | |
4934 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4935 | .write = mem_cgroup_write, |
791badbd | 4936 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4937 | }, |
8cdea7c0 BS |
4938 | { |
4939 | .name = "failcnt", | |
8c7c6e34 | 4940 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4941 | .write = mem_cgroup_reset, |
791badbd | 4942 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4943 | }, |
d2ceb9b7 KH |
4944 | { |
4945 | .name = "stat", | |
2da8ca82 | 4946 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4947 | }, |
c1e862c1 KH |
4948 | { |
4949 | .name = "force_empty", | |
6770c64e | 4950 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4951 | }, |
18f59ea7 BS |
4952 | { |
4953 | .name = "use_hierarchy", | |
4954 | .write_u64 = mem_cgroup_hierarchy_write, | |
4955 | .read_u64 = mem_cgroup_hierarchy_read, | |
4956 | }, | |
79bd9814 | 4957 | { |
3bc942f3 | 4958 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4959 | .write = memcg_write_event_control, |
7dbdb199 | 4960 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4961 | }, |
a7885eb8 KM |
4962 | { |
4963 | .name = "swappiness", | |
4964 | .read_u64 = mem_cgroup_swappiness_read, | |
4965 | .write_u64 = mem_cgroup_swappiness_write, | |
4966 | }, | |
7dc74be0 DN |
4967 | { |
4968 | .name = "move_charge_at_immigrate", | |
4969 | .read_u64 = mem_cgroup_move_charge_read, | |
4970 | .write_u64 = mem_cgroup_move_charge_write, | |
4971 | }, | |
9490ff27 KH |
4972 | { |
4973 | .name = "oom_control", | |
2da8ca82 | 4974 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4975 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 | 4976 | }, |
70ddf637 AV |
4977 | { |
4978 | .name = "pressure_level", | |
70ddf637 | 4979 | }, |
406eb0c9 YH |
4980 | #ifdef CONFIG_NUMA |
4981 | { | |
4982 | .name = "numa_stat", | |
2da8ca82 | 4983 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4984 | }, |
4985 | #endif | |
510fc4e1 GC |
4986 | { |
4987 | .name = "kmem.limit_in_bytes", | |
4988 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4989 | .write = mem_cgroup_write, |
791badbd | 4990 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4991 | }, |
4992 | { | |
4993 | .name = "kmem.usage_in_bytes", | |
4994 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4995 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4996 | }, |
4997 | { | |
4998 | .name = "kmem.failcnt", | |
4999 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5000 | .write = mem_cgroup_reset, |
791badbd | 5001 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5002 | }, |
5003 | { | |
5004 | .name = "kmem.max_usage_in_bytes", | |
5005 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5006 | .write = mem_cgroup_reset, |
791badbd | 5007 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5008 | }, |
a87425a3 YS |
5009 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5010 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5011 | { |
5012 | .name = "kmem.slabinfo", | |
c29b5b3d | 5013 | .seq_show = mem_cgroup_slab_show, |
749c5415 GC |
5014 | }, |
5015 | #endif | |
d55f90bf VD |
5016 | { |
5017 | .name = "kmem.tcp.limit_in_bytes", | |
5018 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5019 | .write = mem_cgroup_write, | |
5020 | .read_u64 = mem_cgroup_read_u64, | |
5021 | }, | |
5022 | { | |
5023 | .name = "kmem.tcp.usage_in_bytes", | |
5024 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5025 | .read_u64 = mem_cgroup_read_u64, | |
5026 | }, | |
5027 | { | |
5028 | .name = "kmem.tcp.failcnt", | |
5029 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5030 | .write = mem_cgroup_reset, | |
5031 | .read_u64 = mem_cgroup_read_u64, | |
5032 | }, | |
5033 | { | |
5034 | .name = "kmem.tcp.max_usage_in_bytes", | |
5035 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5036 | .write = mem_cgroup_reset, | |
5037 | .read_u64 = mem_cgroup_read_u64, | |
5038 | }, | |
6bc10349 | 5039 | { }, /* terminate */ |
af36f906 | 5040 | }; |
8c7c6e34 | 5041 | |
73f576c0 JW |
5042 | /* |
5043 | * Private memory cgroup IDR | |
5044 | * | |
5045 | * Swap-out records and page cache shadow entries need to store memcg | |
5046 | * references in constrained space, so we maintain an ID space that is | |
5047 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5048 | * memory-controlled cgroups to 64k. | |
5049 | * | |
b8f2935f | 5050 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5051 | * the cgroup has been destroyed, such as page cache or reclaimable |
5052 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5053 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5054 | * relatively small ID space and prevent the creation of new cgroups | |
5055 | * even when there are much fewer than 64k cgroups - possibly none. | |
5056 | * | |
5057 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5058 | * be freed and recycled when it's no longer needed, which is usually | |
5059 | * when the CSS is offlined. | |
5060 | * | |
5061 | * The only exception to that are records of swapped out tmpfs/shmem | |
5062 | * pages that need to be attributed to live ancestors on swapin. But | |
5063 | * those references are manageable from userspace. | |
5064 | */ | |
5065 | ||
5066 | static DEFINE_IDR(mem_cgroup_idr); | |
5067 | ||
7e97de0b KT |
5068 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5069 | { | |
5070 | if (memcg->id.id > 0) { | |
5071 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5072 | memcg->id.id = 0; | |
5073 | } | |
5074 | } | |
5075 | ||
c1514c0a VF |
5076 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5077 | unsigned int n) | |
73f576c0 | 5078 | { |
1c2d479a | 5079 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5080 | } |
5081 | ||
615d66c3 | 5082 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5083 | { |
1c2d479a | 5084 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5085 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5086 | |
5087 | /* Memcg ID pins CSS */ | |
5088 | css_put(&memcg->css); | |
5089 | } | |
5090 | } | |
5091 | ||
615d66c3 VD |
5092 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5093 | { | |
5094 | mem_cgroup_id_put_many(memcg, 1); | |
5095 | } | |
5096 | ||
73f576c0 JW |
5097 | /** |
5098 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5099 | * @id: the memcg id to look up | |
5100 | * | |
5101 | * Caller must hold rcu_read_lock(). | |
5102 | */ | |
5103 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5104 | { | |
5105 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5106 | return idr_find(&mem_cgroup_idr, id); | |
5107 | } | |
5108 | ||
c15187a4 RG |
5109 | #ifdef CONFIG_SHRINKER_DEBUG |
5110 | struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino) | |
5111 | { | |
5112 | struct cgroup *cgrp; | |
5113 | struct cgroup_subsys_state *css; | |
5114 | struct mem_cgroup *memcg; | |
5115 | ||
5116 | cgrp = cgroup_get_from_id(ino); | |
5117 | if (!cgrp) | |
5118 | return ERR_PTR(-ENOENT); | |
5119 | ||
5120 | css = cgroup_get_e_css(cgrp, &memory_cgrp_subsys); | |
5121 | if (css) | |
5122 | memcg = container_of(css, struct mem_cgroup, css); | |
5123 | else | |
5124 | memcg = ERR_PTR(-ENOENT); | |
5125 | ||
5126 | cgroup_put(cgrp); | |
5127 | ||
5128 | return memcg; | |
5129 | } | |
5130 | #endif | |
5131 | ||
ef8f2327 | 5132 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5133 | { |
5134 | struct mem_cgroup_per_node *pn; | |
8c9bb398 WY |
5135 | |
5136 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
6d12e2d8 KH |
5137 | if (!pn) |
5138 | return 1; | |
1ecaab2b | 5139 | |
7e1c0d6f SB |
5140 | pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu, |
5141 | GFP_KERNEL_ACCOUNT); | |
5142 | if (!pn->lruvec_stats_percpu) { | |
00f3ca2c JW |
5143 | kfree(pn); |
5144 | return 1; | |
5145 | } | |
5146 | ||
ef8f2327 | 5147 | lruvec_init(&pn->lruvec); |
ef8f2327 MG |
5148 | pn->memcg = memcg; |
5149 | ||
54f72fe0 | 5150 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5151 | return 0; |
5152 | } | |
5153 | ||
ef8f2327 | 5154 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5155 | { |
00f3ca2c JW |
5156 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5157 | ||
4eaf431f MH |
5158 | if (!pn) |
5159 | return; | |
5160 | ||
7e1c0d6f | 5161 | free_percpu(pn->lruvec_stats_percpu); |
00f3ca2c | 5162 | kfree(pn); |
1ecaab2b KH |
5163 | } |
5164 | ||
40e952f9 | 5165 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5166 | { |
c8b2a36f | 5167 | int node; |
59927fb9 | 5168 | |
c8b2a36f | 5169 | for_each_node(node) |
ef8f2327 | 5170 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5171 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 5172 | kfree(memcg); |
59927fb9 | 5173 | } |
3afe36b1 | 5174 | |
40e952f9 TE |
5175 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5176 | { | |
5177 | memcg_wb_domain_exit(memcg); | |
5178 | __mem_cgroup_free(memcg); | |
5179 | } | |
5180 | ||
0b8f73e1 | 5181 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5182 | { |
d142e3e6 | 5183 | struct mem_cgroup *memcg; |
6d12e2d8 | 5184 | int node; |
97b27821 | 5185 | int __maybe_unused i; |
11d67612 | 5186 | long error = -ENOMEM; |
8cdea7c0 | 5187 | |
06b2c3b0 | 5188 | memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL); |
c0ff4b85 | 5189 | if (!memcg) |
11d67612 | 5190 | return ERR_PTR(error); |
0b8f73e1 | 5191 | |
73f576c0 | 5192 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
be740503 | 5193 | 1, MEM_CGROUP_ID_MAX + 1, GFP_KERNEL); |
11d67612 YS |
5194 | if (memcg->id.id < 0) { |
5195 | error = memcg->id.id; | |
73f576c0 | 5196 | goto fail; |
11d67612 | 5197 | } |
73f576c0 | 5198 | |
3e38e0aa RG |
5199 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5200 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5201 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5202 | goto fail; |
78fb7466 | 5203 | |
3ed28fa1 | 5204 | for_each_node(node) |
ef8f2327 | 5205 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5206 | goto fail; |
f64c3f54 | 5207 | |
0b8f73e1 JW |
5208 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5209 | goto fail; | |
28dbc4b6 | 5210 | |
f7e1cb6e | 5211 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5212 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5213 | mutex_init(&memcg->thresholds_lock); |
5214 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5215 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5216 | INIT_LIST_HEAD(&memcg->event_list); |
5217 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5218 | memcg->socket_pressure = jiffies; |
84c07d11 | 5219 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5220 | memcg->kmemcg_id = -1; |
bf4f0599 | 5221 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5222 | #endif |
52ebea74 TH |
5223 | #ifdef CONFIG_CGROUP_WRITEBACK |
5224 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5225 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5226 | memcg->cgwb_frn[i].done = | |
5227 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5228 | #endif |
5229 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5230 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5231 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5232 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5233 | #endif |
73f576c0 | 5234 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5235 | return memcg; |
5236 | fail: | |
7e97de0b | 5237 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5238 | __mem_cgroup_free(memcg); |
11d67612 | 5239 | return ERR_PTR(error); |
d142e3e6 GC |
5240 | } |
5241 | ||
0b8f73e1 JW |
5242 | static struct cgroup_subsys_state * __ref |
5243 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5244 | { |
0b8f73e1 | 5245 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5246 | struct mem_cgroup *memcg, *old_memcg; |
d142e3e6 | 5247 | |
b87d8cef | 5248 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5249 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5250 | set_active_memcg(old_memcg); |
11d67612 YS |
5251 | if (IS_ERR(memcg)) |
5252 | return ERR_CAST(memcg); | |
d142e3e6 | 5253 | |
d1663a90 | 5254 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5255 | memcg->soft_limit = PAGE_COUNTER_MAX; |
f4840ccf JW |
5256 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
5257 | memcg->zswap_max = PAGE_COUNTER_MAX; | |
5258 | #endif | |
4b82ab4f | 5259 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5260 | if (parent) { |
5261 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5262 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
bef8620c | 5263 | |
3e32cb2e | 5264 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5265 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5266 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5267 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5268 | } else { |
bef8620c RG |
5269 | page_counter_init(&memcg->memory, NULL); |
5270 | page_counter_init(&memcg->swap, NULL); | |
5271 | page_counter_init(&memcg->kmem, NULL); | |
5272 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5273 | |
0b8f73e1 JW |
5274 | root_mem_cgroup = memcg; |
5275 | return &memcg->css; | |
5276 | } | |
5277 | ||
f7e1cb6e | 5278 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5279 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5280 | |
0b8f73e1 | 5281 | return &memcg->css; |
0b8f73e1 JW |
5282 | } |
5283 | ||
73f576c0 | 5284 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5285 | { |
58fa2a55 VD |
5286 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5287 | ||
da0efe30 MS |
5288 | if (memcg_online_kmem(memcg)) |
5289 | goto remove_id; | |
5290 | ||
0a4465d3 | 5291 | /* |
e4262c4f | 5292 | * A memcg must be visible for expand_shrinker_info() |
0a4465d3 KT |
5293 | * by the time the maps are allocated. So, we allocate maps |
5294 | * here, when for_each_mem_cgroup() can't skip it. | |
5295 | */ | |
da0efe30 MS |
5296 | if (alloc_shrinker_info(memcg)) |
5297 | goto offline_kmem; | |
0a4465d3 | 5298 | |
73f576c0 | 5299 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5300 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5301 | css_get(css); |
aa48e47e SB |
5302 | |
5303 | if (unlikely(mem_cgroup_is_root(memcg))) | |
5304 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, | |
5305 | 2UL*HZ); | |
2f7dd7a4 | 5306 | return 0; |
da0efe30 MS |
5307 | offline_kmem: |
5308 | memcg_offline_kmem(memcg); | |
5309 | remove_id: | |
5310 | mem_cgroup_id_remove(memcg); | |
5311 | return -ENOMEM; | |
8cdea7c0 BS |
5312 | } |
5313 | ||
eb95419b | 5314 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5315 | { |
eb95419b | 5316 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5317 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5318 | |
5319 | /* | |
5320 | * Unregister events and notify userspace. | |
5321 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5322 | * directory to avoid race between userspace and kernelspace. | |
5323 | */ | |
4ba9515d | 5324 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5325 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
79bd9814 TH |
5326 | list_del_init(&event->list); |
5327 | schedule_work(&event->remove); | |
5328 | } | |
4ba9515d | 5329 | spin_unlock_irq(&memcg->event_list_lock); |
ec64f515 | 5330 | |
bf8d5d52 | 5331 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5332 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5333 | |
567e9ab2 | 5334 | memcg_offline_kmem(memcg); |
a178015c | 5335 | reparent_shrinker_deferred(memcg); |
52ebea74 | 5336 | wb_memcg_offline(memcg); |
73f576c0 | 5337 | |
591edfb1 RG |
5338 | drain_all_stock(memcg); |
5339 | ||
73f576c0 | 5340 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5341 | } |
5342 | ||
6df38689 VD |
5343 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5344 | { | |
5345 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5346 | ||
5347 | invalidate_reclaim_iterators(memcg); | |
5348 | } | |
5349 | ||
eb95419b | 5350 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5351 | { |
eb95419b | 5352 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5353 | int __maybe_unused i; |
c268e994 | 5354 | |
97b27821 TH |
5355 | #ifdef CONFIG_CGROUP_WRITEBACK |
5356 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5357 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5358 | #endif | |
f7e1cb6e | 5359 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5360 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5361 | |
0db15298 | 5362 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5363 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5364 | |
0b8f73e1 JW |
5365 | vmpressure_cleanup(&memcg->vmpressure); |
5366 | cancel_work_sync(&memcg->high_work); | |
5367 | mem_cgroup_remove_from_trees(memcg); | |
e4262c4f | 5368 | free_shrinker_info(memcg); |
0b8f73e1 | 5369 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5370 | } |
5371 | ||
1ced953b TH |
5372 | /** |
5373 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5374 | * @css: the target css | |
5375 | * | |
5376 | * Reset the states of the mem_cgroup associated with @css. This is | |
5377 | * invoked when the userland requests disabling on the default hierarchy | |
5378 | * but the memcg is pinned through dependency. The memcg should stop | |
5379 | * applying policies and should revert to the vanilla state as it may be | |
5380 | * made visible again. | |
5381 | * | |
5382 | * The current implementation only resets the essential configurations. | |
5383 | * This needs to be expanded to cover all the visible parts. | |
5384 | */ | |
5385 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5386 | { | |
5387 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5388 | ||
bbec2e15 RG |
5389 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5390 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5391 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5392 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5393 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5394 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5395 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5396 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5397 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5398 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5399 | } |
5400 | ||
2d146aa3 JW |
5401 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
5402 | { | |
5403 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5404 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5405 | struct memcg_vmstats_percpu *statc; | |
5406 | long delta, v; | |
7e1c0d6f | 5407 | int i, nid; |
2d146aa3 JW |
5408 | |
5409 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5410 | ||
5411 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
5412 | /* | |
5413 | * Collect the aggregated propagation counts of groups | |
5414 | * below us. We're in a per-cpu loop here and this is | |
5415 | * a global counter, so the first cycle will get them. | |
5416 | */ | |
5417 | delta = memcg->vmstats.state_pending[i]; | |
5418 | if (delta) | |
5419 | memcg->vmstats.state_pending[i] = 0; | |
5420 | ||
5421 | /* Add CPU changes on this level since the last flush */ | |
5422 | v = READ_ONCE(statc->state[i]); | |
5423 | if (v != statc->state_prev[i]) { | |
5424 | delta += v - statc->state_prev[i]; | |
5425 | statc->state_prev[i] = v; | |
5426 | } | |
5427 | ||
5428 | if (!delta) | |
5429 | continue; | |
5430 | ||
5431 | /* Aggregate counts on this level and propagate upwards */ | |
5432 | memcg->vmstats.state[i] += delta; | |
5433 | if (parent) | |
5434 | parent->vmstats.state_pending[i] += delta; | |
5435 | } | |
5436 | ||
5437 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { | |
5438 | delta = memcg->vmstats.events_pending[i]; | |
5439 | if (delta) | |
5440 | memcg->vmstats.events_pending[i] = 0; | |
5441 | ||
5442 | v = READ_ONCE(statc->events[i]); | |
5443 | if (v != statc->events_prev[i]) { | |
5444 | delta += v - statc->events_prev[i]; | |
5445 | statc->events_prev[i] = v; | |
5446 | } | |
5447 | ||
5448 | if (!delta) | |
5449 | continue; | |
5450 | ||
5451 | memcg->vmstats.events[i] += delta; | |
5452 | if (parent) | |
5453 | parent->vmstats.events_pending[i] += delta; | |
5454 | } | |
7e1c0d6f SB |
5455 | |
5456 | for_each_node_state(nid, N_MEMORY) { | |
5457 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; | |
5458 | struct mem_cgroup_per_node *ppn = NULL; | |
5459 | struct lruvec_stats_percpu *lstatc; | |
5460 | ||
5461 | if (parent) | |
5462 | ppn = parent->nodeinfo[nid]; | |
5463 | ||
5464 | lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu); | |
5465 | ||
5466 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { | |
5467 | delta = pn->lruvec_stats.state_pending[i]; | |
5468 | if (delta) | |
5469 | pn->lruvec_stats.state_pending[i] = 0; | |
5470 | ||
5471 | v = READ_ONCE(lstatc->state[i]); | |
5472 | if (v != lstatc->state_prev[i]) { | |
5473 | delta += v - lstatc->state_prev[i]; | |
5474 | lstatc->state_prev[i] = v; | |
5475 | } | |
5476 | ||
5477 | if (!delta) | |
5478 | continue; | |
5479 | ||
5480 | pn->lruvec_stats.state[i] += delta; | |
5481 | if (ppn) | |
5482 | ppn->lruvec_stats.state_pending[i] += delta; | |
5483 | } | |
5484 | } | |
2d146aa3 JW |
5485 | } |
5486 | ||
02491447 | 5487 | #ifdef CONFIG_MMU |
7dc74be0 | 5488 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5489 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5490 | { |
05b84301 | 5491 | int ret; |
9476db97 | 5492 | |
d0164adc MG |
5493 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5494 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5495 | if (!ret) { |
854ffa8d | 5496 | mc.precharge += count; |
854ffa8d DN |
5497 | return ret; |
5498 | } | |
9476db97 | 5499 | |
3674534b | 5500 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5501 | while (count--) { |
3674534b | 5502 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5503 | if (ret) |
38c5d72f | 5504 | return ret; |
854ffa8d | 5505 | mc.precharge++; |
9476db97 | 5506 | cond_resched(); |
854ffa8d | 5507 | } |
9476db97 | 5508 | return 0; |
4ffef5fe DN |
5509 | } |
5510 | ||
4ffef5fe DN |
5511 | union mc_target { |
5512 | struct page *page; | |
02491447 | 5513 | swp_entry_t ent; |
4ffef5fe DN |
5514 | }; |
5515 | ||
4ffef5fe | 5516 | enum mc_target_type { |
8d32ff84 | 5517 | MC_TARGET_NONE = 0, |
4ffef5fe | 5518 | MC_TARGET_PAGE, |
02491447 | 5519 | MC_TARGET_SWAP, |
c733a828 | 5520 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5521 | }; |
5522 | ||
90254a65 DN |
5523 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5524 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5525 | { |
25b2995a | 5526 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5527 | |
90254a65 DN |
5528 | if (!page || !page_mapped(page)) |
5529 | return NULL; | |
5530 | if (PageAnon(page)) { | |
1dfab5ab | 5531 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5532 | return NULL; |
1dfab5ab JW |
5533 | } else { |
5534 | if (!(mc.flags & MOVE_FILE)) | |
5535 | return NULL; | |
5536 | } | |
90254a65 DN |
5537 | if (!get_page_unless_zero(page)) |
5538 | return NULL; | |
5539 | ||
5540 | return page; | |
5541 | } | |
5542 | ||
c733a828 | 5543 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5544 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5545 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5546 | { |
90254a65 DN |
5547 | struct page *page = NULL; |
5548 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5549 | ||
9a137153 | 5550 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5551 | return NULL; |
c733a828 JG |
5552 | |
5553 | /* | |
27674ef6 CH |
5554 | * Handle device private pages that are not accessible by the CPU, but |
5555 | * stored as special swap entries in the page table. | |
c733a828 JG |
5556 | */ |
5557 | if (is_device_private_entry(ent)) { | |
af5cdaf8 | 5558 | page = pfn_swap_entry_to_page(ent); |
27674ef6 | 5559 | if (!get_page_unless_zero(page)) |
c733a828 JG |
5560 | return NULL; |
5561 | return page; | |
5562 | } | |
5563 | ||
9a137153 RC |
5564 | if (non_swap_entry(ent)) |
5565 | return NULL; | |
5566 | ||
4b91355e KH |
5567 | /* |
5568 | * Because lookup_swap_cache() updates some statistics counter, | |
5569 | * we call find_get_page() with swapper_space directly. | |
5570 | */ | |
f6ab1f7f | 5571 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5572 | entry->val = ent.val; |
90254a65 DN |
5573 | |
5574 | return page; | |
5575 | } | |
4b91355e KH |
5576 | #else |
5577 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5578 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5579 | { |
5580 | return NULL; | |
5581 | } | |
5582 | #endif | |
90254a65 | 5583 | |
87946a72 | 5584 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
48384b0b | 5585 | unsigned long addr, pte_t ptent) |
87946a72 | 5586 | { |
87946a72 DN |
5587 | if (!vma->vm_file) /* anonymous vma */ |
5588 | return NULL; | |
1dfab5ab | 5589 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5590 | return NULL; |
5591 | ||
87946a72 | 5592 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5593 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5594 | return find_get_incore_page(vma->vm_file->f_mapping, |
5595 | linear_page_index(vma, addr)); | |
87946a72 DN |
5596 | } |
5597 | ||
b1b0deab CG |
5598 | /** |
5599 | * mem_cgroup_move_account - move account of the page | |
5600 | * @page: the page | |
25843c2b | 5601 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5602 | * @from: mem_cgroup which the page is moved from. |
5603 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5604 | * | |
3ac808fd | 5605 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5606 | * |
5607 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5608 | * from old cgroup. | |
5609 | */ | |
5610 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5611 | bool compound, |
b1b0deab CG |
5612 | struct mem_cgroup *from, |
5613 | struct mem_cgroup *to) | |
5614 | { | |
fcce4672 | 5615 | struct folio *folio = page_folio(page); |
ae8af438 KK |
5616 | struct lruvec *from_vec, *to_vec; |
5617 | struct pglist_data *pgdat; | |
fcce4672 | 5618 | unsigned int nr_pages = compound ? folio_nr_pages(folio) : 1; |
8e88bd2d | 5619 | int nid, ret; |
b1b0deab CG |
5620 | |
5621 | VM_BUG_ON(from == to); | |
fcce4672 | 5622 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
9c325215 | 5623 | VM_BUG_ON(compound && !folio_test_large(folio)); |
b1b0deab CG |
5624 | |
5625 | /* | |
6a93ca8f | 5626 | * Prevent mem_cgroup_migrate() from looking at |
bcfe06bf | 5627 | * page's memory cgroup of its source page while we change it. |
b1b0deab | 5628 | */ |
f627c2f5 | 5629 | ret = -EBUSY; |
fcce4672 | 5630 | if (!folio_trylock(folio)) |
b1b0deab CG |
5631 | goto out; |
5632 | ||
5633 | ret = -EINVAL; | |
fcce4672 | 5634 | if (folio_memcg(folio) != from) |
b1b0deab CG |
5635 | goto out_unlock; |
5636 | ||
fcce4672 | 5637 | pgdat = folio_pgdat(folio); |
867e5e1d JW |
5638 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5639 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5640 | |
fcce4672 | 5641 | folio_memcg_lock(folio); |
b1b0deab | 5642 | |
fcce4672 MWO |
5643 | if (folio_test_anon(folio)) { |
5644 | if (folio_mapped(folio)) { | |
be5d0a74 JW |
5645 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); |
5646 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
fcce4672 | 5647 | if (folio_test_transhuge(folio)) { |
69473e5d MS |
5648 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
5649 | -nr_pages); | |
5650 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5651 | nr_pages); | |
468c3982 | 5652 | } |
be5d0a74 JW |
5653 | } |
5654 | } else { | |
0d1c2072 JW |
5655 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5656 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5657 | ||
fcce4672 | 5658 | if (folio_test_swapbacked(folio)) { |
0d1c2072 JW |
5659 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); |
5660 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5661 | } | |
5662 | ||
fcce4672 | 5663 | if (folio_mapped(folio)) { |
49e50d27 JW |
5664 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
5665 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5666 | } | |
b1b0deab | 5667 | |
fcce4672 MWO |
5668 | if (folio_test_dirty(folio)) { |
5669 | struct address_space *mapping = folio_mapping(folio); | |
c4843a75 | 5670 | |
f56753ac | 5671 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5672 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5673 | -nr_pages); | |
5674 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5675 | nr_pages); | |
5676 | } | |
c4843a75 GT |
5677 | } |
5678 | } | |
5679 | ||
fcce4672 | 5680 | if (folio_test_writeback(folio)) { |
ae8af438 KK |
5681 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5682 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5683 | } |
5684 | ||
5685 | /* | |
abb242f5 JW |
5686 | * All state has been migrated, let's switch to the new memcg. |
5687 | * | |
bcfe06bf | 5688 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
5689 | * is referenced, charged, isolated, and locked: we can't race |
5690 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 5691 | * that would rely on a stable page's memory cgroup. |
abb242f5 JW |
5692 | * |
5693 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
bcfe06bf | 5694 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
5695 | * new memcg that isn't locked, the above state can change |
5696 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5697 | */ |
abb242f5 | 5698 | smp_mb(); |
b1b0deab | 5699 | |
1a3e1f40 JW |
5700 | css_get(&to->css); |
5701 | css_put(&from->css); | |
5702 | ||
fcce4672 | 5703 | folio->memcg_data = (unsigned long)to; |
87eaceb3 | 5704 | |
f70ad448 | 5705 | __folio_memcg_unlock(from); |
b1b0deab CG |
5706 | |
5707 | ret = 0; | |
fcce4672 | 5708 | nid = folio_nid(folio); |
b1b0deab CG |
5709 | |
5710 | local_irq_disable(); | |
6e0110c2 | 5711 | mem_cgroup_charge_statistics(to, nr_pages); |
8e88bd2d | 5712 | memcg_check_events(to, nid); |
6e0110c2 | 5713 | mem_cgroup_charge_statistics(from, -nr_pages); |
8e88bd2d | 5714 | memcg_check_events(from, nid); |
b1b0deab CG |
5715 | local_irq_enable(); |
5716 | out_unlock: | |
fcce4672 | 5717 | folio_unlock(folio); |
b1b0deab CG |
5718 | out: |
5719 | return ret; | |
5720 | } | |
5721 | ||
7cf7806c LR |
5722 | /** |
5723 | * get_mctgt_type - get target type of moving charge | |
5724 | * @vma: the vma the pte to be checked belongs | |
5725 | * @addr: the address corresponding to the pte to be checked | |
5726 | * @ptent: the pte to be checked | |
5727 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5728 | * | |
5729 | * Returns | |
5730 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5731 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5732 | * move charge. if @target is not NULL, the page is stored in target->page | |
5733 | * with extra refcnt got(Callers should handle it). | |
5734 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5735 | * target for charge migration. if @target is not NULL, the entry is stored | |
5736 | * in target->ent. | |
f25cbb7a AS |
5737 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is device memory and |
5738 | * thus not on the lru. | |
df6ad698 JG |
5739 | * For now we such page is charge like a regular page would be as for all |
5740 | * intent and purposes it is just special memory taking the place of a | |
5741 | * regular page. | |
c733a828 JG |
5742 | * |
5743 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5744 | * |
5745 | * Called with pte lock held. | |
5746 | */ | |
5747 | ||
8d32ff84 | 5748 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5749 | unsigned long addr, pte_t ptent, union mc_target *target) |
5750 | { | |
5751 | struct page *page = NULL; | |
8d32ff84 | 5752 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5753 | swp_entry_t ent = { .val = 0 }; |
5754 | ||
5755 | if (pte_present(ptent)) | |
5756 | page = mc_handle_present_pte(vma, addr, ptent); | |
5c041f5d PX |
5757 | else if (pte_none_mostly(ptent)) |
5758 | /* | |
5759 | * PTE markers should be treated as a none pte here, separated | |
5760 | * from other swap handling below. | |
5761 | */ | |
5762 | page = mc_handle_file_pte(vma, addr, ptent); | |
90254a65 | 5763 | else if (is_swap_pte(ptent)) |
48406ef8 | 5764 | page = mc_handle_swap_pte(vma, ptent, &ent); |
90254a65 DN |
5765 | |
5766 | if (!page && !ent.val) | |
8d32ff84 | 5767 | return ret; |
02491447 | 5768 | if (page) { |
02491447 | 5769 | /* |
0a31bc97 | 5770 | * Do only loose check w/o serialization. |
1306a85a | 5771 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5772 | * not under LRU exclusion. |
02491447 | 5773 | */ |
bcfe06bf | 5774 | if (page_memcg(page) == mc.from) { |
02491447 | 5775 | ret = MC_TARGET_PAGE; |
f25cbb7a AS |
5776 | if (is_device_private_page(page) || |
5777 | is_device_coherent_page(page)) | |
c733a828 | 5778 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5779 | if (target) |
5780 | target->page = page; | |
5781 | } | |
5782 | if (!ret || !target) | |
5783 | put_page(page); | |
5784 | } | |
3e14a57b HY |
5785 | /* |
5786 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5787 | * But we cannot move a tail-page in a THP. | |
5788 | */ | |
5789 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5790 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5791 | ret = MC_TARGET_SWAP; |
5792 | if (target) | |
5793 | target->ent = ent; | |
4ffef5fe | 5794 | } |
4ffef5fe DN |
5795 | return ret; |
5796 | } | |
5797 | ||
12724850 NH |
5798 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5799 | /* | |
d6810d73 HY |
5800 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5801 | * not support them for now. | |
12724850 NH |
5802 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5803 | */ | |
5804 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5805 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5806 | { | |
5807 | struct page *page = NULL; | |
12724850 NH |
5808 | enum mc_target_type ret = MC_TARGET_NONE; |
5809 | ||
84c3fc4e ZY |
5810 | if (unlikely(is_swap_pmd(pmd))) { |
5811 | VM_BUG_ON(thp_migration_supported() && | |
5812 | !is_pmd_migration_entry(pmd)); | |
5813 | return ret; | |
5814 | } | |
12724850 | 5815 | page = pmd_page(pmd); |
309381fe | 5816 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5817 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5818 | return ret; |
bcfe06bf | 5819 | if (page_memcg(page) == mc.from) { |
12724850 NH |
5820 | ret = MC_TARGET_PAGE; |
5821 | if (target) { | |
5822 | get_page(page); | |
5823 | target->page = page; | |
5824 | } | |
5825 | } | |
5826 | return ret; | |
5827 | } | |
5828 | #else | |
5829 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5830 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5831 | { | |
5832 | return MC_TARGET_NONE; | |
5833 | } | |
5834 | #endif | |
5835 | ||
4ffef5fe DN |
5836 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5837 | unsigned long addr, unsigned long end, | |
5838 | struct mm_walk *walk) | |
5839 | { | |
26bcd64a | 5840 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5841 | pte_t *pte; |
5842 | spinlock_t *ptl; | |
5843 | ||
b6ec57f4 KS |
5844 | ptl = pmd_trans_huge_lock(pmd, vma); |
5845 | if (ptl) { | |
c733a828 JG |
5846 | /* |
5847 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5848 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5849 | * this might change. | |
c733a828 | 5850 | */ |
12724850 NH |
5851 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5852 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5853 | spin_unlock(ptl); |
1a5a9906 | 5854 | return 0; |
12724850 | 5855 | } |
03319327 | 5856 | |
45f83cef AA |
5857 | if (pmd_trans_unstable(pmd)) |
5858 | return 0; | |
4ffef5fe DN |
5859 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5860 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5861 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5862 | mc.precharge++; /* increment precharge temporarily */ |
5863 | pte_unmap_unlock(pte - 1, ptl); | |
5864 | cond_resched(); | |
5865 | ||
7dc74be0 DN |
5866 | return 0; |
5867 | } | |
5868 | ||
7b86ac33 CH |
5869 | static const struct mm_walk_ops precharge_walk_ops = { |
5870 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5871 | }; | |
5872 | ||
4ffef5fe DN |
5873 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5874 | { | |
5875 | unsigned long precharge; | |
4ffef5fe | 5876 | |
d8ed45c5 | 5877 | mmap_read_lock(mm); |
7b86ac33 | 5878 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5879 | mmap_read_unlock(mm); |
4ffef5fe DN |
5880 | |
5881 | precharge = mc.precharge; | |
5882 | mc.precharge = 0; | |
5883 | ||
5884 | return precharge; | |
5885 | } | |
5886 | ||
4ffef5fe DN |
5887 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5888 | { | |
dfe076b0 DN |
5889 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5890 | ||
5891 | VM_BUG_ON(mc.moving_task); | |
5892 | mc.moving_task = current; | |
5893 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5894 | } |
5895 | ||
dfe076b0 DN |
5896 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5897 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5898 | { |
2bd9bb20 KH |
5899 | struct mem_cgroup *from = mc.from; |
5900 | struct mem_cgroup *to = mc.to; | |
5901 | ||
4ffef5fe | 5902 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5903 | if (mc.precharge) { |
00501b53 | 5904 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5905 | mc.precharge = 0; |
5906 | } | |
5907 | /* | |
5908 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5909 | * we must uncharge here. | |
5910 | */ | |
5911 | if (mc.moved_charge) { | |
00501b53 | 5912 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5913 | mc.moved_charge = 0; |
4ffef5fe | 5914 | } |
483c30b5 DN |
5915 | /* we must fixup refcnts and charges */ |
5916 | if (mc.moved_swap) { | |
483c30b5 | 5917 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5918 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5919 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5920 | |
615d66c3 VD |
5921 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5922 | ||
05b84301 | 5923 | /* |
3e32cb2e JW |
5924 | * we charged both to->memory and to->memsw, so we |
5925 | * should uncharge to->memory. | |
05b84301 | 5926 | */ |
ce00a967 | 5927 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5928 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5929 | ||
483c30b5 DN |
5930 | mc.moved_swap = 0; |
5931 | } | |
dfe076b0 DN |
5932 | memcg_oom_recover(from); |
5933 | memcg_oom_recover(to); | |
5934 | wake_up_all(&mc.waitq); | |
5935 | } | |
5936 | ||
5937 | static void mem_cgroup_clear_mc(void) | |
5938 | { | |
264a0ae1 TH |
5939 | struct mm_struct *mm = mc.mm; |
5940 | ||
dfe076b0 DN |
5941 | /* |
5942 | * we must clear moving_task before waking up waiters at the end of | |
5943 | * task migration. | |
5944 | */ | |
5945 | mc.moving_task = NULL; | |
5946 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5947 | spin_lock(&mc.lock); |
4ffef5fe DN |
5948 | mc.from = NULL; |
5949 | mc.to = NULL; | |
264a0ae1 | 5950 | mc.mm = NULL; |
2bd9bb20 | 5951 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5952 | |
5953 | mmput(mm); | |
4ffef5fe DN |
5954 | } |
5955 | ||
1f7dd3e5 | 5956 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5957 | { |
1f7dd3e5 | 5958 | struct cgroup_subsys_state *css; |
eed67d75 | 5959 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5960 | struct mem_cgroup *from; |
4530eddb | 5961 | struct task_struct *leader, *p; |
9f2115f9 | 5962 | struct mm_struct *mm; |
1dfab5ab | 5963 | unsigned long move_flags; |
9f2115f9 | 5964 | int ret = 0; |
7dc74be0 | 5965 | |
1f7dd3e5 TH |
5966 | /* charge immigration isn't supported on the default hierarchy */ |
5967 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5968 | return 0; |
5969 | ||
4530eddb TH |
5970 | /* |
5971 | * Multi-process migrations only happen on the default hierarchy | |
5972 | * where charge immigration is not used. Perform charge | |
5973 | * immigration if @tset contains a leader and whine if there are | |
5974 | * multiple. | |
5975 | */ | |
5976 | p = NULL; | |
1f7dd3e5 | 5977 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5978 | WARN_ON_ONCE(p); |
5979 | p = leader; | |
1f7dd3e5 | 5980 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5981 | } |
5982 | if (!p) | |
5983 | return 0; | |
5984 | ||
1f7dd3e5 | 5985 | /* |
f0953a1b | 5986 | * We are now committed to this value whatever it is. Changes in this |
1f7dd3e5 TH |
5987 | * tunable will only affect upcoming migrations, not the current one. |
5988 | * So we need to save it, and keep it going. | |
5989 | */ | |
5990 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5991 | if (!move_flags) | |
5992 | return 0; | |
5993 | ||
9f2115f9 TH |
5994 | from = mem_cgroup_from_task(p); |
5995 | ||
5996 | VM_BUG_ON(from == memcg); | |
5997 | ||
5998 | mm = get_task_mm(p); | |
5999 | if (!mm) | |
6000 | return 0; | |
6001 | /* We move charges only when we move a owner of the mm */ | |
6002 | if (mm->owner == p) { | |
6003 | VM_BUG_ON(mc.from); | |
6004 | VM_BUG_ON(mc.to); | |
6005 | VM_BUG_ON(mc.precharge); | |
6006 | VM_BUG_ON(mc.moved_charge); | |
6007 | VM_BUG_ON(mc.moved_swap); | |
6008 | ||
6009 | spin_lock(&mc.lock); | |
264a0ae1 | 6010 | mc.mm = mm; |
9f2115f9 TH |
6011 | mc.from = from; |
6012 | mc.to = memcg; | |
6013 | mc.flags = move_flags; | |
6014 | spin_unlock(&mc.lock); | |
6015 | /* We set mc.moving_task later */ | |
6016 | ||
6017 | ret = mem_cgroup_precharge_mc(mm); | |
6018 | if (ret) | |
6019 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6020 | } else { |
6021 | mmput(mm); | |
7dc74be0 DN |
6022 | } |
6023 | return ret; | |
6024 | } | |
6025 | ||
1f7dd3e5 | 6026 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6027 | { |
4e2f245d JW |
6028 | if (mc.to) |
6029 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6030 | } |
6031 | ||
4ffef5fe DN |
6032 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6033 | unsigned long addr, unsigned long end, | |
6034 | struct mm_walk *walk) | |
7dc74be0 | 6035 | { |
4ffef5fe | 6036 | int ret = 0; |
26bcd64a | 6037 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6038 | pte_t *pte; |
6039 | spinlock_t *ptl; | |
12724850 NH |
6040 | enum mc_target_type target_type; |
6041 | union mc_target target; | |
6042 | struct page *page; | |
4ffef5fe | 6043 | |
b6ec57f4 KS |
6044 | ptl = pmd_trans_huge_lock(pmd, vma); |
6045 | if (ptl) { | |
62ade86a | 6046 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6047 | spin_unlock(ptl); |
12724850 NH |
6048 | return 0; |
6049 | } | |
6050 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6051 | if (target_type == MC_TARGET_PAGE) { | |
6052 | page = target.page; | |
6053 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6054 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6055 | mc.from, mc.to)) { |
12724850 NH |
6056 | mc.precharge -= HPAGE_PMD_NR; |
6057 | mc.moved_charge += HPAGE_PMD_NR; | |
6058 | } | |
6059 | putback_lru_page(page); | |
6060 | } | |
6061 | put_page(page); | |
c733a828 JG |
6062 | } else if (target_type == MC_TARGET_DEVICE) { |
6063 | page = target.page; | |
6064 | if (!mem_cgroup_move_account(page, true, | |
6065 | mc.from, mc.to)) { | |
6066 | mc.precharge -= HPAGE_PMD_NR; | |
6067 | mc.moved_charge += HPAGE_PMD_NR; | |
6068 | } | |
6069 | put_page(page); | |
12724850 | 6070 | } |
bf929152 | 6071 | spin_unlock(ptl); |
1a5a9906 | 6072 | return 0; |
12724850 NH |
6073 | } |
6074 | ||
45f83cef AA |
6075 | if (pmd_trans_unstable(pmd)) |
6076 | return 0; | |
4ffef5fe DN |
6077 | retry: |
6078 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6079 | for (; addr != end; addr += PAGE_SIZE) { | |
6080 | pte_t ptent = *(pte++); | |
c733a828 | 6081 | bool device = false; |
02491447 | 6082 | swp_entry_t ent; |
4ffef5fe DN |
6083 | |
6084 | if (!mc.precharge) | |
6085 | break; | |
6086 | ||
8d32ff84 | 6087 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6088 | case MC_TARGET_DEVICE: |
6089 | device = true; | |
e4a9bc58 | 6090 | fallthrough; |
4ffef5fe DN |
6091 | case MC_TARGET_PAGE: |
6092 | page = target.page; | |
53f9263b KS |
6093 | /* |
6094 | * We can have a part of the split pmd here. Moving it | |
6095 | * can be done but it would be too convoluted so simply | |
6096 | * ignore such a partial THP and keep it in original | |
6097 | * memcg. There should be somebody mapping the head. | |
6098 | */ | |
6099 | if (PageTransCompound(page)) | |
6100 | goto put; | |
c733a828 | 6101 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6102 | goto put; |
f627c2f5 KS |
6103 | if (!mem_cgroup_move_account(page, false, |
6104 | mc.from, mc.to)) { | |
4ffef5fe | 6105 | mc.precharge--; |
854ffa8d DN |
6106 | /* we uncharge from mc.from later. */ |
6107 | mc.moved_charge++; | |
4ffef5fe | 6108 | } |
c733a828 JG |
6109 | if (!device) |
6110 | putback_lru_page(page); | |
8d32ff84 | 6111 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6112 | put_page(page); |
6113 | break; | |
02491447 DN |
6114 | case MC_TARGET_SWAP: |
6115 | ent = target.ent; | |
e91cbb42 | 6116 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6117 | mc.precharge--; |
8d22a935 HD |
6118 | mem_cgroup_id_get_many(mc.to, 1); |
6119 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6120 | mc.moved_swap++; |
6121 | } | |
02491447 | 6122 | break; |
4ffef5fe DN |
6123 | default: |
6124 | break; | |
6125 | } | |
6126 | } | |
6127 | pte_unmap_unlock(pte - 1, ptl); | |
6128 | cond_resched(); | |
6129 | ||
6130 | if (addr != end) { | |
6131 | /* | |
6132 | * We have consumed all precharges we got in can_attach(). | |
6133 | * We try charge one by one, but don't do any additional | |
6134 | * charges to mc.to if we have failed in charge once in attach() | |
6135 | * phase. | |
6136 | */ | |
854ffa8d | 6137 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6138 | if (!ret) |
6139 | goto retry; | |
6140 | } | |
6141 | ||
6142 | return ret; | |
6143 | } | |
6144 | ||
7b86ac33 CH |
6145 | static const struct mm_walk_ops charge_walk_ops = { |
6146 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6147 | }; | |
6148 | ||
264a0ae1 | 6149 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6150 | { |
4ffef5fe | 6151 | lru_add_drain_all(); |
312722cb | 6152 | /* |
81f8c3a4 JW |
6153 | * Signal lock_page_memcg() to take the memcg's move_lock |
6154 | * while we're moving its pages to another memcg. Then wait | |
6155 | * for already started RCU-only updates to finish. | |
312722cb JW |
6156 | */ |
6157 | atomic_inc(&mc.from->moving_account); | |
6158 | synchronize_rcu(); | |
dfe076b0 | 6159 | retry: |
d8ed45c5 | 6160 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6161 | /* |
c1e8d7c6 | 6162 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6163 | * waitq. So we cancel all extra charges, wake up all waiters, |
6164 | * and retry. Because we cancel precharges, we might not be able | |
6165 | * to move enough charges, but moving charge is a best-effort | |
6166 | * feature anyway, so it wouldn't be a big problem. | |
6167 | */ | |
6168 | __mem_cgroup_clear_mc(); | |
6169 | cond_resched(); | |
6170 | goto retry; | |
6171 | } | |
26bcd64a NH |
6172 | /* |
6173 | * When we have consumed all precharges and failed in doing | |
6174 | * additional charge, the page walk just aborts. | |
6175 | */ | |
7b86ac33 CH |
6176 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6177 | NULL); | |
0247f3f4 | 6178 | |
d8ed45c5 | 6179 | mmap_read_unlock(mc.mm); |
312722cb | 6180 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6181 | } |
6182 | ||
264a0ae1 | 6183 | static void mem_cgroup_move_task(void) |
67e465a7 | 6184 | { |
264a0ae1 TH |
6185 | if (mc.to) { |
6186 | mem_cgroup_move_charge(); | |
a433658c | 6187 | mem_cgroup_clear_mc(); |
264a0ae1 | 6188 | } |
67e465a7 | 6189 | } |
5cfb80a7 | 6190 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6191 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6192 | { |
6193 | return 0; | |
6194 | } | |
1f7dd3e5 | 6195 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6196 | { |
6197 | } | |
264a0ae1 | 6198 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6199 | { |
6200 | } | |
6201 | #endif | |
67e465a7 | 6202 | |
677dc973 CD |
6203 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6204 | { | |
6205 | if (value == PAGE_COUNTER_MAX) | |
6206 | seq_puts(m, "max\n"); | |
6207 | else | |
6208 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6209 | ||
6210 | return 0; | |
6211 | } | |
6212 | ||
241994ed JW |
6213 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6214 | struct cftype *cft) | |
6215 | { | |
f5fc3c5d JW |
6216 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6217 | ||
6218 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6219 | } |
6220 | ||
8e20d4b3 GR |
6221 | static u64 memory_peak_read(struct cgroup_subsys_state *css, |
6222 | struct cftype *cft) | |
6223 | { | |
6224 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6225 | ||
6226 | return (u64)memcg->memory.watermark * PAGE_SIZE; | |
6227 | } | |
6228 | ||
bf8d5d52 RG |
6229 | static int memory_min_show(struct seq_file *m, void *v) |
6230 | { | |
677dc973 CD |
6231 | return seq_puts_memcg_tunable(m, |
6232 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6233 | } |
6234 | ||
6235 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6236 | char *buf, size_t nbytes, loff_t off) | |
6237 | { | |
6238 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6239 | unsigned long min; | |
6240 | int err; | |
6241 | ||
6242 | buf = strstrip(buf); | |
6243 | err = page_counter_memparse(buf, "max", &min); | |
6244 | if (err) | |
6245 | return err; | |
6246 | ||
6247 | page_counter_set_min(&memcg->memory, min); | |
6248 | ||
6249 | return nbytes; | |
6250 | } | |
6251 | ||
241994ed JW |
6252 | static int memory_low_show(struct seq_file *m, void *v) |
6253 | { | |
677dc973 CD |
6254 | return seq_puts_memcg_tunable(m, |
6255 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6256 | } |
6257 | ||
6258 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6259 | char *buf, size_t nbytes, loff_t off) | |
6260 | { | |
6261 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6262 | unsigned long low; | |
6263 | int err; | |
6264 | ||
6265 | buf = strstrip(buf); | |
d2973697 | 6266 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6267 | if (err) |
6268 | return err; | |
6269 | ||
23067153 | 6270 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6271 | |
6272 | return nbytes; | |
6273 | } | |
6274 | ||
6275 | static int memory_high_show(struct seq_file *m, void *v) | |
6276 | { | |
d1663a90 JK |
6277 | return seq_puts_memcg_tunable(m, |
6278 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6279 | } |
6280 | ||
6281 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6282 | char *buf, size_t nbytes, loff_t off) | |
6283 | { | |
6284 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6285 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6286 | bool drained = false; |
241994ed JW |
6287 | unsigned long high; |
6288 | int err; | |
6289 | ||
6290 | buf = strstrip(buf); | |
d2973697 | 6291 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6292 | if (err) |
6293 | return err; | |
6294 | ||
e82553c1 JW |
6295 | page_counter_set_high(&memcg->memory, high); |
6296 | ||
8c8c383c JW |
6297 | for (;;) { |
6298 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6299 | unsigned long reclaimed; | |
6300 | ||
6301 | if (nr_pages <= high) | |
6302 | break; | |
6303 | ||
6304 | if (signal_pending(current)) | |
6305 | break; | |
6306 | ||
6307 | if (!drained) { | |
6308 | drain_all_stock(memcg); | |
6309 | drained = true; | |
6310 | continue; | |
6311 | } | |
6312 | ||
6313 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
73b73bac | 6314 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP); |
8c8c383c JW |
6315 | |
6316 | if (!reclaimed && !nr_retries--) | |
6317 | break; | |
6318 | } | |
588083bb | 6319 | |
19ce33ac | 6320 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6321 | return nbytes; |
6322 | } | |
6323 | ||
6324 | static int memory_max_show(struct seq_file *m, void *v) | |
6325 | { | |
677dc973 CD |
6326 | return seq_puts_memcg_tunable(m, |
6327 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6328 | } |
6329 | ||
6330 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6331 | char *buf, size_t nbytes, loff_t off) | |
6332 | { | |
6333 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6334 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6335 | bool drained = false; |
241994ed JW |
6336 | unsigned long max; |
6337 | int err; | |
6338 | ||
6339 | buf = strstrip(buf); | |
d2973697 | 6340 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6341 | if (err) |
6342 | return err; | |
6343 | ||
bbec2e15 | 6344 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6345 | |
6346 | for (;;) { | |
6347 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6348 | ||
6349 | if (nr_pages <= max) | |
6350 | break; | |
6351 | ||
7249c9f0 | 6352 | if (signal_pending(current)) |
b6e6edcf | 6353 | break; |
b6e6edcf JW |
6354 | |
6355 | if (!drained) { | |
6356 | drain_all_stock(memcg); | |
6357 | drained = true; | |
6358 | continue; | |
6359 | } | |
6360 | ||
6361 | if (nr_reclaims) { | |
6362 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
73b73bac | 6363 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP)) |
b6e6edcf JW |
6364 | nr_reclaims--; |
6365 | continue; | |
6366 | } | |
6367 | ||
e27be240 | 6368 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6369 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6370 | break; | |
6371 | } | |
241994ed | 6372 | |
2529bb3a | 6373 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6374 | return nbytes; |
6375 | } | |
6376 | ||
1e577f97 SB |
6377 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6378 | { | |
6379 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6380 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6381 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6382 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6383 | seq_printf(m, "oom_kill %lu\n", | |
6384 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
b6bf9abb DS |
6385 | seq_printf(m, "oom_group_kill %lu\n", |
6386 | atomic_long_read(&events[MEMCG_OOM_GROUP_KILL])); | |
1e577f97 SB |
6387 | } |
6388 | ||
241994ed JW |
6389 | static int memory_events_show(struct seq_file *m, void *v) |
6390 | { | |
aa9694bb | 6391 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6392 | |
1e577f97 SB |
6393 | __memory_events_show(m, memcg->memory_events); |
6394 | return 0; | |
6395 | } | |
6396 | ||
6397 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6398 | { | |
6399 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6400 | |
1e577f97 | 6401 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6402 | return 0; |
6403 | } | |
6404 | ||
587d9f72 JW |
6405 | static int memory_stat_show(struct seq_file *m, void *v) |
6406 | { | |
aa9694bb | 6407 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6408 | char *buf; |
1ff9e6e1 | 6409 | |
c8713d0b JW |
6410 | buf = memory_stat_format(memcg); |
6411 | if (!buf) | |
6412 | return -ENOMEM; | |
6413 | seq_puts(m, buf); | |
6414 | kfree(buf); | |
587d9f72 JW |
6415 | return 0; |
6416 | } | |
6417 | ||
5f9a4f4a | 6418 | #ifdef CONFIG_NUMA |
fff66b79 MS |
6419 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
6420 | int item) | |
6421 | { | |
6422 | return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item); | |
6423 | } | |
6424 | ||
5f9a4f4a MS |
6425 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
6426 | { | |
6427 | int i; | |
6428 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6429 | ||
fd25a9e0 | 6430 | mem_cgroup_flush_stats(); |
7e1c0d6f | 6431 | |
5f9a4f4a MS |
6432 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
6433 | int nid; | |
6434 | ||
6435 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6436 | continue; | |
6437 | ||
6438 | seq_printf(m, "%s", memory_stats[i].name); | |
6439 | for_each_node_state(nid, N_MEMORY) { | |
6440 | u64 size; | |
6441 | struct lruvec *lruvec; | |
6442 | ||
6443 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
6444 | size = lruvec_page_state_output(lruvec, |
6445 | memory_stats[i].idx); | |
5f9a4f4a MS |
6446 | seq_printf(m, " N%d=%llu", nid, size); |
6447 | } | |
6448 | seq_putc(m, '\n'); | |
6449 | } | |
6450 | ||
6451 | return 0; | |
6452 | } | |
6453 | #endif | |
6454 | ||
3d8b38eb RG |
6455 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6456 | { | |
aa9694bb | 6457 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6458 | |
6459 | seq_printf(m, "%d\n", memcg->oom_group); | |
6460 | ||
6461 | return 0; | |
6462 | } | |
6463 | ||
6464 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6465 | char *buf, size_t nbytes, loff_t off) | |
6466 | { | |
6467 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6468 | int ret, oom_group; | |
6469 | ||
6470 | buf = strstrip(buf); | |
6471 | if (!buf) | |
6472 | return -EINVAL; | |
6473 | ||
6474 | ret = kstrtoint(buf, 0, &oom_group); | |
6475 | if (ret) | |
6476 | return ret; | |
6477 | ||
6478 | if (oom_group != 0 && oom_group != 1) | |
6479 | return -EINVAL; | |
6480 | ||
6481 | memcg->oom_group = oom_group; | |
6482 | ||
6483 | return nbytes; | |
6484 | } | |
6485 | ||
94968384 SB |
6486 | static ssize_t memory_reclaim(struct kernfs_open_file *of, char *buf, |
6487 | size_t nbytes, loff_t off) | |
6488 | { | |
6489 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6490 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; | |
6491 | unsigned long nr_to_reclaim, nr_reclaimed = 0; | |
73b73bac | 6492 | unsigned int reclaim_options; |
94968384 SB |
6493 | int err; |
6494 | ||
6495 | buf = strstrip(buf); | |
6496 | err = page_counter_memparse(buf, "", &nr_to_reclaim); | |
6497 | if (err) | |
6498 | return err; | |
6499 | ||
73b73bac | 6500 | reclaim_options = MEMCG_RECLAIM_MAY_SWAP | MEMCG_RECLAIM_PROACTIVE; |
94968384 SB |
6501 | while (nr_reclaimed < nr_to_reclaim) { |
6502 | unsigned long reclaimed; | |
6503 | ||
6504 | if (signal_pending(current)) | |
6505 | return -EINTR; | |
6506 | ||
6507 | /* | |
6508 | * This is the final attempt, drain percpu lru caches in the | |
6509 | * hope of introducing more evictable pages for | |
6510 | * try_to_free_mem_cgroup_pages(). | |
6511 | */ | |
6512 | if (!nr_retries) | |
6513 | lru_add_drain_all(); | |
6514 | ||
6515 | reclaimed = try_to_free_mem_cgroup_pages(memcg, | |
6516 | nr_to_reclaim - nr_reclaimed, | |
73b73bac | 6517 | GFP_KERNEL, reclaim_options); |
94968384 SB |
6518 | |
6519 | if (!reclaimed && !nr_retries--) | |
6520 | return -EAGAIN; | |
6521 | ||
6522 | nr_reclaimed += reclaimed; | |
6523 | } | |
6524 | ||
6525 | return nbytes; | |
6526 | } | |
6527 | ||
241994ed JW |
6528 | static struct cftype memory_files[] = { |
6529 | { | |
6530 | .name = "current", | |
f5fc3c5d | 6531 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6532 | .read_u64 = memory_current_read, |
6533 | }, | |
8e20d4b3 GR |
6534 | { |
6535 | .name = "peak", | |
6536 | .flags = CFTYPE_NOT_ON_ROOT, | |
6537 | .read_u64 = memory_peak_read, | |
6538 | }, | |
bf8d5d52 RG |
6539 | { |
6540 | .name = "min", | |
6541 | .flags = CFTYPE_NOT_ON_ROOT, | |
6542 | .seq_show = memory_min_show, | |
6543 | .write = memory_min_write, | |
6544 | }, | |
241994ed JW |
6545 | { |
6546 | .name = "low", | |
6547 | .flags = CFTYPE_NOT_ON_ROOT, | |
6548 | .seq_show = memory_low_show, | |
6549 | .write = memory_low_write, | |
6550 | }, | |
6551 | { | |
6552 | .name = "high", | |
6553 | .flags = CFTYPE_NOT_ON_ROOT, | |
6554 | .seq_show = memory_high_show, | |
6555 | .write = memory_high_write, | |
6556 | }, | |
6557 | { | |
6558 | .name = "max", | |
6559 | .flags = CFTYPE_NOT_ON_ROOT, | |
6560 | .seq_show = memory_max_show, | |
6561 | .write = memory_max_write, | |
6562 | }, | |
6563 | { | |
6564 | .name = "events", | |
6565 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6566 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6567 | .seq_show = memory_events_show, |
6568 | }, | |
1e577f97 SB |
6569 | { |
6570 | .name = "events.local", | |
6571 | .flags = CFTYPE_NOT_ON_ROOT, | |
6572 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6573 | .seq_show = memory_events_local_show, | |
6574 | }, | |
587d9f72 JW |
6575 | { |
6576 | .name = "stat", | |
587d9f72 JW |
6577 | .seq_show = memory_stat_show, |
6578 | }, | |
5f9a4f4a MS |
6579 | #ifdef CONFIG_NUMA |
6580 | { | |
6581 | .name = "numa_stat", | |
6582 | .seq_show = memory_numa_stat_show, | |
6583 | }, | |
6584 | #endif | |
3d8b38eb RG |
6585 | { |
6586 | .name = "oom.group", | |
6587 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6588 | .seq_show = memory_oom_group_show, | |
6589 | .write = memory_oom_group_write, | |
6590 | }, | |
94968384 SB |
6591 | { |
6592 | .name = "reclaim", | |
6593 | .flags = CFTYPE_NS_DELEGATABLE, | |
6594 | .write = memory_reclaim, | |
6595 | }, | |
241994ed JW |
6596 | { } /* terminate */ |
6597 | }; | |
6598 | ||
073219e9 | 6599 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6600 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6601 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6602 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6603 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6604 | .css_free = mem_cgroup_css_free, |
1ced953b | 6605 | .css_reset = mem_cgroup_css_reset, |
2d146aa3 | 6606 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
7dc74be0 DN |
6607 | .can_attach = mem_cgroup_can_attach, |
6608 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6609 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6610 | .dfl_cftypes = memory_files, |
6611 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6612 | .early_init = 0, |
8cdea7c0 | 6613 | }; |
c077719b | 6614 | |
bc50bcc6 JW |
6615 | /* |
6616 | * This function calculates an individual cgroup's effective | |
6617 | * protection which is derived from its own memory.min/low, its | |
6618 | * parent's and siblings' settings, as well as the actual memory | |
6619 | * distribution in the tree. | |
6620 | * | |
6621 | * The following rules apply to the effective protection values: | |
6622 | * | |
6623 | * 1. At the first level of reclaim, effective protection is equal to | |
6624 | * the declared protection in memory.min and memory.low. | |
6625 | * | |
6626 | * 2. To enable safe delegation of the protection configuration, at | |
6627 | * subsequent levels the effective protection is capped to the | |
6628 | * parent's effective protection. | |
6629 | * | |
6630 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6631 | * user is allowed to overcommit the declared protection at a given | |
6632 | * level. If that is the case, the parent's effective protection is | |
6633 | * distributed to the children in proportion to how much protection | |
6634 | * they have declared and how much of it they are utilizing. | |
6635 | * | |
6636 | * This makes distribution proportional, but also work-conserving: | |
6637 | * if one cgroup claims much more protection than it uses memory, | |
6638 | * the unused remainder is available to its siblings. | |
6639 | * | |
6640 | * 4. Conversely, when the declared protection is undercommitted at a | |
6641 | * given level, the distribution of the larger parental protection | |
6642 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6643 | * is capped to its own memory.min/low setting. | |
6644 | * | |
8a931f80 JW |
6645 | * 5. However, to allow protecting recursive subtrees from each other |
6646 | * without having to declare each individual cgroup's fixed share | |
6647 | * of the ancestor's claim to protection, any unutilized - | |
6648 | * "floating" - protection from up the tree is distributed in | |
6649 | * proportion to each cgroup's *usage*. This makes the protection | |
6650 | * neutral wrt sibling cgroups and lets them compete freely over | |
6651 | * the shared parental protection budget, but it protects the | |
6652 | * subtree as a whole from neighboring subtrees. | |
6653 | * | |
6654 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6655 | * against immediate siblings whereas 5. is about protecting against | |
6656 | * neighboring subtrees. | |
bc50bcc6 JW |
6657 | */ |
6658 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6659 | unsigned long parent_usage, |
bc50bcc6 JW |
6660 | unsigned long setting, |
6661 | unsigned long parent_effective, | |
6662 | unsigned long siblings_protected) | |
6663 | { | |
6664 | unsigned long protected; | |
8a931f80 | 6665 | unsigned long ep; |
bc50bcc6 JW |
6666 | |
6667 | protected = min(usage, setting); | |
6668 | /* | |
6669 | * If all cgroups at this level combined claim and use more | |
6670 | * protection then what the parent affords them, distribute | |
6671 | * shares in proportion to utilization. | |
6672 | * | |
6673 | * We are using actual utilization rather than the statically | |
6674 | * claimed protection in order to be work-conserving: claimed | |
6675 | * but unused protection is available to siblings that would | |
6676 | * otherwise get a smaller chunk than what they claimed. | |
6677 | */ | |
6678 | if (siblings_protected > parent_effective) | |
6679 | return protected * parent_effective / siblings_protected; | |
6680 | ||
6681 | /* | |
6682 | * Ok, utilized protection of all children is within what the | |
6683 | * parent affords them, so we know whatever this child claims | |
6684 | * and utilizes is effectively protected. | |
6685 | * | |
6686 | * If there is unprotected usage beyond this value, reclaim | |
6687 | * will apply pressure in proportion to that amount. | |
6688 | * | |
6689 | * If there is unutilized protection, the cgroup will be fully | |
6690 | * shielded from reclaim, but we do return a smaller value for | |
6691 | * protection than what the group could enjoy in theory. This | |
6692 | * is okay. With the overcommit distribution above, effective | |
6693 | * protection is always dependent on how memory is actually | |
6694 | * consumed among the siblings anyway. | |
6695 | */ | |
8a931f80 JW |
6696 | ep = protected; |
6697 | ||
6698 | /* | |
6699 | * If the children aren't claiming (all of) the protection | |
6700 | * afforded to them by the parent, distribute the remainder in | |
6701 | * proportion to the (unprotected) memory of each cgroup. That | |
6702 | * way, cgroups that aren't explicitly prioritized wrt each | |
6703 | * other compete freely over the allowance, but they are | |
6704 | * collectively protected from neighboring trees. | |
6705 | * | |
6706 | * We're using unprotected memory for the weight so that if | |
6707 | * some cgroups DO claim explicit protection, we don't protect | |
6708 | * the same bytes twice. | |
cd324edc JW |
6709 | * |
6710 | * Check both usage and parent_usage against the respective | |
6711 | * protected values. One should imply the other, but they | |
6712 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6713 | */ |
6714 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6715 | return ep; | |
cd324edc JW |
6716 | if (parent_effective > siblings_protected && |
6717 | parent_usage > siblings_protected && | |
6718 | usage > protected) { | |
8a931f80 JW |
6719 | unsigned long unclaimed; |
6720 | ||
6721 | unclaimed = parent_effective - siblings_protected; | |
6722 | unclaimed *= usage - protected; | |
6723 | unclaimed /= parent_usage - siblings_protected; | |
6724 | ||
6725 | ep += unclaimed; | |
6726 | } | |
6727 | ||
6728 | return ep; | |
bc50bcc6 JW |
6729 | } |
6730 | ||
241994ed | 6731 | /** |
05395718 | 6732 | * mem_cgroup_calculate_protection - check if memory consumption is in the normal range |
34c81057 | 6733 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6734 | * @memcg: the memory cgroup to check |
6735 | * | |
23067153 RG |
6736 | * WARNING: This function is not stateless! It can only be used as part |
6737 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6738 | */ |
45c7f7e1 CD |
6739 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6740 | struct mem_cgroup *memcg) | |
241994ed | 6741 | { |
8a931f80 | 6742 | unsigned long usage, parent_usage; |
23067153 RG |
6743 | struct mem_cgroup *parent; |
6744 | ||
241994ed | 6745 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6746 | return; |
241994ed | 6747 | |
34c81057 SC |
6748 | if (!root) |
6749 | root = root_mem_cgroup; | |
22f7496f YS |
6750 | |
6751 | /* | |
6752 | * Effective values of the reclaim targets are ignored so they | |
6753 | * can be stale. Have a look at mem_cgroup_protection for more | |
6754 | * details. | |
6755 | * TODO: calculation should be more robust so that we do not need | |
6756 | * that special casing. | |
6757 | */ | |
34c81057 | 6758 | if (memcg == root) |
45c7f7e1 | 6759 | return; |
241994ed | 6760 | |
23067153 | 6761 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6762 | if (!usage) |
45c7f7e1 | 6763 | return; |
bf8d5d52 | 6764 | |
bf8d5d52 | 6765 | parent = parent_mem_cgroup(memcg); |
df2a4196 | 6766 | |
bc50bcc6 | 6767 | if (parent == root) { |
c3d53200 | 6768 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6769 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6770 | return; |
bf8d5d52 RG |
6771 | } |
6772 | ||
8a931f80 JW |
6773 | parent_usage = page_counter_read(&parent->memory); |
6774 | ||
b3a7822e | 6775 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6776 | READ_ONCE(memcg->memory.min), |
6777 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6778 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6779 | |
b3a7822e | 6780 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6781 | READ_ONCE(memcg->memory.low), |
6782 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6783 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6784 | } |
6785 | ||
8f425e4e MWO |
6786 | static int charge_memcg(struct folio *folio, struct mem_cgroup *memcg, |
6787 | gfp_t gfp) | |
0add0c77 | 6788 | { |
118f2875 | 6789 | long nr_pages = folio_nr_pages(folio); |
0add0c77 SB |
6790 | int ret; |
6791 | ||
6792 | ret = try_charge(memcg, gfp, nr_pages); | |
6793 | if (ret) | |
6794 | goto out; | |
6795 | ||
6796 | css_get(&memcg->css); | |
118f2875 | 6797 | commit_charge(folio, memcg); |
0add0c77 SB |
6798 | |
6799 | local_irq_disable(); | |
6e0110c2 | 6800 | mem_cgroup_charge_statistics(memcg, nr_pages); |
8f425e4e | 6801 | memcg_check_events(memcg, folio_nid(folio)); |
0add0c77 SB |
6802 | local_irq_enable(); |
6803 | out: | |
6804 | return ret; | |
6805 | } | |
6806 | ||
8f425e4e | 6807 | int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp) |
00501b53 | 6808 | { |
0add0c77 SB |
6809 | struct mem_cgroup *memcg; |
6810 | int ret; | |
00501b53 | 6811 | |
0add0c77 | 6812 | memcg = get_mem_cgroup_from_mm(mm); |
8f425e4e | 6813 | ret = charge_memcg(folio, memcg, gfp); |
0add0c77 | 6814 | css_put(&memcg->css); |
2d1c4980 | 6815 | |
0add0c77 SB |
6816 | return ret; |
6817 | } | |
e993d905 | 6818 | |
0add0c77 SB |
6819 | /** |
6820 | * mem_cgroup_swapin_charge_page - charge a newly allocated page for swapin | |
6821 | * @page: page to charge | |
6822 | * @mm: mm context of the victim | |
6823 | * @gfp: reclaim mode | |
6824 | * @entry: swap entry for which the page is allocated | |
6825 | * | |
6826 | * This function charges a page allocated for swapin. Please call this before | |
6827 | * adding the page to the swapcache. | |
6828 | * | |
6829 | * Returns 0 on success. Otherwise, an error code is returned. | |
6830 | */ | |
6831 | int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm, | |
6832 | gfp_t gfp, swp_entry_t entry) | |
6833 | { | |
8f425e4e | 6834 | struct folio *folio = page_folio(page); |
0add0c77 SB |
6835 | struct mem_cgroup *memcg; |
6836 | unsigned short id; | |
6837 | int ret; | |
00501b53 | 6838 | |
0add0c77 SB |
6839 | if (mem_cgroup_disabled()) |
6840 | return 0; | |
00501b53 | 6841 | |
0add0c77 SB |
6842 | id = lookup_swap_cgroup_id(entry); |
6843 | rcu_read_lock(); | |
6844 | memcg = mem_cgroup_from_id(id); | |
6845 | if (!memcg || !css_tryget_online(&memcg->css)) | |
6846 | memcg = get_mem_cgroup_from_mm(mm); | |
6847 | rcu_read_unlock(); | |
00501b53 | 6848 | |
8f425e4e | 6849 | ret = charge_memcg(folio, memcg, gfp); |
6abb5a86 | 6850 | |
0add0c77 SB |
6851 | css_put(&memcg->css); |
6852 | return ret; | |
6853 | } | |
00501b53 | 6854 | |
0add0c77 SB |
6855 | /* |
6856 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot | |
6857 | * @entry: swap entry for which the page is charged | |
6858 | * | |
6859 | * Call this function after successfully adding the charged page to swapcache. | |
6860 | * | |
6861 | * Note: This function assumes the page for which swap slot is being uncharged | |
6862 | * is order 0 page. | |
6863 | */ | |
6864 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) | |
6865 | { | |
cae3af62 MS |
6866 | /* |
6867 | * Cgroup1's unified memory+swap counter has been charged with the | |
6868 | * new swapcache page, finish the transfer by uncharging the swap | |
6869 | * slot. The swap slot would also get uncharged when it dies, but | |
6870 | * it can stick around indefinitely and we'd count the page twice | |
6871 | * the entire time. | |
6872 | * | |
6873 | * Cgroup2 has separate resource counters for memory and swap, | |
6874 | * so this is a non-issue here. Memory and swap charge lifetimes | |
6875 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
6876 | * page to memory here, and uncharge swap when the slot is freed. | |
6877 | */ | |
0add0c77 | 6878 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
00501b53 JW |
6879 | /* |
6880 | * The swap entry might not get freed for a long time, | |
6881 | * let's not wait for it. The page already received a | |
6882 | * memory+swap charge, drop the swap entry duplicate. | |
6883 | */ | |
0add0c77 | 6884 | mem_cgroup_uncharge_swap(entry, 1); |
00501b53 | 6885 | } |
3fea5a49 JW |
6886 | } |
6887 | ||
a9d5adee JG |
6888 | struct uncharge_gather { |
6889 | struct mem_cgroup *memcg; | |
b4e0b68f | 6890 | unsigned long nr_memory; |
a9d5adee | 6891 | unsigned long pgpgout; |
a9d5adee | 6892 | unsigned long nr_kmem; |
8e88bd2d | 6893 | int nid; |
a9d5adee JG |
6894 | }; |
6895 | ||
6896 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6897 | { |
a9d5adee JG |
6898 | memset(ug, 0, sizeof(*ug)); |
6899 | } | |
6900 | ||
6901 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6902 | { | |
747db954 JW |
6903 | unsigned long flags; |
6904 | ||
b4e0b68f MS |
6905 | if (ug->nr_memory) { |
6906 | page_counter_uncharge(&ug->memcg->memory, ug->nr_memory); | |
7941d214 | 6907 | if (do_memsw_account()) |
b4e0b68f | 6908 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory); |
a8c49af3 YA |
6909 | if (ug->nr_kmem) |
6910 | memcg_account_kmem(ug->memcg, -ug->nr_kmem); | |
a9d5adee | 6911 | memcg_oom_recover(ug->memcg); |
ce00a967 | 6912 | } |
747db954 JW |
6913 | |
6914 | local_irq_save(flags); | |
c9019e9b | 6915 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
b4e0b68f | 6916 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory); |
8e88bd2d | 6917 | memcg_check_events(ug->memcg, ug->nid); |
747db954 | 6918 | local_irq_restore(flags); |
f1796544 | 6919 | |
c4ed6ebf | 6920 | /* drop reference from uncharge_folio */ |
f1796544 | 6921 | css_put(&ug->memcg->css); |
a9d5adee JG |
6922 | } |
6923 | ||
c4ed6ebf | 6924 | static void uncharge_folio(struct folio *folio, struct uncharge_gather *ug) |
a9d5adee | 6925 | { |
c4ed6ebf | 6926 | long nr_pages; |
b4e0b68f MS |
6927 | struct mem_cgroup *memcg; |
6928 | struct obj_cgroup *objcg; | |
9f762dbe | 6929 | |
c4ed6ebf | 6930 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
a9d5adee | 6931 | |
a9d5adee JG |
6932 | /* |
6933 | * Nobody should be changing or seriously looking at | |
c4ed6ebf MWO |
6934 | * folio memcg or objcg at this point, we have fully |
6935 | * exclusive access to the folio. | |
a9d5adee | 6936 | */ |
fead2b86 | 6937 | if (folio_memcg_kmem(folio)) { |
1b7e4464 | 6938 | objcg = __folio_objcg(folio); |
b4e0b68f MS |
6939 | /* |
6940 | * This get matches the put at the end of the function and | |
6941 | * kmem pages do not hold memcg references anymore. | |
6942 | */ | |
6943 | memcg = get_mem_cgroup_from_objcg(objcg); | |
6944 | } else { | |
1b7e4464 | 6945 | memcg = __folio_memcg(folio); |
b4e0b68f | 6946 | } |
a9d5adee | 6947 | |
b4e0b68f MS |
6948 | if (!memcg) |
6949 | return; | |
6950 | ||
6951 | if (ug->memcg != memcg) { | |
a9d5adee JG |
6952 | if (ug->memcg) { |
6953 | uncharge_batch(ug); | |
6954 | uncharge_gather_clear(ug); | |
6955 | } | |
b4e0b68f | 6956 | ug->memcg = memcg; |
c4ed6ebf | 6957 | ug->nid = folio_nid(folio); |
f1796544 MH |
6958 | |
6959 | /* pairs with css_put in uncharge_batch */ | |
b4e0b68f | 6960 | css_get(&memcg->css); |
a9d5adee JG |
6961 | } |
6962 | ||
c4ed6ebf | 6963 | nr_pages = folio_nr_pages(folio); |
a9d5adee | 6964 | |
fead2b86 | 6965 | if (folio_memcg_kmem(folio)) { |
b4e0b68f | 6966 | ug->nr_memory += nr_pages; |
9f762dbe | 6967 | ug->nr_kmem += nr_pages; |
b4e0b68f | 6968 | |
c4ed6ebf | 6969 | folio->memcg_data = 0; |
b4e0b68f MS |
6970 | obj_cgroup_put(objcg); |
6971 | } else { | |
6972 | /* LRU pages aren't accounted at the root level */ | |
6973 | if (!mem_cgroup_is_root(memcg)) | |
6974 | ug->nr_memory += nr_pages; | |
18b2db3b | 6975 | ug->pgpgout++; |
a9d5adee | 6976 | |
c4ed6ebf | 6977 | folio->memcg_data = 0; |
b4e0b68f MS |
6978 | } |
6979 | ||
6980 | css_put(&memcg->css); | |
747db954 JW |
6981 | } |
6982 | ||
bbc6b703 | 6983 | void __mem_cgroup_uncharge(struct folio *folio) |
0a31bc97 | 6984 | { |
a9d5adee JG |
6985 | struct uncharge_gather ug; |
6986 | ||
bbc6b703 MWO |
6987 | /* Don't touch folio->lru of any random page, pre-check: */ |
6988 | if (!folio_memcg(folio)) | |
0a31bc97 JW |
6989 | return; |
6990 | ||
a9d5adee | 6991 | uncharge_gather_clear(&ug); |
bbc6b703 | 6992 | uncharge_folio(folio, &ug); |
a9d5adee | 6993 | uncharge_batch(&ug); |
747db954 | 6994 | } |
0a31bc97 | 6995 | |
747db954 | 6996 | /** |
2c8d8f97 | 6997 | * __mem_cgroup_uncharge_list - uncharge a list of page |
747db954 JW |
6998 | * @page_list: list of pages to uncharge |
6999 | * | |
7000 | * Uncharge a list of pages previously charged with | |
2c8d8f97 | 7001 | * __mem_cgroup_charge(). |
747db954 | 7002 | */ |
2c8d8f97 | 7003 | void __mem_cgroup_uncharge_list(struct list_head *page_list) |
747db954 | 7004 | { |
c41a40b6 | 7005 | struct uncharge_gather ug; |
c4ed6ebf | 7006 | struct folio *folio; |
c41a40b6 | 7007 | |
c41a40b6 | 7008 | uncharge_gather_clear(&ug); |
c4ed6ebf MWO |
7009 | list_for_each_entry(folio, page_list, lru) |
7010 | uncharge_folio(folio, &ug); | |
c41a40b6 MS |
7011 | if (ug.memcg) |
7012 | uncharge_batch(&ug); | |
0a31bc97 JW |
7013 | } |
7014 | ||
7015 | /** | |
d21bba2b MWO |
7016 | * mem_cgroup_migrate - Charge a folio's replacement. |
7017 | * @old: Currently circulating folio. | |
7018 | * @new: Replacement folio. | |
0a31bc97 | 7019 | * |
d21bba2b | 7020 | * Charge @new as a replacement folio for @old. @old will |
6a93ca8f | 7021 | * be uncharged upon free. |
0a31bc97 | 7022 | * |
d21bba2b | 7023 | * Both folios must be locked, @new->mapping must be set up. |
0a31bc97 | 7024 | */ |
d21bba2b | 7025 | void mem_cgroup_migrate(struct folio *old, struct folio *new) |
0a31bc97 | 7026 | { |
29833315 | 7027 | struct mem_cgroup *memcg; |
d21bba2b | 7028 | long nr_pages = folio_nr_pages(new); |
d93c4130 | 7029 | unsigned long flags; |
0a31bc97 | 7030 | |
d21bba2b MWO |
7031 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); |
7032 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7033 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7034 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != nr_pages, new); | |
0a31bc97 JW |
7035 | |
7036 | if (mem_cgroup_disabled()) | |
7037 | return; | |
7038 | ||
d21bba2b MWO |
7039 | /* Page cache replacement: new folio already charged? */ |
7040 | if (folio_memcg(new)) | |
0a31bc97 JW |
7041 | return; |
7042 | ||
d21bba2b MWO |
7043 | memcg = folio_memcg(old); |
7044 | VM_WARN_ON_ONCE_FOLIO(!memcg, old); | |
29833315 | 7045 | if (!memcg) |
0a31bc97 JW |
7046 | return; |
7047 | ||
44b7a8d3 | 7048 | /* Force-charge the new page. The old one will be freed soon */ |
8dc87c7d MS |
7049 | if (!mem_cgroup_is_root(memcg)) { |
7050 | page_counter_charge(&memcg->memory, nr_pages); | |
7051 | if (do_memsw_account()) | |
7052 | page_counter_charge(&memcg->memsw, nr_pages); | |
7053 | } | |
0a31bc97 | 7054 | |
1a3e1f40 | 7055 | css_get(&memcg->css); |
d21bba2b | 7056 | commit_charge(new, memcg); |
44b7a8d3 | 7057 | |
d93c4130 | 7058 | local_irq_save(flags); |
6e0110c2 | 7059 | mem_cgroup_charge_statistics(memcg, nr_pages); |
d21bba2b | 7060 | memcg_check_events(memcg, folio_nid(new)); |
d93c4130 | 7061 | local_irq_restore(flags); |
0a31bc97 JW |
7062 | } |
7063 | ||
ef12947c | 7064 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
7065 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
7066 | ||
2d758073 | 7067 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
7068 | { |
7069 | struct mem_cgroup *memcg; | |
7070 | ||
2d758073 JW |
7071 | if (!mem_cgroup_sockets_enabled) |
7072 | return; | |
7073 | ||
e876ecc6 | 7074 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
086f694a | 7075 | if (!in_task()) |
e876ecc6 SB |
7076 | return; |
7077 | ||
11092087 JW |
7078 | rcu_read_lock(); |
7079 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
7080 | if (memcg == root_mem_cgroup) |
7081 | goto out; | |
0db15298 | 7082 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 7083 | goto out; |
8965aa28 | 7084 | if (css_tryget(&memcg->css)) |
11092087 | 7085 | sk->sk_memcg = memcg; |
f7e1cb6e | 7086 | out: |
11092087 JW |
7087 | rcu_read_unlock(); |
7088 | } | |
11092087 | 7089 | |
2d758073 | 7090 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 7091 | { |
2d758073 JW |
7092 | if (sk->sk_memcg) |
7093 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7094 | } |
7095 | ||
7096 | /** | |
7097 | * mem_cgroup_charge_skmem - charge socket memory | |
7098 | * @memcg: memcg to charge | |
7099 | * @nr_pages: number of pages to charge | |
4b1327be | 7100 | * @gfp_mask: reclaim mode |
11092087 JW |
7101 | * |
7102 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
4b1327be | 7103 | * @memcg's configured limit, %false if it doesn't. |
11092087 | 7104 | */ |
4b1327be WW |
7105 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, |
7106 | gfp_t gfp_mask) | |
11092087 | 7107 | { |
f7e1cb6e | 7108 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7109 | struct page_counter *fail; |
f7e1cb6e | 7110 | |
0db15298 JW |
7111 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7112 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7113 | return true; |
7114 | } | |
0db15298 | 7115 | memcg->tcpmem_pressure = 1; |
4b1327be WW |
7116 | if (gfp_mask & __GFP_NOFAIL) { |
7117 | page_counter_charge(&memcg->tcpmem, nr_pages); | |
7118 | return true; | |
7119 | } | |
f7e1cb6e | 7120 | return false; |
11092087 | 7121 | } |
d886f4e4 | 7122 | |
4b1327be WW |
7123 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) { |
7124 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); | |
f7e1cb6e | 7125 | return true; |
4b1327be | 7126 | } |
f7e1cb6e | 7127 | |
11092087 JW |
7128 | return false; |
7129 | } | |
7130 | ||
7131 | /** | |
7132 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7133 | * @memcg: memcg to uncharge |
7134 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7135 | */ |
7136 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7137 | { | |
f7e1cb6e | 7138 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7139 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7140 | return; |
7141 | } | |
d886f4e4 | 7142 | |
c9019e9b | 7143 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7144 | |
475d0487 | 7145 | refill_stock(memcg, nr_pages); |
11092087 JW |
7146 | } |
7147 | ||
f7e1cb6e JW |
7148 | static int __init cgroup_memory(char *s) |
7149 | { | |
7150 | char *token; | |
7151 | ||
7152 | while ((token = strsep(&s, ",")) != NULL) { | |
7153 | if (!*token) | |
7154 | continue; | |
7155 | if (!strcmp(token, "nosocket")) | |
7156 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7157 | if (!strcmp(token, "nokmem")) |
7158 | cgroup_memory_nokmem = true; | |
f7e1cb6e | 7159 | } |
460a79e1 | 7160 | return 1; |
f7e1cb6e JW |
7161 | } |
7162 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7163 | |
2d11085e | 7164 | /* |
1081312f MH |
7165 | * subsys_initcall() for memory controller. |
7166 | * | |
308167fc SAS |
7167 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7168 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7169 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7170 | * should be initialized from here. | |
2d11085e MH |
7171 | */ |
7172 | static int __init mem_cgroup_init(void) | |
7173 | { | |
95a045f6 JW |
7174 | int cpu, node; |
7175 | ||
f3344adf MS |
7176 | /* |
7177 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7178 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7179 | * to work fine, we should make sure that the overfill threshold can't | |
7180 | * exceed S32_MAX / PAGE_SIZE. | |
7181 | */ | |
7182 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7183 | ||
308167fc SAS |
7184 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7185 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7186 | |
7187 | for_each_possible_cpu(cpu) | |
7188 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7189 | drain_local_stock); | |
7190 | ||
7191 | for_each_node(node) { | |
7192 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7193 | |
7194 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7195 | node_online(node) ? node : NUMA_NO_NODE); | |
7196 | ||
ef8f2327 | 7197 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7198 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7199 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7200 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7201 | } | |
7202 | ||
2d11085e MH |
7203 | return 0; |
7204 | } | |
7205 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7206 | |
7207 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7208 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7209 | { | |
1c2d479a | 7210 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7211 | /* |
7212 | * The root cgroup cannot be destroyed, so it's refcount must | |
7213 | * always be >= 1. | |
7214 | */ | |
7215 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7216 | VM_BUG_ON(1); | |
7217 | break; | |
7218 | } | |
7219 | memcg = parent_mem_cgroup(memcg); | |
7220 | if (!memcg) | |
7221 | memcg = root_mem_cgroup; | |
7222 | } | |
7223 | return memcg; | |
7224 | } | |
7225 | ||
21afa38e JW |
7226 | /** |
7227 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
3ecb0087 | 7228 | * @folio: folio whose memsw charge to transfer |
21afa38e JW |
7229 | * @entry: swap entry to move the charge to |
7230 | * | |
3ecb0087 | 7231 | * Transfer the memsw charge of @folio to @entry. |
21afa38e | 7232 | */ |
3ecb0087 | 7233 | void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry) |
21afa38e | 7234 | { |
1f47b61f | 7235 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7236 | unsigned int nr_entries; |
21afa38e JW |
7237 | unsigned short oldid; |
7238 | ||
3ecb0087 MWO |
7239 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
7240 | VM_BUG_ON_FOLIO(folio_ref_count(folio), folio); | |
21afa38e | 7241 | |
76358ab5 AS |
7242 | if (mem_cgroup_disabled()) |
7243 | return; | |
7244 | ||
2d1c4980 | 7245 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7246 | return; |
7247 | ||
3ecb0087 | 7248 | memcg = folio_memcg(folio); |
21afa38e | 7249 | |
3ecb0087 | 7250 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
21afa38e JW |
7251 | if (!memcg) |
7252 | return; | |
7253 | ||
1f47b61f VD |
7254 | /* |
7255 | * In case the memcg owning these pages has been offlined and doesn't | |
7256 | * have an ID allocated to it anymore, charge the closest online | |
7257 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7258 | */ | |
7259 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
3ecb0087 | 7260 | nr_entries = folio_nr_pages(folio); |
d6810d73 HY |
7261 | /* Get references for the tail pages, too */ |
7262 | if (nr_entries > 1) | |
7263 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7264 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7265 | nr_entries); | |
3ecb0087 | 7266 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 7267 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 7268 | |
3ecb0087 | 7269 | folio->memcg_data = 0; |
21afa38e JW |
7270 | |
7271 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7272 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7273 | |
2d1c4980 | 7274 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7275 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7276 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7277 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7278 | } |
7279 | ||
ce9ce665 SAS |
7280 | /* |
7281 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7282 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7283 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7284 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 | 7285 | */ |
be3e67b5 | 7286 | memcg_stats_lock(); |
6e0110c2 | 7287 | mem_cgroup_charge_statistics(memcg, -nr_entries); |
be3e67b5 | 7288 | memcg_stats_unlock(); |
3ecb0087 | 7289 | memcg_check_events(memcg, folio_nid(folio)); |
73f576c0 | 7290 | |
1a3e1f40 | 7291 | css_put(&memcg->css); |
21afa38e JW |
7292 | } |
7293 | ||
38d8b4e6 | 7294 | /** |
e2e3fdc7 MWO |
7295 | * __mem_cgroup_try_charge_swap - try charging swap space for a folio |
7296 | * @folio: folio being added to swap | |
37e84351 VD |
7297 | * @entry: swap entry to charge |
7298 | * | |
e2e3fdc7 | 7299 | * Try to charge @folio's memcg for the swap space at @entry. |
37e84351 VD |
7300 | * |
7301 | * Returns 0 on success, -ENOMEM on failure. | |
7302 | */ | |
e2e3fdc7 | 7303 | int __mem_cgroup_try_charge_swap(struct folio *folio, swp_entry_t entry) |
37e84351 | 7304 | { |
e2e3fdc7 | 7305 | unsigned int nr_pages = folio_nr_pages(folio); |
37e84351 | 7306 | struct page_counter *counter; |
38d8b4e6 | 7307 | struct mem_cgroup *memcg; |
37e84351 VD |
7308 | unsigned short oldid; |
7309 | ||
2d1c4980 | 7310 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7311 | return 0; |
7312 | ||
e2e3fdc7 | 7313 | memcg = folio_memcg(folio); |
37e84351 | 7314 | |
e2e3fdc7 | 7315 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
37e84351 VD |
7316 | if (!memcg) |
7317 | return 0; | |
7318 | ||
f3a53a3a TH |
7319 | if (!entry.val) { |
7320 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7321 | return 0; |
f3a53a3a | 7322 | } |
bb98f2c5 | 7323 | |
1f47b61f VD |
7324 | memcg = mem_cgroup_id_get_online(memcg); |
7325 | ||
2d1c4980 | 7326 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7327 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7328 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7329 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7330 | mem_cgroup_id_put(memcg); |
37e84351 | 7331 | return -ENOMEM; |
1f47b61f | 7332 | } |
37e84351 | 7333 | |
38d8b4e6 HY |
7334 | /* Get references for the tail pages, too */ |
7335 | if (nr_pages > 1) | |
7336 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7337 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
e2e3fdc7 | 7338 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 7339 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7340 | |
37e84351 VD |
7341 | return 0; |
7342 | } | |
7343 | ||
21afa38e | 7344 | /** |
01c4b28c | 7345 | * __mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7346 | * @entry: swap entry to uncharge |
38d8b4e6 | 7347 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7348 | */ |
01c4b28c | 7349 | void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7350 | { |
7351 | struct mem_cgroup *memcg; | |
7352 | unsigned short id; | |
7353 | ||
38d8b4e6 | 7354 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7355 | rcu_read_lock(); |
adbe427b | 7356 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7357 | if (memcg) { |
2d1c4980 | 7358 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7359 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7360 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7361 | else |
38d8b4e6 | 7362 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7363 | } |
c9019e9b | 7364 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7365 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7366 | } |
7367 | rcu_read_unlock(); | |
7368 | } | |
7369 | ||
d8b38438 VD |
7370 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7371 | { | |
7372 | long nr_swap_pages = get_nr_swap_pages(); | |
7373 | ||
eccb52e7 | 7374 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7375 | return nr_swap_pages; |
7376 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7377 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7378 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7379 | page_counter_read(&memcg->swap)); |
7380 | return nr_swap_pages; | |
7381 | } | |
7382 | ||
5ccc5aba VD |
7383 | bool mem_cgroup_swap_full(struct page *page) |
7384 | { | |
7385 | struct mem_cgroup *memcg; | |
7386 | ||
7387 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7388 | ||
7389 | if (vm_swap_full()) | |
7390 | return true; | |
eccb52e7 | 7391 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7392 | return false; |
7393 | ||
bcfe06bf | 7394 | memcg = page_memcg(page); |
5ccc5aba VD |
7395 | if (!memcg) |
7396 | return false; | |
7397 | ||
4b82ab4f JK |
7398 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7399 | unsigned long usage = page_counter_read(&memcg->swap); | |
7400 | ||
7401 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7402 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7403 | return true; |
4b82ab4f | 7404 | } |
5ccc5aba VD |
7405 | |
7406 | return false; | |
7407 | } | |
7408 | ||
eccb52e7 | 7409 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7410 | { |
7411 | if (!strcmp(s, "1")) | |
5ab92901 | 7412 | cgroup_memory_noswap = false; |
21afa38e | 7413 | else if (!strcmp(s, "0")) |
5ab92901 | 7414 | cgroup_memory_noswap = true; |
21afa38e JW |
7415 | return 1; |
7416 | } | |
eccb52e7 | 7417 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7418 | |
37e84351 VD |
7419 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7420 | struct cftype *cft) | |
7421 | { | |
7422 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7423 | ||
7424 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7425 | } | |
7426 | ||
4b82ab4f JK |
7427 | static int swap_high_show(struct seq_file *m, void *v) |
7428 | { | |
7429 | return seq_puts_memcg_tunable(m, | |
7430 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7431 | } | |
7432 | ||
7433 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7434 | char *buf, size_t nbytes, loff_t off) | |
7435 | { | |
7436 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7437 | unsigned long high; | |
7438 | int err; | |
7439 | ||
7440 | buf = strstrip(buf); | |
7441 | err = page_counter_memparse(buf, "max", &high); | |
7442 | if (err) | |
7443 | return err; | |
7444 | ||
7445 | page_counter_set_high(&memcg->swap, high); | |
7446 | ||
7447 | return nbytes; | |
7448 | } | |
7449 | ||
37e84351 VD |
7450 | static int swap_max_show(struct seq_file *m, void *v) |
7451 | { | |
677dc973 CD |
7452 | return seq_puts_memcg_tunable(m, |
7453 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7454 | } |
7455 | ||
7456 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7457 | char *buf, size_t nbytes, loff_t off) | |
7458 | { | |
7459 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7460 | unsigned long max; | |
7461 | int err; | |
7462 | ||
7463 | buf = strstrip(buf); | |
7464 | err = page_counter_memparse(buf, "max", &max); | |
7465 | if (err) | |
7466 | return err; | |
7467 | ||
be09102b | 7468 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7469 | |
7470 | return nbytes; | |
7471 | } | |
7472 | ||
f3a53a3a TH |
7473 | static int swap_events_show(struct seq_file *m, void *v) |
7474 | { | |
aa9694bb | 7475 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7476 | |
4b82ab4f JK |
7477 | seq_printf(m, "high %lu\n", |
7478 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7479 | seq_printf(m, "max %lu\n", |
7480 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7481 | seq_printf(m, "fail %lu\n", | |
7482 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7483 | ||
7484 | return 0; | |
7485 | } | |
7486 | ||
37e84351 VD |
7487 | static struct cftype swap_files[] = { |
7488 | { | |
7489 | .name = "swap.current", | |
7490 | .flags = CFTYPE_NOT_ON_ROOT, | |
7491 | .read_u64 = swap_current_read, | |
7492 | }, | |
4b82ab4f JK |
7493 | { |
7494 | .name = "swap.high", | |
7495 | .flags = CFTYPE_NOT_ON_ROOT, | |
7496 | .seq_show = swap_high_show, | |
7497 | .write = swap_high_write, | |
7498 | }, | |
37e84351 VD |
7499 | { |
7500 | .name = "swap.max", | |
7501 | .flags = CFTYPE_NOT_ON_ROOT, | |
7502 | .seq_show = swap_max_show, | |
7503 | .write = swap_max_write, | |
7504 | }, | |
f3a53a3a TH |
7505 | { |
7506 | .name = "swap.events", | |
7507 | .flags = CFTYPE_NOT_ON_ROOT, | |
7508 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7509 | .seq_show = swap_events_show, | |
7510 | }, | |
37e84351 VD |
7511 | { } /* terminate */ |
7512 | }; | |
7513 | ||
eccb52e7 | 7514 | static struct cftype memsw_files[] = { |
21afa38e JW |
7515 | { |
7516 | .name = "memsw.usage_in_bytes", | |
7517 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7518 | .read_u64 = mem_cgroup_read_u64, | |
7519 | }, | |
7520 | { | |
7521 | .name = "memsw.max_usage_in_bytes", | |
7522 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7523 | .write = mem_cgroup_reset, | |
7524 | .read_u64 = mem_cgroup_read_u64, | |
7525 | }, | |
7526 | { | |
7527 | .name = "memsw.limit_in_bytes", | |
7528 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7529 | .write = mem_cgroup_write, | |
7530 | .read_u64 = mem_cgroup_read_u64, | |
7531 | }, | |
7532 | { | |
7533 | .name = "memsw.failcnt", | |
7534 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7535 | .write = mem_cgroup_reset, | |
7536 | .read_u64 = mem_cgroup_read_u64, | |
7537 | }, | |
7538 | { }, /* terminate */ | |
7539 | }; | |
7540 | ||
f4840ccf JW |
7541 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
7542 | /** | |
7543 | * obj_cgroup_may_zswap - check if this cgroup can zswap | |
7544 | * @objcg: the object cgroup | |
7545 | * | |
7546 | * Check if the hierarchical zswap limit has been reached. | |
7547 | * | |
7548 | * This doesn't check for specific headroom, and it is not atomic | |
7549 | * either. But with zswap, the size of the allocation is only known | |
7550 | * once compression has occured, and this optimistic pre-check avoids | |
7551 | * spending cycles on compression when there is already no room left | |
7552 | * or zswap is disabled altogether somewhere in the hierarchy. | |
7553 | */ | |
7554 | bool obj_cgroup_may_zswap(struct obj_cgroup *objcg) | |
7555 | { | |
7556 | struct mem_cgroup *memcg, *original_memcg; | |
7557 | bool ret = true; | |
7558 | ||
7559 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7560 | return true; | |
7561 | ||
7562 | original_memcg = get_mem_cgroup_from_objcg(objcg); | |
7563 | for (memcg = original_memcg; memcg != root_mem_cgroup; | |
7564 | memcg = parent_mem_cgroup(memcg)) { | |
7565 | unsigned long max = READ_ONCE(memcg->zswap_max); | |
7566 | unsigned long pages; | |
7567 | ||
7568 | if (max == PAGE_COUNTER_MAX) | |
7569 | continue; | |
7570 | if (max == 0) { | |
7571 | ret = false; | |
7572 | break; | |
7573 | } | |
7574 | ||
7575 | cgroup_rstat_flush(memcg->css.cgroup); | |
7576 | pages = memcg_page_state(memcg, MEMCG_ZSWAP_B) / PAGE_SIZE; | |
7577 | if (pages < max) | |
7578 | continue; | |
7579 | ret = false; | |
7580 | break; | |
7581 | } | |
7582 | mem_cgroup_put(original_memcg); | |
7583 | return ret; | |
7584 | } | |
7585 | ||
7586 | /** | |
7587 | * obj_cgroup_charge_zswap - charge compression backend memory | |
7588 | * @objcg: the object cgroup | |
7589 | * @size: size of compressed object | |
7590 | * | |
7591 | * This forces the charge after obj_cgroup_may_swap() allowed | |
7592 | * compression and storage in zwap for this cgroup to go ahead. | |
7593 | */ | |
7594 | void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size) | |
7595 | { | |
7596 | struct mem_cgroup *memcg; | |
7597 | ||
7598 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7599 | return; | |
7600 | ||
7601 | VM_WARN_ON_ONCE(!(current->flags & PF_MEMALLOC)); | |
7602 | ||
7603 | /* PF_MEMALLOC context, charging must succeed */ | |
7604 | if (obj_cgroup_charge(objcg, GFP_KERNEL, size)) | |
7605 | VM_WARN_ON_ONCE(1); | |
7606 | ||
7607 | rcu_read_lock(); | |
7608 | memcg = obj_cgroup_memcg(objcg); | |
7609 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, size); | |
7610 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, 1); | |
7611 | rcu_read_unlock(); | |
7612 | } | |
7613 | ||
7614 | /** | |
7615 | * obj_cgroup_uncharge_zswap - uncharge compression backend memory | |
7616 | * @objcg: the object cgroup | |
7617 | * @size: size of compressed object | |
7618 | * | |
7619 | * Uncharges zswap memory on page in. | |
7620 | */ | |
7621 | void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size) | |
7622 | { | |
7623 | struct mem_cgroup *memcg; | |
7624 | ||
7625 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7626 | return; | |
7627 | ||
7628 | obj_cgroup_uncharge(objcg, size); | |
7629 | ||
7630 | rcu_read_lock(); | |
7631 | memcg = obj_cgroup_memcg(objcg); | |
7632 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, -size); | |
7633 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, -1); | |
7634 | rcu_read_unlock(); | |
7635 | } | |
7636 | ||
7637 | static u64 zswap_current_read(struct cgroup_subsys_state *css, | |
7638 | struct cftype *cft) | |
7639 | { | |
7640 | cgroup_rstat_flush(css->cgroup); | |
7641 | return memcg_page_state(mem_cgroup_from_css(css), MEMCG_ZSWAP_B); | |
7642 | } | |
7643 | ||
7644 | static int zswap_max_show(struct seq_file *m, void *v) | |
7645 | { | |
7646 | return seq_puts_memcg_tunable(m, | |
7647 | READ_ONCE(mem_cgroup_from_seq(m)->zswap_max)); | |
7648 | } | |
7649 | ||
7650 | static ssize_t zswap_max_write(struct kernfs_open_file *of, | |
7651 | char *buf, size_t nbytes, loff_t off) | |
7652 | { | |
7653 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7654 | unsigned long max; | |
7655 | int err; | |
7656 | ||
7657 | buf = strstrip(buf); | |
7658 | err = page_counter_memparse(buf, "max", &max); | |
7659 | if (err) | |
7660 | return err; | |
7661 | ||
7662 | xchg(&memcg->zswap_max, max); | |
7663 | ||
7664 | return nbytes; | |
7665 | } | |
7666 | ||
7667 | static struct cftype zswap_files[] = { | |
7668 | { | |
7669 | .name = "zswap.current", | |
7670 | .flags = CFTYPE_NOT_ON_ROOT, | |
7671 | .read_u64 = zswap_current_read, | |
7672 | }, | |
7673 | { | |
7674 | .name = "zswap.max", | |
7675 | .flags = CFTYPE_NOT_ON_ROOT, | |
7676 | .seq_show = zswap_max_show, | |
7677 | .write = zswap_max_write, | |
7678 | }, | |
7679 | { } /* terminate */ | |
7680 | }; | |
7681 | #endif /* CONFIG_MEMCG_KMEM && CONFIG_ZSWAP */ | |
7682 | ||
82ff165c BS |
7683 | /* |
7684 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7685 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7686 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7687 | * boot parameter. This may result in premature OOPS inside | |
7688 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7689 | */ | |
21afa38e JW |
7690 | static int __init mem_cgroup_swap_init(void) |
7691 | { | |
2d1c4980 JW |
7692 | /* No memory control -> no swap control */ |
7693 | if (mem_cgroup_disabled()) | |
7694 | cgroup_memory_noswap = true; | |
7695 | ||
7696 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7697 | return 0; |
7698 | ||
7699 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7700 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
f4840ccf JW |
7701 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
7702 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, zswap_files)); | |
7703 | #endif | |
21afa38e JW |
7704 | return 0; |
7705 | } | |
82ff165c | 7706 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7707 | |
7708 | #endif /* CONFIG_MEMCG_SWAP */ |