]>
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 | |
b6c1a8af YS |
91 | /* BPF memory accounting disabled? */ |
92 | static bool cgroup_memory_nobpf __ro_after_init; | |
93 | ||
97b27821 TH |
94 | #ifdef CONFIG_CGROUP_WRITEBACK |
95 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
96 | #endif | |
97 | ||
7941d214 JW |
98 | /* Whether legacy memory+swap accounting is active */ |
99 | static bool do_memsw_account(void) | |
100 | { | |
b25806dc | 101 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys); |
7941d214 JW |
102 | } |
103 | ||
a0db00fc KS |
104 | #define THRESHOLDS_EVENTS_TARGET 128 |
105 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 106 | |
bb4cc1a8 AM |
107 | /* |
108 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
109 | * their hierarchy representation | |
110 | */ | |
111 | ||
ef8f2327 | 112 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 113 | struct rb_root rb_root; |
fa90b2fd | 114 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
115 | spinlock_t lock; |
116 | }; | |
117 | ||
bb4cc1a8 AM |
118 | struct mem_cgroup_tree { |
119 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
120 | }; | |
121 | ||
122 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
123 | ||
9490ff27 KH |
124 | /* for OOM */ |
125 | struct mem_cgroup_eventfd_list { | |
126 | struct list_head list; | |
127 | struct eventfd_ctx *eventfd; | |
128 | }; | |
2e72b634 | 129 | |
79bd9814 TH |
130 | /* |
131 | * cgroup_event represents events which userspace want to receive. | |
132 | */ | |
3bc942f3 | 133 | struct mem_cgroup_event { |
79bd9814 | 134 | /* |
59b6f873 | 135 | * memcg which the event belongs to. |
79bd9814 | 136 | */ |
59b6f873 | 137 | struct mem_cgroup *memcg; |
79bd9814 TH |
138 | /* |
139 | * eventfd to signal userspace about the event. | |
140 | */ | |
141 | struct eventfd_ctx *eventfd; | |
142 | /* | |
143 | * Each of these stored in a list by the cgroup. | |
144 | */ | |
145 | struct list_head list; | |
fba94807 TH |
146 | /* |
147 | * register_event() callback will be used to add new userspace | |
148 | * waiter for changes related to this event. Use eventfd_signal() | |
149 | * on eventfd to send notification to userspace. | |
150 | */ | |
59b6f873 | 151 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 152 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
153 | /* |
154 | * unregister_event() callback will be called when userspace closes | |
155 | * the eventfd or on cgroup removing. This callback must be set, | |
156 | * if you want provide notification functionality. | |
157 | */ | |
59b6f873 | 158 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 159 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
160 | /* |
161 | * All fields below needed to unregister event when | |
162 | * userspace closes eventfd. | |
163 | */ | |
164 | poll_table pt; | |
165 | wait_queue_head_t *wqh; | |
ac6424b9 | 166 | wait_queue_entry_t wait; |
79bd9814 TH |
167 | struct work_struct remove; |
168 | }; | |
169 | ||
c0ff4b85 R |
170 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
171 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 172 | |
7dc74be0 DN |
173 | /* Stuffs for move charges at task migration. */ |
174 | /* | |
1dfab5ab | 175 | * Types of charges to be moved. |
7dc74be0 | 176 | */ |
1dfab5ab JW |
177 | #define MOVE_ANON 0x1U |
178 | #define MOVE_FILE 0x2U | |
179 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 180 | |
4ffef5fe DN |
181 | /* "mc" and its members are protected by cgroup_mutex */ |
182 | static struct move_charge_struct { | |
b1dd693e | 183 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 184 | struct mm_struct *mm; |
4ffef5fe DN |
185 | struct mem_cgroup *from; |
186 | struct mem_cgroup *to; | |
1dfab5ab | 187 | unsigned long flags; |
4ffef5fe | 188 | unsigned long precharge; |
854ffa8d | 189 | unsigned long moved_charge; |
483c30b5 | 190 | unsigned long moved_swap; |
8033b97c DN |
191 | struct task_struct *moving_task; /* a task moving charges */ |
192 | wait_queue_head_t waitq; /* a waitq for other context */ | |
193 | } mc = { | |
2bd9bb20 | 194 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
195 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
196 | }; | |
4ffef5fe | 197 | |
4e416953 BS |
198 | /* |
199 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
200 | * limit reclaim to prevent infinite loops, if they ever occur. | |
201 | */ | |
a0db00fc | 202 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 203 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 204 | |
8c7c6e34 | 205 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
206 | enum res_type { |
207 | _MEM, | |
208 | _MEMSWAP, | |
510fc4e1 | 209 | _KMEM, |
d55f90bf | 210 | _TCP, |
86ae53e1 GC |
211 | }; |
212 | ||
a0db00fc KS |
213 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
214 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 KH |
215 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
216 | ||
b05706f1 KT |
217 | /* |
218 | * Iteration constructs for visiting all cgroups (under a tree). If | |
219 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
220 | * be used for reference counting. | |
221 | */ | |
222 | #define for_each_mem_cgroup_tree(iter, root) \ | |
223 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
224 | iter != NULL; \ | |
225 | iter = mem_cgroup_iter(root, iter, NULL)) | |
226 | ||
227 | #define for_each_mem_cgroup(iter) \ | |
228 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
229 | iter != NULL; \ | |
230 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
231 | ||
a4ebf1b6 | 232 | static inline bool task_is_dying(void) |
7775face TH |
233 | { |
234 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
235 | (current->flags & PF_EXITING); | |
236 | } | |
237 | ||
70ddf637 AV |
238 | /* Some nice accessors for the vmpressure. */ |
239 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
240 | { | |
241 | if (!memcg) | |
242 | memcg = root_mem_cgroup; | |
243 | return &memcg->vmpressure; | |
244 | } | |
245 | ||
9647875b | 246 | struct mem_cgroup *vmpressure_to_memcg(struct vmpressure *vmpr) |
70ddf637 | 247 | { |
9647875b | 248 | return container_of(vmpr, struct mem_cgroup, vmpressure); |
70ddf637 AV |
249 | } |
250 | ||
84c07d11 | 251 | #ifdef CONFIG_MEMCG_KMEM |
0764db9b | 252 | static DEFINE_SPINLOCK(objcg_lock); |
bf4f0599 | 253 | |
4d5c8aed RG |
254 | bool mem_cgroup_kmem_disabled(void) |
255 | { | |
256 | return cgroup_memory_nokmem; | |
257 | } | |
258 | ||
f1286fae MS |
259 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
260 | unsigned int nr_pages); | |
c1a660de | 261 | |
bf4f0599 RG |
262 | static void obj_cgroup_release(struct percpu_ref *ref) |
263 | { | |
264 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
bf4f0599 RG |
265 | unsigned int nr_bytes; |
266 | unsigned int nr_pages; | |
267 | unsigned long flags; | |
268 | ||
269 | /* | |
270 | * At this point all allocated objects are freed, and | |
271 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
272 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
273 | * | |
274 | * The following sequence can lead to it: | |
275 | * 1) CPU0: objcg == stock->cached_objcg | |
276 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
277 | * PAGE_SIZE bytes are charged | |
278 | * 3) CPU1: a process from another memcg is allocating something, | |
279 | * the stock if flushed, | |
280 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
281 | * 5) CPU0: we do release this object, | |
282 | * 92 bytes are added to stock->nr_bytes | |
283 | * 6) CPU0: stock is flushed, | |
284 | * 92 bytes are added to objcg->nr_charged_bytes | |
285 | * | |
286 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
287 | * This page will be uncharged in obj_cgroup_release(). | |
288 | */ | |
289 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
290 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
291 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
292 | ||
bf4f0599 | 293 | if (nr_pages) |
f1286fae | 294 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
271dd6b1 | 295 | |
0764db9b | 296 | spin_lock_irqsave(&objcg_lock, flags); |
bf4f0599 | 297 | list_del(&objcg->list); |
0764db9b | 298 | spin_unlock_irqrestore(&objcg_lock, flags); |
bf4f0599 RG |
299 | |
300 | percpu_ref_exit(ref); | |
301 | kfree_rcu(objcg, rcu); | |
302 | } | |
303 | ||
304 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
305 | { | |
306 | struct obj_cgroup *objcg; | |
307 | int ret; | |
308 | ||
309 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
310 | if (!objcg) | |
311 | return NULL; | |
312 | ||
313 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
314 | GFP_KERNEL); | |
315 | if (ret) { | |
316 | kfree(objcg); | |
317 | return NULL; | |
318 | } | |
319 | INIT_LIST_HEAD(&objcg->list); | |
320 | return objcg; | |
321 | } | |
322 | ||
323 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
324 | struct mem_cgroup *parent) | |
325 | { | |
326 | struct obj_cgroup *objcg, *iter; | |
327 | ||
328 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
329 | ||
0764db9b | 330 | spin_lock_irq(&objcg_lock); |
bf4f0599 | 331 | |
9838354e MS |
332 | /* 1) Ready to reparent active objcg. */ |
333 | list_add(&objcg->list, &memcg->objcg_list); | |
334 | /* 2) Reparent active objcg and already reparented objcgs to parent. */ | |
335 | list_for_each_entry(iter, &memcg->objcg_list, list) | |
336 | WRITE_ONCE(iter->memcg, parent); | |
337 | /* 3) Move already reparented objcgs to the parent's list */ | |
bf4f0599 RG |
338 | list_splice(&memcg->objcg_list, &parent->objcg_list); |
339 | ||
0764db9b | 340 | spin_unlock_irq(&objcg_lock); |
bf4f0599 RG |
341 | |
342 | percpu_ref_kill(&objcg->refcnt); | |
343 | } | |
344 | ||
d7f25f8a GC |
345 | /* |
346 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 347 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
348 | * conditional to this static branch, we'll have to allow modules that does |
349 | * kmem_cache_alloc and the such to see this symbol as well | |
350 | */ | |
f7a449f7 RG |
351 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_online_key); |
352 | EXPORT_SYMBOL(memcg_kmem_online_key); | |
b6c1a8af YS |
353 | |
354 | DEFINE_STATIC_KEY_FALSE(memcg_bpf_enabled_key); | |
355 | EXPORT_SYMBOL(memcg_bpf_enabled_key); | |
0a432dcb | 356 | #endif |
17cc4dfe | 357 | |
ad7fa852 | 358 | /** |
75376c6f MWO |
359 | * mem_cgroup_css_from_folio - css of the memcg associated with a folio |
360 | * @folio: folio of interest | |
ad7fa852 TH |
361 | * |
362 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
75376c6f | 363 | * with @folio is returned. The returned css remains associated with @folio |
ad7fa852 TH |
364 | * until it is released. |
365 | * | |
366 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
367 | * is returned. | |
ad7fa852 | 368 | */ |
75376c6f | 369 | struct cgroup_subsys_state *mem_cgroup_css_from_folio(struct folio *folio) |
ad7fa852 | 370 | { |
75376c6f | 371 | struct mem_cgroup *memcg = folio_memcg(folio); |
ad7fa852 | 372 | |
9e10a130 | 373 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
374 | memcg = root_mem_cgroup; |
375 | ||
ad7fa852 TH |
376 | return &memcg->css; |
377 | } | |
378 | ||
2fc04524 VD |
379 | /** |
380 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
381 | * @page: the page | |
382 | * | |
383 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
384 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
385 | * is safe to call this function without holding a reference to @page. | |
386 | * | |
387 | * Note, this function is inherently racy, because there is nothing to prevent | |
388 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
389 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
390 | * do not care (such as procfs interfaces). | |
391 | */ | |
392 | ino_t page_cgroup_ino(struct page *page) | |
393 | { | |
394 | struct mem_cgroup *memcg; | |
395 | unsigned long ino = 0; | |
396 | ||
397 | rcu_read_lock(); | |
bcfe06bf | 398 | memcg = page_memcg_check(page); |
286e04b8 | 399 | |
2fc04524 VD |
400 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
401 | memcg = parent_mem_cgroup(memcg); | |
402 | if (memcg) | |
403 | ino = cgroup_ino(memcg->css.cgroup); | |
404 | rcu_read_unlock(); | |
405 | return ino; | |
406 | } | |
407 | ||
ef8f2327 MG |
408 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
409 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 410 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
411 | { |
412 | struct rb_node **p = &mctz->rb_root.rb_node; | |
413 | struct rb_node *parent = NULL; | |
ef8f2327 | 414 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 415 | bool rightmost = true; |
bb4cc1a8 AM |
416 | |
417 | if (mz->on_tree) | |
418 | return; | |
419 | ||
420 | mz->usage_in_excess = new_usage_in_excess; | |
421 | if (!mz->usage_in_excess) | |
422 | return; | |
423 | while (*p) { | |
424 | parent = *p; | |
ef8f2327 | 425 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 426 | tree_node); |
fa90b2fd | 427 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 428 | p = &(*p)->rb_left; |
fa90b2fd | 429 | rightmost = false; |
378876b0 | 430 | } else { |
bb4cc1a8 | 431 | p = &(*p)->rb_right; |
378876b0 | 432 | } |
bb4cc1a8 | 433 | } |
fa90b2fd DB |
434 | |
435 | if (rightmost) | |
436 | mctz->rb_rightmost = &mz->tree_node; | |
437 | ||
bb4cc1a8 AM |
438 | rb_link_node(&mz->tree_node, parent, p); |
439 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
440 | mz->on_tree = true; | |
441 | } | |
442 | ||
ef8f2327 MG |
443 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
444 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
445 | { |
446 | if (!mz->on_tree) | |
447 | return; | |
fa90b2fd DB |
448 | |
449 | if (&mz->tree_node == mctz->rb_rightmost) | |
450 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
451 | ||
bb4cc1a8 AM |
452 | rb_erase(&mz->tree_node, &mctz->rb_root); |
453 | mz->on_tree = false; | |
454 | } | |
455 | ||
ef8f2327 MG |
456 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
457 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 458 | { |
0a31bc97 JW |
459 | unsigned long flags; |
460 | ||
461 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 462 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 463 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
464 | } |
465 | ||
3e32cb2e JW |
466 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
467 | { | |
468 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 469 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
470 | unsigned long excess = 0; |
471 | ||
472 | if (nr_pages > soft_limit) | |
473 | excess = nr_pages - soft_limit; | |
474 | ||
475 | return excess; | |
476 | } | |
bb4cc1a8 | 477 | |
658b69c9 | 478 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, int nid) |
bb4cc1a8 | 479 | { |
3e32cb2e | 480 | unsigned long excess; |
ef8f2327 MG |
481 | struct mem_cgroup_per_node *mz; |
482 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 483 | |
e4dde56c | 484 | if (lru_gen_enabled()) { |
36c7b4db A |
485 | if (soft_limit_excess(memcg)) |
486 | lru_gen_soft_reclaim(&memcg->nodeinfo[nid]->lruvec); | |
e4dde56c YZ |
487 | return; |
488 | } | |
489 | ||
2ab082ba | 490 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
491 | if (!mctz) |
492 | return; | |
bb4cc1a8 AM |
493 | /* |
494 | * Necessary to update all ancestors when hierarchy is used. | |
495 | * because their event counter is not touched. | |
496 | */ | |
497 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
658b69c9 | 498 | mz = memcg->nodeinfo[nid]; |
3e32cb2e | 499 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
500 | /* |
501 | * We have to update the tree if mz is on RB-tree or | |
502 | * mem is over its softlimit. | |
503 | */ | |
504 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
505 | unsigned long flags; |
506 | ||
507 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
508 | /* if on-tree, remove it */ |
509 | if (mz->on_tree) | |
cf2c8127 | 510 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
511 | /* |
512 | * Insert again. mz->usage_in_excess will be updated. | |
513 | * If excess is 0, no tree ops. | |
514 | */ | |
cf2c8127 | 515 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 516 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
517 | } |
518 | } | |
519 | } | |
520 | ||
521 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
522 | { | |
ef8f2327 MG |
523 | struct mem_cgroup_tree_per_node *mctz; |
524 | struct mem_cgroup_per_node *mz; | |
525 | int nid; | |
bb4cc1a8 | 526 | |
e231875b | 527 | for_each_node(nid) { |
a3747b53 | 528 | mz = memcg->nodeinfo[nid]; |
2ab082ba | 529 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
530 | if (mctz) |
531 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
532 | } |
533 | } | |
534 | ||
ef8f2327 MG |
535 | static struct mem_cgroup_per_node * |
536 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 537 | { |
ef8f2327 | 538 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
539 | |
540 | retry: | |
541 | mz = NULL; | |
fa90b2fd | 542 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
543 | goto done; /* Nothing to reclaim from */ |
544 | ||
fa90b2fd DB |
545 | mz = rb_entry(mctz->rb_rightmost, |
546 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
547 | /* |
548 | * Remove the node now but someone else can add it back, | |
549 | * we will to add it back at the end of reclaim to its correct | |
550 | * position in the tree. | |
551 | */ | |
cf2c8127 | 552 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 553 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 554 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
555 | goto retry; |
556 | done: | |
557 | return mz; | |
558 | } | |
559 | ||
ef8f2327 MG |
560 | static struct mem_cgroup_per_node * |
561 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 562 | { |
ef8f2327 | 563 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 564 | |
0a31bc97 | 565 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 566 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 567 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
568 | return mz; |
569 | } | |
570 | ||
11192d9c SB |
571 | /* |
572 | * memcg and lruvec stats flushing | |
573 | * | |
574 | * Many codepaths leading to stats update or read are performance sensitive and | |
575 | * adding stats flushing in such codepaths is not desirable. So, to optimize the | |
576 | * flushing the kernel does: | |
577 | * | |
578 | * 1) Periodically and asynchronously flush the stats every 2 seconds to not let | |
579 | * rstat update tree grow unbounded. | |
580 | * | |
581 | * 2) Flush the stats synchronously on reader side only when there are more than | |
582 | * (MEMCG_CHARGE_BATCH * nr_cpus) update events. Though this optimization | |
583 | * will let stats be out of sync by atmost (MEMCG_CHARGE_BATCH * nr_cpus) but | |
584 | * only for 2 seconds due to (1). | |
585 | */ | |
586 | static void flush_memcg_stats_dwork(struct work_struct *w); | |
587 | static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork); | |
588 | static DEFINE_SPINLOCK(stats_flush_lock); | |
589 | static DEFINE_PER_CPU(unsigned int, stats_updates); | |
590 | static atomic_t stats_flush_threshold = ATOMIC_INIT(0); | |
9b301615 SB |
591 | static u64 flush_next_time; |
592 | ||
593 | #define FLUSH_TIME (2UL*HZ) | |
11192d9c | 594 | |
be3e67b5 SAS |
595 | /* |
596 | * Accessors to ensure that preemption is disabled on PREEMPT_RT because it can | |
597 | * not rely on this as part of an acquired spinlock_t lock. These functions are | |
598 | * never used in hardirq context on PREEMPT_RT and therefore disabling preemtion | |
599 | * is sufficient. | |
600 | */ | |
601 | static void memcg_stats_lock(void) | |
602 | { | |
e575d401 TG |
603 | preempt_disable_nested(); |
604 | VM_WARN_ON_IRQS_ENABLED(); | |
be3e67b5 SAS |
605 | } |
606 | ||
607 | static void __memcg_stats_lock(void) | |
608 | { | |
e575d401 | 609 | preempt_disable_nested(); |
be3e67b5 SAS |
610 | } |
611 | ||
612 | static void memcg_stats_unlock(void) | |
613 | { | |
e575d401 | 614 | preempt_enable_nested(); |
be3e67b5 SAS |
615 | } |
616 | ||
5b3be698 | 617 | static inline void memcg_rstat_updated(struct mem_cgroup *memcg, int val) |
11192d9c | 618 | { |
5b3be698 SB |
619 | unsigned int x; |
620 | ||
11192d9c | 621 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); |
5b3be698 SB |
622 | |
623 | x = __this_cpu_add_return(stats_updates, abs(val)); | |
624 | if (x > MEMCG_CHARGE_BATCH) { | |
873f64b7 JS |
625 | /* |
626 | * If stats_flush_threshold exceeds the threshold | |
627 | * (>num_online_cpus()), cgroup stats update will be triggered | |
628 | * in __mem_cgroup_flush_stats(). Increasing this var further | |
629 | * is redundant and simply adds overhead in atomic update. | |
630 | */ | |
631 | if (atomic_read(&stats_flush_threshold) <= num_online_cpus()) | |
632 | atomic_add(x / MEMCG_CHARGE_BATCH, &stats_flush_threshold); | |
5b3be698 SB |
633 | __this_cpu_write(stats_updates, 0); |
634 | } | |
11192d9c SB |
635 | } |
636 | ||
637 | static void __mem_cgroup_flush_stats(void) | |
638 | { | |
fd25a9e0 SB |
639 | unsigned long flag; |
640 | ||
641 | if (!spin_trylock_irqsave(&stats_flush_lock, flag)) | |
11192d9c SB |
642 | return; |
643 | ||
9b301615 | 644 | flush_next_time = jiffies_64 + 2*FLUSH_TIME; |
11192d9c SB |
645 | cgroup_rstat_flush_irqsafe(root_mem_cgroup->css.cgroup); |
646 | atomic_set(&stats_flush_threshold, 0); | |
fd25a9e0 | 647 | spin_unlock_irqrestore(&stats_flush_lock, flag); |
11192d9c SB |
648 | } |
649 | ||
650 | void mem_cgroup_flush_stats(void) | |
651 | { | |
652 | if (atomic_read(&stats_flush_threshold) > num_online_cpus()) | |
653 | __mem_cgroup_flush_stats(); | |
654 | } | |
655 | ||
9b301615 SB |
656 | void mem_cgroup_flush_stats_delayed(void) |
657 | { | |
658 | if (time_after64(jiffies_64, flush_next_time)) | |
659 | mem_cgroup_flush_stats(); | |
660 | } | |
661 | ||
11192d9c SB |
662 | static void flush_memcg_stats_dwork(struct work_struct *w) |
663 | { | |
5b3be698 | 664 | __mem_cgroup_flush_stats(); |
9b301615 | 665 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, FLUSH_TIME); |
11192d9c SB |
666 | } |
667 | ||
d396def5 SB |
668 | /* Subset of vm_event_item to report for memcg event stats */ |
669 | static const unsigned int memcg_vm_event_stat[] = { | |
8278f1c7 SB |
670 | PGPGIN, |
671 | PGPGOUT, | |
d396def5 SB |
672 | PGSCAN_KSWAPD, |
673 | PGSCAN_DIRECT, | |
57e9cc50 | 674 | PGSCAN_KHUGEPAGED, |
d396def5 SB |
675 | PGSTEAL_KSWAPD, |
676 | PGSTEAL_DIRECT, | |
57e9cc50 | 677 | PGSTEAL_KHUGEPAGED, |
d396def5 SB |
678 | PGFAULT, |
679 | PGMAJFAULT, | |
680 | PGREFILL, | |
681 | PGACTIVATE, | |
682 | PGDEACTIVATE, | |
683 | PGLAZYFREE, | |
684 | PGLAZYFREED, | |
685 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) | |
686 | ZSWPIN, | |
687 | ZSWPOUT, | |
688 | #endif | |
689 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
690 | THP_FAULT_ALLOC, | |
691 | THP_COLLAPSE_ALLOC, | |
692 | #endif | |
693 | }; | |
694 | ||
8278f1c7 SB |
695 | #define NR_MEMCG_EVENTS ARRAY_SIZE(memcg_vm_event_stat) |
696 | static int mem_cgroup_events_index[NR_VM_EVENT_ITEMS] __read_mostly; | |
697 | ||
698 | static void init_memcg_events(void) | |
699 | { | |
700 | int i; | |
701 | ||
702 | for (i = 0; i < NR_MEMCG_EVENTS; ++i) | |
703 | mem_cgroup_events_index[memcg_vm_event_stat[i]] = i + 1; | |
704 | } | |
705 | ||
706 | static inline int memcg_events_index(enum vm_event_item idx) | |
707 | { | |
708 | return mem_cgroup_events_index[idx] - 1; | |
709 | } | |
710 | ||
410f8e82 SB |
711 | struct memcg_vmstats_percpu { |
712 | /* Local (CPU and cgroup) page state & events */ | |
713 | long state[MEMCG_NR_STAT]; | |
8278f1c7 | 714 | unsigned long events[NR_MEMCG_EVENTS]; |
410f8e82 SB |
715 | |
716 | /* Delta calculation for lockless upward propagation */ | |
717 | long state_prev[MEMCG_NR_STAT]; | |
8278f1c7 | 718 | unsigned long events_prev[NR_MEMCG_EVENTS]; |
410f8e82 SB |
719 | |
720 | /* Cgroup1: threshold notifications & softlimit tree updates */ | |
721 | unsigned long nr_page_events; | |
722 | unsigned long targets[MEM_CGROUP_NTARGETS]; | |
723 | }; | |
724 | ||
725 | struct memcg_vmstats { | |
726 | /* Aggregated (CPU and subtree) page state & events */ | |
727 | long state[MEMCG_NR_STAT]; | |
8278f1c7 | 728 | unsigned long events[NR_MEMCG_EVENTS]; |
410f8e82 SB |
729 | |
730 | /* Pending child counts during tree propagation */ | |
731 | long state_pending[MEMCG_NR_STAT]; | |
8278f1c7 | 732 | unsigned long events_pending[NR_MEMCG_EVENTS]; |
410f8e82 SB |
733 | }; |
734 | ||
735 | unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) | |
736 | { | |
737 | long x = READ_ONCE(memcg->vmstats->state[idx]); | |
738 | #ifdef CONFIG_SMP | |
739 | if (x < 0) | |
740 | x = 0; | |
741 | #endif | |
742 | return x; | |
743 | } | |
744 | ||
db9adbcb JW |
745 | /** |
746 | * __mod_memcg_state - update cgroup memory statistics | |
747 | * @memcg: the memory cgroup | |
748 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
749 | * @val: delta to add to the counter, can be negative | |
750 | */ | |
751 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
752 | { | |
db9adbcb JW |
753 | if (mem_cgroup_disabled()) |
754 | return; | |
755 | ||
2d146aa3 | 756 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
5b3be698 | 757 | memcg_rstat_updated(memcg, val); |
db9adbcb JW |
758 | } |
759 | ||
2d146aa3 | 760 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
a18e6e6e JW |
761 | static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx) |
762 | { | |
763 | long x = 0; | |
764 | int cpu; | |
765 | ||
766 | for_each_possible_cpu(cpu) | |
2d146aa3 | 767 | x += per_cpu(memcg->vmstats_percpu->state[idx], cpu); |
a18e6e6e JW |
768 | #ifdef CONFIG_SMP |
769 | if (x < 0) | |
770 | x = 0; | |
771 | #endif | |
772 | return x; | |
773 | } | |
774 | ||
eedc4e5a RG |
775 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
776 | int val) | |
db9adbcb JW |
777 | { |
778 | struct mem_cgroup_per_node *pn; | |
42a30035 | 779 | struct mem_cgroup *memcg; |
db9adbcb | 780 | |
db9adbcb | 781 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 782 | memcg = pn->memcg; |
db9adbcb | 783 | |
be3e67b5 SAS |
784 | /* |
785 | * The caller from rmap relay on disabled preemption becase they never | |
786 | * update their counter from in-interrupt context. For these two | |
787 | * counters we check that the update is never performed from an | |
788 | * interrupt context while other caller need to have disabled interrupt. | |
789 | */ | |
790 | __memcg_stats_lock(); | |
e575d401 | 791 | if (IS_ENABLED(CONFIG_DEBUG_VM)) { |
be3e67b5 SAS |
792 | switch (idx) { |
793 | case NR_ANON_MAPPED: | |
794 | case NR_FILE_MAPPED: | |
795 | case NR_ANON_THPS: | |
796 | case NR_SHMEM_PMDMAPPED: | |
797 | case NR_FILE_PMDMAPPED: | |
798 | WARN_ON_ONCE(!in_task()); | |
799 | break; | |
800 | default: | |
e575d401 | 801 | VM_WARN_ON_IRQS_ENABLED(); |
be3e67b5 SAS |
802 | } |
803 | } | |
804 | ||
db9adbcb | 805 | /* Update memcg */ |
11192d9c | 806 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
db9adbcb | 807 | |
b4c46484 | 808 | /* Update lruvec */ |
7e1c0d6f | 809 | __this_cpu_add(pn->lruvec_stats_percpu->state[idx], val); |
11192d9c | 810 | |
5b3be698 | 811 | memcg_rstat_updated(memcg, val); |
be3e67b5 | 812 | memcg_stats_unlock(); |
db9adbcb JW |
813 | } |
814 | ||
eedc4e5a RG |
815 | /** |
816 | * __mod_lruvec_state - update lruvec memory statistics | |
817 | * @lruvec: the lruvec | |
818 | * @idx: the stat item | |
819 | * @val: delta to add to the counter, can be negative | |
820 | * | |
821 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
822 | * function updates the all three counters that are affected by a | |
823 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
824 | */ | |
825 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
826 | int val) | |
827 | { | |
828 | /* Update node */ | |
829 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
830 | ||
831 | /* Update memcg and lruvec */ | |
832 | if (!mem_cgroup_disabled()) | |
833 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
834 | } | |
835 | ||
c47d5032 SB |
836 | void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx, |
837 | int val) | |
838 | { | |
839 | struct page *head = compound_head(page); /* rmap on tail pages */ | |
b4e0b68f | 840 | struct mem_cgroup *memcg; |
c47d5032 SB |
841 | pg_data_t *pgdat = page_pgdat(page); |
842 | struct lruvec *lruvec; | |
843 | ||
b4e0b68f MS |
844 | rcu_read_lock(); |
845 | memcg = page_memcg(head); | |
c47d5032 | 846 | /* Untracked pages have no memcg, no lruvec. Update only the node */ |
d635a69d | 847 | if (!memcg) { |
b4e0b68f | 848 | rcu_read_unlock(); |
c47d5032 SB |
849 | __mod_node_page_state(pgdat, idx, val); |
850 | return; | |
851 | } | |
852 | ||
d635a69d | 853 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 | 854 | __mod_lruvec_state(lruvec, idx, val); |
b4e0b68f | 855 | rcu_read_unlock(); |
c47d5032 | 856 | } |
f0c0c115 | 857 | EXPORT_SYMBOL(__mod_lruvec_page_state); |
c47d5032 | 858 | |
da3ceeff | 859 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 860 | { |
4f103c63 | 861 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
862 | struct mem_cgroup *memcg; |
863 | struct lruvec *lruvec; | |
864 | ||
865 | rcu_read_lock(); | |
fc4db90f | 866 | memcg = mem_cgroup_from_slab_obj(p); |
ec9f0238 | 867 | |
8faeb1ff MS |
868 | /* |
869 | * Untracked pages have no memcg, no lruvec. Update only the | |
870 | * node. If we reparent the slab objects to the root memcg, | |
871 | * when we free the slab object, we need to update the per-memcg | |
872 | * vmstats to keep it correct for the root memcg. | |
873 | */ | |
874 | if (!memcg) { | |
ec9f0238 RG |
875 | __mod_node_page_state(pgdat, idx, val); |
876 | } else { | |
867e5e1d | 877 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
878 | __mod_lruvec_state(lruvec, idx, val); |
879 | } | |
880 | rcu_read_unlock(); | |
881 | } | |
882 | ||
db9adbcb JW |
883 | /** |
884 | * __count_memcg_events - account VM events in a cgroup | |
885 | * @memcg: the memory cgroup | |
886 | * @idx: the event item | |
f0953a1b | 887 | * @count: the number of events that occurred |
db9adbcb JW |
888 | */ |
889 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
890 | unsigned long count) | |
891 | { | |
8278f1c7 SB |
892 | int index = memcg_events_index(idx); |
893 | ||
894 | if (mem_cgroup_disabled() || index < 0) | |
db9adbcb JW |
895 | return; |
896 | ||
be3e67b5 | 897 | memcg_stats_lock(); |
8278f1c7 | 898 | __this_cpu_add(memcg->vmstats_percpu->events[index], count); |
5b3be698 | 899 | memcg_rstat_updated(memcg, count); |
be3e67b5 | 900 | memcg_stats_unlock(); |
db9adbcb JW |
901 | } |
902 | ||
42a30035 | 903 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 904 | { |
8278f1c7 SB |
905 | int index = memcg_events_index(event); |
906 | ||
907 | if (index < 0) | |
908 | return 0; | |
909 | return READ_ONCE(memcg->vmstats->events[index]); | |
e9f8974f JW |
910 | } |
911 | ||
42a30035 JW |
912 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
913 | { | |
815744d7 JW |
914 | long x = 0; |
915 | int cpu; | |
8278f1c7 SB |
916 | int index = memcg_events_index(event); |
917 | ||
918 | if (index < 0) | |
919 | return 0; | |
815744d7 JW |
920 | |
921 | for_each_possible_cpu(cpu) | |
8278f1c7 | 922 | x += per_cpu(memcg->vmstats_percpu->events[index], cpu); |
815744d7 | 923 | return x; |
42a30035 JW |
924 | } |
925 | ||
c0ff4b85 | 926 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
3fba69a5 | 927 | int nr_pages) |
d52aa412 | 928 | { |
e401f176 KH |
929 | /* pagein of a big page is an event. So, ignore page size */ |
930 | if (nr_pages > 0) | |
c9019e9b | 931 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 932 | else { |
c9019e9b | 933 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
934 | nr_pages = -nr_pages; /* for event */ |
935 | } | |
e401f176 | 936 | |
871789d4 | 937 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
938 | } |
939 | ||
f53d7ce3 JW |
940 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
941 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
942 | { |
943 | unsigned long val, next; | |
944 | ||
871789d4 CD |
945 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
946 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 947 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 948 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
949 | switch (target) { |
950 | case MEM_CGROUP_TARGET_THRESH: | |
951 | next = val + THRESHOLDS_EVENTS_TARGET; | |
952 | break; | |
bb4cc1a8 AM |
953 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
954 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
955 | break; | |
f53d7ce3 JW |
956 | default: |
957 | break; | |
958 | } | |
871789d4 | 959 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 960 | return true; |
7a159cc9 | 961 | } |
f53d7ce3 | 962 | return false; |
d2265e6f KH |
963 | } |
964 | ||
965 | /* | |
966 | * Check events in order. | |
967 | * | |
968 | */ | |
8e88bd2d | 969 | static void memcg_check_events(struct mem_cgroup *memcg, int nid) |
d2265e6f | 970 | { |
2343e88d SAS |
971 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
972 | return; | |
973 | ||
d2265e6f | 974 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
975 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
976 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 977 | bool do_softlimit; |
f53d7ce3 | 978 | |
bb4cc1a8 AM |
979 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
980 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 981 | mem_cgroup_threshold(memcg); |
bb4cc1a8 | 982 | if (unlikely(do_softlimit)) |
8e88bd2d | 983 | mem_cgroup_update_tree(memcg, nid); |
0a31bc97 | 984 | } |
d2265e6f KH |
985 | } |
986 | ||
cf475ad2 | 987 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 988 | { |
31a78f23 BS |
989 | /* |
990 | * mm_update_next_owner() may clear mm->owner to NULL | |
991 | * if it races with swapoff, page migration, etc. | |
992 | * So this can be called with p == NULL. | |
993 | */ | |
994 | if (unlikely(!p)) | |
995 | return NULL; | |
996 | ||
073219e9 | 997 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 998 | } |
33398cf2 | 999 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1000 | |
04f94e3f DS |
1001 | static __always_inline struct mem_cgroup *active_memcg(void) |
1002 | { | |
55a68c82 | 1003 | if (!in_task()) |
04f94e3f DS |
1004 | return this_cpu_read(int_active_memcg); |
1005 | else | |
1006 | return current->active_memcg; | |
1007 | } | |
1008 | ||
d46eb14b SB |
1009 | /** |
1010 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1011 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1012 | * | |
04f94e3f DS |
1013 | * Obtain a reference on mm->memcg and returns it if successful. If mm |
1014 | * is NULL, then the memcg is chosen as follows: | |
1015 | * 1) The active memcg, if set. | |
1016 | * 2) current->mm->memcg, if available | |
1017 | * 3) root memcg | |
1018 | * If mem_cgroup is disabled, NULL is returned. | |
d46eb14b SB |
1019 | */ |
1020 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1021 | { |
d46eb14b SB |
1022 | struct mem_cgroup *memcg; |
1023 | ||
1024 | if (mem_cgroup_disabled()) | |
1025 | return NULL; | |
0b7f569e | 1026 | |
2884b6b7 MS |
1027 | /* |
1028 | * Page cache insertions can happen without an | |
1029 | * actual mm context, e.g. during disk probing | |
1030 | * on boot, loopback IO, acct() writes etc. | |
1031 | * | |
1032 | * No need to css_get on root memcg as the reference | |
1033 | * counting is disabled on the root level in the | |
1034 | * cgroup core. See CSS_NO_REF. | |
1035 | */ | |
04f94e3f DS |
1036 | if (unlikely(!mm)) { |
1037 | memcg = active_memcg(); | |
1038 | if (unlikely(memcg)) { | |
1039 | /* remote memcg must hold a ref */ | |
1040 | css_get(&memcg->css); | |
1041 | return memcg; | |
1042 | } | |
1043 | mm = current->mm; | |
1044 | if (unlikely(!mm)) | |
1045 | return root_mem_cgroup; | |
1046 | } | |
2884b6b7 | 1047 | |
54595fe2 KH |
1048 | rcu_read_lock(); |
1049 | do { | |
2884b6b7 MS |
1050 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1051 | if (unlikely(!memcg)) | |
df381975 | 1052 | memcg = root_mem_cgroup; |
00d484f3 | 1053 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1054 | rcu_read_unlock(); |
c0ff4b85 | 1055 | return memcg; |
54595fe2 | 1056 | } |
d46eb14b SB |
1057 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1058 | ||
4127c650 RG |
1059 | static __always_inline bool memcg_kmem_bypass(void) |
1060 | { | |
1061 | /* Allow remote memcg charging from any context. */ | |
1062 | if (unlikely(active_memcg())) | |
1063 | return false; | |
1064 | ||
1065 | /* Memcg to charge can't be determined. */ | |
6126891c | 1066 | if (!in_task() || !current->mm || (current->flags & PF_KTHREAD)) |
4127c650 RG |
1067 | return true; |
1068 | ||
1069 | return false; | |
1070 | } | |
1071 | ||
5660048c JW |
1072 | /** |
1073 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1074 | * @root: hierarchy root | |
1075 | * @prev: previously returned memcg, NULL on first invocation | |
1076 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1077 | * | |
1078 | * Returns references to children of the hierarchy below @root, or | |
1079 | * @root itself, or %NULL after a full round-trip. | |
1080 | * | |
1081 | * Caller must pass the return value in @prev on subsequent | |
1082 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1083 | * to cancel a hierarchy walk before the round-trip is complete. | |
1084 | * | |
05bdc520 ML |
1085 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1086 | * in the hierarchy among all concurrent reclaimers operating on the | |
1087 | * same node. | |
5660048c | 1088 | */ |
694fbc0f | 1089 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1090 | struct mem_cgroup *prev, |
694fbc0f | 1091 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1092 | { |
3f649ab7 | 1093 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1094 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1095 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1096 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1097 | |
694fbc0f AM |
1098 | if (mem_cgroup_disabled()) |
1099 | return NULL; | |
5660048c | 1100 | |
9f3a0d09 JW |
1101 | if (!root) |
1102 | root = root_mem_cgroup; | |
7d74b06f | 1103 | |
542f85f9 | 1104 | rcu_read_lock(); |
5f578161 | 1105 | |
5ac8fb31 | 1106 | if (reclaim) { |
ef8f2327 | 1107 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1108 | |
a3747b53 | 1109 | mz = root->nodeinfo[reclaim->pgdat->node_id]; |
9da83f3f | 1110 | iter = &mz->iter; |
5ac8fb31 | 1111 | |
a9320aae WY |
1112 | /* |
1113 | * On start, join the current reclaim iteration cycle. | |
1114 | * Exit when a concurrent walker completes it. | |
1115 | */ | |
1116 | if (!prev) | |
1117 | reclaim->generation = iter->generation; | |
1118 | else if (reclaim->generation != iter->generation) | |
5ac8fb31 JW |
1119 | goto out_unlock; |
1120 | ||
6df38689 | 1121 | while (1) { |
4db0c3c2 | 1122 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1123 | if (!pos || css_tryget(&pos->css)) |
1124 | break; | |
5ac8fb31 | 1125 | /* |
6df38689 VD |
1126 | * css reference reached zero, so iter->position will |
1127 | * be cleared by ->css_released. However, we should not | |
1128 | * rely on this happening soon, because ->css_released | |
1129 | * is called from a work queue, and by busy-waiting we | |
1130 | * might block it. So we clear iter->position right | |
1131 | * away. | |
5ac8fb31 | 1132 | */ |
6df38689 VD |
1133 | (void)cmpxchg(&iter->position, pos, NULL); |
1134 | } | |
89d8330c WY |
1135 | } else if (prev) { |
1136 | pos = prev; | |
5ac8fb31 JW |
1137 | } |
1138 | ||
1139 | if (pos) | |
1140 | css = &pos->css; | |
1141 | ||
1142 | for (;;) { | |
1143 | css = css_next_descendant_pre(css, &root->css); | |
1144 | if (!css) { | |
1145 | /* | |
1146 | * Reclaimers share the hierarchy walk, and a | |
1147 | * new one might jump in right at the end of | |
1148 | * the hierarchy - make sure they see at least | |
1149 | * one group and restart from the beginning. | |
1150 | */ | |
1151 | if (!prev) | |
1152 | continue; | |
1153 | break; | |
527a5ec9 | 1154 | } |
7d74b06f | 1155 | |
5ac8fb31 JW |
1156 | /* |
1157 | * Verify the css and acquire a reference. The root | |
1158 | * is provided by the caller, so we know it's alive | |
1159 | * and kicking, and don't take an extra reference. | |
1160 | */ | |
41555dad WY |
1161 | if (css == &root->css || css_tryget(css)) { |
1162 | memcg = mem_cgroup_from_css(css); | |
0b8f73e1 | 1163 | break; |
41555dad | 1164 | } |
9f3a0d09 | 1165 | } |
5ac8fb31 JW |
1166 | |
1167 | if (reclaim) { | |
5ac8fb31 | 1168 | /* |
6df38689 VD |
1169 | * The position could have already been updated by a competing |
1170 | * thread, so check that the value hasn't changed since we read | |
1171 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1172 | */ |
6df38689 VD |
1173 | (void)cmpxchg(&iter->position, pos, memcg); |
1174 | ||
5ac8fb31 JW |
1175 | if (pos) |
1176 | css_put(&pos->css); | |
1177 | ||
1178 | if (!memcg) | |
1179 | iter->generation++; | |
9f3a0d09 | 1180 | } |
5ac8fb31 | 1181 | |
542f85f9 MH |
1182 | out_unlock: |
1183 | rcu_read_unlock(); | |
c40046f3 MH |
1184 | if (prev && prev != root) |
1185 | css_put(&prev->css); | |
1186 | ||
9f3a0d09 | 1187 | return memcg; |
14067bb3 | 1188 | } |
7d74b06f | 1189 | |
5660048c JW |
1190 | /** |
1191 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1192 | * @root: hierarchy root | |
1193 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1194 | */ | |
1195 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1196 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1197 | { |
1198 | if (!root) | |
1199 | root = root_mem_cgroup; | |
1200 | if (prev && prev != root) | |
1201 | css_put(&prev->css); | |
1202 | } | |
7d74b06f | 1203 | |
54a83d6b MC |
1204 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1205 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1206 | { |
6df38689 | 1207 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1208 | struct mem_cgroup_per_node *mz; |
1209 | int nid; | |
6df38689 | 1210 | |
54a83d6b | 1211 | for_each_node(nid) { |
a3747b53 | 1212 | mz = from->nodeinfo[nid]; |
9da83f3f YS |
1213 | iter = &mz->iter; |
1214 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1215 | } |
1216 | } | |
1217 | ||
54a83d6b MC |
1218 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1219 | { | |
1220 | struct mem_cgroup *memcg = dead_memcg; | |
1221 | struct mem_cgroup *last; | |
1222 | ||
1223 | do { | |
1224 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1225 | last = memcg; | |
1226 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1227 | ||
1228 | /* | |
b8dd3ee9 | 1229 | * When cgroup1 non-hierarchy mode is used, |
54a83d6b MC |
1230 | * parent_mem_cgroup() does not walk all the way up to the |
1231 | * cgroup root (root_mem_cgroup). So we have to handle | |
1232 | * dead_memcg from cgroup root separately. | |
1233 | */ | |
7848ed62 | 1234 | if (!mem_cgroup_is_root(last)) |
54a83d6b MC |
1235 | __invalidate_reclaim_iterators(root_mem_cgroup, |
1236 | dead_memcg); | |
1237 | } | |
1238 | ||
7c5f64f8 VD |
1239 | /** |
1240 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1241 | * @memcg: hierarchy root | |
1242 | * @fn: function to call for each task | |
1243 | * @arg: argument passed to @fn | |
1244 | * | |
1245 | * This function iterates over tasks attached to @memcg or to any of its | |
1246 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1247 | * value, the function breaks the iteration loop and returns the value. | |
1248 | * Otherwise, it will iterate over all tasks and return 0. | |
1249 | * | |
1250 | * This function must not be called for the root memory cgroup. | |
1251 | */ | |
1252 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1253 | int (*fn)(struct task_struct *, void *), void *arg) | |
1254 | { | |
1255 | struct mem_cgroup *iter; | |
1256 | int ret = 0; | |
1257 | ||
7848ed62 | 1258 | BUG_ON(mem_cgroup_is_root(memcg)); |
7c5f64f8 VD |
1259 | |
1260 | for_each_mem_cgroup_tree(iter, memcg) { | |
1261 | struct css_task_iter it; | |
1262 | struct task_struct *task; | |
1263 | ||
f168a9a5 | 1264 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1265 | while (!ret && (task = css_task_iter_next(&it))) |
1266 | ret = fn(task, arg); | |
1267 | css_task_iter_end(&it); | |
1268 | if (ret) { | |
1269 | mem_cgroup_iter_break(memcg, iter); | |
1270 | break; | |
1271 | } | |
1272 | } | |
1273 | return ret; | |
1274 | } | |
1275 | ||
6168d0da | 1276 | #ifdef CONFIG_DEBUG_VM |
e809c3fe | 1277 | void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio) |
6168d0da AS |
1278 | { |
1279 | struct mem_cgroup *memcg; | |
1280 | ||
1281 | if (mem_cgroup_disabled()) | |
1282 | return; | |
1283 | ||
e809c3fe | 1284 | memcg = folio_memcg(folio); |
6168d0da AS |
1285 | |
1286 | if (!memcg) | |
7848ed62 | 1287 | VM_BUG_ON_FOLIO(!mem_cgroup_is_root(lruvec_memcg(lruvec)), folio); |
6168d0da | 1288 | else |
e809c3fe | 1289 | VM_BUG_ON_FOLIO(lruvec_memcg(lruvec) != memcg, folio); |
6168d0da AS |
1290 | } |
1291 | #endif | |
1292 | ||
6168d0da | 1293 | /** |
e809c3fe MWO |
1294 | * folio_lruvec_lock - Lock the lruvec for a folio. |
1295 | * @folio: Pointer to the folio. | |
6168d0da | 1296 | * |
d7e3aba5 | 1297 | * These functions are safe to use under any of the following conditions: |
e809c3fe MWO |
1298 | * - folio locked |
1299 | * - folio_test_lru false | |
1300 | * - folio_memcg_lock() | |
1301 | * - folio frozen (refcount of 0) | |
1302 | * | |
1303 | * Return: The lruvec this folio is on with its lock held. | |
6168d0da | 1304 | */ |
e809c3fe | 1305 | struct lruvec *folio_lruvec_lock(struct folio *folio) |
6168d0da | 1306 | { |
e809c3fe | 1307 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1308 | |
6168d0da | 1309 | spin_lock(&lruvec->lru_lock); |
e809c3fe | 1310 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1311 | |
1312 | return lruvec; | |
1313 | } | |
1314 | ||
e809c3fe MWO |
1315 | /** |
1316 | * folio_lruvec_lock_irq - Lock the lruvec for a folio. | |
1317 | * @folio: Pointer to the folio. | |
1318 | * | |
1319 | * These functions are safe to use under any of the following conditions: | |
1320 | * - folio locked | |
1321 | * - folio_test_lru false | |
1322 | * - folio_memcg_lock() | |
1323 | * - folio frozen (refcount of 0) | |
1324 | * | |
1325 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1326 | * disabled. | |
1327 | */ | |
1328 | struct lruvec *folio_lruvec_lock_irq(struct folio *folio) | |
6168d0da | 1329 | { |
e809c3fe | 1330 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1331 | |
6168d0da | 1332 | spin_lock_irq(&lruvec->lru_lock); |
e809c3fe | 1333 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1334 | |
1335 | return lruvec; | |
1336 | } | |
1337 | ||
e809c3fe MWO |
1338 | /** |
1339 | * folio_lruvec_lock_irqsave - Lock the lruvec for a folio. | |
1340 | * @folio: Pointer to the folio. | |
1341 | * @flags: Pointer to irqsave flags. | |
1342 | * | |
1343 | * These functions are safe to use under any of the following conditions: | |
1344 | * - folio locked | |
1345 | * - folio_test_lru false | |
1346 | * - folio_memcg_lock() | |
1347 | * - folio frozen (refcount of 0) | |
1348 | * | |
1349 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1350 | * disabled. | |
1351 | */ | |
1352 | struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio, | |
1353 | unsigned long *flags) | |
6168d0da | 1354 | { |
e809c3fe | 1355 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1356 | |
6168d0da | 1357 | spin_lock_irqsave(&lruvec->lru_lock, *flags); |
e809c3fe | 1358 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1359 | |
1360 | return lruvec; | |
1361 | } | |
1362 | ||
925b7673 | 1363 | /** |
fa9add64 HD |
1364 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1365 | * @lruvec: mem_cgroup per zone lru vector | |
1366 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1367 | * @zid: zone id of the accounted pages |
fa9add64 | 1368 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1369 | * |
ca707239 | 1370 | * This function must be called under lru_lock, just before a page is added |
07ca7606 | 1371 | * to or just after a page is removed from an lru list. |
3f58a829 | 1372 | */ |
fa9add64 | 1373 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1374 | int zid, int nr_pages) |
3f58a829 | 1375 | { |
ef8f2327 | 1376 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1377 | unsigned long *lru_size; |
ca707239 | 1378 | long size; |
3f58a829 MK |
1379 | |
1380 | if (mem_cgroup_disabled()) | |
1381 | return; | |
1382 | ||
ef8f2327 | 1383 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1384 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1385 | |
1386 | if (nr_pages < 0) | |
1387 | *lru_size += nr_pages; | |
1388 | ||
1389 | size = *lru_size; | |
b4536f0c MH |
1390 | if (WARN_ONCE(size < 0, |
1391 | "%s(%p, %d, %d): lru_size %ld\n", | |
1392 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1393 | VM_BUG_ON(1); |
1394 | *lru_size = 0; | |
1395 | } | |
1396 | ||
1397 | if (nr_pages > 0) | |
1398 | *lru_size += nr_pages; | |
08e552c6 | 1399 | } |
544122e5 | 1400 | |
19942822 | 1401 | /** |
9d11ea9f | 1402 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1403 | * @memcg: the memory cgroup |
19942822 | 1404 | * |
9d11ea9f | 1405 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1406 | * pages. |
19942822 | 1407 | */ |
c0ff4b85 | 1408 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1409 | { |
3e32cb2e JW |
1410 | unsigned long margin = 0; |
1411 | unsigned long count; | |
1412 | unsigned long limit; | |
9d11ea9f | 1413 | |
3e32cb2e | 1414 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1415 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1416 | if (count < limit) |
1417 | margin = limit - count; | |
1418 | ||
7941d214 | 1419 | if (do_memsw_account()) { |
3e32cb2e | 1420 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1421 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1422 | if (count < limit) |
3e32cb2e | 1423 | margin = min(margin, limit - count); |
cbedbac3 LR |
1424 | else |
1425 | margin = 0; | |
3e32cb2e JW |
1426 | } |
1427 | ||
1428 | return margin; | |
19942822 JW |
1429 | } |
1430 | ||
32047e2a | 1431 | /* |
bdcbb659 | 1432 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1433 | * |
bdcbb659 QH |
1434 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1435 | * moving cgroups. This is for waiting at high-memory pressure | |
1436 | * caused by "move". | |
32047e2a | 1437 | */ |
c0ff4b85 | 1438 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1439 | { |
2bd9bb20 KH |
1440 | struct mem_cgroup *from; |
1441 | struct mem_cgroup *to; | |
4b534334 | 1442 | bool ret = false; |
2bd9bb20 KH |
1443 | /* |
1444 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1445 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1446 | */ | |
1447 | spin_lock(&mc.lock); | |
1448 | from = mc.from; | |
1449 | to = mc.to; | |
1450 | if (!from) | |
1451 | goto unlock; | |
3e92041d | 1452 | |
2314b42d JW |
1453 | ret = mem_cgroup_is_descendant(from, memcg) || |
1454 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1455 | unlock: |
1456 | spin_unlock(&mc.lock); | |
4b534334 KH |
1457 | return ret; |
1458 | } | |
1459 | ||
c0ff4b85 | 1460 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1461 | { |
1462 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1463 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1464 | DEFINE_WAIT(wait); |
1465 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1466 | /* moving charge context might have finished. */ | |
1467 | if (mc.moving_task) | |
1468 | schedule(); | |
1469 | finish_wait(&mc.waitq, &wait); | |
1470 | return true; | |
1471 | } | |
1472 | } | |
1473 | return false; | |
1474 | } | |
1475 | ||
5f9a4f4a MS |
1476 | struct memory_stat { |
1477 | const char *name; | |
5f9a4f4a MS |
1478 | unsigned int idx; |
1479 | }; | |
1480 | ||
57b2847d | 1481 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1482 | { "anon", NR_ANON_MAPPED }, |
1483 | { "file", NR_FILE_PAGES }, | |
a8c49af3 | 1484 | { "kernel", MEMCG_KMEM }, |
fff66b79 MS |
1485 | { "kernel_stack", NR_KERNEL_STACK_KB }, |
1486 | { "pagetables", NR_PAGETABLE }, | |
ebc97a52 | 1487 | { "sec_pagetables", NR_SECONDARY_PAGETABLE }, |
fff66b79 MS |
1488 | { "percpu", MEMCG_PERCPU_B }, |
1489 | { "sock", MEMCG_SOCK }, | |
4e5aa1f4 | 1490 | { "vmalloc", MEMCG_VMALLOC }, |
fff66b79 | 1491 | { "shmem", NR_SHMEM }, |
f4840ccf JW |
1492 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
1493 | { "zswap", MEMCG_ZSWAP_B }, | |
1494 | { "zswapped", MEMCG_ZSWAPPED }, | |
1495 | #endif | |
fff66b79 MS |
1496 | { "file_mapped", NR_FILE_MAPPED }, |
1497 | { "file_dirty", NR_FILE_DIRTY }, | |
1498 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1499 | #ifdef CONFIG_SWAP |
1500 | { "swapcached", NR_SWAPCACHE }, | |
1501 | #endif | |
5f9a4f4a | 1502 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1503 | { "anon_thp", NR_ANON_THPS }, |
1504 | { "file_thp", NR_FILE_THPS }, | |
1505 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1506 | #endif |
fff66b79 MS |
1507 | { "inactive_anon", NR_INACTIVE_ANON }, |
1508 | { "active_anon", NR_ACTIVE_ANON }, | |
1509 | { "inactive_file", NR_INACTIVE_FILE }, | |
1510 | { "active_file", NR_ACTIVE_FILE }, | |
1511 | { "unevictable", NR_UNEVICTABLE }, | |
1512 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1513 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1514 | |
1515 | /* The memory events */ | |
fff66b79 MS |
1516 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1517 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1518 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1519 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1520 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1521 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1522 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1523 | }; |
1524 | ||
fff66b79 MS |
1525 | /* Translate stat items to the correct unit for memory.stat output */ |
1526 | static int memcg_page_state_unit(int item) | |
1527 | { | |
1528 | switch (item) { | |
1529 | case MEMCG_PERCPU_B: | |
f4840ccf | 1530 | case MEMCG_ZSWAP_B: |
fff66b79 MS |
1531 | case NR_SLAB_RECLAIMABLE_B: |
1532 | case NR_SLAB_UNRECLAIMABLE_B: | |
1533 | case WORKINGSET_REFAULT_ANON: | |
1534 | case WORKINGSET_REFAULT_FILE: | |
1535 | case WORKINGSET_ACTIVATE_ANON: | |
1536 | case WORKINGSET_ACTIVATE_FILE: | |
1537 | case WORKINGSET_RESTORE_ANON: | |
1538 | case WORKINGSET_RESTORE_FILE: | |
1539 | case WORKINGSET_NODERECLAIM: | |
1540 | return 1; | |
1541 | case NR_KERNEL_STACK_KB: | |
1542 | return SZ_1K; | |
1543 | default: | |
1544 | return PAGE_SIZE; | |
1545 | } | |
1546 | } | |
1547 | ||
1548 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1549 | int item) | |
1550 | { | |
1551 | return memcg_page_state(memcg, item) * memcg_page_state_unit(item); | |
1552 | } | |
1553 | ||
68aaee14 | 1554 | static void memory_stat_format(struct mem_cgroup *memcg, char *buf, int bufsize) |
c8713d0b JW |
1555 | { |
1556 | struct seq_buf s; | |
1557 | int i; | |
71cd3113 | 1558 | |
68aaee14 | 1559 | seq_buf_init(&s, buf, bufsize); |
c8713d0b JW |
1560 | |
1561 | /* | |
1562 | * Provide statistics on the state of the memory subsystem as | |
1563 | * well as cumulative event counters that show past behavior. | |
1564 | * | |
1565 | * This list is ordered following a combination of these gradients: | |
1566 | * 1) generic big picture -> specifics and details | |
1567 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1568 | * | |
1569 | * Current memory state: | |
1570 | */ | |
fd25a9e0 | 1571 | mem_cgroup_flush_stats(); |
c8713d0b | 1572 | |
5f9a4f4a MS |
1573 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1574 | u64 size; | |
c8713d0b | 1575 | |
fff66b79 | 1576 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5f9a4f4a | 1577 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1578 | |
5f9a4f4a | 1579 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1580 | size += memcg_page_state_output(memcg, |
1581 | NR_SLAB_RECLAIMABLE_B); | |
5f9a4f4a MS |
1582 | seq_buf_printf(&s, "slab %llu\n", size); |
1583 | } | |
1584 | } | |
c8713d0b JW |
1585 | |
1586 | /* Accumulated memory events */ | |
c8713d0b JW |
1587 | seq_buf_printf(&s, "pgscan %lu\n", |
1588 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
57e9cc50 JW |
1589 | memcg_events(memcg, PGSCAN_DIRECT) + |
1590 | memcg_events(memcg, PGSCAN_KHUGEPAGED)); | |
c8713d0b JW |
1591 | seq_buf_printf(&s, "pgsteal %lu\n", |
1592 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
57e9cc50 JW |
1593 | memcg_events(memcg, PGSTEAL_DIRECT) + |
1594 | memcg_events(memcg, PGSTEAL_KHUGEPAGED)); | |
c8713d0b | 1595 | |
8278f1c7 SB |
1596 | for (i = 0; i < ARRAY_SIZE(memcg_vm_event_stat); i++) { |
1597 | if (memcg_vm_event_stat[i] == PGPGIN || | |
1598 | memcg_vm_event_stat[i] == PGPGOUT) | |
1599 | continue; | |
1600 | ||
673520f8 QZ |
1601 | seq_buf_printf(&s, "%s %lu\n", |
1602 | vm_event_name(memcg_vm_event_stat[i]), | |
1603 | memcg_events(memcg, memcg_vm_event_stat[i])); | |
8278f1c7 | 1604 | } |
c8713d0b JW |
1605 | |
1606 | /* The above should easily fit into one page */ | |
1607 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
c8713d0b | 1608 | } |
71cd3113 | 1609 | |
58cf188e | 1610 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1611 | /** |
f0c867d9 | 1612 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1613 | * memory controller. | |
e222432b BS |
1614 | * @memcg: The memory cgroup that went over limit |
1615 | * @p: Task that is going to be killed | |
1616 | * | |
1617 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1618 | * enabled | |
1619 | */ | |
f0c867d9 | 1620 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1621 | { |
e222432b BS |
1622 | rcu_read_lock(); |
1623 | ||
f0c867d9 | 1624 | if (memcg) { |
1625 | pr_cont(",oom_memcg="); | |
1626 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1627 | } else | |
1628 | pr_cont(",global_oom"); | |
2415b9f5 | 1629 | if (p) { |
f0c867d9 | 1630 | pr_cont(",task_memcg="); |
2415b9f5 | 1631 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1632 | } |
e222432b | 1633 | rcu_read_unlock(); |
f0c867d9 | 1634 | } |
1635 | ||
1636 | /** | |
1637 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1638 | * memory controller. | |
1639 | * @memcg: The memory cgroup that went over limit | |
1640 | */ | |
1641 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1642 | { | |
68aaee14 TH |
1643 | /* Use static buffer, for the caller is holding oom_lock. */ |
1644 | static char buf[PAGE_SIZE]; | |
1645 | ||
1646 | lockdep_assert_held(&oom_lock); | |
e222432b | 1647 | |
3e32cb2e JW |
1648 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1649 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1650 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1651 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1652 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1653 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1654 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1655 | else { |
1656 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1657 | K((u64)page_counter_read(&memcg->memsw)), | |
1658 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1659 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1660 | K((u64)page_counter_read(&memcg->kmem)), | |
1661 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1662 | } |
c8713d0b JW |
1663 | |
1664 | pr_info("Memory cgroup stats for "); | |
1665 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1666 | pr_cont(":"); | |
68aaee14 | 1667 | memory_stat_format(memcg, buf, sizeof(buf)); |
c8713d0b | 1668 | pr_info("%s", buf); |
e222432b BS |
1669 | } |
1670 | ||
a63d83f4 DR |
1671 | /* |
1672 | * Return the memory (and swap, if configured) limit for a memcg. | |
1673 | */ | |
bbec2e15 | 1674 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1675 | { |
8d387a5f WL |
1676 | unsigned long max = READ_ONCE(memcg->memory.max); |
1677 | ||
b94c4e94 | 1678 | if (do_memsw_account()) { |
8d387a5f WL |
1679 | if (mem_cgroup_swappiness(memcg)) { |
1680 | /* Calculate swap excess capacity from memsw limit */ | |
1681 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1682 | ||
1683 | max += min(swap, (unsigned long)total_swap_pages); | |
1684 | } | |
b94c4e94 JW |
1685 | } else { |
1686 | if (mem_cgroup_swappiness(memcg)) | |
1687 | max += min(READ_ONCE(memcg->swap.max), | |
1688 | (unsigned long)total_swap_pages); | |
9a5a8f19 | 1689 | } |
bbec2e15 | 1690 | return max; |
a63d83f4 DR |
1691 | } |
1692 | ||
9783aa99 CD |
1693 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1694 | { | |
1695 | return page_counter_read(&memcg->memory); | |
1696 | } | |
1697 | ||
b6e6edcf | 1698 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1699 | int order) |
9cbb78bb | 1700 | { |
6e0fc46d DR |
1701 | struct oom_control oc = { |
1702 | .zonelist = NULL, | |
1703 | .nodemask = NULL, | |
2a966b77 | 1704 | .memcg = memcg, |
6e0fc46d DR |
1705 | .gfp_mask = gfp_mask, |
1706 | .order = order, | |
6e0fc46d | 1707 | }; |
1378b37d | 1708 | bool ret = true; |
9cbb78bb | 1709 | |
7775face TH |
1710 | if (mutex_lock_killable(&oom_lock)) |
1711 | return true; | |
1378b37d YS |
1712 | |
1713 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1714 | goto unlock; | |
1715 | ||
7775face TH |
1716 | /* |
1717 | * A few threads which were not waiting at mutex_lock_killable() can | |
1718 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1719 | */ | |
a4ebf1b6 | 1720 | ret = task_is_dying() || out_of_memory(&oc); |
1378b37d YS |
1721 | |
1722 | unlock: | |
dc56401f | 1723 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1724 | return ret; |
9cbb78bb DR |
1725 | } |
1726 | ||
0608f43d | 1727 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1728 | pg_data_t *pgdat, |
0608f43d AM |
1729 | gfp_t gfp_mask, |
1730 | unsigned long *total_scanned) | |
1731 | { | |
1732 | struct mem_cgroup *victim = NULL; | |
1733 | int total = 0; | |
1734 | int loop = 0; | |
1735 | unsigned long excess; | |
1736 | unsigned long nr_scanned; | |
1737 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1738 | .pgdat = pgdat, |
0608f43d AM |
1739 | }; |
1740 | ||
3e32cb2e | 1741 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1742 | |
1743 | while (1) { | |
1744 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1745 | if (!victim) { | |
1746 | loop++; | |
1747 | if (loop >= 2) { | |
1748 | /* | |
1749 | * If we have not been able to reclaim | |
1750 | * anything, it might because there are | |
1751 | * no reclaimable pages under this hierarchy | |
1752 | */ | |
1753 | if (!total) | |
1754 | break; | |
1755 | /* | |
1756 | * We want to do more targeted reclaim. | |
1757 | * excess >> 2 is not to excessive so as to | |
1758 | * reclaim too much, nor too less that we keep | |
1759 | * coming back to reclaim from this cgroup | |
1760 | */ | |
1761 | if (total >= (excess >> 2) || | |
1762 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1763 | break; | |
1764 | } | |
1765 | continue; | |
1766 | } | |
a9dd0a83 | 1767 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1768 | pgdat, &nr_scanned); |
0608f43d | 1769 | *total_scanned += nr_scanned; |
3e32cb2e | 1770 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1771 | break; |
6d61ef40 | 1772 | } |
0608f43d AM |
1773 | mem_cgroup_iter_break(root_memcg, victim); |
1774 | return total; | |
6d61ef40 BS |
1775 | } |
1776 | ||
0056f4e6 JW |
1777 | #ifdef CONFIG_LOCKDEP |
1778 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1779 | .name = "memcg_oom_lock", | |
1780 | }; | |
1781 | #endif | |
1782 | ||
fb2a6fc5 JW |
1783 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1784 | ||
867578cb KH |
1785 | /* |
1786 | * Check OOM-Killer is already running under our hierarchy. | |
1787 | * If someone is running, return false. | |
1788 | */ | |
fb2a6fc5 | 1789 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1790 | { |
79dfdacc | 1791 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1792 | |
fb2a6fc5 JW |
1793 | spin_lock(&memcg_oom_lock); |
1794 | ||
9f3a0d09 | 1795 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1796 | if (iter->oom_lock) { |
79dfdacc MH |
1797 | /* |
1798 | * this subtree of our hierarchy is already locked | |
1799 | * so we cannot give a lock. | |
1800 | */ | |
79dfdacc | 1801 | failed = iter; |
9f3a0d09 JW |
1802 | mem_cgroup_iter_break(memcg, iter); |
1803 | break; | |
23751be0 JW |
1804 | } else |
1805 | iter->oom_lock = true; | |
7d74b06f | 1806 | } |
867578cb | 1807 | |
fb2a6fc5 JW |
1808 | if (failed) { |
1809 | /* | |
1810 | * OK, we failed to lock the whole subtree so we have | |
1811 | * to clean up what we set up to the failing subtree | |
1812 | */ | |
1813 | for_each_mem_cgroup_tree(iter, memcg) { | |
1814 | if (iter == failed) { | |
1815 | mem_cgroup_iter_break(memcg, iter); | |
1816 | break; | |
1817 | } | |
1818 | iter->oom_lock = false; | |
79dfdacc | 1819 | } |
0056f4e6 JW |
1820 | } else |
1821 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1822 | |
1823 | spin_unlock(&memcg_oom_lock); | |
1824 | ||
1825 | return !failed; | |
a636b327 | 1826 | } |
0b7f569e | 1827 | |
fb2a6fc5 | 1828 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1829 | { |
7d74b06f KH |
1830 | struct mem_cgroup *iter; |
1831 | ||
fb2a6fc5 | 1832 | spin_lock(&memcg_oom_lock); |
5facae4f | 1833 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1834 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1835 | iter->oom_lock = false; |
fb2a6fc5 | 1836 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1837 | } |
1838 | ||
c0ff4b85 | 1839 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1840 | { |
1841 | struct mem_cgroup *iter; | |
1842 | ||
c2b42d3c | 1843 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1844 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1845 | iter->under_oom++; |
1846 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1847 | } |
1848 | ||
c0ff4b85 | 1849 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1850 | { |
1851 | struct mem_cgroup *iter; | |
1852 | ||
867578cb | 1853 | /* |
f0953a1b | 1854 | * Be careful about under_oom underflows because a child memcg |
7a52d4d8 | 1855 | * could have been added after mem_cgroup_mark_under_oom. |
867578cb | 1856 | */ |
c2b42d3c | 1857 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1858 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1859 | if (iter->under_oom > 0) |
1860 | iter->under_oom--; | |
1861 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1862 | } |
1863 | ||
867578cb KH |
1864 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1865 | ||
dc98df5a | 1866 | struct oom_wait_info { |
d79154bb | 1867 | struct mem_cgroup *memcg; |
ac6424b9 | 1868 | wait_queue_entry_t wait; |
dc98df5a KH |
1869 | }; |
1870 | ||
ac6424b9 | 1871 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1872 | unsigned mode, int sync, void *arg) |
1873 | { | |
d79154bb HD |
1874 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1875 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1876 | struct oom_wait_info *oom_wait_info; |
1877 | ||
1878 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1879 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1880 | |
2314b42d JW |
1881 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1882 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1883 | return 0; |
dc98df5a KH |
1884 | return autoremove_wake_function(wait, mode, sync, arg); |
1885 | } | |
1886 | ||
c0ff4b85 | 1887 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1888 | { |
c2b42d3c TH |
1889 | /* |
1890 | * For the following lockless ->under_oom test, the only required | |
1891 | * guarantee is that it must see the state asserted by an OOM when | |
1892 | * this function is called as a result of userland actions | |
1893 | * triggered by the notification of the OOM. This is trivially | |
1894 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1895 | * triggering notification. | |
1896 | */ | |
1897 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1898 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1899 | } |
1900 | ||
becdf89d SB |
1901 | /* |
1902 | * Returns true if successfully killed one or more processes. Though in some | |
1903 | * corner cases it can return true even without killing any process. | |
1904 | */ | |
1905 | static bool mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1906 | { |
becdf89d | 1907 | bool locked, ret; |
7056d3a3 | 1908 | |
29ef680a | 1909 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
becdf89d | 1910 | return false; |
29ef680a | 1911 | |
7a1adfdd RG |
1912 | memcg_memory_event(memcg, MEMCG_OOM); |
1913 | ||
867578cb | 1914 | /* |
49426420 JW |
1915 | * We are in the middle of the charge context here, so we |
1916 | * don't want to block when potentially sitting on a callstack | |
1917 | * that holds all kinds of filesystem and mm locks. | |
1918 | * | |
29ef680a MH |
1919 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1920 | * handling until the charge can succeed; remember the context and put | |
1921 | * the task to sleep at the end of the page fault when all locks are | |
1922 | * released. | |
49426420 | 1923 | * |
29ef680a MH |
1924 | * On the other hand, in-kernel OOM killer allows for an async victim |
1925 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1926 | * relying on the oom victim to make a forward progress and we can | |
1927 | * invoke the oom killer here. | |
1928 | * | |
1929 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1930 | * victim and then we have to bail out from the charge path. | |
867578cb | 1931 | */ |
17c56de6 | 1932 | if (READ_ONCE(memcg->oom_kill_disable)) { |
becdf89d SB |
1933 | if (current->in_user_fault) { |
1934 | css_get(&memcg->css); | |
1935 | current->memcg_in_oom = memcg; | |
1936 | current->memcg_oom_gfp_mask = mask; | |
1937 | current->memcg_oom_order = order; | |
1938 | } | |
1939 | return false; | |
29ef680a MH |
1940 | } |
1941 | ||
7056d3a3 MH |
1942 | mem_cgroup_mark_under_oom(memcg); |
1943 | ||
1944 | locked = mem_cgroup_oom_trylock(memcg); | |
1945 | ||
1946 | if (locked) | |
1947 | mem_cgroup_oom_notify(memcg); | |
1948 | ||
1949 | mem_cgroup_unmark_under_oom(memcg); | |
becdf89d | 1950 | ret = mem_cgroup_out_of_memory(memcg, mask, order); |
7056d3a3 MH |
1951 | |
1952 | if (locked) | |
1953 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1954 | |
7056d3a3 | 1955 | return ret; |
3812c8c8 JW |
1956 | } |
1957 | ||
1958 | /** | |
1959 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1960 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1961 | * |
49426420 JW |
1962 | * This has to be called at the end of a page fault if the memcg OOM |
1963 | * handler was enabled. | |
3812c8c8 | 1964 | * |
49426420 | 1965 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1966 | * sleep on a waitqueue until the userspace task resolves the |
1967 | * situation. Sleeping directly in the charge context with all kinds | |
1968 | * of locks held is not a good idea, instead we remember an OOM state | |
1969 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1970 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1971 | * |
1972 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1973 | * completed, %false otherwise. |
3812c8c8 | 1974 | */ |
49426420 | 1975 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1976 | { |
626ebc41 | 1977 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1978 | struct oom_wait_info owait; |
49426420 | 1979 | bool locked; |
3812c8c8 JW |
1980 | |
1981 | /* OOM is global, do not handle */ | |
3812c8c8 | 1982 | if (!memcg) |
49426420 | 1983 | return false; |
3812c8c8 | 1984 | |
7c5f64f8 | 1985 | if (!handle) |
49426420 | 1986 | goto cleanup; |
3812c8c8 JW |
1987 | |
1988 | owait.memcg = memcg; | |
1989 | owait.wait.flags = 0; | |
1990 | owait.wait.func = memcg_oom_wake_function; | |
1991 | owait.wait.private = current; | |
2055da97 | 1992 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1993 | |
3812c8c8 | 1994 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1995 | mem_cgroup_mark_under_oom(memcg); |
1996 | ||
1997 | locked = mem_cgroup_oom_trylock(memcg); | |
1998 | ||
1999 | if (locked) | |
2000 | mem_cgroup_oom_notify(memcg); | |
2001 | ||
17c56de6 | 2002 | if (locked && !READ_ONCE(memcg->oom_kill_disable)) { |
49426420 JW |
2003 | mem_cgroup_unmark_under_oom(memcg); |
2004 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
2005 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
2006 | current->memcg_oom_order); | |
49426420 | 2007 | } else { |
3812c8c8 | 2008 | schedule(); |
49426420 JW |
2009 | mem_cgroup_unmark_under_oom(memcg); |
2010 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2011 | } | |
2012 | ||
2013 | if (locked) { | |
fb2a6fc5 JW |
2014 | mem_cgroup_oom_unlock(memcg); |
2015 | /* | |
2016 | * There is no guarantee that an OOM-lock contender | |
2017 | * sees the wakeups triggered by the OOM kill | |
f0953a1b | 2018 | * uncharges. Wake any sleepers explicitly. |
fb2a6fc5 JW |
2019 | */ |
2020 | memcg_oom_recover(memcg); | |
2021 | } | |
49426420 | 2022 | cleanup: |
626ebc41 | 2023 | current->memcg_in_oom = NULL; |
3812c8c8 | 2024 | css_put(&memcg->css); |
867578cb | 2025 | return true; |
0b7f569e KH |
2026 | } |
2027 | ||
3d8b38eb RG |
2028 | /** |
2029 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2030 | * @victim: task to be killed by the OOM killer | |
2031 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2032 | * | |
2033 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2034 | * by killing all belonging OOM-killable tasks. | |
2035 | * | |
2036 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2037 | */ | |
2038 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2039 | struct mem_cgroup *oom_domain) | |
2040 | { | |
2041 | struct mem_cgroup *oom_group = NULL; | |
2042 | struct mem_cgroup *memcg; | |
2043 | ||
2044 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2045 | return NULL; | |
2046 | ||
2047 | if (!oom_domain) | |
2048 | oom_domain = root_mem_cgroup; | |
2049 | ||
2050 | rcu_read_lock(); | |
2051 | ||
2052 | memcg = mem_cgroup_from_task(victim); | |
7848ed62 | 2053 | if (mem_cgroup_is_root(memcg)) |
3d8b38eb RG |
2054 | goto out; |
2055 | ||
48fe267c RG |
2056 | /* |
2057 | * If the victim task has been asynchronously moved to a different | |
2058 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2059 | * In this case it's better to ignore memory.group.oom. | |
2060 | */ | |
2061 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2062 | goto out; | |
2063 | ||
3d8b38eb RG |
2064 | /* |
2065 | * Traverse the memory cgroup hierarchy from the victim task's | |
2066 | * cgroup up to the OOMing cgroup (or root) to find the | |
2067 | * highest-level memory cgroup with oom.group set. | |
2068 | */ | |
2069 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
eaf7b66b | 2070 | if (READ_ONCE(memcg->oom_group)) |
3d8b38eb RG |
2071 | oom_group = memcg; |
2072 | ||
2073 | if (memcg == oom_domain) | |
2074 | break; | |
2075 | } | |
2076 | ||
2077 | if (oom_group) | |
2078 | css_get(&oom_group->css); | |
2079 | out: | |
2080 | rcu_read_unlock(); | |
2081 | ||
2082 | return oom_group; | |
2083 | } | |
2084 | ||
2085 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2086 | { | |
2087 | pr_info("Tasks in "); | |
2088 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2089 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2090 | } | |
2091 | ||
d7365e78 | 2092 | /** |
f70ad448 MWO |
2093 | * folio_memcg_lock - Bind a folio to its memcg. |
2094 | * @folio: The folio. | |
32047e2a | 2095 | * |
f70ad448 | 2096 | * This function prevents unlocked LRU folios from being moved to |
739f79fc JW |
2097 | * another cgroup. |
2098 | * | |
f70ad448 MWO |
2099 | * It ensures lifetime of the bound memcg. The caller is responsible |
2100 | * for the lifetime of the folio. | |
d69b042f | 2101 | */ |
f70ad448 | 2102 | void folio_memcg_lock(struct folio *folio) |
89c06bd5 KH |
2103 | { |
2104 | struct mem_cgroup *memcg; | |
6de22619 | 2105 | unsigned long flags; |
89c06bd5 | 2106 | |
6de22619 JW |
2107 | /* |
2108 | * The RCU lock is held throughout the transaction. The fast | |
2109 | * path can get away without acquiring the memcg->move_lock | |
2110 | * because page moving starts with an RCU grace period. | |
739f79fc | 2111 | */ |
d7365e78 JW |
2112 | rcu_read_lock(); |
2113 | ||
2114 | if (mem_cgroup_disabled()) | |
1c824a68 | 2115 | return; |
89c06bd5 | 2116 | again: |
f70ad448 | 2117 | memcg = folio_memcg(folio); |
29833315 | 2118 | if (unlikely(!memcg)) |
1c824a68 | 2119 | return; |
d7365e78 | 2120 | |
20ad50d6 AS |
2121 | #ifdef CONFIG_PROVE_LOCKING |
2122 | local_irq_save(flags); | |
2123 | might_lock(&memcg->move_lock); | |
2124 | local_irq_restore(flags); | |
2125 | #endif | |
2126 | ||
bdcbb659 | 2127 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 2128 | return; |
89c06bd5 | 2129 | |
6de22619 | 2130 | spin_lock_irqsave(&memcg->move_lock, flags); |
f70ad448 | 2131 | if (memcg != folio_memcg(folio)) { |
6de22619 | 2132 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2133 | goto again; |
2134 | } | |
6de22619 JW |
2135 | |
2136 | /* | |
1c824a68 JW |
2137 | * When charge migration first begins, we can have multiple |
2138 | * critical sections holding the fast-path RCU lock and one | |
2139 | * holding the slowpath move_lock. Track the task who has the | |
2140 | * move_lock for unlock_page_memcg(). | |
6de22619 JW |
2141 | */ |
2142 | memcg->move_lock_task = current; | |
2143 | memcg->move_lock_flags = flags; | |
89c06bd5 | 2144 | } |
f70ad448 MWO |
2145 | |
2146 | void lock_page_memcg(struct page *page) | |
2147 | { | |
2148 | folio_memcg_lock(page_folio(page)); | |
2149 | } | |
89c06bd5 | 2150 | |
f70ad448 | 2151 | static void __folio_memcg_unlock(struct mem_cgroup *memcg) |
89c06bd5 | 2152 | { |
6de22619 JW |
2153 | if (memcg && memcg->move_lock_task == current) { |
2154 | unsigned long flags = memcg->move_lock_flags; | |
2155 | ||
2156 | memcg->move_lock_task = NULL; | |
2157 | memcg->move_lock_flags = 0; | |
2158 | ||
2159 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2160 | } | |
89c06bd5 | 2161 | |
d7365e78 | 2162 | rcu_read_unlock(); |
89c06bd5 | 2163 | } |
739f79fc JW |
2164 | |
2165 | /** | |
f70ad448 MWO |
2166 | * folio_memcg_unlock - Release the binding between a folio and its memcg. |
2167 | * @folio: The folio. | |
2168 | * | |
2169 | * This releases the binding created by folio_memcg_lock(). This does | |
2170 | * not change the accounting of this folio to its memcg, but it does | |
2171 | * permit others to change it. | |
739f79fc | 2172 | */ |
f70ad448 | 2173 | void folio_memcg_unlock(struct folio *folio) |
739f79fc | 2174 | { |
f70ad448 MWO |
2175 | __folio_memcg_unlock(folio_memcg(folio)); |
2176 | } | |
9da7b521 | 2177 | |
f70ad448 MWO |
2178 | void unlock_page_memcg(struct page *page) |
2179 | { | |
2180 | folio_memcg_unlock(page_folio(page)); | |
739f79fc | 2181 | } |
89c06bd5 | 2182 | |
fead2b86 | 2183 | struct memcg_stock_pcp { |
56751146 | 2184 | local_lock_t stock_lock; |
fead2b86 MH |
2185 | struct mem_cgroup *cached; /* this never be root cgroup */ |
2186 | unsigned int nr_pages; | |
2187 | ||
bf4f0599 RG |
2188 | #ifdef CONFIG_MEMCG_KMEM |
2189 | struct obj_cgroup *cached_objcg; | |
68ac5b3c | 2190 | struct pglist_data *cached_pgdat; |
bf4f0599 | 2191 | unsigned int nr_bytes; |
68ac5b3c WL |
2192 | int nr_slab_reclaimable_b; |
2193 | int nr_slab_unreclaimable_b; | |
bf4f0599 RG |
2194 | #endif |
2195 | ||
cdec2e42 | 2196 | struct work_struct work; |
26fe6168 | 2197 | unsigned long flags; |
a0db00fc | 2198 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 | 2199 | }; |
56751146 SAS |
2200 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = { |
2201 | .stock_lock = INIT_LOCAL_LOCK(stock_lock), | |
2202 | }; | |
9f50fad6 | 2203 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2204 | |
bf4f0599 | 2205 | #ifdef CONFIG_MEMCG_KMEM |
56751146 | 2206 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock); |
bf4f0599 RG |
2207 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
2208 | struct mem_cgroup *root_memcg); | |
a8c49af3 | 2209 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages); |
bf4f0599 RG |
2210 | |
2211 | #else | |
56751146 | 2212 | static inline struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 2213 | { |
56751146 | 2214 | return NULL; |
bf4f0599 RG |
2215 | } |
2216 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2217 | struct mem_cgroup *root_memcg) | |
2218 | { | |
2219 | return false; | |
2220 | } | |
a8c49af3 YA |
2221 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
2222 | { | |
2223 | } | |
bf4f0599 RG |
2224 | #endif |
2225 | ||
a0956d54 SS |
2226 | /** |
2227 | * consume_stock: Try to consume stocked charge on this cpu. | |
2228 | * @memcg: memcg to consume from. | |
2229 | * @nr_pages: how many pages to charge. | |
2230 | * | |
2231 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2232 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2233 | * service an allocation will refill the stock. | |
2234 | * | |
2235 | * returns true if successful, false otherwise. | |
cdec2e42 | 2236 | */ |
a0956d54 | 2237 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2238 | { |
2239 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2240 | unsigned long flags; |
3e32cb2e | 2241 | bool ret = false; |
cdec2e42 | 2242 | |
a983b5eb | 2243 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2244 | return ret; |
a0956d54 | 2245 | |
56751146 | 2246 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2247 | |
2248 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2249 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2250 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2251 | ret = true; |
2252 | } | |
db2ba40c | 2253 | |
56751146 | 2254 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
db2ba40c | 2255 | |
cdec2e42 KH |
2256 | return ret; |
2257 | } | |
2258 | ||
2259 | /* | |
3e32cb2e | 2260 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2261 | */ |
2262 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2263 | { | |
2264 | struct mem_cgroup *old = stock->cached; | |
2265 | ||
1a3e1f40 JW |
2266 | if (!old) |
2267 | return; | |
2268 | ||
11c9ea4e | 2269 | if (stock->nr_pages) { |
3e32cb2e | 2270 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2271 | if (do_memsw_account()) |
3e32cb2e | 2272 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2273 | stock->nr_pages = 0; |
cdec2e42 | 2274 | } |
1a3e1f40 JW |
2275 | |
2276 | css_put(&old->css); | |
cdec2e42 | 2277 | stock->cached = NULL; |
cdec2e42 KH |
2278 | } |
2279 | ||
cdec2e42 KH |
2280 | static void drain_local_stock(struct work_struct *dummy) |
2281 | { | |
db2ba40c | 2282 | struct memcg_stock_pcp *stock; |
56751146 | 2283 | struct obj_cgroup *old = NULL; |
db2ba40c JW |
2284 | unsigned long flags; |
2285 | ||
72f0184c | 2286 | /* |
5c49cf9a MH |
2287 | * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs. |
2288 | * drain_stock races is that we always operate on local CPU stock | |
2289 | * here with IRQ disabled | |
72f0184c | 2290 | */ |
56751146 | 2291 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2292 | |
2293 | stock = this_cpu_ptr(&memcg_stock); | |
56751146 | 2294 | old = drain_obj_stock(stock); |
cdec2e42 | 2295 | drain_stock(stock); |
26fe6168 | 2296 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c | 2297 | |
56751146 SAS |
2298 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
2299 | if (old) | |
2300 | obj_cgroup_put(old); | |
cdec2e42 KH |
2301 | } |
2302 | ||
2303 | /* | |
3e32cb2e | 2304 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2305 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2306 | */ |
af9a3b69 | 2307 | static void __refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2308 | { |
db2ba40c | 2309 | struct memcg_stock_pcp *stock; |
cdec2e42 | 2310 | |
db2ba40c | 2311 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2312 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2313 | drain_stock(stock); |
1a3e1f40 | 2314 | css_get(&memcg->css); |
c0ff4b85 | 2315 | stock->cached = memcg; |
cdec2e42 | 2316 | } |
11c9ea4e | 2317 | stock->nr_pages += nr_pages; |
db2ba40c | 2318 | |
a983b5eb | 2319 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 | 2320 | drain_stock(stock); |
af9a3b69 JW |
2321 | } |
2322 | ||
2323 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) | |
2324 | { | |
2325 | unsigned long flags; | |
475d0487 | 2326 | |
56751146 | 2327 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
af9a3b69 | 2328 | __refill_stock(memcg, nr_pages); |
56751146 | 2329 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
cdec2e42 KH |
2330 | } |
2331 | ||
2332 | /* | |
c0ff4b85 | 2333 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2334 | * of the hierarchy under it. |
cdec2e42 | 2335 | */ |
6d3d6aa2 | 2336 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2337 | { |
26fe6168 | 2338 | int cpu, curcpu; |
d38144b7 | 2339 | |
6d3d6aa2 JW |
2340 | /* If someone's already draining, avoid adding running more workers. */ |
2341 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2342 | return; | |
72f0184c MH |
2343 | /* |
2344 | * Notify other cpus that system-wide "drain" is running | |
2345 | * We do not care about races with the cpu hotplug because cpu down | |
2346 | * as well as workers from this path always operate on the local | |
2347 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2348 | */ | |
0790ed62 SAS |
2349 | migrate_disable(); |
2350 | curcpu = smp_processor_id(); | |
cdec2e42 KH |
2351 | for_each_online_cpu(cpu) { |
2352 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2353 | struct mem_cgroup *memcg; |
e1a366be | 2354 | bool flush = false; |
26fe6168 | 2355 | |
e1a366be | 2356 | rcu_read_lock(); |
c0ff4b85 | 2357 | memcg = stock->cached; |
e1a366be RG |
2358 | if (memcg && stock->nr_pages && |
2359 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2360 | flush = true; | |
27fb0956 | 2361 | else if (obj_stock_flush_required(stock, root_memcg)) |
bf4f0599 | 2362 | flush = true; |
e1a366be RG |
2363 | rcu_read_unlock(); |
2364 | ||
2365 | if (flush && | |
2366 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2367 | if (cpu == curcpu) |
2368 | drain_local_stock(&stock->work); | |
2369 | else | |
2370 | schedule_work_on(cpu, &stock->work); | |
2371 | } | |
cdec2e42 | 2372 | } |
0790ed62 | 2373 | migrate_enable(); |
9f50fad6 | 2374 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2375 | } |
2376 | ||
2cd21c89 JW |
2377 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
2378 | { | |
2379 | struct memcg_stock_pcp *stock; | |
a3d4c05a | 2380 | |
2cd21c89 JW |
2381 | stock = &per_cpu(memcg_stock, cpu); |
2382 | drain_stock(stock); | |
a3d4c05a | 2383 | |
308167fc | 2384 | return 0; |
cdec2e42 KH |
2385 | } |
2386 | ||
b3ff9291 CD |
2387 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2388 | unsigned int nr_pages, | |
2389 | gfp_t gfp_mask) | |
f7e1cb6e | 2390 | { |
b3ff9291 CD |
2391 | unsigned long nr_reclaimed = 0; |
2392 | ||
f7e1cb6e | 2393 | do { |
e22c6ed9 JW |
2394 | unsigned long pflags; |
2395 | ||
d1663a90 JK |
2396 | if (page_counter_read(&memcg->memory) <= |
2397 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2398 | continue; |
e22c6ed9 | 2399 | |
e27be240 | 2400 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2401 | |
2402 | psi_memstall_enter(&pflags); | |
b3ff9291 | 2403 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
73b73bac | 2404 | gfp_mask, |
55ab834a | 2405 | MEMCG_RECLAIM_MAY_SWAP); |
e22c6ed9 | 2406 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2407 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2408 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2409 | |
2410 | return nr_reclaimed; | |
f7e1cb6e JW |
2411 | } |
2412 | ||
2413 | static void high_work_func(struct work_struct *work) | |
2414 | { | |
2415 | struct mem_cgroup *memcg; | |
2416 | ||
2417 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2418 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2419 | } |
2420 | ||
0e4b01df CD |
2421 | /* |
2422 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2423 | * enough to still cause a significant slowdown in most cases, while still | |
2424 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2425 | */ | |
2426 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2427 | ||
2428 | /* | |
2429 | * When calculating the delay, we use these either side of the exponentiation to | |
2430 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2431 | * below. | |
2432 | * | |
2433 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2434 | * overage ratio to a delay. | |
ac5ddd0f | 2435 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2436 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2437 | * to produce a reasonable delay curve. | |
2438 | * | |
2439 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2440 | * reasonable delay curve compared to precision-adjusted overage, not | |
2441 | * penalising heavily at first, but still making sure that growth beyond the | |
2442 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2443 | * example, with a high of 100 megabytes: | |
2444 | * | |
2445 | * +-------+------------------------+ | |
2446 | * | usage | time to allocate in ms | | |
2447 | * +-------+------------------------+ | |
2448 | * | 100M | 0 | | |
2449 | * | 101M | 6 | | |
2450 | * | 102M | 25 | | |
2451 | * | 103M | 57 | | |
2452 | * | 104M | 102 | | |
2453 | * | 105M | 159 | | |
2454 | * | 106M | 230 | | |
2455 | * | 107M | 313 | | |
2456 | * | 108M | 409 | | |
2457 | * | 109M | 518 | | |
2458 | * | 110M | 639 | | |
2459 | * | 111M | 774 | | |
2460 | * | 112M | 921 | | |
2461 | * | 113M | 1081 | | |
2462 | * | 114M | 1254 | | |
2463 | * | 115M | 1439 | | |
2464 | * | 116M | 1638 | | |
2465 | * | 117M | 1849 | | |
2466 | * | 118M | 2000 | | |
2467 | * | 119M | 2000 | | |
2468 | * | 120M | 2000 | | |
2469 | * +-------+------------------------+ | |
2470 | */ | |
2471 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2472 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2473 | ||
8a5dbc65 | 2474 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2475 | { |
8a5dbc65 | 2476 | u64 overage; |
b23afb93 | 2477 | |
8a5dbc65 JK |
2478 | if (usage <= high) |
2479 | return 0; | |
e26733e0 | 2480 | |
8a5dbc65 JK |
2481 | /* |
2482 | * Prevent division by 0 in overage calculation by acting as if | |
2483 | * it was a threshold of 1 page | |
2484 | */ | |
2485 | high = max(high, 1UL); | |
9b8b1754 | 2486 | |
8a5dbc65 JK |
2487 | overage = usage - high; |
2488 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2489 | return div64_u64(overage, high); | |
2490 | } | |
e26733e0 | 2491 | |
8a5dbc65 JK |
2492 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2493 | { | |
2494 | u64 overage, max_overage = 0; | |
e26733e0 | 2495 | |
8a5dbc65 JK |
2496 | do { |
2497 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2498 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2499 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2500 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2501 | !mem_cgroup_is_root(memcg)); | |
2502 | ||
8a5dbc65 JK |
2503 | return max_overage; |
2504 | } | |
2505 | ||
4b82ab4f JK |
2506 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2507 | { | |
2508 | u64 overage, max_overage = 0; | |
2509 | ||
2510 | do { | |
2511 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2512 | READ_ONCE(memcg->swap.high)); | |
2513 | if (overage) | |
2514 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2515 | max_overage = max(overage, max_overage); | |
2516 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2517 | !mem_cgroup_is_root(memcg)); | |
2518 | ||
2519 | return max_overage; | |
2520 | } | |
2521 | ||
8a5dbc65 JK |
2522 | /* |
2523 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2524 | * is exceeding its memory.high by checking both it and its ancestors. | |
2525 | */ | |
2526 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2527 | unsigned int nr_pages, | |
2528 | u64 max_overage) | |
2529 | { | |
2530 | unsigned long penalty_jiffies; | |
2531 | ||
e26733e0 CD |
2532 | if (!max_overage) |
2533 | return 0; | |
0e4b01df CD |
2534 | |
2535 | /* | |
0e4b01df CD |
2536 | * We use overage compared to memory.high to calculate the number of |
2537 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2538 | * fairly lenient on small overages, and increasingly harsh when the | |
2539 | * memcg in question makes it clear that it has no intention of stopping | |
2540 | * its crazy behaviour, so we exponentially increase the delay based on | |
2541 | * overage amount. | |
2542 | */ | |
e26733e0 CD |
2543 | penalty_jiffies = max_overage * max_overage * HZ; |
2544 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2545 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2546 | |
2547 | /* | |
2548 | * Factor in the task's own contribution to the overage, such that four | |
2549 | * N-sized allocations are throttled approximately the same as one | |
2550 | * 4N-sized allocation. | |
2551 | * | |
2552 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2553 | * larger the current charge patch is than that. | |
2554 | */ | |
ff144e69 | 2555 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2556 | } |
2557 | ||
2558 | /* | |
2559 | * Scheduled by try_charge() to be executed from the userland return path | |
2560 | * and reclaims memory over the high limit. | |
2561 | */ | |
2562 | void mem_cgroup_handle_over_high(void) | |
2563 | { | |
2564 | unsigned long penalty_jiffies; | |
2565 | unsigned long pflags; | |
b3ff9291 | 2566 | unsigned long nr_reclaimed; |
e26733e0 | 2567 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2568 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2569 | struct mem_cgroup *memcg; |
b3ff9291 | 2570 | bool in_retry = false; |
e26733e0 CD |
2571 | |
2572 | if (likely(!nr_pages)) | |
2573 | return; | |
2574 | ||
2575 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2576 | current->memcg_nr_pages_over_high = 0; |
2577 | ||
b3ff9291 CD |
2578 | retry_reclaim: |
2579 | /* | |
2580 | * The allocating task should reclaim at least the batch size, but for | |
2581 | * subsequent retries we only want to do what's necessary to prevent oom | |
2582 | * or breaching resource isolation. | |
2583 | * | |
2584 | * This is distinct from memory.max or page allocator behaviour because | |
2585 | * memory.high is currently batched, whereas memory.max and the page | |
2586 | * allocator run every time an allocation is made. | |
2587 | */ | |
2588 | nr_reclaimed = reclaim_high(memcg, | |
2589 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2590 | GFP_KERNEL); | |
2591 | ||
e26733e0 CD |
2592 | /* |
2593 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2594 | * allocators proactively to slow down excessive growth. | |
2595 | */ | |
8a5dbc65 JK |
2596 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2597 | mem_find_max_overage(memcg)); | |
0e4b01df | 2598 | |
4b82ab4f JK |
2599 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2600 | swap_find_max_overage(memcg)); | |
2601 | ||
ff144e69 JK |
2602 | /* |
2603 | * Clamp the max delay per usermode return so as to still keep the | |
2604 | * application moving forwards and also permit diagnostics, albeit | |
2605 | * extremely slowly. | |
2606 | */ | |
2607 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2608 | ||
0e4b01df CD |
2609 | /* |
2610 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2611 | * that it's not even worth doing, in an attempt to be nice to those who | |
2612 | * go only a small amount over their memory.high value and maybe haven't | |
2613 | * been aggressively reclaimed enough yet. | |
2614 | */ | |
2615 | if (penalty_jiffies <= HZ / 100) | |
2616 | goto out; | |
2617 | ||
b3ff9291 CD |
2618 | /* |
2619 | * If reclaim is making forward progress but we're still over | |
2620 | * memory.high, we want to encourage that rather than doing allocator | |
2621 | * throttling. | |
2622 | */ | |
2623 | if (nr_reclaimed || nr_retries--) { | |
2624 | in_retry = true; | |
2625 | goto retry_reclaim; | |
2626 | } | |
2627 | ||
0e4b01df CD |
2628 | /* |
2629 | * If we exit early, we're guaranteed to die (since | |
2630 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2631 | * need to account for any ill-begotten jiffies to pay them off later. | |
2632 | */ | |
2633 | psi_memstall_enter(&pflags); | |
2634 | schedule_timeout_killable(penalty_jiffies); | |
2635 | psi_memstall_leave(&pflags); | |
2636 | ||
2637 | out: | |
2638 | css_put(&memcg->css); | |
b23afb93 TH |
2639 | } |
2640 | ||
c5c8b16b MS |
2641 | static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2642 | unsigned int nr_pages) | |
8a9f3ccd | 2643 | { |
a983b5eb | 2644 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2645 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2646 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2647 | struct page_counter *counter; |
6539cc05 | 2648 | unsigned long nr_reclaimed; |
a4ebf1b6 | 2649 | bool passed_oom = false; |
73b73bac | 2650 | unsigned int reclaim_options = MEMCG_RECLAIM_MAY_SWAP; |
b70a2a21 | 2651 | bool drained = false; |
d6e103a7 | 2652 | bool raised_max_event = false; |
e22c6ed9 | 2653 | unsigned long pflags; |
a636b327 | 2654 | |
6539cc05 | 2655 | retry: |
b6b6cc72 | 2656 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2657 | return 0; |
8a9f3ccd | 2658 | |
7941d214 | 2659 | if (!do_memsw_account() || |
6071ca52 JW |
2660 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2661 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2662 | goto done_restock; |
7941d214 | 2663 | if (do_memsw_account()) |
3e32cb2e JW |
2664 | page_counter_uncharge(&memcg->memsw, batch); |
2665 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2666 | } else { |
3e32cb2e | 2667 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
73b73bac | 2668 | reclaim_options &= ~MEMCG_RECLAIM_MAY_SWAP; |
3fbe7244 | 2669 | } |
7a81b88c | 2670 | |
6539cc05 JW |
2671 | if (batch > nr_pages) { |
2672 | batch = nr_pages; | |
2673 | goto retry; | |
2674 | } | |
6d61ef40 | 2675 | |
89a28483 JW |
2676 | /* |
2677 | * Prevent unbounded recursion when reclaim operations need to | |
2678 | * allocate memory. This might exceed the limits temporarily, | |
2679 | * but we prefer facilitating memory reclaim and getting back | |
2680 | * under the limit over triggering OOM kills in these cases. | |
2681 | */ | |
2682 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2683 | goto force; | |
2684 | ||
06b078fc JW |
2685 | if (unlikely(task_in_memcg_oom(current))) |
2686 | goto nomem; | |
2687 | ||
d0164adc | 2688 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2689 | goto nomem; |
4b534334 | 2690 | |
e27be240 | 2691 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
d6e103a7 | 2692 | raised_max_event = true; |
241994ed | 2693 | |
e22c6ed9 | 2694 | psi_memstall_enter(&pflags); |
b70a2a21 | 2695 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
55ab834a | 2696 | gfp_mask, reclaim_options); |
e22c6ed9 | 2697 | psi_memstall_leave(&pflags); |
6539cc05 | 2698 | |
61e02c74 | 2699 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2700 | goto retry; |
28c34c29 | 2701 | |
b70a2a21 | 2702 | if (!drained) { |
6d3d6aa2 | 2703 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2704 | drained = true; |
2705 | goto retry; | |
2706 | } | |
2707 | ||
28c34c29 JW |
2708 | if (gfp_mask & __GFP_NORETRY) |
2709 | goto nomem; | |
6539cc05 JW |
2710 | /* |
2711 | * Even though the limit is exceeded at this point, reclaim | |
2712 | * may have been able to free some pages. Retry the charge | |
2713 | * before killing the task. | |
2714 | * | |
2715 | * Only for regular pages, though: huge pages are rather | |
2716 | * unlikely to succeed so close to the limit, and we fall back | |
2717 | * to regular pages anyway in case of failure. | |
2718 | */ | |
61e02c74 | 2719 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2720 | goto retry; |
2721 | /* | |
2722 | * At task move, charge accounts can be doubly counted. So, it's | |
2723 | * better to wait until the end of task_move if something is going on. | |
2724 | */ | |
2725 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2726 | goto retry; | |
2727 | ||
9b130619 JW |
2728 | if (nr_retries--) |
2729 | goto retry; | |
2730 | ||
38d38493 | 2731 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2732 | goto nomem; |
2733 | ||
a4ebf1b6 VA |
2734 | /* Avoid endless loop for tasks bypassed by the oom killer */ |
2735 | if (passed_oom && task_is_dying()) | |
2736 | goto nomem; | |
6539cc05 | 2737 | |
29ef680a MH |
2738 | /* |
2739 | * keep retrying as long as the memcg oom killer is able to make | |
2740 | * a forward progress or bypass the charge if the oom killer | |
2741 | * couldn't make any progress. | |
2742 | */ | |
becdf89d SB |
2743 | if (mem_cgroup_oom(mem_over_limit, gfp_mask, |
2744 | get_order(nr_pages * PAGE_SIZE))) { | |
a4ebf1b6 | 2745 | passed_oom = true; |
d977aa93 | 2746 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a | 2747 | goto retry; |
29ef680a | 2748 | } |
7a81b88c | 2749 | nomem: |
1461e8c2 SB |
2750 | /* |
2751 | * Memcg doesn't have a dedicated reserve for atomic | |
2752 | * allocations. But like the global atomic pool, we need to | |
2753 | * put the burden of reclaim on regular allocation requests | |
2754 | * and let these go through as privileged allocations. | |
2755 | */ | |
2756 | if (!(gfp_mask & (__GFP_NOFAIL | __GFP_HIGH))) | |
3168ecbe | 2757 | return -ENOMEM; |
10d53c74 | 2758 | force: |
d6e103a7 RG |
2759 | /* |
2760 | * If the allocation has to be enforced, don't forget to raise | |
2761 | * a MEMCG_MAX event. | |
2762 | */ | |
2763 | if (!raised_max_event) | |
2764 | memcg_memory_event(mem_over_limit, MEMCG_MAX); | |
2765 | ||
10d53c74 TH |
2766 | /* |
2767 | * The allocation either can't fail or will lead to more memory | |
2768 | * being freed very soon. Allow memory usage go over the limit | |
2769 | * temporarily by force charging it. | |
2770 | */ | |
2771 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2772 | if (do_memsw_account()) |
10d53c74 | 2773 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2774 | |
2775 | return 0; | |
6539cc05 JW |
2776 | |
2777 | done_restock: | |
2778 | if (batch > nr_pages) | |
2779 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2780 | |
241994ed | 2781 | /* |
b23afb93 TH |
2782 | * If the hierarchy is above the normal consumption range, schedule |
2783 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2784 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2785 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2786 | * not recorded as it most likely matches current's and won't | |
2787 | * change in the meantime. As high limit is checked again before | |
2788 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2789 | */ |
2790 | do { | |
4b82ab4f JK |
2791 | bool mem_high, swap_high; |
2792 | ||
2793 | mem_high = page_counter_read(&memcg->memory) > | |
2794 | READ_ONCE(memcg->memory.high); | |
2795 | swap_high = page_counter_read(&memcg->swap) > | |
2796 | READ_ONCE(memcg->swap.high); | |
2797 | ||
2798 | /* Don't bother a random interrupted task */ | |
086f694a | 2799 | if (!in_task()) { |
4b82ab4f | 2800 | if (mem_high) { |
f7e1cb6e JW |
2801 | schedule_work(&memcg->high_work); |
2802 | break; | |
2803 | } | |
4b82ab4f JK |
2804 | continue; |
2805 | } | |
2806 | ||
2807 | if (mem_high || swap_high) { | |
2808 | /* | |
2809 | * The allocating tasks in this cgroup will need to do | |
2810 | * reclaim or be throttled to prevent further growth | |
2811 | * of the memory or swap footprints. | |
2812 | * | |
2813 | * Target some best-effort fairness between the tasks, | |
2814 | * and distribute reclaim work and delay penalties | |
2815 | * based on how much each task is actually allocating. | |
2816 | */ | |
9516a18a | 2817 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2818 | set_notify_resume(current); |
2819 | break; | |
2820 | } | |
241994ed | 2821 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 | 2822 | |
c9afe31e SB |
2823 | if (current->memcg_nr_pages_over_high > MEMCG_CHARGE_BATCH && |
2824 | !(current->flags & PF_MEMALLOC) && | |
2825 | gfpflags_allow_blocking(gfp_mask)) { | |
2826 | mem_cgroup_handle_over_high(); | |
2827 | } | |
10d53c74 | 2828 | return 0; |
7a81b88c | 2829 | } |
8a9f3ccd | 2830 | |
c5c8b16b MS |
2831 | static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2832 | unsigned int nr_pages) | |
2833 | { | |
2834 | if (mem_cgroup_is_root(memcg)) | |
2835 | return 0; | |
2836 | ||
2837 | return try_charge_memcg(memcg, gfp_mask, nr_pages); | |
2838 | } | |
2839 | ||
58056f77 | 2840 | static inline void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2841 | { |
ce00a967 JW |
2842 | if (mem_cgroup_is_root(memcg)) |
2843 | return; | |
2844 | ||
3e32cb2e | 2845 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2846 | if (do_memsw_account()) |
3e32cb2e | 2847 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f KH |
2848 | } |
2849 | ||
118f2875 | 2850 | static void commit_charge(struct folio *folio, struct mem_cgroup *memcg) |
0a31bc97 | 2851 | { |
118f2875 | 2852 | VM_BUG_ON_FOLIO(folio_memcg(folio), folio); |
0a31bc97 | 2853 | /* |
a5eb011a | 2854 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2855 | * |
a0b5b414 JW |
2856 | * - the page lock |
2857 | * - LRU isolation | |
2858 | * - lock_page_memcg() | |
2859 | * - exclusive reference | |
018ee47f | 2860 | * - mem_cgroup_trylock_pages() |
0a31bc97 | 2861 | */ |
118f2875 | 2862 | folio->memcg_data = (unsigned long)memcg; |
7a81b88c | 2863 | } |
66e1707b | 2864 | |
84c07d11 | 2865 | #ifdef CONFIG_MEMCG_KMEM |
41eb5df1 WL |
2866 | /* |
2867 | * The allocated objcg pointers array is not accounted directly. | |
2868 | * Moreover, it should not come from DMA buffer and is not readily | |
2869 | * reclaimable. So those GFP bits should be masked off. | |
2870 | */ | |
2871 | #define OBJCGS_CLEAR_MASK (__GFP_DMA | __GFP_RECLAIMABLE | __GFP_ACCOUNT) | |
2872 | ||
a7ebf564 WL |
2873 | /* |
2874 | * mod_objcg_mlstate() may be called with irq enabled, so | |
2875 | * mod_memcg_lruvec_state() should be used. | |
2876 | */ | |
2877 | static inline void mod_objcg_mlstate(struct obj_cgroup *objcg, | |
2878 | struct pglist_data *pgdat, | |
2879 | enum node_stat_item idx, int nr) | |
2880 | { | |
2881 | struct mem_cgroup *memcg; | |
2882 | struct lruvec *lruvec; | |
2883 | ||
2884 | rcu_read_lock(); | |
2885 | memcg = obj_cgroup_memcg(objcg); | |
2886 | lruvec = mem_cgroup_lruvec(memcg, pgdat); | |
2887 | mod_memcg_lruvec_state(lruvec, idx, nr); | |
2888 | rcu_read_unlock(); | |
2889 | } | |
2890 | ||
4b5f8d9a VB |
2891 | int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s, |
2892 | gfp_t gfp, bool new_slab) | |
10befea9 | 2893 | { |
4b5f8d9a | 2894 | unsigned int objects = objs_per_slab(s, slab); |
2e9bd483 | 2895 | unsigned long memcg_data; |
10befea9 RG |
2896 | void *vec; |
2897 | ||
41eb5df1 | 2898 | gfp &= ~OBJCGS_CLEAR_MASK; |
10befea9 | 2899 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, |
4b5f8d9a | 2900 | slab_nid(slab)); |
10befea9 RG |
2901 | if (!vec) |
2902 | return -ENOMEM; | |
2903 | ||
2e9bd483 | 2904 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
4b5f8d9a | 2905 | if (new_slab) { |
2e9bd483 | 2906 | /* |
4b5f8d9a VB |
2907 | * If the slab is brand new and nobody can yet access its |
2908 | * memcg_data, no synchronization is required and memcg_data can | |
2909 | * be simply assigned. | |
2e9bd483 | 2910 | */ |
4b5f8d9a VB |
2911 | slab->memcg_data = memcg_data; |
2912 | } else if (cmpxchg(&slab->memcg_data, 0, memcg_data)) { | |
2e9bd483 | 2913 | /* |
4b5f8d9a VB |
2914 | * If the slab is already in use, somebody can allocate and |
2915 | * assign obj_cgroups in parallel. In this case the existing | |
2e9bd483 RG |
2916 | * objcg vector should be reused. |
2917 | */ | |
10befea9 | 2918 | kfree(vec); |
2e9bd483 RG |
2919 | return 0; |
2920 | } | |
10befea9 | 2921 | |
2e9bd483 | 2922 | kmemleak_not_leak(vec); |
10befea9 RG |
2923 | return 0; |
2924 | } | |
2925 | ||
fc4db90f RG |
2926 | static __always_inline |
2927 | struct mem_cgroup *mem_cgroup_from_obj_folio(struct folio *folio, void *p) | |
8380ce47 | 2928 | { |
8380ce47 | 2929 | /* |
9855609b RG |
2930 | * Slab objects are accounted individually, not per-page. |
2931 | * Memcg membership data for each individual object is saved in | |
4b5f8d9a | 2932 | * slab->memcg_data. |
8380ce47 | 2933 | */ |
4b5f8d9a VB |
2934 | if (folio_test_slab(folio)) { |
2935 | struct obj_cgroup **objcgs; | |
2936 | struct slab *slab; | |
9855609b RG |
2937 | unsigned int off; |
2938 | ||
4b5f8d9a VB |
2939 | slab = folio_slab(folio); |
2940 | objcgs = slab_objcgs(slab); | |
2941 | if (!objcgs) | |
2942 | return NULL; | |
2943 | ||
2944 | off = obj_to_index(slab->slab_cache, slab, p); | |
2945 | if (objcgs[off]) | |
2946 | return obj_cgroup_memcg(objcgs[off]); | |
10befea9 RG |
2947 | |
2948 | return NULL; | |
9855609b | 2949 | } |
8380ce47 | 2950 | |
bcfe06bf | 2951 | /* |
becacb04 | 2952 | * folio_memcg_check() is used here, because in theory we can encounter |
4b5f8d9a VB |
2953 | * a folio where the slab flag has been cleared already, but |
2954 | * slab->memcg_data has not been freed yet | |
becacb04 | 2955 | * folio_memcg_check() will guarantee that a proper memory |
bcfe06bf RG |
2956 | * cgroup pointer or NULL will be returned. |
2957 | */ | |
becacb04 | 2958 | return folio_memcg_check(folio); |
8380ce47 RG |
2959 | } |
2960 | ||
fc4db90f RG |
2961 | /* |
2962 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2963 | * | |
2964 | * A passed kernel object can be a slab object, vmalloc object or a generic | |
2965 | * kernel page, so different mechanisms for getting the memory cgroup pointer | |
2966 | * should be used. | |
2967 | * | |
2968 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
2969 | * can not know for sure how the kernel object is implemented. | |
2970 | * mem_cgroup_from_obj() can be safely used in such cases. | |
2971 | * | |
2972 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2973 | * cgroup_mutex, etc. | |
2974 | */ | |
2975 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2976 | { | |
2977 | struct folio *folio; | |
2978 | ||
2979 | if (mem_cgroup_disabled()) | |
2980 | return NULL; | |
2981 | ||
2982 | if (unlikely(is_vmalloc_addr(p))) | |
2983 | folio = page_folio(vmalloc_to_page(p)); | |
2984 | else | |
2985 | folio = virt_to_folio(p); | |
2986 | ||
2987 | return mem_cgroup_from_obj_folio(folio, p); | |
2988 | } | |
2989 | ||
2990 | /* | |
2991 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2992 | * Similar to mem_cgroup_from_obj(), but faster and not suitable for objects, | |
2993 | * allocated using vmalloc(). | |
2994 | * | |
2995 | * A passed kernel object must be a slab object or a generic kernel page. | |
2996 | * | |
2997 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2998 | * cgroup_mutex, etc. | |
2999 | */ | |
3000 | struct mem_cgroup *mem_cgroup_from_slab_obj(void *p) | |
3001 | { | |
3002 | if (mem_cgroup_disabled()) | |
3003 | return NULL; | |
3004 | ||
3005 | return mem_cgroup_from_obj_folio(virt_to_folio(p), p); | |
3006 | } | |
3007 | ||
f4840ccf JW |
3008 | static struct obj_cgroup *__get_obj_cgroup_from_memcg(struct mem_cgroup *memcg) |
3009 | { | |
3010 | struct obj_cgroup *objcg = NULL; | |
3011 | ||
7848ed62 | 3012 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
f4840ccf JW |
3013 | objcg = rcu_dereference(memcg->objcg); |
3014 | if (objcg && obj_cgroup_tryget(objcg)) | |
3015 | break; | |
3016 | objcg = NULL; | |
3017 | } | |
3018 | return objcg; | |
3019 | } | |
3020 | ||
bf4f0599 RG |
3021 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
3022 | { | |
3023 | struct obj_cgroup *objcg = NULL; | |
3024 | struct mem_cgroup *memcg; | |
3025 | ||
279c3393 RG |
3026 | if (memcg_kmem_bypass()) |
3027 | return NULL; | |
3028 | ||
bf4f0599 | 3029 | rcu_read_lock(); |
37d5985c RG |
3030 | if (unlikely(active_memcg())) |
3031 | memcg = active_memcg(); | |
bf4f0599 RG |
3032 | else |
3033 | memcg = mem_cgroup_from_task(current); | |
f4840ccf | 3034 | objcg = __get_obj_cgroup_from_memcg(memcg); |
bf4f0599 | 3035 | rcu_read_unlock(); |
f4840ccf JW |
3036 | return objcg; |
3037 | } | |
3038 | ||
3039 | struct obj_cgroup *get_obj_cgroup_from_page(struct page *page) | |
3040 | { | |
3041 | struct obj_cgroup *objcg; | |
3042 | ||
f7a449f7 | 3043 | if (!memcg_kmem_online()) |
f4840ccf JW |
3044 | return NULL; |
3045 | ||
3046 | if (PageMemcgKmem(page)) { | |
3047 | objcg = __folio_objcg(page_folio(page)); | |
3048 | obj_cgroup_get(objcg); | |
3049 | } else { | |
3050 | struct mem_cgroup *memcg; | |
bf4f0599 | 3051 | |
f4840ccf JW |
3052 | rcu_read_lock(); |
3053 | memcg = __folio_memcg(page_folio(page)); | |
3054 | if (memcg) | |
3055 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
3056 | else | |
3057 | objcg = NULL; | |
3058 | rcu_read_unlock(); | |
3059 | } | |
bf4f0599 RG |
3060 | return objcg; |
3061 | } | |
3062 | ||
a8c49af3 YA |
3063 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
3064 | { | |
3065 | mod_memcg_state(memcg, MEMCG_KMEM, nr_pages); | |
3066 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
3067 | if (nr_pages > 0) | |
3068 | page_counter_charge(&memcg->kmem, nr_pages); | |
3069 | else | |
3070 | page_counter_uncharge(&memcg->kmem, -nr_pages); | |
3071 | } | |
3072 | } | |
3073 | ||
3074 | ||
f1286fae MS |
3075 | /* |
3076 | * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg | |
3077 | * @objcg: object cgroup to uncharge | |
3078 | * @nr_pages: number of pages to uncharge | |
3079 | */ | |
e74d2259 MS |
3080 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
3081 | unsigned int nr_pages) | |
3082 | { | |
3083 | struct mem_cgroup *memcg; | |
3084 | ||
3085 | memcg = get_mem_cgroup_from_objcg(objcg); | |
e74d2259 | 3086 | |
a8c49af3 | 3087 | memcg_account_kmem(memcg, -nr_pages); |
f1286fae | 3088 | refill_stock(memcg, nr_pages); |
e74d2259 | 3089 | |
e74d2259 | 3090 | css_put(&memcg->css); |
e74d2259 MS |
3091 | } |
3092 | ||
f1286fae MS |
3093 | /* |
3094 | * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg | |
3095 | * @objcg: object cgroup to charge | |
45264778 | 3096 | * @gfp: reclaim mode |
92d0510c | 3097 | * @nr_pages: number of pages to charge |
45264778 VD |
3098 | * |
3099 | * Returns 0 on success, an error code on failure. | |
3100 | */ | |
f1286fae MS |
3101 | static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp, |
3102 | unsigned int nr_pages) | |
7ae1e1d0 | 3103 | { |
f1286fae | 3104 | struct mem_cgroup *memcg; |
7ae1e1d0 GC |
3105 | int ret; |
3106 | ||
f1286fae MS |
3107 | memcg = get_mem_cgroup_from_objcg(objcg); |
3108 | ||
c5c8b16b | 3109 | ret = try_charge_memcg(memcg, gfp, nr_pages); |
52c29b04 | 3110 | if (ret) |
f1286fae | 3111 | goto out; |
52c29b04 | 3112 | |
a8c49af3 | 3113 | memcg_account_kmem(memcg, nr_pages); |
f1286fae MS |
3114 | out: |
3115 | css_put(&memcg->css); | |
4b13f64d | 3116 | |
f1286fae | 3117 | return ret; |
4b13f64d RG |
3118 | } |
3119 | ||
45264778 | 3120 | /** |
f4b00eab | 3121 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3122 | * @page: page to charge |
3123 | * @gfp: reclaim mode | |
3124 | * @order: allocation order | |
3125 | * | |
3126 | * Returns 0 on success, an error code on failure. | |
3127 | */ | |
f4b00eab | 3128 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3129 | { |
b4e0b68f | 3130 | struct obj_cgroup *objcg; |
fcff7d7e | 3131 | int ret = 0; |
7ae1e1d0 | 3132 | |
b4e0b68f MS |
3133 | objcg = get_obj_cgroup_from_current(); |
3134 | if (objcg) { | |
3135 | ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order); | |
4d96ba35 | 3136 | if (!ret) { |
b4e0b68f | 3137 | page->memcg_data = (unsigned long)objcg | |
18b2db3b | 3138 | MEMCG_DATA_KMEM; |
1a3e1f40 | 3139 | return 0; |
4d96ba35 | 3140 | } |
b4e0b68f | 3141 | obj_cgroup_put(objcg); |
c4159a75 | 3142 | } |
d05e83a6 | 3143 | return ret; |
7ae1e1d0 | 3144 | } |
49a18eae | 3145 | |
45264778 | 3146 | /** |
f4b00eab | 3147 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3148 | * @page: page to uncharge |
3149 | * @order: allocation order | |
3150 | */ | |
f4b00eab | 3151 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3152 | { |
1b7e4464 | 3153 | struct folio *folio = page_folio(page); |
b4e0b68f | 3154 | struct obj_cgroup *objcg; |
f3ccb2c4 | 3155 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3156 | |
1b7e4464 | 3157 | if (!folio_memcg_kmem(folio)) |
7ae1e1d0 GC |
3158 | return; |
3159 | ||
1b7e4464 | 3160 | objcg = __folio_objcg(folio); |
b4e0b68f | 3161 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
1b7e4464 | 3162 | folio->memcg_data = 0; |
b4e0b68f | 3163 | obj_cgroup_put(objcg); |
60d3fd32 | 3164 | } |
bf4f0599 | 3165 | |
68ac5b3c WL |
3166 | void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
3167 | enum node_stat_item idx, int nr) | |
3168 | { | |
fead2b86 | 3169 | struct memcg_stock_pcp *stock; |
56751146 | 3170 | struct obj_cgroup *old = NULL; |
68ac5b3c WL |
3171 | unsigned long flags; |
3172 | int *bytes; | |
3173 | ||
56751146 | 3174 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3175 | stock = this_cpu_ptr(&memcg_stock); |
3176 | ||
68ac5b3c WL |
3177 | /* |
3178 | * Save vmstat data in stock and skip vmstat array update unless | |
3179 | * accumulating over a page of vmstat data or when pgdat or idx | |
3180 | * changes. | |
3181 | */ | |
3182 | if (stock->cached_objcg != objcg) { | |
56751146 | 3183 | old = drain_obj_stock(stock); |
68ac5b3c WL |
3184 | obj_cgroup_get(objcg); |
3185 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) | |
3186 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3187 | stock->cached_objcg = objcg; | |
3188 | stock->cached_pgdat = pgdat; | |
3189 | } else if (stock->cached_pgdat != pgdat) { | |
3190 | /* Flush the existing cached vmstat data */ | |
7fa0dacb WL |
3191 | struct pglist_data *oldpg = stock->cached_pgdat; |
3192 | ||
68ac5b3c | 3193 | if (stock->nr_slab_reclaimable_b) { |
7fa0dacb | 3194 | mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B, |
68ac5b3c WL |
3195 | stock->nr_slab_reclaimable_b); |
3196 | stock->nr_slab_reclaimable_b = 0; | |
3197 | } | |
3198 | if (stock->nr_slab_unreclaimable_b) { | |
7fa0dacb | 3199 | mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B, |
68ac5b3c WL |
3200 | stock->nr_slab_unreclaimable_b); |
3201 | stock->nr_slab_unreclaimable_b = 0; | |
3202 | } | |
3203 | stock->cached_pgdat = pgdat; | |
3204 | } | |
3205 | ||
3206 | bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b | |
3207 | : &stock->nr_slab_unreclaimable_b; | |
3208 | /* | |
3209 | * Even for large object >= PAGE_SIZE, the vmstat data will still be | |
3210 | * cached locally at least once before pushing it out. | |
3211 | */ | |
3212 | if (!*bytes) { | |
3213 | *bytes = nr; | |
3214 | nr = 0; | |
3215 | } else { | |
3216 | *bytes += nr; | |
3217 | if (abs(*bytes) > PAGE_SIZE) { | |
3218 | nr = *bytes; | |
3219 | *bytes = 0; | |
3220 | } else { | |
3221 | nr = 0; | |
3222 | } | |
3223 | } | |
3224 | if (nr) | |
3225 | mod_objcg_mlstate(objcg, pgdat, idx, nr); | |
3226 | ||
56751146 SAS |
3227 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
3228 | if (old) | |
3229 | obj_cgroup_put(old); | |
68ac5b3c WL |
3230 | } |
3231 | ||
bf4f0599 RG |
3232 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
3233 | { | |
fead2b86 | 3234 | struct memcg_stock_pcp *stock; |
bf4f0599 RG |
3235 | unsigned long flags; |
3236 | bool ret = false; | |
3237 | ||
56751146 | 3238 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3239 | |
3240 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 RG |
3241 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { |
3242 | stock->nr_bytes -= nr_bytes; | |
3243 | ret = true; | |
3244 | } | |
3245 | ||
56751146 | 3246 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
bf4f0599 RG |
3247 | |
3248 | return ret; | |
3249 | } | |
3250 | ||
56751146 | 3251 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 RG |
3252 | { |
3253 | struct obj_cgroup *old = stock->cached_objcg; | |
3254 | ||
3255 | if (!old) | |
56751146 | 3256 | return NULL; |
bf4f0599 RG |
3257 | |
3258 | if (stock->nr_bytes) { | |
3259 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3260 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3261 | ||
af9a3b69 JW |
3262 | if (nr_pages) { |
3263 | struct mem_cgroup *memcg; | |
3264 | ||
3265 | memcg = get_mem_cgroup_from_objcg(old); | |
3266 | ||
3267 | memcg_account_kmem(memcg, -nr_pages); | |
3268 | __refill_stock(memcg, nr_pages); | |
3269 | ||
3270 | css_put(&memcg->css); | |
3271 | } | |
bf4f0599 RG |
3272 | |
3273 | /* | |
3274 | * The leftover is flushed to the centralized per-memcg value. | |
3275 | * On the next attempt to refill obj stock it will be moved | |
3276 | * to a per-cpu stock (probably, on an other CPU), see | |
3277 | * refill_obj_stock(). | |
3278 | * | |
3279 | * How often it's flushed is a trade-off between the memory | |
3280 | * limit enforcement accuracy and potential CPU contention, | |
3281 | * so it might be changed in the future. | |
3282 | */ | |
3283 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3284 | stock->nr_bytes = 0; | |
3285 | } | |
3286 | ||
68ac5b3c WL |
3287 | /* |
3288 | * Flush the vmstat data in current stock | |
3289 | */ | |
3290 | if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) { | |
3291 | if (stock->nr_slab_reclaimable_b) { | |
3292 | mod_objcg_mlstate(old, stock->cached_pgdat, | |
3293 | NR_SLAB_RECLAIMABLE_B, | |
3294 | stock->nr_slab_reclaimable_b); | |
3295 | stock->nr_slab_reclaimable_b = 0; | |
3296 | } | |
3297 | if (stock->nr_slab_unreclaimable_b) { | |
3298 | mod_objcg_mlstate(old, stock->cached_pgdat, | |
3299 | NR_SLAB_UNRECLAIMABLE_B, | |
3300 | stock->nr_slab_unreclaimable_b); | |
3301 | stock->nr_slab_unreclaimable_b = 0; | |
3302 | } | |
3303 | stock->cached_pgdat = NULL; | |
3304 | } | |
3305 | ||
bf4f0599 | 3306 | stock->cached_objcg = NULL; |
56751146 SAS |
3307 | /* |
3308 | * The `old' objects needs to be released by the caller via | |
3309 | * obj_cgroup_put() outside of memcg_stock_pcp::stock_lock. | |
3310 | */ | |
3311 | return old; | |
bf4f0599 RG |
3312 | } |
3313 | ||
3314 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3315 | struct mem_cgroup *root_memcg) | |
3316 | { | |
3317 | struct mem_cgroup *memcg; | |
3318 | ||
fead2b86 MH |
3319 | if (stock->cached_objcg) { |
3320 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
bf4f0599 RG |
3321 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) |
3322 | return true; | |
3323 | } | |
3324 | ||
3325 | return false; | |
3326 | } | |
3327 | ||
5387c904 WL |
3328 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes, |
3329 | bool allow_uncharge) | |
bf4f0599 | 3330 | { |
fead2b86 | 3331 | struct memcg_stock_pcp *stock; |
56751146 | 3332 | struct obj_cgroup *old = NULL; |
bf4f0599 | 3333 | unsigned long flags; |
5387c904 | 3334 | unsigned int nr_pages = 0; |
bf4f0599 | 3335 | |
56751146 | 3336 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3337 | |
3338 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 3339 | if (stock->cached_objcg != objcg) { /* reset if necessary */ |
56751146 | 3340 | old = drain_obj_stock(stock); |
bf4f0599 RG |
3341 | obj_cgroup_get(objcg); |
3342 | stock->cached_objcg = objcg; | |
5387c904 WL |
3343 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) |
3344 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3345 | allow_uncharge = true; /* Allow uncharge when objcg changes */ | |
bf4f0599 RG |
3346 | } |
3347 | stock->nr_bytes += nr_bytes; | |
3348 | ||
5387c904 WL |
3349 | if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) { |
3350 | nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3351 | stock->nr_bytes &= (PAGE_SIZE - 1); | |
3352 | } | |
bf4f0599 | 3353 | |
56751146 SAS |
3354 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
3355 | if (old) | |
3356 | obj_cgroup_put(old); | |
5387c904 WL |
3357 | |
3358 | if (nr_pages) | |
3359 | obj_cgroup_uncharge_pages(objcg, nr_pages); | |
bf4f0599 RG |
3360 | } |
3361 | ||
3362 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3363 | { | |
bf4f0599 RG |
3364 | unsigned int nr_pages, nr_bytes; |
3365 | int ret; | |
3366 | ||
3367 | if (consume_obj_stock(objcg, size)) | |
3368 | return 0; | |
3369 | ||
3370 | /* | |
5387c904 | 3371 | * In theory, objcg->nr_charged_bytes can have enough |
bf4f0599 | 3372 | * pre-charged bytes to satisfy the allocation. However, |
5387c904 WL |
3373 | * flushing objcg->nr_charged_bytes requires two atomic |
3374 | * operations, and objcg->nr_charged_bytes can't be big. | |
3375 | * The shared objcg->nr_charged_bytes can also become a | |
3376 | * performance bottleneck if all tasks of the same memcg are | |
3377 | * trying to update it. So it's better to ignore it and try | |
3378 | * grab some new pages. The stock's nr_bytes will be flushed to | |
3379 | * objcg->nr_charged_bytes later on when objcg changes. | |
3380 | * | |
3381 | * The stock's nr_bytes may contain enough pre-charged bytes | |
3382 | * to allow one less page from being charged, but we can't rely | |
3383 | * on the pre-charged bytes not being changed outside of | |
3384 | * consume_obj_stock() or refill_obj_stock(). So ignore those | |
3385 | * pre-charged bytes as well when charging pages. To avoid a | |
3386 | * page uncharge right after a page charge, we set the | |
3387 | * allow_uncharge flag to false when calling refill_obj_stock() | |
3388 | * to temporarily allow the pre-charged bytes to exceed the page | |
3389 | * size limit. The maximum reachable value of the pre-charged | |
3390 | * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data | |
3391 | * race. | |
bf4f0599 | 3392 | */ |
bf4f0599 RG |
3393 | nr_pages = size >> PAGE_SHIFT; |
3394 | nr_bytes = size & (PAGE_SIZE - 1); | |
3395 | ||
3396 | if (nr_bytes) | |
3397 | nr_pages += 1; | |
3398 | ||
e74d2259 | 3399 | ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages); |
bf4f0599 | 3400 | if (!ret && nr_bytes) |
5387c904 | 3401 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false); |
bf4f0599 | 3402 | |
bf4f0599 RG |
3403 | return ret; |
3404 | } | |
3405 | ||
3406 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3407 | { | |
5387c904 | 3408 | refill_obj_stock(objcg, size, true); |
bf4f0599 RG |
3409 | } |
3410 | ||
84c07d11 | 3411 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3412 | |
ca3e0214 | 3413 | /* |
be6c8982 | 3414 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3415 | */ |
be6c8982 | 3416 | void split_page_memcg(struct page *head, unsigned int nr) |
ca3e0214 | 3417 | { |
1b7e4464 MWO |
3418 | struct folio *folio = page_folio(head); |
3419 | struct mem_cgroup *memcg = folio_memcg(folio); | |
e94c8a9c | 3420 | int i; |
ca3e0214 | 3421 | |
be6c8982 | 3422 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3423 | return; |
b070e65c | 3424 | |
be6c8982 | 3425 | for (i = 1; i < nr; i++) |
1b7e4464 | 3426 | folio_page(folio, i)->memcg_data = folio->memcg_data; |
b4e0b68f | 3427 | |
1b7e4464 MWO |
3428 | if (folio_memcg_kmem(folio)) |
3429 | obj_cgroup_get_many(__folio_objcg(folio), nr - 1); | |
b4e0b68f MS |
3430 | else |
3431 | css_get_many(&memcg->css, nr - 1); | |
ca3e0214 | 3432 | } |
ca3e0214 | 3433 | |
e55b9f96 | 3434 | #ifdef CONFIG_SWAP |
02491447 DN |
3435 | /** |
3436 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3437 | * @entry: swap entry to be moved | |
3438 | * @from: mem_cgroup which the entry is moved from | |
3439 | * @to: mem_cgroup which the entry is moved to | |
3440 | * | |
3441 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3442 | * as the mem_cgroup's id of @from. | |
3443 | * | |
3444 | * Returns 0 on success, -EINVAL on failure. | |
3445 | * | |
3e32cb2e | 3446 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3447 | * both res and memsw, and called css_get(). |
3448 | */ | |
3449 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3450 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3451 | { |
3452 | unsigned short old_id, new_id; | |
3453 | ||
34c00c31 LZ |
3454 | old_id = mem_cgroup_id(from); |
3455 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3456 | |
3457 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3458 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3459 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3460 | return 0; |
3461 | } | |
3462 | return -EINVAL; | |
3463 | } | |
3464 | #else | |
3465 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3466 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3467 | { |
3468 | return -EINVAL; | |
3469 | } | |
8c7c6e34 | 3470 | #endif |
d13d1443 | 3471 | |
bbec2e15 | 3472 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3473 | |
bbec2e15 RG |
3474 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3475 | unsigned long max, bool memsw) | |
628f4235 | 3476 | { |
3e32cb2e | 3477 | bool enlarge = false; |
bb4a7ea2 | 3478 | bool drained = false; |
3e32cb2e | 3479 | int ret; |
c054a78c YZ |
3480 | bool limits_invariant; |
3481 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3482 | |
3e32cb2e | 3483 | do { |
628f4235 KH |
3484 | if (signal_pending(current)) { |
3485 | ret = -EINTR; | |
3486 | break; | |
3487 | } | |
3e32cb2e | 3488 | |
bbec2e15 | 3489 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3490 | /* |
3491 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3492 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3493 | */ |
15b42562 | 3494 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3495 | max <= memcg->memsw.max; |
c054a78c | 3496 | if (!limits_invariant) { |
bbec2e15 | 3497 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3498 | ret = -EINVAL; |
8c7c6e34 KH |
3499 | break; |
3500 | } | |
bbec2e15 | 3501 | if (max > counter->max) |
3e32cb2e | 3502 | enlarge = true; |
bbec2e15 RG |
3503 | ret = page_counter_set_max(counter, max); |
3504 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3505 | |
3506 | if (!ret) | |
3507 | break; | |
3508 | ||
bb4a7ea2 SB |
3509 | if (!drained) { |
3510 | drain_all_stock(memcg); | |
3511 | drained = true; | |
3512 | continue; | |
3513 | } | |
3514 | ||
73b73bac | 3515 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
55ab834a | 3516 | memsw ? 0 : MEMCG_RECLAIM_MAY_SWAP)) { |
1ab5c056 AR |
3517 | ret = -EBUSY; |
3518 | break; | |
3519 | } | |
3520 | } while (true); | |
3e32cb2e | 3521 | |
3c11ecf4 KH |
3522 | if (!ret && enlarge) |
3523 | memcg_oom_recover(memcg); | |
3e32cb2e | 3524 | |
628f4235 KH |
3525 | return ret; |
3526 | } | |
3527 | ||
ef8f2327 | 3528 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3529 | gfp_t gfp_mask, |
3530 | unsigned long *total_scanned) | |
3531 | { | |
3532 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3533 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3534 | unsigned long reclaimed; |
3535 | int loop = 0; | |
ef8f2327 | 3536 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3537 | unsigned long excess; |
0608f43d | 3538 | |
e4dde56c YZ |
3539 | if (lru_gen_enabled()) |
3540 | return 0; | |
3541 | ||
0608f43d AM |
3542 | if (order > 0) |
3543 | return 0; | |
3544 | ||
2ab082ba | 3545 | mctz = soft_limit_tree.rb_tree_per_node[pgdat->node_id]; |
d6507ff5 MH |
3546 | |
3547 | /* | |
3548 | * Do not even bother to check the largest node if the root | |
3549 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3550 | * are acceptable as soft limit is best effort anyway. | |
3551 | */ | |
bfc7228b | 3552 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3553 | return 0; |
3554 | ||
0608f43d AM |
3555 | /* |
3556 | * This loop can run a while, specially if mem_cgroup's continuously | |
3557 | * keep exceeding their soft limit and putting the system under | |
3558 | * pressure | |
3559 | */ | |
3560 | do { | |
3561 | if (next_mz) | |
3562 | mz = next_mz; | |
3563 | else | |
3564 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3565 | if (!mz) | |
3566 | break; | |
3567 | ||
ef8f2327 | 3568 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
d8f65338 | 3569 | gfp_mask, total_scanned); |
0608f43d | 3570 | nr_reclaimed += reclaimed; |
0a31bc97 | 3571 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3572 | |
3573 | /* | |
3574 | * If we failed to reclaim anything from this memory cgroup | |
3575 | * it is time to move on to the next cgroup | |
3576 | */ | |
3577 | next_mz = NULL; | |
bc2f2e7f VD |
3578 | if (!reclaimed) |
3579 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3580 | ||
3e32cb2e | 3581 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3582 | /* |
3583 | * One school of thought says that we should not add | |
3584 | * back the node to the tree if reclaim returns 0. | |
3585 | * But our reclaim could return 0, simply because due | |
3586 | * to priority we are exposing a smaller subset of | |
3587 | * memory to reclaim from. Consider this as a longer | |
3588 | * term TODO. | |
3589 | */ | |
3590 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3591 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3592 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3593 | css_put(&mz->memcg->css); |
3594 | loop++; | |
3595 | /* | |
3596 | * Could not reclaim anything and there are no more | |
3597 | * mem cgroups to try or we seem to be looping without | |
3598 | * reclaiming anything. | |
3599 | */ | |
3600 | if (!nr_reclaimed && | |
3601 | (next_mz == NULL || | |
3602 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3603 | break; | |
3604 | } while (!nr_reclaimed); | |
3605 | if (next_mz) | |
3606 | css_put(&next_mz->memcg->css); | |
3607 | return nr_reclaimed; | |
3608 | } | |
3609 | ||
c26251f9 | 3610 | /* |
51038171 | 3611 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3612 | * |
3613 | * Caller is responsible for holding css reference for memcg. | |
3614 | */ | |
3615 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3616 | { | |
d977aa93 | 3617 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3618 | |
c1e862c1 KH |
3619 | /* we call try-to-free pages for make this cgroup empty */ |
3620 | lru_add_drain_all(); | |
d12c60f6 JS |
3621 | |
3622 | drain_all_stock(memcg); | |
3623 | ||
f817ed48 | 3624 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3625 | while (nr_retries && page_counter_read(&memcg->memory)) { |
c26251f9 MH |
3626 | if (signal_pending(current)) |
3627 | return -EINTR; | |
3628 | ||
73b73bac | 3629 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
55ab834a | 3630 | MEMCG_RECLAIM_MAY_SWAP)) |
f817ed48 | 3631 | nr_retries--; |
f817ed48 | 3632 | } |
ab5196c2 MH |
3633 | |
3634 | return 0; | |
cc847582 KH |
3635 | } |
3636 | ||
6770c64e TH |
3637 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3638 | char *buf, size_t nbytes, | |
3639 | loff_t off) | |
c1e862c1 | 3640 | { |
6770c64e | 3641 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3642 | |
d8423011 MH |
3643 | if (mem_cgroup_is_root(memcg)) |
3644 | return -EINVAL; | |
6770c64e | 3645 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3646 | } |
3647 | ||
182446d0 TH |
3648 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3649 | struct cftype *cft) | |
18f59ea7 | 3650 | { |
bef8620c | 3651 | return 1; |
18f59ea7 BS |
3652 | } |
3653 | ||
182446d0 TH |
3654 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3655 | struct cftype *cft, u64 val) | |
18f59ea7 | 3656 | { |
bef8620c | 3657 | if (val == 1) |
0b8f73e1 | 3658 | return 0; |
567fb435 | 3659 | |
bef8620c RG |
3660 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3661 | "Please report your usecase to linux-mm@kvack.org if you " | |
3662 | "depend on this functionality.\n"); | |
567fb435 | 3663 | |
bef8620c | 3664 | return -EINVAL; |
18f59ea7 BS |
3665 | } |
3666 | ||
6f646156 | 3667 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3668 | { |
42a30035 | 3669 | unsigned long val; |
ce00a967 | 3670 | |
3e32cb2e | 3671 | if (mem_cgroup_is_root(memcg)) { |
fd25a9e0 | 3672 | mem_cgroup_flush_stats(); |
0d1c2072 | 3673 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3674 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3675 | if (swap) |
3676 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3677 | } else { |
ce00a967 | 3678 | if (!swap) |
3e32cb2e | 3679 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3680 | else |
3e32cb2e | 3681 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3682 | } |
c12176d3 | 3683 | return val; |
ce00a967 JW |
3684 | } |
3685 | ||
3e32cb2e JW |
3686 | enum { |
3687 | RES_USAGE, | |
3688 | RES_LIMIT, | |
3689 | RES_MAX_USAGE, | |
3690 | RES_FAILCNT, | |
3691 | RES_SOFT_LIMIT, | |
3692 | }; | |
ce00a967 | 3693 | |
791badbd | 3694 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3695 | struct cftype *cft) |
8cdea7c0 | 3696 | { |
182446d0 | 3697 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3698 | struct page_counter *counter; |
af36f906 | 3699 | |
3e32cb2e | 3700 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3701 | case _MEM: |
3e32cb2e JW |
3702 | counter = &memcg->memory; |
3703 | break; | |
8c7c6e34 | 3704 | case _MEMSWAP: |
3e32cb2e JW |
3705 | counter = &memcg->memsw; |
3706 | break; | |
510fc4e1 | 3707 | case _KMEM: |
3e32cb2e | 3708 | counter = &memcg->kmem; |
510fc4e1 | 3709 | break; |
d55f90bf | 3710 | case _TCP: |
0db15298 | 3711 | counter = &memcg->tcpmem; |
d55f90bf | 3712 | break; |
8c7c6e34 KH |
3713 | default: |
3714 | BUG(); | |
8c7c6e34 | 3715 | } |
3e32cb2e JW |
3716 | |
3717 | switch (MEMFILE_ATTR(cft->private)) { | |
3718 | case RES_USAGE: | |
3719 | if (counter == &memcg->memory) | |
c12176d3 | 3720 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3721 | if (counter == &memcg->memsw) |
c12176d3 | 3722 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3723 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3724 | case RES_LIMIT: | |
bbec2e15 | 3725 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3726 | case RES_MAX_USAGE: |
3727 | return (u64)counter->watermark * PAGE_SIZE; | |
3728 | case RES_FAILCNT: | |
3729 | return counter->failcnt; | |
3730 | case RES_SOFT_LIMIT: | |
2178e20c | 3731 | return (u64)READ_ONCE(memcg->soft_limit) * PAGE_SIZE; |
3e32cb2e JW |
3732 | default: |
3733 | BUG(); | |
3734 | } | |
8cdea7c0 | 3735 | } |
510fc4e1 | 3736 | |
84c07d11 | 3737 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3738 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3739 | { |
bf4f0599 | 3740 | struct obj_cgroup *objcg; |
d6441637 | 3741 | |
9c94bef9 | 3742 | if (mem_cgroup_kmem_disabled()) |
b313aeee VD |
3743 | return 0; |
3744 | ||
da0efe30 MS |
3745 | if (unlikely(mem_cgroup_is_root(memcg))) |
3746 | return 0; | |
d6441637 | 3747 | |
bf4f0599 | 3748 | objcg = obj_cgroup_alloc(); |
f9c69d63 | 3749 | if (!objcg) |
bf4f0599 | 3750 | return -ENOMEM; |
f9c69d63 | 3751 | |
bf4f0599 RG |
3752 | objcg->memcg = memcg; |
3753 | rcu_assign_pointer(memcg->objcg, objcg); | |
3754 | ||
f7a449f7 | 3755 | static_branch_enable(&memcg_kmem_online_key); |
d648bcc7 | 3756 | |
f9c69d63 | 3757 | memcg->kmemcg_id = memcg->id.id; |
0b8f73e1 JW |
3758 | |
3759 | return 0; | |
d6441637 VD |
3760 | } |
3761 | ||
8e0a8912 JW |
3762 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3763 | { | |
64268868 | 3764 | struct mem_cgroup *parent; |
8e0a8912 | 3765 | |
9c94bef9 | 3766 | if (mem_cgroup_kmem_disabled()) |
da0efe30 MS |
3767 | return; |
3768 | ||
3769 | if (unlikely(mem_cgroup_is_root(memcg))) | |
8e0a8912 | 3770 | return; |
9855609b | 3771 | |
8e0a8912 JW |
3772 | parent = parent_mem_cgroup(memcg); |
3773 | if (!parent) | |
3774 | parent = root_mem_cgroup; | |
3775 | ||
bf4f0599 | 3776 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a | 3777 | |
8e0a8912 | 3778 | /* |
64268868 MS |
3779 | * After we have finished memcg_reparent_objcgs(), all list_lrus |
3780 | * corresponding to this cgroup are guaranteed to remain empty. | |
3781 | * The ordering is imposed by list_lru_node->lock taken by | |
1f391eb2 | 3782 | * memcg_reparent_list_lrus(). |
8e0a8912 | 3783 | */ |
1f391eb2 | 3784 | memcg_reparent_list_lrus(memcg, parent); |
8e0a8912 | 3785 | } |
d6441637 | 3786 | #else |
0b8f73e1 | 3787 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3788 | { |
3789 | return 0; | |
3790 | } | |
3791 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3792 | { | |
3793 | } | |
84c07d11 | 3794 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3795 | |
bbec2e15 | 3796 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3797 | { |
3798 | int ret; | |
3799 | ||
bbec2e15 | 3800 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3801 | |
bbec2e15 | 3802 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3803 | if (ret) |
3804 | goto out; | |
3805 | ||
0db15298 | 3806 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3807 | /* |
3808 | * The active flag needs to be written after the static_key | |
3809 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3810 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3811 | * for details, and note that we don't mark any socket as | |
3812 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3813 | * |
3814 | * We need to do this, because static_keys will span multiple | |
3815 | * sites, but we can't control their order. If we mark a socket | |
3816 | * as accounted, but the accounting functions are not patched in | |
3817 | * yet, we'll lose accounting. | |
3818 | * | |
2d758073 | 3819 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3820 | * because when this value change, the code to process it is not |
3821 | * patched in yet. | |
3822 | */ | |
3823 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3824 | memcg->tcpmem_active = true; |
d55f90bf VD |
3825 | } |
3826 | out: | |
bbec2e15 | 3827 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3828 | return ret; |
3829 | } | |
d55f90bf | 3830 | |
628f4235 KH |
3831 | /* |
3832 | * The user of this function is... | |
3833 | * RES_LIMIT. | |
3834 | */ | |
451af504 TH |
3835 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3836 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3837 | { |
451af504 | 3838 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3839 | unsigned long nr_pages; |
628f4235 KH |
3840 | int ret; |
3841 | ||
451af504 | 3842 | buf = strstrip(buf); |
650c5e56 | 3843 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3844 | if (ret) |
3845 | return ret; | |
af36f906 | 3846 | |
3e32cb2e | 3847 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3848 | case RES_LIMIT: |
4b3bde4c BS |
3849 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3850 | ret = -EINVAL; | |
3851 | break; | |
3852 | } | |
3e32cb2e JW |
3853 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3854 | case _MEM: | |
bbec2e15 | 3855 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3856 | break; |
3e32cb2e | 3857 | case _MEMSWAP: |
bbec2e15 | 3858 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3859 | break; |
3e32cb2e | 3860 | case _KMEM: |
58056f77 SB |
3861 | /* kmem.limit_in_bytes is deprecated. */ |
3862 | ret = -EOPNOTSUPP; | |
3e32cb2e | 3863 | break; |
d55f90bf | 3864 | case _TCP: |
bbec2e15 | 3865 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3866 | break; |
3e32cb2e | 3867 | } |
296c81d8 | 3868 | break; |
3e32cb2e | 3869 | case RES_SOFT_LIMIT: |
2343e88d SAS |
3870 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
3871 | ret = -EOPNOTSUPP; | |
3872 | } else { | |
2178e20c | 3873 | WRITE_ONCE(memcg->soft_limit, nr_pages); |
2343e88d SAS |
3874 | ret = 0; |
3875 | } | |
628f4235 KH |
3876 | break; |
3877 | } | |
451af504 | 3878 | return ret ?: nbytes; |
8cdea7c0 BS |
3879 | } |
3880 | ||
6770c64e TH |
3881 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3882 | size_t nbytes, loff_t off) | |
c84872e1 | 3883 | { |
6770c64e | 3884 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3885 | struct page_counter *counter; |
c84872e1 | 3886 | |
3e32cb2e JW |
3887 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3888 | case _MEM: | |
3889 | counter = &memcg->memory; | |
3890 | break; | |
3891 | case _MEMSWAP: | |
3892 | counter = &memcg->memsw; | |
3893 | break; | |
3894 | case _KMEM: | |
3895 | counter = &memcg->kmem; | |
3896 | break; | |
d55f90bf | 3897 | case _TCP: |
0db15298 | 3898 | counter = &memcg->tcpmem; |
d55f90bf | 3899 | break; |
3e32cb2e JW |
3900 | default: |
3901 | BUG(); | |
3902 | } | |
af36f906 | 3903 | |
3e32cb2e | 3904 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3905 | case RES_MAX_USAGE: |
3e32cb2e | 3906 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3907 | break; |
3908 | case RES_FAILCNT: | |
3e32cb2e | 3909 | counter->failcnt = 0; |
29f2a4da | 3910 | break; |
3e32cb2e JW |
3911 | default: |
3912 | BUG(); | |
29f2a4da | 3913 | } |
f64c3f54 | 3914 | |
6770c64e | 3915 | return nbytes; |
c84872e1 PE |
3916 | } |
3917 | ||
182446d0 | 3918 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3919 | struct cftype *cft) |
3920 | { | |
182446d0 | 3921 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3922 | } |
3923 | ||
02491447 | 3924 | #ifdef CONFIG_MMU |
182446d0 | 3925 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3926 | struct cftype *cft, u64 val) |
3927 | { | |
182446d0 | 3928 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3929 | |
da34a848 JW |
3930 | pr_warn_once("Cgroup memory moving (move_charge_at_immigrate) is deprecated. " |
3931 | "Please report your usecase to linux-mm@kvack.org if you " | |
3932 | "depend on this functionality.\n"); | |
3933 | ||
1dfab5ab | 3934 | if (val & ~MOVE_MASK) |
7dc74be0 | 3935 | return -EINVAL; |
ee5e8472 | 3936 | |
7dc74be0 | 3937 | /* |
ee5e8472 GC |
3938 | * No kind of locking is needed in here, because ->can_attach() will |
3939 | * check this value once in the beginning of the process, and then carry | |
3940 | * on with stale data. This means that changes to this value will only | |
3941 | * affect task migrations starting after the change. | |
7dc74be0 | 3942 | */ |
c0ff4b85 | 3943 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3944 | return 0; |
3945 | } | |
02491447 | 3946 | #else |
182446d0 | 3947 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3948 | struct cftype *cft, u64 val) |
3949 | { | |
3950 | return -ENOSYS; | |
3951 | } | |
3952 | #endif | |
7dc74be0 | 3953 | |
406eb0c9 | 3954 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3955 | |
3956 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3957 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3958 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3959 | ||
3960 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3961 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3962 | { |
867e5e1d | 3963 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3964 | unsigned long nr = 0; |
3965 | enum lru_list lru; | |
3966 | ||
3967 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3968 | ||
3969 | for_each_lru(lru) { | |
3970 | if (!(BIT(lru) & lru_mask)) | |
3971 | continue; | |
dd8657b6 SB |
3972 | if (tree) |
3973 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3974 | else | |
3975 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3976 | } |
3977 | return nr; | |
3978 | } | |
3979 | ||
3980 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3981 | unsigned int lru_mask, |
3982 | bool tree) | |
113b7dfd JW |
3983 | { |
3984 | unsigned long nr = 0; | |
3985 | enum lru_list lru; | |
3986 | ||
3987 | for_each_lru(lru) { | |
3988 | if (!(BIT(lru) & lru_mask)) | |
3989 | continue; | |
dd8657b6 SB |
3990 | if (tree) |
3991 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3992 | else | |
3993 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3994 | } |
3995 | return nr; | |
3996 | } | |
3997 | ||
2da8ca82 | 3998 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3999 | { |
25485de6 GT |
4000 | struct numa_stat { |
4001 | const char *name; | |
4002 | unsigned int lru_mask; | |
4003 | }; | |
4004 | ||
4005 | static const struct numa_stat stats[] = { | |
4006 | { "total", LRU_ALL }, | |
4007 | { "file", LRU_ALL_FILE }, | |
4008 | { "anon", LRU_ALL_ANON }, | |
4009 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4010 | }; | |
4011 | const struct numa_stat *stat; | |
406eb0c9 | 4012 | int nid; |
aa9694bb | 4013 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 4014 | |
fd25a9e0 | 4015 | mem_cgroup_flush_stats(); |
2d146aa3 | 4016 | |
25485de6 | 4017 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4018 | seq_printf(m, "%s=%lu", stat->name, |
4019 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4020 | false)); | |
4021 | for_each_node_state(nid, N_MEMORY) | |
4022 | seq_printf(m, " N%d=%lu", nid, | |
4023 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4024 | stat->lru_mask, false)); | |
25485de6 | 4025 | seq_putc(m, '\n'); |
406eb0c9 | 4026 | } |
406eb0c9 | 4027 | |
071aee13 | 4028 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4029 | |
4030 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
4031 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4032 | true)); | |
4033 | for_each_node_state(nid, N_MEMORY) | |
4034 | seq_printf(m, " N%d=%lu", nid, | |
4035 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4036 | stat->lru_mask, true)); | |
071aee13 | 4037 | seq_putc(m, '\n'); |
406eb0c9 | 4038 | } |
406eb0c9 | 4039 | |
406eb0c9 YH |
4040 | return 0; |
4041 | } | |
4042 | #endif /* CONFIG_NUMA */ | |
4043 | ||
c8713d0b | 4044 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4045 | NR_FILE_PAGES, |
be5d0a74 | 4046 | NR_ANON_MAPPED, |
468c3982 JW |
4047 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4048 | NR_ANON_THPS, | |
4049 | #endif | |
c8713d0b JW |
4050 | NR_SHMEM, |
4051 | NR_FILE_MAPPED, | |
4052 | NR_FILE_DIRTY, | |
4053 | NR_WRITEBACK, | |
e09b0b61 YS |
4054 | WORKINGSET_REFAULT_ANON, |
4055 | WORKINGSET_REFAULT_FILE, | |
c8713d0b JW |
4056 | MEMCG_SWAP, |
4057 | }; | |
4058 | ||
4059 | static const char *const memcg1_stat_names[] = { | |
4060 | "cache", | |
4061 | "rss", | |
468c3982 | 4062 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4063 | "rss_huge", |
468c3982 | 4064 | #endif |
c8713d0b JW |
4065 | "shmem", |
4066 | "mapped_file", | |
4067 | "dirty", | |
4068 | "writeback", | |
e09b0b61 YS |
4069 | "workingset_refault_anon", |
4070 | "workingset_refault_file", | |
c8713d0b JW |
4071 | "swap", |
4072 | }; | |
4073 | ||
df0e53d0 | 4074 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4075 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4076 | PGPGIN, |
4077 | PGPGOUT, | |
4078 | PGFAULT, | |
4079 | PGMAJFAULT, | |
4080 | }; | |
4081 | ||
2da8ca82 | 4082 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4083 | { |
aa9694bb | 4084 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4085 | unsigned long memory, memsw; |
af7c4b0e JW |
4086 | struct mem_cgroup *mi; |
4087 | unsigned int i; | |
406eb0c9 | 4088 | |
71cd3113 | 4089 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4090 | |
fd25a9e0 | 4091 | mem_cgroup_flush_stats(); |
2d146aa3 | 4092 | |
71cd3113 | 4093 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4094 | unsigned long nr; |
4095 | ||
71cd3113 | 4096 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4097 | continue; |
468c3982 | 4098 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
e09b0b61 YS |
4099 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
4100 | nr * memcg_page_state_unit(memcg1_stats[i])); | |
1dd3a273 | 4101 | } |
7b854121 | 4102 | |
df0e53d0 | 4103 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4104 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4105 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4106 | |
4107 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4108 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4109 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4110 | PAGE_SIZE); |
af7c4b0e | 4111 | |
14067bb3 | 4112 | /* Hierarchical information */ |
3e32cb2e JW |
4113 | memory = memsw = PAGE_COUNTER_MAX; |
4114 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4115 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4116 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4117 | } |
3e32cb2e JW |
4118 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4119 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4120 | if (do_memsw_account()) |
3e32cb2e JW |
4121 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4122 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4123 | |
8de7ecc6 | 4124 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4125 | unsigned long nr; |
4126 | ||
71cd3113 | 4127 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4128 | continue; |
7de2e9f1 | 4129 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
8de7ecc6 | 4130 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
e09b0b61 | 4131 | (u64)nr * memcg_page_state_unit(memcg1_stats[i])); |
af7c4b0e JW |
4132 | } |
4133 | ||
8de7ecc6 | 4134 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4135 | seq_printf(m, "total_%s %llu\n", |
4136 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4137 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4138 | |
8de7ecc6 | 4139 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4140 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4141 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4142 | PAGE_SIZE); | |
14067bb3 | 4143 | |
7f016ee8 | 4144 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4145 | { |
ef8f2327 MG |
4146 | pg_data_t *pgdat; |
4147 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4148 | unsigned long anon_cost = 0; |
4149 | unsigned long file_cost = 0; | |
7f016ee8 | 4150 | |
ef8f2327 | 4151 | for_each_online_pgdat(pgdat) { |
a3747b53 | 4152 | mz = memcg->nodeinfo[pgdat->node_id]; |
7f016ee8 | 4153 | |
1431d4d1 JW |
4154 | anon_cost += mz->lruvec.anon_cost; |
4155 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4156 | } |
1431d4d1 JW |
4157 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4158 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4159 | } |
4160 | #endif | |
4161 | ||
d2ceb9b7 KH |
4162 | return 0; |
4163 | } | |
4164 | ||
182446d0 TH |
4165 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4166 | struct cftype *cft) | |
a7885eb8 | 4167 | { |
182446d0 | 4168 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4169 | |
1f4c025b | 4170 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4171 | } |
4172 | ||
182446d0 TH |
4173 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4174 | struct cftype *cft, u64 val) | |
a7885eb8 | 4175 | { |
182446d0 | 4176 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4177 | |
37bc3cb9 | 4178 | if (val > 200) |
a7885eb8 KM |
4179 | return -EINVAL; |
4180 | ||
a4792030 | 4181 | if (!mem_cgroup_is_root(memcg)) |
82b3aa26 | 4182 | WRITE_ONCE(memcg->swappiness, val); |
3dae7fec | 4183 | else |
82b3aa26 | 4184 | WRITE_ONCE(vm_swappiness, val); |
068b38c1 | 4185 | |
a7885eb8 KM |
4186 | return 0; |
4187 | } | |
4188 | ||
2e72b634 KS |
4189 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4190 | { | |
4191 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4192 | unsigned long usage; |
2e72b634 KS |
4193 | int i; |
4194 | ||
4195 | rcu_read_lock(); | |
4196 | if (!swap) | |
2c488db2 | 4197 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4198 | else |
2c488db2 | 4199 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4200 | |
4201 | if (!t) | |
4202 | goto unlock; | |
4203 | ||
ce00a967 | 4204 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4205 | |
4206 | /* | |
748dad36 | 4207 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4208 | * If it's not true, a threshold was crossed after last |
4209 | * call of __mem_cgroup_threshold(). | |
4210 | */ | |
5407a562 | 4211 | i = t->current_threshold; |
2e72b634 KS |
4212 | |
4213 | /* | |
4214 | * Iterate backward over array of thresholds starting from | |
4215 | * current_threshold and check if a threshold is crossed. | |
4216 | * If none of thresholds below usage is crossed, we read | |
4217 | * only one element of the array here. | |
4218 | */ | |
4219 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4220 | eventfd_signal(t->entries[i].eventfd, 1); | |
4221 | ||
4222 | /* i = current_threshold + 1 */ | |
4223 | i++; | |
4224 | ||
4225 | /* | |
4226 | * Iterate forward over array of thresholds starting from | |
4227 | * current_threshold+1 and check if a threshold is crossed. | |
4228 | * If none of thresholds above usage is crossed, we read | |
4229 | * only one element of the array here. | |
4230 | */ | |
4231 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4232 | eventfd_signal(t->entries[i].eventfd, 1); | |
4233 | ||
4234 | /* Update current_threshold */ | |
5407a562 | 4235 | t->current_threshold = i - 1; |
2e72b634 KS |
4236 | unlock: |
4237 | rcu_read_unlock(); | |
4238 | } | |
4239 | ||
4240 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4241 | { | |
ad4ca5f4 KS |
4242 | while (memcg) { |
4243 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4244 | if (do_memsw_account()) |
ad4ca5f4 KS |
4245 | __mem_cgroup_threshold(memcg, true); |
4246 | ||
4247 | memcg = parent_mem_cgroup(memcg); | |
4248 | } | |
2e72b634 KS |
4249 | } |
4250 | ||
4251 | static int compare_thresholds(const void *a, const void *b) | |
4252 | { | |
4253 | const struct mem_cgroup_threshold *_a = a; | |
4254 | const struct mem_cgroup_threshold *_b = b; | |
4255 | ||
2bff24a3 GT |
4256 | if (_a->threshold > _b->threshold) |
4257 | return 1; | |
4258 | ||
4259 | if (_a->threshold < _b->threshold) | |
4260 | return -1; | |
4261 | ||
4262 | return 0; | |
2e72b634 KS |
4263 | } |
4264 | ||
c0ff4b85 | 4265 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4266 | { |
4267 | struct mem_cgroup_eventfd_list *ev; | |
4268 | ||
2bcf2e92 MH |
4269 | spin_lock(&memcg_oom_lock); |
4270 | ||
c0ff4b85 | 4271 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4272 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4273 | |
4274 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4275 | return 0; |
4276 | } | |
4277 | ||
c0ff4b85 | 4278 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4279 | { |
7d74b06f KH |
4280 | struct mem_cgroup *iter; |
4281 | ||
c0ff4b85 | 4282 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4283 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4284 | } |
4285 | ||
59b6f873 | 4286 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4287 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4288 | { |
2c488db2 KS |
4289 | struct mem_cgroup_thresholds *thresholds; |
4290 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4291 | unsigned long threshold; |
4292 | unsigned long usage; | |
2c488db2 | 4293 | int i, size, ret; |
2e72b634 | 4294 | |
650c5e56 | 4295 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4296 | if (ret) |
4297 | return ret; | |
4298 | ||
4299 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4300 | |
05b84301 | 4301 | if (type == _MEM) { |
2c488db2 | 4302 | thresholds = &memcg->thresholds; |
ce00a967 | 4303 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4304 | } else if (type == _MEMSWAP) { |
2c488db2 | 4305 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4306 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4307 | } else |
2e72b634 KS |
4308 | BUG(); |
4309 | ||
2e72b634 | 4310 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4311 | if (thresholds->primary) |
2e72b634 KS |
4312 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4313 | ||
2c488db2 | 4314 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4315 | |
4316 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4317 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4318 | if (!new) { |
2e72b634 KS |
4319 | ret = -ENOMEM; |
4320 | goto unlock; | |
4321 | } | |
2c488db2 | 4322 | new->size = size; |
2e72b634 KS |
4323 | |
4324 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4325 | if (thresholds->primary) |
4326 | memcpy(new->entries, thresholds->primary->entries, | |
4327 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4328 | |
2e72b634 | 4329 | /* Add new threshold */ |
2c488db2 KS |
4330 | new->entries[size - 1].eventfd = eventfd; |
4331 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4332 | |
4333 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4334 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4335 | compare_thresholds, NULL); |
4336 | ||
4337 | /* Find current threshold */ | |
2c488db2 | 4338 | new->current_threshold = -1; |
2e72b634 | 4339 | for (i = 0; i < size; i++) { |
748dad36 | 4340 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4341 | /* |
2c488db2 KS |
4342 | * new->current_threshold will not be used until |
4343 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4344 | * it here. |
4345 | */ | |
2c488db2 | 4346 | ++new->current_threshold; |
748dad36 SZ |
4347 | } else |
4348 | break; | |
2e72b634 KS |
4349 | } |
4350 | ||
2c488db2 KS |
4351 | /* Free old spare buffer and save old primary buffer as spare */ |
4352 | kfree(thresholds->spare); | |
4353 | thresholds->spare = thresholds->primary; | |
4354 | ||
4355 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4356 | |
907860ed | 4357 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4358 | synchronize_rcu(); |
4359 | ||
2e72b634 KS |
4360 | unlock: |
4361 | mutex_unlock(&memcg->thresholds_lock); | |
4362 | ||
4363 | return ret; | |
4364 | } | |
4365 | ||
59b6f873 | 4366 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4367 | struct eventfd_ctx *eventfd, const char *args) |
4368 | { | |
59b6f873 | 4369 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4370 | } |
4371 | ||
59b6f873 | 4372 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4373 | struct eventfd_ctx *eventfd, const char *args) |
4374 | { | |
59b6f873 | 4375 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4376 | } |
4377 | ||
59b6f873 | 4378 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4379 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4380 | { |
2c488db2 KS |
4381 | struct mem_cgroup_thresholds *thresholds; |
4382 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4383 | unsigned long usage; |
7d36665a | 4384 | int i, j, size, entries; |
2e72b634 KS |
4385 | |
4386 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4387 | |
4388 | if (type == _MEM) { | |
2c488db2 | 4389 | thresholds = &memcg->thresholds; |
ce00a967 | 4390 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4391 | } else if (type == _MEMSWAP) { |
2c488db2 | 4392 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4393 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4394 | } else |
2e72b634 KS |
4395 | BUG(); |
4396 | ||
371528ca AV |
4397 | if (!thresholds->primary) |
4398 | goto unlock; | |
4399 | ||
2e72b634 KS |
4400 | /* Check if a threshold crossed before removing */ |
4401 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4402 | ||
4403 | /* Calculate new number of threshold */ | |
7d36665a | 4404 | size = entries = 0; |
2c488db2 KS |
4405 | for (i = 0; i < thresholds->primary->size; i++) { |
4406 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4407 | size++; |
7d36665a CX |
4408 | else |
4409 | entries++; | |
2e72b634 KS |
4410 | } |
4411 | ||
2c488db2 | 4412 | new = thresholds->spare; |
907860ed | 4413 | |
7d36665a CX |
4414 | /* If no items related to eventfd have been cleared, nothing to do */ |
4415 | if (!entries) | |
4416 | goto unlock; | |
4417 | ||
2e72b634 KS |
4418 | /* Set thresholds array to NULL if we don't have thresholds */ |
4419 | if (!size) { | |
2c488db2 KS |
4420 | kfree(new); |
4421 | new = NULL; | |
907860ed | 4422 | goto swap_buffers; |
2e72b634 KS |
4423 | } |
4424 | ||
2c488db2 | 4425 | new->size = size; |
2e72b634 KS |
4426 | |
4427 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4428 | new->current_threshold = -1; |
4429 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4430 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4431 | continue; |
4432 | ||
2c488db2 | 4433 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4434 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4435 | /* |
2c488db2 | 4436 | * new->current_threshold will not be used |
2e72b634 KS |
4437 | * until rcu_assign_pointer(), so it's safe to increment |
4438 | * it here. | |
4439 | */ | |
2c488db2 | 4440 | ++new->current_threshold; |
2e72b634 KS |
4441 | } |
4442 | j++; | |
4443 | } | |
4444 | ||
907860ed | 4445 | swap_buffers: |
2c488db2 KS |
4446 | /* Swap primary and spare array */ |
4447 | thresholds->spare = thresholds->primary; | |
8c757763 | 4448 | |
2c488db2 | 4449 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4450 | |
907860ed | 4451 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4452 | synchronize_rcu(); |
6611d8d7 MC |
4453 | |
4454 | /* If all events are unregistered, free the spare array */ | |
4455 | if (!new) { | |
4456 | kfree(thresholds->spare); | |
4457 | thresholds->spare = NULL; | |
4458 | } | |
371528ca | 4459 | unlock: |
2e72b634 | 4460 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4461 | } |
c1e862c1 | 4462 | |
59b6f873 | 4463 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4464 | struct eventfd_ctx *eventfd) |
4465 | { | |
59b6f873 | 4466 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4467 | } |
4468 | ||
59b6f873 | 4469 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4470 | struct eventfd_ctx *eventfd) |
4471 | { | |
59b6f873 | 4472 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4473 | } |
4474 | ||
59b6f873 | 4475 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4476 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4477 | { |
9490ff27 | 4478 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4479 | |
9490ff27 KH |
4480 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4481 | if (!event) | |
4482 | return -ENOMEM; | |
4483 | ||
1af8efe9 | 4484 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4485 | |
4486 | event->eventfd = eventfd; | |
4487 | list_add(&event->list, &memcg->oom_notify); | |
4488 | ||
4489 | /* already in OOM ? */ | |
c2b42d3c | 4490 | if (memcg->under_oom) |
9490ff27 | 4491 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4492 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4493 | |
4494 | return 0; | |
4495 | } | |
4496 | ||
59b6f873 | 4497 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4498 | struct eventfd_ctx *eventfd) |
9490ff27 | 4499 | { |
9490ff27 | 4500 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4501 | |
1af8efe9 | 4502 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4503 | |
c0ff4b85 | 4504 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4505 | if (ev->eventfd == eventfd) { |
4506 | list_del(&ev->list); | |
4507 | kfree(ev); | |
4508 | } | |
4509 | } | |
4510 | ||
1af8efe9 | 4511 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4512 | } |
4513 | ||
2da8ca82 | 4514 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4515 | { |
aa9694bb | 4516 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4517 | |
17c56de6 | 4518 | seq_printf(sf, "oom_kill_disable %d\n", READ_ONCE(memcg->oom_kill_disable)); |
c2b42d3c | 4519 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4520 | seq_printf(sf, "oom_kill %lu\n", |
4521 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4522 | return 0; |
4523 | } | |
4524 | ||
182446d0 | 4525 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4526 | struct cftype *cft, u64 val) |
4527 | { | |
182446d0 | 4528 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4529 | |
4530 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4531 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4532 | return -EINVAL; |
4533 | ||
17c56de6 | 4534 | WRITE_ONCE(memcg->oom_kill_disable, val); |
4d845ebf | 4535 | if (!val) |
c0ff4b85 | 4536 | memcg_oom_recover(memcg); |
3dae7fec | 4537 | |
3c11ecf4 KH |
4538 | return 0; |
4539 | } | |
4540 | ||
52ebea74 TH |
4541 | #ifdef CONFIG_CGROUP_WRITEBACK |
4542 | ||
3a8e9ac8 TH |
4543 | #include <trace/events/writeback.h> |
4544 | ||
841710aa TH |
4545 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4546 | { | |
4547 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4548 | } | |
4549 | ||
4550 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4551 | { | |
4552 | wb_domain_exit(&memcg->cgwb_domain); | |
4553 | } | |
4554 | ||
2529bb3a TH |
4555 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4556 | { | |
4557 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4558 | } | |
4559 | ||
841710aa TH |
4560 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4561 | { | |
4562 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4563 | ||
4564 | if (!memcg->css.parent) | |
4565 | return NULL; | |
4566 | ||
4567 | return &memcg->cgwb_domain; | |
4568 | } | |
4569 | ||
c2aa723a TH |
4570 | /** |
4571 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4572 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4573 | * @pfilepages: out parameter for number of file pages |
4574 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4575 | * @pdirty: out parameter for number of dirty pages |
4576 | * @pwriteback: out parameter for number of pages under writeback | |
4577 | * | |
c5edf9cd TH |
4578 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4579 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4580 | * is a bit more involved. | |
c2aa723a | 4581 | * |
c5edf9cd TH |
4582 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4583 | * headroom is calculated as the lowest headroom of itself and the | |
4584 | * ancestors. Note that this doesn't consider the actual amount of | |
4585 | * available memory in the system. The caller should further cap | |
4586 | * *@pheadroom accordingly. | |
c2aa723a | 4587 | */ |
c5edf9cd TH |
4588 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4589 | unsigned long *pheadroom, unsigned long *pdirty, | |
4590 | unsigned long *pwriteback) | |
c2aa723a TH |
4591 | { |
4592 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4593 | struct mem_cgroup *parent; | |
c2aa723a | 4594 | |
fd25a9e0 | 4595 | mem_cgroup_flush_stats(); |
c2aa723a | 4596 | |
2d146aa3 JW |
4597 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
4598 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); | |
4599 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + | |
4600 | memcg_page_state(memcg, NR_ACTIVE_FILE); | |
c2aa723a | 4601 | |
2d146aa3 | 4602 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4603 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4604 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4605 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4606 | unsigned long used = page_counter_read(&memcg->memory); |
4607 | ||
c5edf9cd | 4608 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4609 | memcg = parent; |
4610 | } | |
c2aa723a TH |
4611 | } |
4612 | ||
97b27821 TH |
4613 | /* |
4614 | * Foreign dirty flushing | |
4615 | * | |
4616 | * There's an inherent mismatch between memcg and writeback. The former | |
f0953a1b | 4617 | * tracks ownership per-page while the latter per-inode. This was a |
97b27821 TH |
4618 | * deliberate design decision because honoring per-page ownership in the |
4619 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4620 | * and deemed unnecessary given that write-sharing an inode across | |
4621 | * different cgroups isn't a common use-case. | |
4622 | * | |
4623 | * Combined with inode majority-writer ownership switching, this works well | |
4624 | * enough in most cases but there are some pathological cases. For | |
4625 | * example, let's say there are two cgroups A and B which keep writing to | |
4626 | * different but confined parts of the same inode. B owns the inode and | |
4627 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4628 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4629 | * triggering background writeback. A will be slowed down without a way to | |
4630 | * make writeback of the dirty pages happen. | |
4631 | * | |
f0953a1b | 4632 | * Conditions like the above can lead to a cgroup getting repeatedly and |
97b27821 | 4633 | * severely throttled after making some progress after each |
f0953a1b | 4634 | * dirty_expire_interval while the underlying IO device is almost |
97b27821 TH |
4635 | * completely idle. |
4636 | * | |
4637 | * Solving this problem completely requires matching the ownership tracking | |
4638 | * granularities between memcg and writeback in either direction. However, | |
4639 | * the more egregious behaviors can be avoided by simply remembering the | |
4640 | * most recent foreign dirtying events and initiating remote flushes on | |
4641 | * them when local writeback isn't enough to keep the memory clean enough. | |
4642 | * | |
4643 | * The following two functions implement such mechanism. When a foreign | |
4644 | * page - a page whose memcg and writeback ownerships don't match - is | |
4645 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4646 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4647 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4648 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4649 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4650 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4651 | * limited to MEMCG_CGWB_FRN_CNT. | |
4652 | * | |
4653 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4654 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4655 | * records are lockless and racy. | |
4656 | */ | |
9d8053fc | 4657 | void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio, |
97b27821 TH |
4658 | struct bdi_writeback *wb) |
4659 | { | |
9d8053fc | 4660 | struct mem_cgroup *memcg = folio_memcg(folio); |
97b27821 TH |
4661 | struct memcg_cgwb_frn *frn; |
4662 | u64 now = get_jiffies_64(); | |
4663 | u64 oldest_at = now; | |
4664 | int oldest = -1; | |
4665 | int i; | |
4666 | ||
9d8053fc | 4667 | trace_track_foreign_dirty(folio, wb); |
3a8e9ac8 | 4668 | |
97b27821 TH |
4669 | /* |
4670 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4671 | * using it. If not replace the oldest one which isn't being | |
4672 | * written out. | |
4673 | */ | |
4674 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4675 | frn = &memcg->cgwb_frn[i]; | |
4676 | if (frn->bdi_id == wb->bdi->id && | |
4677 | frn->memcg_id == wb->memcg_css->id) | |
4678 | break; | |
4679 | if (time_before64(frn->at, oldest_at) && | |
4680 | atomic_read(&frn->done.cnt) == 1) { | |
4681 | oldest = i; | |
4682 | oldest_at = frn->at; | |
4683 | } | |
4684 | } | |
4685 | ||
4686 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4687 | /* | |
4688 | * Re-using an existing one. Update timestamp lazily to | |
4689 | * avoid making the cacheline hot. We want them to be | |
4690 | * reasonably up-to-date and significantly shorter than | |
4691 | * dirty_expire_interval as that's what expires the record. | |
4692 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4693 | */ | |
4694 | unsigned long update_intv = | |
4695 | min_t(unsigned long, HZ, | |
4696 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4697 | ||
4698 | if (time_before64(frn->at, now - update_intv)) | |
4699 | frn->at = now; | |
4700 | } else if (oldest >= 0) { | |
4701 | /* replace the oldest free one */ | |
4702 | frn = &memcg->cgwb_frn[oldest]; | |
4703 | frn->bdi_id = wb->bdi->id; | |
4704 | frn->memcg_id = wb->memcg_css->id; | |
4705 | frn->at = now; | |
4706 | } | |
4707 | } | |
4708 | ||
4709 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4710 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4711 | { | |
4712 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4713 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4714 | u64 now = jiffies_64; | |
4715 | int i; | |
4716 | ||
4717 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4718 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4719 | ||
4720 | /* | |
4721 | * If the record is older than dirty_expire_interval, | |
4722 | * writeback on it has already started. No need to kick it | |
4723 | * off again. Also, don't start a new one if there's | |
4724 | * already one in flight. | |
4725 | */ | |
4726 | if (time_after64(frn->at, now - intv) && | |
4727 | atomic_read(&frn->done.cnt) == 1) { | |
4728 | frn->at = 0; | |
3a8e9ac8 | 4729 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
7490a2d2 | 4730 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, |
97b27821 TH |
4731 | WB_REASON_FOREIGN_FLUSH, |
4732 | &frn->done); | |
4733 | } | |
4734 | } | |
4735 | } | |
4736 | ||
841710aa TH |
4737 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4738 | ||
4739 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4740 | { | |
4741 | return 0; | |
4742 | } | |
4743 | ||
4744 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4745 | { | |
4746 | } | |
4747 | ||
2529bb3a TH |
4748 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4749 | { | |
4750 | } | |
4751 | ||
52ebea74 TH |
4752 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4753 | ||
3bc942f3 TH |
4754 | /* |
4755 | * DO NOT USE IN NEW FILES. | |
4756 | * | |
4757 | * "cgroup.event_control" implementation. | |
4758 | * | |
4759 | * This is way over-engineered. It tries to support fully configurable | |
4760 | * events for each user. Such level of flexibility is completely | |
4761 | * unnecessary especially in the light of the planned unified hierarchy. | |
4762 | * | |
4763 | * Please deprecate this and replace with something simpler if at all | |
4764 | * possible. | |
4765 | */ | |
4766 | ||
79bd9814 TH |
4767 | /* |
4768 | * Unregister event and free resources. | |
4769 | * | |
4770 | * Gets called from workqueue. | |
4771 | */ | |
3bc942f3 | 4772 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4773 | { |
3bc942f3 TH |
4774 | struct mem_cgroup_event *event = |
4775 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4776 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4777 | |
4778 | remove_wait_queue(event->wqh, &event->wait); | |
4779 | ||
59b6f873 | 4780 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4781 | |
4782 | /* Notify userspace the event is going away. */ | |
4783 | eventfd_signal(event->eventfd, 1); | |
4784 | ||
4785 | eventfd_ctx_put(event->eventfd); | |
4786 | kfree(event); | |
59b6f873 | 4787 | css_put(&memcg->css); |
79bd9814 TH |
4788 | } |
4789 | ||
4790 | /* | |
a9a08845 | 4791 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4792 | * |
4793 | * Called with wqh->lock held and interrupts disabled. | |
4794 | */ | |
ac6424b9 | 4795 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4796 | int sync, void *key) |
79bd9814 | 4797 | { |
3bc942f3 TH |
4798 | struct mem_cgroup_event *event = |
4799 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4800 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4801 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4802 | |
a9a08845 | 4803 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4804 | /* |
4805 | * If the event has been detached at cgroup removal, we | |
4806 | * can simply return knowing the other side will cleanup | |
4807 | * for us. | |
4808 | * | |
4809 | * We can't race against event freeing since the other | |
4810 | * side will require wqh->lock via remove_wait_queue(), | |
4811 | * which we hold. | |
4812 | */ | |
fba94807 | 4813 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4814 | if (!list_empty(&event->list)) { |
4815 | list_del_init(&event->list); | |
4816 | /* | |
4817 | * We are in atomic context, but cgroup_event_remove() | |
4818 | * may sleep, so we have to call it in workqueue. | |
4819 | */ | |
4820 | schedule_work(&event->remove); | |
4821 | } | |
fba94807 | 4822 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4823 | } |
4824 | ||
4825 | return 0; | |
4826 | } | |
4827 | ||
3bc942f3 | 4828 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4829 | wait_queue_head_t *wqh, poll_table *pt) |
4830 | { | |
3bc942f3 TH |
4831 | struct mem_cgroup_event *event = |
4832 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4833 | |
4834 | event->wqh = wqh; | |
4835 | add_wait_queue(wqh, &event->wait); | |
4836 | } | |
4837 | ||
4838 | /* | |
3bc942f3 TH |
4839 | * DO NOT USE IN NEW FILES. |
4840 | * | |
79bd9814 TH |
4841 | * Parse input and register new cgroup event handler. |
4842 | * | |
4843 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4844 | * Interpretation of args is defined by control file implementation. | |
4845 | */ | |
451af504 TH |
4846 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4847 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4848 | { |
451af504 | 4849 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4850 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4851 | struct mem_cgroup_event *event; |
79bd9814 TH |
4852 | struct cgroup_subsys_state *cfile_css; |
4853 | unsigned int efd, cfd; | |
4854 | struct fd efile; | |
4855 | struct fd cfile; | |
4a7ba45b | 4856 | struct dentry *cdentry; |
fba94807 | 4857 | const char *name; |
79bd9814 TH |
4858 | char *endp; |
4859 | int ret; | |
4860 | ||
2343e88d SAS |
4861 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
4862 | return -EOPNOTSUPP; | |
4863 | ||
451af504 TH |
4864 | buf = strstrip(buf); |
4865 | ||
4866 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4867 | if (*endp != ' ') |
4868 | return -EINVAL; | |
451af504 | 4869 | buf = endp + 1; |
79bd9814 | 4870 | |
451af504 | 4871 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4872 | if ((*endp != ' ') && (*endp != '\0')) |
4873 | return -EINVAL; | |
451af504 | 4874 | buf = endp + 1; |
79bd9814 TH |
4875 | |
4876 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4877 | if (!event) | |
4878 | return -ENOMEM; | |
4879 | ||
59b6f873 | 4880 | event->memcg = memcg; |
79bd9814 | 4881 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4882 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4883 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4884 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4885 | |
4886 | efile = fdget(efd); | |
4887 | if (!efile.file) { | |
4888 | ret = -EBADF; | |
4889 | goto out_kfree; | |
4890 | } | |
4891 | ||
4892 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4893 | if (IS_ERR(event->eventfd)) { | |
4894 | ret = PTR_ERR(event->eventfd); | |
4895 | goto out_put_efile; | |
4896 | } | |
4897 | ||
4898 | cfile = fdget(cfd); | |
4899 | if (!cfile.file) { | |
4900 | ret = -EBADF; | |
4901 | goto out_put_eventfd; | |
4902 | } | |
4903 | ||
4904 | /* the process need read permission on control file */ | |
4905 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 4906 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
4907 | if (ret < 0) |
4908 | goto out_put_cfile; | |
4909 | ||
4a7ba45b TH |
4910 | /* |
4911 | * The control file must be a regular cgroup1 file. As a regular cgroup | |
4912 | * file can't be renamed, it's safe to access its name afterwards. | |
4913 | */ | |
4914 | cdentry = cfile.file->f_path.dentry; | |
4915 | if (cdentry->d_sb->s_type != &cgroup_fs_type || !d_is_reg(cdentry)) { | |
4916 | ret = -EINVAL; | |
4917 | goto out_put_cfile; | |
4918 | } | |
4919 | ||
fba94807 TH |
4920 | /* |
4921 | * Determine the event callbacks and set them in @event. This used | |
4922 | * to be done via struct cftype but cgroup core no longer knows | |
4923 | * about these events. The following is crude but the whole thing | |
4924 | * is for compatibility anyway. | |
3bc942f3 TH |
4925 | * |
4926 | * DO NOT ADD NEW FILES. | |
fba94807 | 4927 | */ |
4a7ba45b | 4928 | name = cdentry->d_name.name; |
fba94807 TH |
4929 | |
4930 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4931 | event->register_event = mem_cgroup_usage_register_event; | |
4932 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4933 | } else if (!strcmp(name, "memory.oom_control")) { | |
4934 | event->register_event = mem_cgroup_oom_register_event; | |
4935 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4936 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4937 | event->register_event = vmpressure_register_event; | |
4938 | event->unregister_event = vmpressure_unregister_event; | |
4939 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4940 | event->register_event = memsw_cgroup_usage_register_event; |
4941 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4942 | } else { |
4943 | ret = -EINVAL; | |
4944 | goto out_put_cfile; | |
4945 | } | |
4946 | ||
79bd9814 | 4947 | /* |
b5557c4c TH |
4948 | * Verify @cfile should belong to @css. Also, remaining events are |
4949 | * automatically removed on cgroup destruction but the removal is | |
4950 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4951 | */ |
4a7ba45b | 4952 | cfile_css = css_tryget_online_from_dir(cdentry->d_parent, |
ec903c0c | 4953 | &memory_cgrp_subsys); |
79bd9814 | 4954 | ret = -EINVAL; |
5a17f543 | 4955 | if (IS_ERR(cfile_css)) |
79bd9814 | 4956 | goto out_put_cfile; |
5a17f543 TH |
4957 | if (cfile_css != css) { |
4958 | css_put(cfile_css); | |
79bd9814 | 4959 | goto out_put_cfile; |
5a17f543 | 4960 | } |
79bd9814 | 4961 | |
451af504 | 4962 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4963 | if (ret) |
4964 | goto out_put_css; | |
4965 | ||
9965ed17 | 4966 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4967 | |
4ba9515d | 4968 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 4969 | list_add(&event->list, &memcg->event_list); |
4ba9515d | 4970 | spin_unlock_irq(&memcg->event_list_lock); |
79bd9814 TH |
4971 | |
4972 | fdput(cfile); | |
4973 | fdput(efile); | |
4974 | ||
451af504 | 4975 | return nbytes; |
79bd9814 TH |
4976 | |
4977 | out_put_css: | |
b5557c4c | 4978 | css_put(css); |
79bd9814 TH |
4979 | out_put_cfile: |
4980 | fdput(cfile); | |
4981 | out_put_eventfd: | |
4982 | eventfd_ctx_put(event->eventfd); | |
4983 | out_put_efile: | |
4984 | fdput(efile); | |
4985 | out_kfree: | |
4986 | kfree(event); | |
4987 | ||
4988 | return ret; | |
4989 | } | |
4990 | ||
c29b5b3d MS |
4991 | #if defined(CONFIG_MEMCG_KMEM) && (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) |
4992 | static int mem_cgroup_slab_show(struct seq_file *m, void *p) | |
4993 | { | |
4994 | /* | |
4995 | * Deprecated. | |
df4ae285 | 4996 | * Please, take a look at tools/cgroup/memcg_slabinfo.py . |
c29b5b3d MS |
4997 | */ |
4998 | return 0; | |
4999 | } | |
5000 | #endif | |
5001 | ||
241994ed | 5002 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 5003 | { |
0eea1030 | 5004 | .name = "usage_in_bytes", |
8c7c6e34 | 5005 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5006 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5007 | }, |
c84872e1 PE |
5008 | { |
5009 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5010 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5011 | .write = mem_cgroup_reset, |
791badbd | 5012 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5013 | }, |
8cdea7c0 | 5014 | { |
0eea1030 | 5015 | .name = "limit_in_bytes", |
8c7c6e34 | 5016 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5017 | .write = mem_cgroup_write, |
791badbd | 5018 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5019 | }, |
296c81d8 BS |
5020 | { |
5021 | .name = "soft_limit_in_bytes", | |
5022 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5023 | .write = mem_cgroup_write, |
791badbd | 5024 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5025 | }, |
8cdea7c0 BS |
5026 | { |
5027 | .name = "failcnt", | |
8c7c6e34 | 5028 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5029 | .write = mem_cgroup_reset, |
791badbd | 5030 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5031 | }, |
d2ceb9b7 KH |
5032 | { |
5033 | .name = "stat", | |
2da8ca82 | 5034 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5035 | }, |
c1e862c1 KH |
5036 | { |
5037 | .name = "force_empty", | |
6770c64e | 5038 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5039 | }, |
18f59ea7 BS |
5040 | { |
5041 | .name = "use_hierarchy", | |
5042 | .write_u64 = mem_cgroup_hierarchy_write, | |
5043 | .read_u64 = mem_cgroup_hierarchy_read, | |
5044 | }, | |
79bd9814 | 5045 | { |
3bc942f3 | 5046 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5047 | .write = memcg_write_event_control, |
7dbdb199 | 5048 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 5049 | }, |
a7885eb8 KM |
5050 | { |
5051 | .name = "swappiness", | |
5052 | .read_u64 = mem_cgroup_swappiness_read, | |
5053 | .write_u64 = mem_cgroup_swappiness_write, | |
5054 | }, | |
7dc74be0 DN |
5055 | { |
5056 | .name = "move_charge_at_immigrate", | |
5057 | .read_u64 = mem_cgroup_move_charge_read, | |
5058 | .write_u64 = mem_cgroup_move_charge_write, | |
5059 | }, | |
9490ff27 KH |
5060 | { |
5061 | .name = "oom_control", | |
2da8ca82 | 5062 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5063 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 | 5064 | }, |
70ddf637 AV |
5065 | { |
5066 | .name = "pressure_level", | |
70ddf637 | 5067 | }, |
406eb0c9 YH |
5068 | #ifdef CONFIG_NUMA |
5069 | { | |
5070 | .name = "numa_stat", | |
2da8ca82 | 5071 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5072 | }, |
5073 | #endif | |
510fc4e1 GC |
5074 | { |
5075 | .name = "kmem.limit_in_bytes", | |
5076 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5077 | .write = mem_cgroup_write, |
791badbd | 5078 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5079 | }, |
5080 | { | |
5081 | .name = "kmem.usage_in_bytes", | |
5082 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5083 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5084 | }, |
5085 | { | |
5086 | .name = "kmem.failcnt", | |
5087 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5088 | .write = mem_cgroup_reset, |
791badbd | 5089 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5090 | }, |
5091 | { | |
5092 | .name = "kmem.max_usage_in_bytes", | |
5093 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5094 | .write = mem_cgroup_reset, |
791badbd | 5095 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5096 | }, |
a87425a3 YS |
5097 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5098 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5099 | { |
5100 | .name = "kmem.slabinfo", | |
c29b5b3d | 5101 | .seq_show = mem_cgroup_slab_show, |
749c5415 GC |
5102 | }, |
5103 | #endif | |
d55f90bf VD |
5104 | { |
5105 | .name = "kmem.tcp.limit_in_bytes", | |
5106 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5107 | .write = mem_cgroup_write, | |
5108 | .read_u64 = mem_cgroup_read_u64, | |
5109 | }, | |
5110 | { | |
5111 | .name = "kmem.tcp.usage_in_bytes", | |
5112 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5113 | .read_u64 = mem_cgroup_read_u64, | |
5114 | }, | |
5115 | { | |
5116 | .name = "kmem.tcp.failcnt", | |
5117 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5118 | .write = mem_cgroup_reset, | |
5119 | .read_u64 = mem_cgroup_read_u64, | |
5120 | }, | |
5121 | { | |
5122 | .name = "kmem.tcp.max_usage_in_bytes", | |
5123 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5124 | .write = mem_cgroup_reset, | |
5125 | .read_u64 = mem_cgroup_read_u64, | |
5126 | }, | |
6bc10349 | 5127 | { }, /* terminate */ |
af36f906 | 5128 | }; |
8c7c6e34 | 5129 | |
73f576c0 JW |
5130 | /* |
5131 | * Private memory cgroup IDR | |
5132 | * | |
5133 | * Swap-out records and page cache shadow entries need to store memcg | |
5134 | * references in constrained space, so we maintain an ID space that is | |
5135 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5136 | * memory-controlled cgroups to 64k. | |
5137 | * | |
b8f2935f | 5138 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5139 | * the cgroup has been destroyed, such as page cache or reclaimable |
5140 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5141 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5142 | * relatively small ID space and prevent the creation of new cgroups | |
5143 | * even when there are much fewer than 64k cgroups - possibly none. | |
5144 | * | |
5145 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5146 | * be freed and recycled when it's no longer needed, which is usually | |
5147 | * when the CSS is offlined. | |
5148 | * | |
5149 | * The only exception to that are records of swapped out tmpfs/shmem | |
5150 | * pages that need to be attributed to live ancestors on swapin. But | |
5151 | * those references are manageable from userspace. | |
5152 | */ | |
5153 | ||
5154 | static DEFINE_IDR(mem_cgroup_idr); | |
5155 | ||
7e97de0b KT |
5156 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5157 | { | |
5158 | if (memcg->id.id > 0) { | |
5159 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5160 | memcg->id.id = 0; | |
5161 | } | |
5162 | } | |
5163 | ||
c1514c0a VF |
5164 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5165 | unsigned int n) | |
73f576c0 | 5166 | { |
1c2d479a | 5167 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5168 | } |
5169 | ||
615d66c3 | 5170 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5171 | { |
1c2d479a | 5172 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5173 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5174 | |
5175 | /* Memcg ID pins CSS */ | |
5176 | css_put(&memcg->css); | |
5177 | } | |
5178 | } | |
5179 | ||
615d66c3 VD |
5180 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5181 | { | |
5182 | mem_cgroup_id_put_many(memcg, 1); | |
5183 | } | |
5184 | ||
73f576c0 JW |
5185 | /** |
5186 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5187 | * @id: the memcg id to look up | |
5188 | * | |
5189 | * Caller must hold rcu_read_lock(). | |
5190 | */ | |
5191 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5192 | { | |
5193 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5194 | return idr_find(&mem_cgroup_idr, id); | |
5195 | } | |
5196 | ||
c15187a4 RG |
5197 | #ifdef CONFIG_SHRINKER_DEBUG |
5198 | struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino) | |
5199 | { | |
5200 | struct cgroup *cgrp; | |
5201 | struct cgroup_subsys_state *css; | |
5202 | struct mem_cgroup *memcg; | |
5203 | ||
5204 | cgrp = cgroup_get_from_id(ino); | |
fa7e439c | 5205 | if (IS_ERR(cgrp)) |
c0f2df49 | 5206 | return ERR_CAST(cgrp); |
c15187a4 RG |
5207 | |
5208 | css = cgroup_get_e_css(cgrp, &memory_cgrp_subsys); | |
5209 | if (css) | |
5210 | memcg = container_of(css, struct mem_cgroup, css); | |
5211 | else | |
5212 | memcg = ERR_PTR(-ENOENT); | |
5213 | ||
5214 | cgroup_put(cgrp); | |
5215 | ||
5216 | return memcg; | |
5217 | } | |
5218 | #endif | |
5219 | ||
ef8f2327 | 5220 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5221 | { |
5222 | struct mem_cgroup_per_node *pn; | |
8c9bb398 WY |
5223 | |
5224 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
6d12e2d8 KH |
5225 | if (!pn) |
5226 | return 1; | |
1ecaab2b | 5227 | |
7e1c0d6f SB |
5228 | pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu, |
5229 | GFP_KERNEL_ACCOUNT); | |
5230 | if (!pn->lruvec_stats_percpu) { | |
00f3ca2c JW |
5231 | kfree(pn); |
5232 | return 1; | |
5233 | } | |
5234 | ||
ef8f2327 | 5235 | lruvec_init(&pn->lruvec); |
ef8f2327 MG |
5236 | pn->memcg = memcg; |
5237 | ||
54f72fe0 | 5238 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5239 | return 0; |
5240 | } | |
5241 | ||
ef8f2327 | 5242 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5243 | { |
00f3ca2c JW |
5244 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5245 | ||
4eaf431f MH |
5246 | if (!pn) |
5247 | return; | |
5248 | ||
7e1c0d6f | 5249 | free_percpu(pn->lruvec_stats_percpu); |
00f3ca2c | 5250 | kfree(pn); |
1ecaab2b KH |
5251 | } |
5252 | ||
40e952f9 | 5253 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5254 | { |
c8b2a36f | 5255 | int node; |
59927fb9 | 5256 | |
c8b2a36f | 5257 | for_each_node(node) |
ef8f2327 | 5258 | free_mem_cgroup_per_node_info(memcg, node); |
410f8e82 | 5259 | kfree(memcg->vmstats); |
871789d4 | 5260 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 5261 | kfree(memcg); |
59927fb9 | 5262 | } |
3afe36b1 | 5263 | |
40e952f9 TE |
5264 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5265 | { | |
ec1c86b2 | 5266 | lru_gen_exit_memcg(memcg); |
40e952f9 TE |
5267 | memcg_wb_domain_exit(memcg); |
5268 | __mem_cgroup_free(memcg); | |
5269 | } | |
5270 | ||
0b8f73e1 | 5271 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5272 | { |
d142e3e6 | 5273 | struct mem_cgroup *memcg; |
6d12e2d8 | 5274 | int node; |
97b27821 | 5275 | int __maybe_unused i; |
11d67612 | 5276 | long error = -ENOMEM; |
8cdea7c0 | 5277 | |
06b2c3b0 | 5278 | memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL); |
c0ff4b85 | 5279 | if (!memcg) |
11d67612 | 5280 | return ERR_PTR(error); |
0b8f73e1 | 5281 | |
73f576c0 | 5282 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
be740503 | 5283 | 1, MEM_CGROUP_ID_MAX + 1, GFP_KERNEL); |
11d67612 YS |
5284 | if (memcg->id.id < 0) { |
5285 | error = memcg->id.id; | |
73f576c0 | 5286 | goto fail; |
11d67612 | 5287 | } |
73f576c0 | 5288 | |
410f8e82 SB |
5289 | memcg->vmstats = kzalloc(sizeof(struct memcg_vmstats), GFP_KERNEL); |
5290 | if (!memcg->vmstats) | |
5291 | goto fail; | |
5292 | ||
3e38e0aa RG |
5293 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5294 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5295 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5296 | goto fail; |
78fb7466 | 5297 | |
3ed28fa1 | 5298 | for_each_node(node) |
ef8f2327 | 5299 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5300 | goto fail; |
f64c3f54 | 5301 | |
0b8f73e1 JW |
5302 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5303 | goto fail; | |
28dbc4b6 | 5304 | |
f7e1cb6e | 5305 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5306 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5307 | mutex_init(&memcg->thresholds_lock); |
5308 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5309 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5310 | INIT_LIST_HEAD(&memcg->event_list); |
5311 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5312 | memcg->socket_pressure = jiffies; |
84c07d11 | 5313 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5314 | memcg->kmemcg_id = -1; |
bf4f0599 | 5315 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5316 | #endif |
52ebea74 TH |
5317 | #ifdef CONFIG_CGROUP_WRITEBACK |
5318 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5319 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5320 | memcg->cgwb_frn[i].done = | |
5321 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5322 | #endif |
5323 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5324 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5325 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5326 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5327 | #endif |
73f576c0 | 5328 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
ec1c86b2 | 5329 | lru_gen_init_memcg(memcg); |
0b8f73e1 JW |
5330 | return memcg; |
5331 | fail: | |
7e97de0b | 5332 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5333 | __mem_cgroup_free(memcg); |
11d67612 | 5334 | return ERR_PTR(error); |
d142e3e6 GC |
5335 | } |
5336 | ||
0b8f73e1 JW |
5337 | static struct cgroup_subsys_state * __ref |
5338 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5339 | { |
0b8f73e1 | 5340 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5341 | struct mem_cgroup *memcg, *old_memcg; |
d142e3e6 | 5342 | |
b87d8cef | 5343 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5344 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5345 | set_active_memcg(old_memcg); |
11d67612 YS |
5346 | if (IS_ERR(memcg)) |
5347 | return ERR_CAST(memcg); | |
d142e3e6 | 5348 | |
d1663a90 | 5349 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
2178e20c | 5350 | WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX); |
f4840ccf JW |
5351 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
5352 | memcg->zswap_max = PAGE_COUNTER_MAX; | |
5353 | #endif | |
4b82ab4f | 5354 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 | 5355 | if (parent) { |
82b3aa26 | 5356 | WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent)); |
17c56de6 | 5357 | WRITE_ONCE(memcg->oom_kill_disable, READ_ONCE(parent->oom_kill_disable)); |
bef8620c | 5358 | |
3e32cb2e | 5359 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5360 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5361 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5362 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5363 | } else { |
8278f1c7 | 5364 | init_memcg_events(); |
bef8620c RG |
5365 | page_counter_init(&memcg->memory, NULL); |
5366 | page_counter_init(&memcg->swap, NULL); | |
5367 | page_counter_init(&memcg->kmem, NULL); | |
5368 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5369 | |
0b8f73e1 JW |
5370 | root_mem_cgroup = memcg; |
5371 | return &memcg->css; | |
5372 | } | |
5373 | ||
f7e1cb6e | 5374 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5375 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5376 | |
b6c1a8af YS |
5377 | #if defined(CONFIG_MEMCG_KMEM) |
5378 | if (!cgroup_memory_nobpf) | |
5379 | static_branch_inc(&memcg_bpf_enabled_key); | |
5380 | #endif | |
5381 | ||
0b8f73e1 | 5382 | return &memcg->css; |
0b8f73e1 JW |
5383 | } |
5384 | ||
73f576c0 | 5385 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5386 | { |
58fa2a55 VD |
5387 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5388 | ||
da0efe30 MS |
5389 | if (memcg_online_kmem(memcg)) |
5390 | goto remove_id; | |
5391 | ||
0a4465d3 | 5392 | /* |
e4262c4f | 5393 | * A memcg must be visible for expand_shrinker_info() |
0a4465d3 KT |
5394 | * by the time the maps are allocated. So, we allocate maps |
5395 | * here, when for_each_mem_cgroup() can't skip it. | |
5396 | */ | |
da0efe30 MS |
5397 | if (alloc_shrinker_info(memcg)) |
5398 | goto offline_kmem; | |
0a4465d3 | 5399 | |
73f576c0 | 5400 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5401 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5402 | css_get(css); |
aa48e47e SB |
5403 | |
5404 | if (unlikely(mem_cgroup_is_root(memcg))) | |
5405 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, | |
5406 | 2UL*HZ); | |
e4dde56c | 5407 | lru_gen_online_memcg(memcg); |
2f7dd7a4 | 5408 | return 0; |
da0efe30 MS |
5409 | offline_kmem: |
5410 | memcg_offline_kmem(memcg); | |
5411 | remove_id: | |
5412 | mem_cgroup_id_remove(memcg); | |
5413 | return -ENOMEM; | |
8cdea7c0 BS |
5414 | } |
5415 | ||
eb95419b | 5416 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5417 | { |
eb95419b | 5418 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5419 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5420 | |
5421 | /* | |
5422 | * Unregister events and notify userspace. | |
5423 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5424 | * directory to avoid race between userspace and kernelspace. | |
5425 | */ | |
4ba9515d | 5426 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5427 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
79bd9814 TH |
5428 | list_del_init(&event->list); |
5429 | schedule_work(&event->remove); | |
5430 | } | |
4ba9515d | 5431 | spin_unlock_irq(&memcg->event_list_lock); |
ec64f515 | 5432 | |
bf8d5d52 | 5433 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5434 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5435 | |
567e9ab2 | 5436 | memcg_offline_kmem(memcg); |
a178015c | 5437 | reparent_shrinker_deferred(memcg); |
52ebea74 | 5438 | wb_memcg_offline(memcg); |
e4dde56c | 5439 | lru_gen_offline_memcg(memcg); |
73f576c0 | 5440 | |
591edfb1 RG |
5441 | drain_all_stock(memcg); |
5442 | ||
73f576c0 | 5443 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5444 | } |
5445 | ||
6df38689 VD |
5446 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5447 | { | |
5448 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5449 | ||
5450 | invalidate_reclaim_iterators(memcg); | |
e4dde56c | 5451 | lru_gen_release_memcg(memcg); |
6df38689 VD |
5452 | } |
5453 | ||
eb95419b | 5454 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5455 | { |
eb95419b | 5456 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5457 | int __maybe_unused i; |
c268e994 | 5458 | |
97b27821 TH |
5459 | #ifdef CONFIG_CGROUP_WRITEBACK |
5460 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5461 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5462 | #endif | |
f7e1cb6e | 5463 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5464 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5465 | |
0db15298 | 5466 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5467 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5468 | |
b6c1a8af YS |
5469 | #if defined(CONFIG_MEMCG_KMEM) |
5470 | if (!cgroup_memory_nobpf) | |
5471 | static_branch_dec(&memcg_bpf_enabled_key); | |
5472 | #endif | |
5473 | ||
0b8f73e1 JW |
5474 | vmpressure_cleanup(&memcg->vmpressure); |
5475 | cancel_work_sync(&memcg->high_work); | |
5476 | mem_cgroup_remove_from_trees(memcg); | |
e4262c4f | 5477 | free_shrinker_info(memcg); |
0b8f73e1 | 5478 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5479 | } |
5480 | ||
1ced953b TH |
5481 | /** |
5482 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5483 | * @css: the target css | |
5484 | * | |
5485 | * Reset the states of the mem_cgroup associated with @css. This is | |
5486 | * invoked when the userland requests disabling on the default hierarchy | |
5487 | * but the memcg is pinned through dependency. The memcg should stop | |
5488 | * applying policies and should revert to the vanilla state as it may be | |
5489 | * made visible again. | |
5490 | * | |
5491 | * The current implementation only resets the essential configurations. | |
5492 | * This needs to be expanded to cover all the visible parts. | |
5493 | */ | |
5494 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5495 | { | |
5496 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5497 | ||
bbec2e15 RG |
5498 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5499 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5500 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5501 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5502 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5503 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5504 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
2178e20c | 5505 | WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX); |
4b82ab4f | 5506 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5507 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5508 | } |
5509 | ||
2d146aa3 JW |
5510 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
5511 | { | |
5512 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5513 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5514 | struct memcg_vmstats_percpu *statc; | |
5515 | long delta, v; | |
7e1c0d6f | 5516 | int i, nid; |
2d146aa3 JW |
5517 | |
5518 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5519 | ||
5520 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
5521 | /* | |
5522 | * Collect the aggregated propagation counts of groups | |
5523 | * below us. We're in a per-cpu loop here and this is | |
5524 | * a global counter, so the first cycle will get them. | |
5525 | */ | |
410f8e82 | 5526 | delta = memcg->vmstats->state_pending[i]; |
2d146aa3 | 5527 | if (delta) |
410f8e82 | 5528 | memcg->vmstats->state_pending[i] = 0; |
2d146aa3 JW |
5529 | |
5530 | /* Add CPU changes on this level since the last flush */ | |
5531 | v = READ_ONCE(statc->state[i]); | |
5532 | if (v != statc->state_prev[i]) { | |
5533 | delta += v - statc->state_prev[i]; | |
5534 | statc->state_prev[i] = v; | |
5535 | } | |
5536 | ||
5537 | if (!delta) | |
5538 | continue; | |
5539 | ||
5540 | /* Aggregate counts on this level and propagate upwards */ | |
410f8e82 | 5541 | memcg->vmstats->state[i] += delta; |
2d146aa3 | 5542 | if (parent) |
410f8e82 | 5543 | parent->vmstats->state_pending[i] += delta; |
2d146aa3 JW |
5544 | } |
5545 | ||
8278f1c7 | 5546 | for (i = 0; i < NR_MEMCG_EVENTS; i++) { |
410f8e82 | 5547 | delta = memcg->vmstats->events_pending[i]; |
2d146aa3 | 5548 | if (delta) |
410f8e82 | 5549 | memcg->vmstats->events_pending[i] = 0; |
2d146aa3 JW |
5550 | |
5551 | v = READ_ONCE(statc->events[i]); | |
5552 | if (v != statc->events_prev[i]) { | |
5553 | delta += v - statc->events_prev[i]; | |
5554 | statc->events_prev[i] = v; | |
5555 | } | |
5556 | ||
5557 | if (!delta) | |
5558 | continue; | |
5559 | ||
410f8e82 | 5560 | memcg->vmstats->events[i] += delta; |
2d146aa3 | 5561 | if (parent) |
410f8e82 | 5562 | parent->vmstats->events_pending[i] += delta; |
2d146aa3 | 5563 | } |
7e1c0d6f SB |
5564 | |
5565 | for_each_node_state(nid, N_MEMORY) { | |
5566 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; | |
5567 | struct mem_cgroup_per_node *ppn = NULL; | |
5568 | struct lruvec_stats_percpu *lstatc; | |
5569 | ||
5570 | if (parent) | |
5571 | ppn = parent->nodeinfo[nid]; | |
5572 | ||
5573 | lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu); | |
5574 | ||
5575 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { | |
5576 | delta = pn->lruvec_stats.state_pending[i]; | |
5577 | if (delta) | |
5578 | pn->lruvec_stats.state_pending[i] = 0; | |
5579 | ||
5580 | v = READ_ONCE(lstatc->state[i]); | |
5581 | if (v != lstatc->state_prev[i]) { | |
5582 | delta += v - lstatc->state_prev[i]; | |
5583 | lstatc->state_prev[i] = v; | |
5584 | } | |
5585 | ||
5586 | if (!delta) | |
5587 | continue; | |
5588 | ||
5589 | pn->lruvec_stats.state[i] += delta; | |
5590 | if (ppn) | |
5591 | ppn->lruvec_stats.state_pending[i] += delta; | |
5592 | } | |
5593 | } | |
2d146aa3 JW |
5594 | } |
5595 | ||
02491447 | 5596 | #ifdef CONFIG_MMU |
7dc74be0 | 5597 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5598 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5599 | { |
05b84301 | 5600 | int ret; |
9476db97 | 5601 | |
d0164adc MG |
5602 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5603 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5604 | if (!ret) { |
854ffa8d | 5605 | mc.precharge += count; |
854ffa8d DN |
5606 | return ret; |
5607 | } | |
9476db97 | 5608 | |
3674534b | 5609 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5610 | while (count--) { |
3674534b | 5611 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5612 | if (ret) |
38c5d72f | 5613 | return ret; |
854ffa8d | 5614 | mc.precharge++; |
9476db97 | 5615 | cond_resched(); |
854ffa8d | 5616 | } |
9476db97 | 5617 | return 0; |
4ffef5fe DN |
5618 | } |
5619 | ||
4ffef5fe DN |
5620 | union mc_target { |
5621 | struct page *page; | |
02491447 | 5622 | swp_entry_t ent; |
4ffef5fe DN |
5623 | }; |
5624 | ||
4ffef5fe | 5625 | enum mc_target_type { |
8d32ff84 | 5626 | MC_TARGET_NONE = 0, |
4ffef5fe | 5627 | MC_TARGET_PAGE, |
02491447 | 5628 | MC_TARGET_SWAP, |
c733a828 | 5629 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5630 | }; |
5631 | ||
90254a65 DN |
5632 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5633 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5634 | { |
25b2995a | 5635 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5636 | |
90254a65 DN |
5637 | if (!page || !page_mapped(page)) |
5638 | return NULL; | |
5639 | if (PageAnon(page)) { | |
1dfab5ab | 5640 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5641 | return NULL; |
1dfab5ab JW |
5642 | } else { |
5643 | if (!(mc.flags & MOVE_FILE)) | |
5644 | return NULL; | |
5645 | } | |
90254a65 DN |
5646 | if (!get_page_unless_zero(page)) |
5647 | return NULL; | |
5648 | ||
5649 | return page; | |
5650 | } | |
5651 | ||
c733a828 | 5652 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5653 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5654 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5655 | { |
90254a65 DN |
5656 | struct page *page = NULL; |
5657 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5658 | ||
9a137153 | 5659 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5660 | return NULL; |
c733a828 JG |
5661 | |
5662 | /* | |
27674ef6 CH |
5663 | * Handle device private pages that are not accessible by the CPU, but |
5664 | * stored as special swap entries in the page table. | |
c733a828 JG |
5665 | */ |
5666 | if (is_device_private_entry(ent)) { | |
af5cdaf8 | 5667 | page = pfn_swap_entry_to_page(ent); |
27674ef6 | 5668 | if (!get_page_unless_zero(page)) |
c733a828 JG |
5669 | return NULL; |
5670 | return page; | |
5671 | } | |
5672 | ||
9a137153 RC |
5673 | if (non_swap_entry(ent)) |
5674 | return NULL; | |
5675 | ||
4b91355e | 5676 | /* |
cb691e2f | 5677 | * Because swap_cache_get_folio() updates some statistics counter, |
4b91355e KH |
5678 | * we call find_get_page() with swapper_space directly. |
5679 | */ | |
f6ab1f7f | 5680 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5681 | entry->val = ent.val; |
90254a65 DN |
5682 | |
5683 | return page; | |
5684 | } | |
4b91355e KH |
5685 | #else |
5686 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5687 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5688 | { |
5689 | return NULL; | |
5690 | } | |
5691 | #endif | |
90254a65 | 5692 | |
87946a72 | 5693 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
48384b0b | 5694 | unsigned long addr, pte_t ptent) |
87946a72 | 5695 | { |
524984ff MWO |
5696 | unsigned long index; |
5697 | struct folio *folio; | |
5698 | ||
87946a72 DN |
5699 | if (!vma->vm_file) /* anonymous vma */ |
5700 | return NULL; | |
1dfab5ab | 5701 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5702 | return NULL; |
5703 | ||
524984ff | 5704 | /* folio is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5705 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
524984ff MWO |
5706 | index = linear_page_index(vma, addr); |
5707 | folio = filemap_get_incore_folio(vma->vm_file->f_mapping, index); | |
66dabbb6 | 5708 | if (IS_ERR(folio)) |
524984ff MWO |
5709 | return NULL; |
5710 | return folio_file_page(folio, index); | |
87946a72 DN |
5711 | } |
5712 | ||
b1b0deab CG |
5713 | /** |
5714 | * mem_cgroup_move_account - move account of the page | |
5715 | * @page: the page | |
25843c2b | 5716 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5717 | * @from: mem_cgroup which the page is moved from. |
5718 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5719 | * | |
4e0cf05f | 5720 | * The page must be locked and not on the LRU. |
b1b0deab CG |
5721 | * |
5722 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5723 | * from old cgroup. | |
5724 | */ | |
5725 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5726 | bool compound, |
b1b0deab CG |
5727 | struct mem_cgroup *from, |
5728 | struct mem_cgroup *to) | |
5729 | { | |
fcce4672 | 5730 | struct folio *folio = page_folio(page); |
ae8af438 KK |
5731 | struct lruvec *from_vec, *to_vec; |
5732 | struct pglist_data *pgdat; | |
fcce4672 | 5733 | unsigned int nr_pages = compound ? folio_nr_pages(folio) : 1; |
8e88bd2d | 5734 | int nid, ret; |
b1b0deab CG |
5735 | |
5736 | VM_BUG_ON(from == to); | |
4e0cf05f | 5737 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
fcce4672 | 5738 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
9c325215 | 5739 | VM_BUG_ON(compound && !folio_test_large(folio)); |
b1b0deab | 5740 | |
b1b0deab | 5741 | ret = -EINVAL; |
fcce4672 | 5742 | if (folio_memcg(folio) != from) |
4e0cf05f | 5743 | goto out; |
b1b0deab | 5744 | |
fcce4672 | 5745 | pgdat = folio_pgdat(folio); |
867e5e1d JW |
5746 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5747 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5748 | |
fcce4672 | 5749 | folio_memcg_lock(folio); |
b1b0deab | 5750 | |
fcce4672 MWO |
5751 | if (folio_test_anon(folio)) { |
5752 | if (folio_mapped(folio)) { | |
be5d0a74 JW |
5753 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); |
5754 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
fcce4672 | 5755 | if (folio_test_transhuge(folio)) { |
69473e5d MS |
5756 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
5757 | -nr_pages); | |
5758 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5759 | nr_pages); | |
468c3982 | 5760 | } |
be5d0a74 JW |
5761 | } |
5762 | } else { | |
0d1c2072 JW |
5763 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5764 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5765 | ||
fcce4672 | 5766 | if (folio_test_swapbacked(folio)) { |
0d1c2072 JW |
5767 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); |
5768 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5769 | } | |
5770 | ||
fcce4672 | 5771 | if (folio_mapped(folio)) { |
49e50d27 JW |
5772 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
5773 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5774 | } | |
b1b0deab | 5775 | |
fcce4672 MWO |
5776 | if (folio_test_dirty(folio)) { |
5777 | struct address_space *mapping = folio_mapping(folio); | |
c4843a75 | 5778 | |
f56753ac | 5779 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5780 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5781 | -nr_pages); | |
5782 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5783 | nr_pages); | |
5784 | } | |
c4843a75 GT |
5785 | } |
5786 | } | |
5787 | ||
c449deb2 HD |
5788 | #ifdef CONFIG_SWAP |
5789 | if (folio_test_swapcache(folio)) { | |
5790 | __mod_lruvec_state(from_vec, NR_SWAPCACHE, -nr_pages); | |
5791 | __mod_lruvec_state(to_vec, NR_SWAPCACHE, nr_pages); | |
5792 | } | |
5793 | #endif | |
fcce4672 | 5794 | if (folio_test_writeback(folio)) { |
ae8af438 KK |
5795 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5796 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5797 | } |
5798 | ||
5799 | /* | |
abb242f5 JW |
5800 | * All state has been migrated, let's switch to the new memcg. |
5801 | * | |
bcfe06bf | 5802 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
5803 | * is referenced, charged, isolated, and locked: we can't race |
5804 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 5805 | * that would rely on a stable page's memory cgroup. |
abb242f5 JW |
5806 | * |
5807 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
bcfe06bf | 5808 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
5809 | * new memcg that isn't locked, the above state can change |
5810 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5811 | */ |
abb242f5 | 5812 | smp_mb(); |
b1b0deab | 5813 | |
1a3e1f40 JW |
5814 | css_get(&to->css); |
5815 | css_put(&from->css); | |
5816 | ||
fcce4672 | 5817 | folio->memcg_data = (unsigned long)to; |
87eaceb3 | 5818 | |
f70ad448 | 5819 | __folio_memcg_unlock(from); |
b1b0deab CG |
5820 | |
5821 | ret = 0; | |
fcce4672 | 5822 | nid = folio_nid(folio); |
b1b0deab CG |
5823 | |
5824 | local_irq_disable(); | |
6e0110c2 | 5825 | mem_cgroup_charge_statistics(to, nr_pages); |
8e88bd2d | 5826 | memcg_check_events(to, nid); |
6e0110c2 | 5827 | mem_cgroup_charge_statistics(from, -nr_pages); |
8e88bd2d | 5828 | memcg_check_events(from, nid); |
b1b0deab | 5829 | local_irq_enable(); |
b1b0deab CG |
5830 | out: |
5831 | return ret; | |
5832 | } | |
5833 | ||
7cf7806c LR |
5834 | /** |
5835 | * get_mctgt_type - get target type of moving charge | |
5836 | * @vma: the vma the pte to be checked belongs | |
5837 | * @addr: the address corresponding to the pte to be checked | |
5838 | * @ptent: the pte to be checked | |
5839 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5840 | * | |
5841 | * Returns | |
5842 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5843 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5844 | * move charge. if @target is not NULL, the page is stored in target->page | |
5845 | * with extra refcnt got(Callers should handle it). | |
5846 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5847 | * target for charge migration. if @target is not NULL, the entry is stored | |
5848 | * in target->ent. | |
f25cbb7a AS |
5849 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is device memory and |
5850 | * thus not on the lru. | |
df6ad698 JG |
5851 | * For now we such page is charge like a regular page would be as for all |
5852 | * intent and purposes it is just special memory taking the place of a | |
5853 | * regular page. | |
c733a828 JG |
5854 | * |
5855 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5856 | * |
5857 | * Called with pte lock held. | |
5858 | */ | |
5859 | ||
8d32ff84 | 5860 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5861 | unsigned long addr, pte_t ptent, union mc_target *target) |
5862 | { | |
5863 | struct page *page = NULL; | |
8d32ff84 | 5864 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5865 | swp_entry_t ent = { .val = 0 }; |
5866 | ||
5867 | if (pte_present(ptent)) | |
5868 | page = mc_handle_present_pte(vma, addr, ptent); | |
5c041f5d PX |
5869 | else if (pte_none_mostly(ptent)) |
5870 | /* | |
5871 | * PTE markers should be treated as a none pte here, separated | |
5872 | * from other swap handling below. | |
5873 | */ | |
5874 | page = mc_handle_file_pte(vma, addr, ptent); | |
90254a65 | 5875 | else if (is_swap_pte(ptent)) |
48406ef8 | 5876 | page = mc_handle_swap_pte(vma, ptent, &ent); |
90254a65 | 5877 | |
4e0cf05f JW |
5878 | if (target && page) { |
5879 | if (!trylock_page(page)) { | |
5880 | put_page(page); | |
5881 | return ret; | |
5882 | } | |
5883 | /* | |
5884 | * page_mapped() must be stable during the move. This | |
5885 | * pte is locked, so if it's present, the page cannot | |
5886 | * become unmapped. If it isn't, we have only partial | |
5887 | * control over the mapped state: the page lock will | |
5888 | * prevent new faults against pagecache and swapcache, | |
5889 | * so an unmapped page cannot become mapped. However, | |
5890 | * if the page is already mapped elsewhere, it can | |
5891 | * unmap, and there is nothing we can do about it. | |
5892 | * Alas, skip moving the page in this case. | |
5893 | */ | |
5894 | if (!pte_present(ptent) && page_mapped(page)) { | |
5895 | unlock_page(page); | |
5896 | put_page(page); | |
5897 | return ret; | |
5898 | } | |
5899 | } | |
5900 | ||
90254a65 | 5901 | if (!page && !ent.val) |
8d32ff84 | 5902 | return ret; |
02491447 | 5903 | if (page) { |
02491447 | 5904 | /* |
0a31bc97 | 5905 | * Do only loose check w/o serialization. |
1306a85a | 5906 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5907 | * not under LRU exclusion. |
02491447 | 5908 | */ |
bcfe06bf | 5909 | if (page_memcg(page) == mc.from) { |
02491447 | 5910 | ret = MC_TARGET_PAGE; |
f25cbb7a AS |
5911 | if (is_device_private_page(page) || |
5912 | is_device_coherent_page(page)) | |
c733a828 | 5913 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5914 | if (target) |
5915 | target->page = page; | |
5916 | } | |
4e0cf05f JW |
5917 | if (!ret || !target) { |
5918 | if (target) | |
5919 | unlock_page(page); | |
02491447 | 5920 | put_page(page); |
4e0cf05f | 5921 | } |
02491447 | 5922 | } |
3e14a57b HY |
5923 | /* |
5924 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5925 | * But we cannot move a tail-page in a THP. | |
5926 | */ | |
5927 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5928 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5929 | ret = MC_TARGET_SWAP; |
5930 | if (target) | |
5931 | target->ent = ent; | |
4ffef5fe | 5932 | } |
4ffef5fe DN |
5933 | return ret; |
5934 | } | |
5935 | ||
12724850 NH |
5936 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5937 | /* | |
d6810d73 HY |
5938 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5939 | * not support them for now. | |
12724850 NH |
5940 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5941 | */ | |
5942 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5943 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5944 | { | |
5945 | struct page *page = NULL; | |
12724850 NH |
5946 | enum mc_target_type ret = MC_TARGET_NONE; |
5947 | ||
84c3fc4e ZY |
5948 | if (unlikely(is_swap_pmd(pmd))) { |
5949 | VM_BUG_ON(thp_migration_supported() && | |
5950 | !is_pmd_migration_entry(pmd)); | |
5951 | return ret; | |
5952 | } | |
12724850 | 5953 | page = pmd_page(pmd); |
309381fe | 5954 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5955 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5956 | return ret; |
bcfe06bf | 5957 | if (page_memcg(page) == mc.from) { |
12724850 NH |
5958 | ret = MC_TARGET_PAGE; |
5959 | if (target) { | |
5960 | get_page(page); | |
4e0cf05f JW |
5961 | if (!trylock_page(page)) { |
5962 | put_page(page); | |
5963 | return MC_TARGET_NONE; | |
5964 | } | |
12724850 NH |
5965 | target->page = page; |
5966 | } | |
5967 | } | |
5968 | return ret; | |
5969 | } | |
5970 | #else | |
5971 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5972 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5973 | { | |
5974 | return MC_TARGET_NONE; | |
5975 | } | |
5976 | #endif | |
5977 | ||
4ffef5fe DN |
5978 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5979 | unsigned long addr, unsigned long end, | |
5980 | struct mm_walk *walk) | |
5981 | { | |
26bcd64a | 5982 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5983 | pte_t *pte; |
5984 | spinlock_t *ptl; | |
5985 | ||
b6ec57f4 KS |
5986 | ptl = pmd_trans_huge_lock(pmd, vma); |
5987 | if (ptl) { | |
c733a828 JG |
5988 | /* |
5989 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5990 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5991 | * this might change. | |
c733a828 | 5992 | */ |
12724850 NH |
5993 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5994 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5995 | spin_unlock(ptl); |
1a5a9906 | 5996 | return 0; |
12724850 | 5997 | } |
03319327 | 5998 | |
45f83cef AA |
5999 | if (pmd_trans_unstable(pmd)) |
6000 | return 0; | |
4ffef5fe DN |
6001 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6002 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6003 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6004 | mc.precharge++; /* increment precharge temporarily */ |
6005 | pte_unmap_unlock(pte - 1, ptl); | |
6006 | cond_resched(); | |
6007 | ||
7dc74be0 DN |
6008 | return 0; |
6009 | } | |
6010 | ||
7b86ac33 CH |
6011 | static const struct mm_walk_ops precharge_walk_ops = { |
6012 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6013 | }; | |
6014 | ||
4ffef5fe DN |
6015 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6016 | { | |
6017 | unsigned long precharge; | |
4ffef5fe | 6018 | |
d8ed45c5 | 6019 | mmap_read_lock(mm); |
ba0aff8e | 6020 | walk_page_range(mm, 0, ULONG_MAX, &precharge_walk_ops, NULL); |
d8ed45c5 | 6021 | mmap_read_unlock(mm); |
4ffef5fe DN |
6022 | |
6023 | precharge = mc.precharge; | |
6024 | mc.precharge = 0; | |
6025 | ||
6026 | return precharge; | |
6027 | } | |
6028 | ||
4ffef5fe DN |
6029 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6030 | { | |
dfe076b0 DN |
6031 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6032 | ||
6033 | VM_BUG_ON(mc.moving_task); | |
6034 | mc.moving_task = current; | |
6035 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6036 | } |
6037 | ||
dfe076b0 DN |
6038 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6039 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6040 | { |
2bd9bb20 KH |
6041 | struct mem_cgroup *from = mc.from; |
6042 | struct mem_cgroup *to = mc.to; | |
6043 | ||
4ffef5fe | 6044 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 6045 | if (mc.precharge) { |
00501b53 | 6046 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
6047 | mc.precharge = 0; |
6048 | } | |
6049 | /* | |
6050 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6051 | * we must uncharge here. | |
6052 | */ | |
6053 | if (mc.moved_charge) { | |
00501b53 | 6054 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 6055 | mc.moved_charge = 0; |
4ffef5fe | 6056 | } |
483c30b5 DN |
6057 | /* we must fixup refcnts and charges */ |
6058 | if (mc.moved_swap) { | |
483c30b5 | 6059 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 6060 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 6061 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 6062 | |
615d66c3 VD |
6063 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
6064 | ||
05b84301 | 6065 | /* |
3e32cb2e JW |
6066 | * we charged both to->memory and to->memsw, so we |
6067 | * should uncharge to->memory. | |
05b84301 | 6068 | */ |
ce00a967 | 6069 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
6070 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
6071 | ||
483c30b5 DN |
6072 | mc.moved_swap = 0; |
6073 | } | |
dfe076b0 DN |
6074 | memcg_oom_recover(from); |
6075 | memcg_oom_recover(to); | |
6076 | wake_up_all(&mc.waitq); | |
6077 | } | |
6078 | ||
6079 | static void mem_cgroup_clear_mc(void) | |
6080 | { | |
264a0ae1 TH |
6081 | struct mm_struct *mm = mc.mm; |
6082 | ||
dfe076b0 DN |
6083 | /* |
6084 | * we must clear moving_task before waking up waiters at the end of | |
6085 | * task migration. | |
6086 | */ | |
6087 | mc.moving_task = NULL; | |
6088 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6089 | spin_lock(&mc.lock); |
4ffef5fe DN |
6090 | mc.from = NULL; |
6091 | mc.to = NULL; | |
264a0ae1 | 6092 | mc.mm = NULL; |
2bd9bb20 | 6093 | spin_unlock(&mc.lock); |
264a0ae1 TH |
6094 | |
6095 | mmput(mm); | |
4ffef5fe DN |
6096 | } |
6097 | ||
1f7dd3e5 | 6098 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6099 | { |
1f7dd3e5 | 6100 | struct cgroup_subsys_state *css; |
eed67d75 | 6101 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 6102 | struct mem_cgroup *from; |
4530eddb | 6103 | struct task_struct *leader, *p; |
9f2115f9 | 6104 | struct mm_struct *mm; |
1dfab5ab | 6105 | unsigned long move_flags; |
9f2115f9 | 6106 | int ret = 0; |
7dc74be0 | 6107 | |
1f7dd3e5 TH |
6108 | /* charge immigration isn't supported on the default hierarchy */ |
6109 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
6110 | return 0; |
6111 | ||
4530eddb TH |
6112 | /* |
6113 | * Multi-process migrations only happen on the default hierarchy | |
6114 | * where charge immigration is not used. Perform charge | |
6115 | * immigration if @tset contains a leader and whine if there are | |
6116 | * multiple. | |
6117 | */ | |
6118 | p = NULL; | |
1f7dd3e5 | 6119 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
6120 | WARN_ON_ONCE(p); |
6121 | p = leader; | |
1f7dd3e5 | 6122 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
6123 | } |
6124 | if (!p) | |
6125 | return 0; | |
6126 | ||
1f7dd3e5 | 6127 | /* |
f0953a1b | 6128 | * We are now committed to this value whatever it is. Changes in this |
1f7dd3e5 TH |
6129 | * tunable will only affect upcoming migrations, not the current one. |
6130 | * So we need to save it, and keep it going. | |
6131 | */ | |
6132 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
6133 | if (!move_flags) | |
6134 | return 0; | |
6135 | ||
9f2115f9 TH |
6136 | from = mem_cgroup_from_task(p); |
6137 | ||
6138 | VM_BUG_ON(from == memcg); | |
6139 | ||
6140 | mm = get_task_mm(p); | |
6141 | if (!mm) | |
6142 | return 0; | |
6143 | /* We move charges only when we move a owner of the mm */ | |
6144 | if (mm->owner == p) { | |
6145 | VM_BUG_ON(mc.from); | |
6146 | VM_BUG_ON(mc.to); | |
6147 | VM_BUG_ON(mc.precharge); | |
6148 | VM_BUG_ON(mc.moved_charge); | |
6149 | VM_BUG_ON(mc.moved_swap); | |
6150 | ||
6151 | spin_lock(&mc.lock); | |
264a0ae1 | 6152 | mc.mm = mm; |
9f2115f9 TH |
6153 | mc.from = from; |
6154 | mc.to = memcg; | |
6155 | mc.flags = move_flags; | |
6156 | spin_unlock(&mc.lock); | |
6157 | /* We set mc.moving_task later */ | |
6158 | ||
6159 | ret = mem_cgroup_precharge_mc(mm); | |
6160 | if (ret) | |
6161 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6162 | } else { |
6163 | mmput(mm); | |
7dc74be0 DN |
6164 | } |
6165 | return ret; | |
6166 | } | |
6167 | ||
1f7dd3e5 | 6168 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6169 | { |
4e2f245d JW |
6170 | if (mc.to) |
6171 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6172 | } |
6173 | ||
4ffef5fe DN |
6174 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6175 | unsigned long addr, unsigned long end, | |
6176 | struct mm_walk *walk) | |
7dc74be0 | 6177 | { |
4ffef5fe | 6178 | int ret = 0; |
26bcd64a | 6179 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6180 | pte_t *pte; |
6181 | spinlock_t *ptl; | |
12724850 NH |
6182 | enum mc_target_type target_type; |
6183 | union mc_target target; | |
6184 | struct page *page; | |
4ffef5fe | 6185 | |
b6ec57f4 KS |
6186 | ptl = pmd_trans_huge_lock(pmd, vma); |
6187 | if (ptl) { | |
62ade86a | 6188 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6189 | spin_unlock(ptl); |
12724850 NH |
6190 | return 0; |
6191 | } | |
6192 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6193 | if (target_type == MC_TARGET_PAGE) { | |
6194 | page = target.page; | |
f7f9c00d | 6195 | if (isolate_lru_page(page)) { |
f627c2f5 | 6196 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6197 | mc.from, mc.to)) { |
12724850 NH |
6198 | mc.precharge -= HPAGE_PMD_NR; |
6199 | mc.moved_charge += HPAGE_PMD_NR; | |
6200 | } | |
6201 | putback_lru_page(page); | |
6202 | } | |
4e0cf05f | 6203 | unlock_page(page); |
12724850 | 6204 | put_page(page); |
c733a828 JG |
6205 | } else if (target_type == MC_TARGET_DEVICE) { |
6206 | page = target.page; | |
6207 | if (!mem_cgroup_move_account(page, true, | |
6208 | mc.from, mc.to)) { | |
6209 | mc.precharge -= HPAGE_PMD_NR; | |
6210 | mc.moved_charge += HPAGE_PMD_NR; | |
6211 | } | |
4e0cf05f | 6212 | unlock_page(page); |
c733a828 | 6213 | put_page(page); |
12724850 | 6214 | } |
bf929152 | 6215 | spin_unlock(ptl); |
1a5a9906 | 6216 | return 0; |
12724850 NH |
6217 | } |
6218 | ||
45f83cef AA |
6219 | if (pmd_trans_unstable(pmd)) |
6220 | return 0; | |
4ffef5fe DN |
6221 | retry: |
6222 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6223 | for (; addr != end; addr += PAGE_SIZE) { | |
6224 | pte_t ptent = *(pte++); | |
c733a828 | 6225 | bool device = false; |
02491447 | 6226 | swp_entry_t ent; |
4ffef5fe DN |
6227 | |
6228 | if (!mc.precharge) | |
6229 | break; | |
6230 | ||
8d32ff84 | 6231 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6232 | case MC_TARGET_DEVICE: |
6233 | device = true; | |
e4a9bc58 | 6234 | fallthrough; |
4ffef5fe DN |
6235 | case MC_TARGET_PAGE: |
6236 | page = target.page; | |
53f9263b KS |
6237 | /* |
6238 | * We can have a part of the split pmd here. Moving it | |
6239 | * can be done but it would be too convoluted so simply | |
6240 | * ignore such a partial THP and keep it in original | |
6241 | * memcg. There should be somebody mapping the head. | |
6242 | */ | |
6243 | if (PageTransCompound(page)) | |
6244 | goto put; | |
f7f9c00d | 6245 | if (!device && !isolate_lru_page(page)) |
4ffef5fe | 6246 | goto put; |
f627c2f5 KS |
6247 | if (!mem_cgroup_move_account(page, false, |
6248 | mc.from, mc.to)) { | |
4ffef5fe | 6249 | mc.precharge--; |
854ffa8d DN |
6250 | /* we uncharge from mc.from later. */ |
6251 | mc.moved_charge++; | |
4ffef5fe | 6252 | } |
c733a828 JG |
6253 | if (!device) |
6254 | putback_lru_page(page); | |
4e0cf05f JW |
6255 | put: /* get_mctgt_type() gets & locks the page */ |
6256 | unlock_page(page); | |
4ffef5fe DN |
6257 | put_page(page); |
6258 | break; | |
02491447 DN |
6259 | case MC_TARGET_SWAP: |
6260 | ent = target.ent; | |
e91cbb42 | 6261 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6262 | mc.precharge--; |
8d22a935 HD |
6263 | mem_cgroup_id_get_many(mc.to, 1); |
6264 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6265 | mc.moved_swap++; |
6266 | } | |
02491447 | 6267 | break; |
4ffef5fe DN |
6268 | default: |
6269 | break; | |
6270 | } | |
6271 | } | |
6272 | pte_unmap_unlock(pte - 1, ptl); | |
6273 | cond_resched(); | |
6274 | ||
6275 | if (addr != end) { | |
6276 | /* | |
6277 | * We have consumed all precharges we got in can_attach(). | |
6278 | * We try charge one by one, but don't do any additional | |
6279 | * charges to mc.to if we have failed in charge once in attach() | |
6280 | * phase. | |
6281 | */ | |
854ffa8d | 6282 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6283 | if (!ret) |
6284 | goto retry; | |
6285 | } | |
6286 | ||
6287 | return ret; | |
6288 | } | |
6289 | ||
7b86ac33 CH |
6290 | static const struct mm_walk_ops charge_walk_ops = { |
6291 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6292 | }; | |
6293 | ||
264a0ae1 | 6294 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6295 | { |
4ffef5fe | 6296 | lru_add_drain_all(); |
312722cb | 6297 | /* |
81f8c3a4 JW |
6298 | * Signal lock_page_memcg() to take the memcg's move_lock |
6299 | * while we're moving its pages to another memcg. Then wait | |
6300 | * for already started RCU-only updates to finish. | |
312722cb JW |
6301 | */ |
6302 | atomic_inc(&mc.from->moving_account); | |
6303 | synchronize_rcu(); | |
dfe076b0 | 6304 | retry: |
d8ed45c5 | 6305 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6306 | /* |
c1e8d7c6 | 6307 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6308 | * waitq. So we cancel all extra charges, wake up all waiters, |
6309 | * and retry. Because we cancel precharges, we might not be able | |
6310 | * to move enough charges, but moving charge is a best-effort | |
6311 | * feature anyway, so it wouldn't be a big problem. | |
6312 | */ | |
6313 | __mem_cgroup_clear_mc(); | |
6314 | cond_resched(); | |
6315 | goto retry; | |
6316 | } | |
26bcd64a NH |
6317 | /* |
6318 | * When we have consumed all precharges and failed in doing | |
6319 | * additional charge, the page walk just aborts. | |
6320 | */ | |
ba0aff8e | 6321 | walk_page_range(mc.mm, 0, ULONG_MAX, &charge_walk_ops, NULL); |
d8ed45c5 | 6322 | mmap_read_unlock(mc.mm); |
312722cb | 6323 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6324 | } |
6325 | ||
264a0ae1 | 6326 | static void mem_cgroup_move_task(void) |
67e465a7 | 6327 | { |
264a0ae1 TH |
6328 | if (mc.to) { |
6329 | mem_cgroup_move_charge(); | |
a433658c | 6330 | mem_cgroup_clear_mc(); |
264a0ae1 | 6331 | } |
67e465a7 | 6332 | } |
5cfb80a7 | 6333 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6334 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6335 | { |
6336 | return 0; | |
6337 | } | |
1f7dd3e5 | 6338 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6339 | { |
6340 | } | |
264a0ae1 | 6341 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6342 | { |
6343 | } | |
6344 | #endif | |
67e465a7 | 6345 | |
bd74fdae YZ |
6346 | #ifdef CONFIG_LRU_GEN |
6347 | static void mem_cgroup_attach(struct cgroup_taskset *tset) | |
6348 | { | |
6349 | struct task_struct *task; | |
6350 | struct cgroup_subsys_state *css; | |
6351 | ||
6352 | /* find the first leader if there is any */ | |
6353 | cgroup_taskset_for_each_leader(task, css, tset) | |
6354 | break; | |
6355 | ||
6356 | if (!task) | |
6357 | return; | |
6358 | ||
6359 | task_lock(task); | |
6360 | if (task->mm && READ_ONCE(task->mm->owner) == task) | |
6361 | lru_gen_migrate_mm(task->mm); | |
6362 | task_unlock(task); | |
6363 | } | |
6364 | #else | |
6365 | static void mem_cgroup_attach(struct cgroup_taskset *tset) | |
6366 | { | |
6367 | } | |
6368 | #endif /* CONFIG_LRU_GEN */ | |
6369 | ||
677dc973 CD |
6370 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6371 | { | |
6372 | if (value == PAGE_COUNTER_MAX) | |
6373 | seq_puts(m, "max\n"); | |
6374 | else | |
6375 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6376 | ||
6377 | return 0; | |
6378 | } | |
6379 | ||
241994ed JW |
6380 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6381 | struct cftype *cft) | |
6382 | { | |
f5fc3c5d JW |
6383 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6384 | ||
6385 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6386 | } |
6387 | ||
8e20d4b3 GR |
6388 | static u64 memory_peak_read(struct cgroup_subsys_state *css, |
6389 | struct cftype *cft) | |
6390 | { | |
6391 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6392 | ||
6393 | return (u64)memcg->memory.watermark * PAGE_SIZE; | |
6394 | } | |
6395 | ||
bf8d5d52 RG |
6396 | static int memory_min_show(struct seq_file *m, void *v) |
6397 | { | |
677dc973 CD |
6398 | return seq_puts_memcg_tunable(m, |
6399 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6400 | } |
6401 | ||
6402 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6403 | char *buf, size_t nbytes, loff_t off) | |
6404 | { | |
6405 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6406 | unsigned long min; | |
6407 | int err; | |
6408 | ||
6409 | buf = strstrip(buf); | |
6410 | err = page_counter_memparse(buf, "max", &min); | |
6411 | if (err) | |
6412 | return err; | |
6413 | ||
6414 | page_counter_set_min(&memcg->memory, min); | |
6415 | ||
6416 | return nbytes; | |
6417 | } | |
6418 | ||
241994ed JW |
6419 | static int memory_low_show(struct seq_file *m, void *v) |
6420 | { | |
677dc973 CD |
6421 | return seq_puts_memcg_tunable(m, |
6422 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6423 | } |
6424 | ||
6425 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6426 | char *buf, size_t nbytes, loff_t off) | |
6427 | { | |
6428 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6429 | unsigned long low; | |
6430 | int err; | |
6431 | ||
6432 | buf = strstrip(buf); | |
d2973697 | 6433 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6434 | if (err) |
6435 | return err; | |
6436 | ||
23067153 | 6437 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6438 | |
6439 | return nbytes; | |
6440 | } | |
6441 | ||
6442 | static int memory_high_show(struct seq_file *m, void *v) | |
6443 | { | |
d1663a90 JK |
6444 | return seq_puts_memcg_tunable(m, |
6445 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6446 | } |
6447 | ||
6448 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6449 | char *buf, size_t nbytes, loff_t off) | |
6450 | { | |
6451 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6452 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6453 | bool drained = false; |
241994ed JW |
6454 | unsigned long high; |
6455 | int err; | |
6456 | ||
6457 | buf = strstrip(buf); | |
d2973697 | 6458 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6459 | if (err) |
6460 | return err; | |
6461 | ||
e82553c1 JW |
6462 | page_counter_set_high(&memcg->memory, high); |
6463 | ||
8c8c383c JW |
6464 | for (;;) { |
6465 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6466 | unsigned long reclaimed; | |
6467 | ||
6468 | if (nr_pages <= high) | |
6469 | break; | |
6470 | ||
6471 | if (signal_pending(current)) | |
6472 | break; | |
6473 | ||
6474 | if (!drained) { | |
6475 | drain_all_stock(memcg); | |
6476 | drained = true; | |
6477 | continue; | |
6478 | } | |
6479 | ||
6480 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
55ab834a | 6481 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP); |
8c8c383c JW |
6482 | |
6483 | if (!reclaimed && !nr_retries--) | |
6484 | break; | |
6485 | } | |
588083bb | 6486 | |
19ce33ac | 6487 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6488 | return nbytes; |
6489 | } | |
6490 | ||
6491 | static int memory_max_show(struct seq_file *m, void *v) | |
6492 | { | |
677dc973 CD |
6493 | return seq_puts_memcg_tunable(m, |
6494 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6495 | } |
6496 | ||
6497 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6498 | char *buf, size_t nbytes, loff_t off) | |
6499 | { | |
6500 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6501 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6502 | bool drained = false; |
241994ed JW |
6503 | unsigned long max; |
6504 | int err; | |
6505 | ||
6506 | buf = strstrip(buf); | |
d2973697 | 6507 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6508 | if (err) |
6509 | return err; | |
6510 | ||
bbec2e15 | 6511 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6512 | |
6513 | for (;;) { | |
6514 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6515 | ||
6516 | if (nr_pages <= max) | |
6517 | break; | |
6518 | ||
7249c9f0 | 6519 | if (signal_pending(current)) |
b6e6edcf | 6520 | break; |
b6e6edcf JW |
6521 | |
6522 | if (!drained) { | |
6523 | drain_all_stock(memcg); | |
6524 | drained = true; | |
6525 | continue; | |
6526 | } | |
6527 | ||
6528 | if (nr_reclaims) { | |
6529 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
55ab834a | 6530 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP)) |
b6e6edcf JW |
6531 | nr_reclaims--; |
6532 | continue; | |
6533 | } | |
6534 | ||
e27be240 | 6535 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6536 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6537 | break; | |
6538 | } | |
241994ed | 6539 | |
2529bb3a | 6540 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6541 | return nbytes; |
6542 | } | |
6543 | ||
1e577f97 SB |
6544 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6545 | { | |
6546 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6547 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6548 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6549 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6550 | seq_printf(m, "oom_kill %lu\n", | |
6551 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
b6bf9abb DS |
6552 | seq_printf(m, "oom_group_kill %lu\n", |
6553 | atomic_long_read(&events[MEMCG_OOM_GROUP_KILL])); | |
1e577f97 SB |
6554 | } |
6555 | ||
241994ed JW |
6556 | static int memory_events_show(struct seq_file *m, void *v) |
6557 | { | |
aa9694bb | 6558 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6559 | |
1e577f97 SB |
6560 | __memory_events_show(m, memcg->memory_events); |
6561 | return 0; | |
6562 | } | |
6563 | ||
6564 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6565 | { | |
6566 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6567 | |
1e577f97 | 6568 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6569 | return 0; |
6570 | } | |
6571 | ||
587d9f72 JW |
6572 | static int memory_stat_show(struct seq_file *m, void *v) |
6573 | { | |
aa9694bb | 6574 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
68aaee14 | 6575 | char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
1ff9e6e1 | 6576 | |
c8713d0b JW |
6577 | if (!buf) |
6578 | return -ENOMEM; | |
68aaee14 | 6579 | memory_stat_format(memcg, buf, PAGE_SIZE); |
c8713d0b JW |
6580 | seq_puts(m, buf); |
6581 | kfree(buf); | |
587d9f72 JW |
6582 | return 0; |
6583 | } | |
6584 | ||
5f9a4f4a | 6585 | #ifdef CONFIG_NUMA |
fff66b79 MS |
6586 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
6587 | int item) | |
6588 | { | |
6589 | return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item); | |
6590 | } | |
6591 | ||
5f9a4f4a MS |
6592 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
6593 | { | |
6594 | int i; | |
6595 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6596 | ||
fd25a9e0 | 6597 | mem_cgroup_flush_stats(); |
7e1c0d6f | 6598 | |
5f9a4f4a MS |
6599 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
6600 | int nid; | |
6601 | ||
6602 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6603 | continue; | |
6604 | ||
6605 | seq_printf(m, "%s", memory_stats[i].name); | |
6606 | for_each_node_state(nid, N_MEMORY) { | |
6607 | u64 size; | |
6608 | struct lruvec *lruvec; | |
6609 | ||
6610 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
6611 | size = lruvec_page_state_output(lruvec, |
6612 | memory_stats[i].idx); | |
5f9a4f4a MS |
6613 | seq_printf(m, " N%d=%llu", nid, size); |
6614 | } | |
6615 | seq_putc(m, '\n'); | |
6616 | } | |
6617 | ||
6618 | return 0; | |
6619 | } | |
6620 | #endif | |
6621 | ||
3d8b38eb RG |
6622 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6623 | { | |
aa9694bb | 6624 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb | 6625 | |
eaf7b66b | 6626 | seq_printf(m, "%d\n", READ_ONCE(memcg->oom_group)); |
3d8b38eb RG |
6627 | |
6628 | return 0; | |
6629 | } | |
6630 | ||
6631 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6632 | char *buf, size_t nbytes, loff_t off) | |
6633 | { | |
6634 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6635 | int ret, oom_group; | |
6636 | ||
6637 | buf = strstrip(buf); | |
6638 | if (!buf) | |
6639 | return -EINVAL; | |
6640 | ||
6641 | ret = kstrtoint(buf, 0, &oom_group); | |
6642 | if (ret) | |
6643 | return ret; | |
6644 | ||
6645 | if (oom_group != 0 && oom_group != 1) | |
6646 | return -EINVAL; | |
6647 | ||
eaf7b66b | 6648 | WRITE_ONCE(memcg->oom_group, oom_group); |
3d8b38eb RG |
6649 | |
6650 | return nbytes; | |
6651 | } | |
6652 | ||
94968384 SB |
6653 | static ssize_t memory_reclaim(struct kernfs_open_file *of, char *buf, |
6654 | size_t nbytes, loff_t off) | |
6655 | { | |
6656 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6657 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; | |
6658 | unsigned long nr_to_reclaim, nr_reclaimed = 0; | |
55ab834a MH |
6659 | unsigned int reclaim_options; |
6660 | int err; | |
12a5d395 MA |
6661 | |
6662 | buf = strstrip(buf); | |
55ab834a MH |
6663 | err = page_counter_memparse(buf, "", &nr_to_reclaim); |
6664 | if (err) | |
6665 | return err; | |
12a5d395 | 6666 | |
55ab834a | 6667 | reclaim_options = MEMCG_RECLAIM_MAY_SWAP | MEMCG_RECLAIM_PROACTIVE; |
94968384 SB |
6668 | while (nr_reclaimed < nr_to_reclaim) { |
6669 | unsigned long reclaimed; | |
6670 | ||
6671 | if (signal_pending(current)) | |
6672 | return -EINTR; | |
6673 | ||
6674 | /* | |
6675 | * This is the final attempt, drain percpu lru caches in the | |
6676 | * hope of introducing more evictable pages for | |
6677 | * try_to_free_mem_cgroup_pages(). | |
6678 | */ | |
6679 | if (!nr_retries) | |
6680 | lru_add_drain_all(); | |
6681 | ||
6682 | reclaimed = try_to_free_mem_cgroup_pages(memcg, | |
6683 | nr_to_reclaim - nr_reclaimed, | |
55ab834a | 6684 | GFP_KERNEL, reclaim_options); |
94968384 SB |
6685 | |
6686 | if (!reclaimed && !nr_retries--) | |
6687 | return -EAGAIN; | |
6688 | ||
6689 | nr_reclaimed += reclaimed; | |
6690 | } | |
6691 | ||
6692 | return nbytes; | |
6693 | } | |
6694 | ||
241994ed JW |
6695 | static struct cftype memory_files[] = { |
6696 | { | |
6697 | .name = "current", | |
f5fc3c5d | 6698 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6699 | .read_u64 = memory_current_read, |
6700 | }, | |
8e20d4b3 GR |
6701 | { |
6702 | .name = "peak", | |
6703 | .flags = CFTYPE_NOT_ON_ROOT, | |
6704 | .read_u64 = memory_peak_read, | |
6705 | }, | |
bf8d5d52 RG |
6706 | { |
6707 | .name = "min", | |
6708 | .flags = CFTYPE_NOT_ON_ROOT, | |
6709 | .seq_show = memory_min_show, | |
6710 | .write = memory_min_write, | |
6711 | }, | |
241994ed JW |
6712 | { |
6713 | .name = "low", | |
6714 | .flags = CFTYPE_NOT_ON_ROOT, | |
6715 | .seq_show = memory_low_show, | |
6716 | .write = memory_low_write, | |
6717 | }, | |
6718 | { | |
6719 | .name = "high", | |
6720 | .flags = CFTYPE_NOT_ON_ROOT, | |
6721 | .seq_show = memory_high_show, | |
6722 | .write = memory_high_write, | |
6723 | }, | |
6724 | { | |
6725 | .name = "max", | |
6726 | .flags = CFTYPE_NOT_ON_ROOT, | |
6727 | .seq_show = memory_max_show, | |
6728 | .write = memory_max_write, | |
6729 | }, | |
6730 | { | |
6731 | .name = "events", | |
6732 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6733 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6734 | .seq_show = memory_events_show, |
6735 | }, | |
1e577f97 SB |
6736 | { |
6737 | .name = "events.local", | |
6738 | .flags = CFTYPE_NOT_ON_ROOT, | |
6739 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6740 | .seq_show = memory_events_local_show, | |
6741 | }, | |
587d9f72 JW |
6742 | { |
6743 | .name = "stat", | |
587d9f72 JW |
6744 | .seq_show = memory_stat_show, |
6745 | }, | |
5f9a4f4a MS |
6746 | #ifdef CONFIG_NUMA |
6747 | { | |
6748 | .name = "numa_stat", | |
6749 | .seq_show = memory_numa_stat_show, | |
6750 | }, | |
6751 | #endif | |
3d8b38eb RG |
6752 | { |
6753 | .name = "oom.group", | |
6754 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6755 | .seq_show = memory_oom_group_show, | |
6756 | .write = memory_oom_group_write, | |
6757 | }, | |
94968384 SB |
6758 | { |
6759 | .name = "reclaim", | |
6760 | .flags = CFTYPE_NS_DELEGATABLE, | |
6761 | .write = memory_reclaim, | |
6762 | }, | |
241994ed JW |
6763 | { } /* terminate */ |
6764 | }; | |
6765 | ||
073219e9 | 6766 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6767 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6768 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6769 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6770 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6771 | .css_free = mem_cgroup_css_free, |
1ced953b | 6772 | .css_reset = mem_cgroup_css_reset, |
2d146aa3 | 6773 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
7dc74be0 | 6774 | .can_attach = mem_cgroup_can_attach, |
bd74fdae | 6775 | .attach = mem_cgroup_attach, |
7dc74be0 | 6776 | .cancel_attach = mem_cgroup_cancel_attach, |
264a0ae1 | 6777 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6778 | .dfl_cftypes = memory_files, |
6779 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6780 | .early_init = 0, |
8cdea7c0 | 6781 | }; |
c077719b | 6782 | |
bc50bcc6 JW |
6783 | /* |
6784 | * This function calculates an individual cgroup's effective | |
6785 | * protection which is derived from its own memory.min/low, its | |
6786 | * parent's and siblings' settings, as well as the actual memory | |
6787 | * distribution in the tree. | |
6788 | * | |
6789 | * The following rules apply to the effective protection values: | |
6790 | * | |
6791 | * 1. At the first level of reclaim, effective protection is equal to | |
6792 | * the declared protection in memory.min and memory.low. | |
6793 | * | |
6794 | * 2. To enable safe delegation of the protection configuration, at | |
6795 | * subsequent levels the effective protection is capped to the | |
6796 | * parent's effective protection. | |
6797 | * | |
6798 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6799 | * user is allowed to overcommit the declared protection at a given | |
6800 | * level. If that is the case, the parent's effective protection is | |
6801 | * distributed to the children in proportion to how much protection | |
6802 | * they have declared and how much of it they are utilizing. | |
6803 | * | |
6804 | * This makes distribution proportional, but also work-conserving: | |
6805 | * if one cgroup claims much more protection than it uses memory, | |
6806 | * the unused remainder is available to its siblings. | |
6807 | * | |
6808 | * 4. Conversely, when the declared protection is undercommitted at a | |
6809 | * given level, the distribution of the larger parental protection | |
6810 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6811 | * is capped to its own memory.min/low setting. | |
6812 | * | |
8a931f80 JW |
6813 | * 5. However, to allow protecting recursive subtrees from each other |
6814 | * without having to declare each individual cgroup's fixed share | |
6815 | * of the ancestor's claim to protection, any unutilized - | |
6816 | * "floating" - protection from up the tree is distributed in | |
6817 | * proportion to each cgroup's *usage*. This makes the protection | |
6818 | * neutral wrt sibling cgroups and lets them compete freely over | |
6819 | * the shared parental protection budget, but it protects the | |
6820 | * subtree as a whole from neighboring subtrees. | |
6821 | * | |
6822 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6823 | * against immediate siblings whereas 5. is about protecting against | |
6824 | * neighboring subtrees. | |
bc50bcc6 JW |
6825 | */ |
6826 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6827 | unsigned long parent_usage, |
bc50bcc6 JW |
6828 | unsigned long setting, |
6829 | unsigned long parent_effective, | |
6830 | unsigned long siblings_protected) | |
6831 | { | |
6832 | unsigned long protected; | |
8a931f80 | 6833 | unsigned long ep; |
bc50bcc6 JW |
6834 | |
6835 | protected = min(usage, setting); | |
6836 | /* | |
6837 | * If all cgroups at this level combined claim and use more | |
6838 | * protection then what the parent affords them, distribute | |
6839 | * shares in proportion to utilization. | |
6840 | * | |
6841 | * We are using actual utilization rather than the statically | |
6842 | * claimed protection in order to be work-conserving: claimed | |
6843 | * but unused protection is available to siblings that would | |
6844 | * otherwise get a smaller chunk than what they claimed. | |
6845 | */ | |
6846 | if (siblings_protected > parent_effective) | |
6847 | return protected * parent_effective / siblings_protected; | |
6848 | ||
6849 | /* | |
6850 | * Ok, utilized protection of all children is within what the | |
6851 | * parent affords them, so we know whatever this child claims | |
6852 | * and utilizes is effectively protected. | |
6853 | * | |
6854 | * If there is unprotected usage beyond this value, reclaim | |
6855 | * will apply pressure in proportion to that amount. | |
6856 | * | |
6857 | * If there is unutilized protection, the cgroup will be fully | |
6858 | * shielded from reclaim, but we do return a smaller value for | |
6859 | * protection than what the group could enjoy in theory. This | |
6860 | * is okay. With the overcommit distribution above, effective | |
6861 | * protection is always dependent on how memory is actually | |
6862 | * consumed among the siblings anyway. | |
6863 | */ | |
8a931f80 JW |
6864 | ep = protected; |
6865 | ||
6866 | /* | |
6867 | * If the children aren't claiming (all of) the protection | |
6868 | * afforded to them by the parent, distribute the remainder in | |
6869 | * proportion to the (unprotected) memory of each cgroup. That | |
6870 | * way, cgroups that aren't explicitly prioritized wrt each | |
6871 | * other compete freely over the allowance, but they are | |
6872 | * collectively protected from neighboring trees. | |
6873 | * | |
6874 | * We're using unprotected memory for the weight so that if | |
6875 | * some cgroups DO claim explicit protection, we don't protect | |
6876 | * the same bytes twice. | |
cd324edc JW |
6877 | * |
6878 | * Check both usage and parent_usage against the respective | |
6879 | * protected values. One should imply the other, but they | |
6880 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6881 | */ |
6882 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6883 | return ep; | |
cd324edc JW |
6884 | if (parent_effective > siblings_protected && |
6885 | parent_usage > siblings_protected && | |
6886 | usage > protected) { | |
8a931f80 JW |
6887 | unsigned long unclaimed; |
6888 | ||
6889 | unclaimed = parent_effective - siblings_protected; | |
6890 | unclaimed *= usage - protected; | |
6891 | unclaimed /= parent_usage - siblings_protected; | |
6892 | ||
6893 | ep += unclaimed; | |
6894 | } | |
6895 | ||
6896 | return ep; | |
bc50bcc6 JW |
6897 | } |
6898 | ||
241994ed | 6899 | /** |
05395718 | 6900 | * mem_cgroup_calculate_protection - check if memory consumption is in the normal range |
34c81057 | 6901 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6902 | * @memcg: the memory cgroup to check |
6903 | * | |
23067153 RG |
6904 | * WARNING: This function is not stateless! It can only be used as part |
6905 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6906 | */ |
45c7f7e1 CD |
6907 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6908 | struct mem_cgroup *memcg) | |
241994ed | 6909 | { |
8a931f80 | 6910 | unsigned long usage, parent_usage; |
23067153 RG |
6911 | struct mem_cgroup *parent; |
6912 | ||
241994ed | 6913 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6914 | return; |
241994ed | 6915 | |
34c81057 SC |
6916 | if (!root) |
6917 | root = root_mem_cgroup; | |
22f7496f YS |
6918 | |
6919 | /* | |
6920 | * Effective values of the reclaim targets are ignored so they | |
6921 | * can be stale. Have a look at mem_cgroup_protection for more | |
6922 | * details. | |
6923 | * TODO: calculation should be more robust so that we do not need | |
6924 | * that special casing. | |
6925 | */ | |
34c81057 | 6926 | if (memcg == root) |
45c7f7e1 | 6927 | return; |
241994ed | 6928 | |
23067153 | 6929 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6930 | if (!usage) |
45c7f7e1 | 6931 | return; |
bf8d5d52 | 6932 | |
bf8d5d52 | 6933 | parent = parent_mem_cgroup(memcg); |
df2a4196 | 6934 | |
bc50bcc6 | 6935 | if (parent == root) { |
c3d53200 | 6936 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6937 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6938 | return; |
bf8d5d52 RG |
6939 | } |
6940 | ||
8a931f80 JW |
6941 | parent_usage = page_counter_read(&parent->memory); |
6942 | ||
b3a7822e | 6943 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6944 | READ_ONCE(memcg->memory.min), |
6945 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6946 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6947 | |
b3a7822e | 6948 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6949 | READ_ONCE(memcg->memory.low), |
6950 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6951 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6952 | } |
6953 | ||
8f425e4e MWO |
6954 | static int charge_memcg(struct folio *folio, struct mem_cgroup *memcg, |
6955 | gfp_t gfp) | |
0add0c77 | 6956 | { |
118f2875 | 6957 | long nr_pages = folio_nr_pages(folio); |
0add0c77 SB |
6958 | int ret; |
6959 | ||
6960 | ret = try_charge(memcg, gfp, nr_pages); | |
6961 | if (ret) | |
6962 | goto out; | |
6963 | ||
6964 | css_get(&memcg->css); | |
118f2875 | 6965 | commit_charge(folio, memcg); |
0add0c77 SB |
6966 | |
6967 | local_irq_disable(); | |
6e0110c2 | 6968 | mem_cgroup_charge_statistics(memcg, nr_pages); |
8f425e4e | 6969 | memcg_check_events(memcg, folio_nid(folio)); |
0add0c77 SB |
6970 | local_irq_enable(); |
6971 | out: | |
6972 | return ret; | |
6973 | } | |
6974 | ||
8f425e4e | 6975 | int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp) |
00501b53 | 6976 | { |
0add0c77 SB |
6977 | struct mem_cgroup *memcg; |
6978 | int ret; | |
00501b53 | 6979 | |
0add0c77 | 6980 | memcg = get_mem_cgroup_from_mm(mm); |
8f425e4e | 6981 | ret = charge_memcg(folio, memcg, gfp); |
0add0c77 | 6982 | css_put(&memcg->css); |
2d1c4980 | 6983 | |
0add0c77 SB |
6984 | return ret; |
6985 | } | |
e993d905 | 6986 | |
0add0c77 | 6987 | /** |
65995918 MWO |
6988 | * mem_cgroup_swapin_charge_folio - Charge a newly allocated folio for swapin. |
6989 | * @folio: folio to charge. | |
0add0c77 SB |
6990 | * @mm: mm context of the victim |
6991 | * @gfp: reclaim mode | |
65995918 | 6992 | * @entry: swap entry for which the folio is allocated |
0add0c77 | 6993 | * |
65995918 MWO |
6994 | * This function charges a folio allocated for swapin. Please call this before |
6995 | * adding the folio to the swapcache. | |
0add0c77 SB |
6996 | * |
6997 | * Returns 0 on success. Otherwise, an error code is returned. | |
6998 | */ | |
65995918 | 6999 | int mem_cgroup_swapin_charge_folio(struct folio *folio, struct mm_struct *mm, |
0add0c77 SB |
7000 | gfp_t gfp, swp_entry_t entry) |
7001 | { | |
7002 | struct mem_cgroup *memcg; | |
7003 | unsigned short id; | |
7004 | int ret; | |
00501b53 | 7005 | |
0add0c77 SB |
7006 | if (mem_cgroup_disabled()) |
7007 | return 0; | |
00501b53 | 7008 | |
0add0c77 SB |
7009 | id = lookup_swap_cgroup_id(entry); |
7010 | rcu_read_lock(); | |
7011 | memcg = mem_cgroup_from_id(id); | |
7012 | if (!memcg || !css_tryget_online(&memcg->css)) | |
7013 | memcg = get_mem_cgroup_from_mm(mm); | |
7014 | rcu_read_unlock(); | |
00501b53 | 7015 | |
8f425e4e | 7016 | ret = charge_memcg(folio, memcg, gfp); |
6abb5a86 | 7017 | |
0add0c77 SB |
7018 | css_put(&memcg->css); |
7019 | return ret; | |
7020 | } | |
00501b53 | 7021 | |
0add0c77 SB |
7022 | /* |
7023 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot | |
7024 | * @entry: swap entry for which the page is charged | |
7025 | * | |
7026 | * Call this function after successfully adding the charged page to swapcache. | |
7027 | * | |
7028 | * Note: This function assumes the page for which swap slot is being uncharged | |
7029 | * is order 0 page. | |
7030 | */ | |
7031 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) | |
7032 | { | |
cae3af62 MS |
7033 | /* |
7034 | * Cgroup1's unified memory+swap counter has been charged with the | |
7035 | * new swapcache page, finish the transfer by uncharging the swap | |
7036 | * slot. The swap slot would also get uncharged when it dies, but | |
7037 | * it can stick around indefinitely and we'd count the page twice | |
7038 | * the entire time. | |
7039 | * | |
7040 | * Cgroup2 has separate resource counters for memory and swap, | |
7041 | * so this is a non-issue here. Memory and swap charge lifetimes | |
7042 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
7043 | * page to memory here, and uncharge swap when the slot is freed. | |
7044 | */ | |
0add0c77 | 7045 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
00501b53 JW |
7046 | /* |
7047 | * The swap entry might not get freed for a long time, | |
7048 | * let's not wait for it. The page already received a | |
7049 | * memory+swap charge, drop the swap entry duplicate. | |
7050 | */ | |
0add0c77 | 7051 | mem_cgroup_uncharge_swap(entry, 1); |
00501b53 | 7052 | } |
3fea5a49 JW |
7053 | } |
7054 | ||
a9d5adee JG |
7055 | struct uncharge_gather { |
7056 | struct mem_cgroup *memcg; | |
b4e0b68f | 7057 | unsigned long nr_memory; |
a9d5adee | 7058 | unsigned long pgpgout; |
a9d5adee | 7059 | unsigned long nr_kmem; |
8e88bd2d | 7060 | int nid; |
a9d5adee JG |
7061 | }; |
7062 | ||
7063 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 7064 | { |
a9d5adee JG |
7065 | memset(ug, 0, sizeof(*ug)); |
7066 | } | |
7067 | ||
7068 | static void uncharge_batch(const struct uncharge_gather *ug) | |
7069 | { | |
747db954 JW |
7070 | unsigned long flags; |
7071 | ||
b4e0b68f MS |
7072 | if (ug->nr_memory) { |
7073 | page_counter_uncharge(&ug->memcg->memory, ug->nr_memory); | |
7941d214 | 7074 | if (do_memsw_account()) |
b4e0b68f | 7075 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory); |
a8c49af3 YA |
7076 | if (ug->nr_kmem) |
7077 | memcg_account_kmem(ug->memcg, -ug->nr_kmem); | |
a9d5adee | 7078 | memcg_oom_recover(ug->memcg); |
ce00a967 | 7079 | } |
747db954 JW |
7080 | |
7081 | local_irq_save(flags); | |
c9019e9b | 7082 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
b4e0b68f | 7083 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory); |
8e88bd2d | 7084 | memcg_check_events(ug->memcg, ug->nid); |
747db954 | 7085 | local_irq_restore(flags); |
f1796544 | 7086 | |
c4ed6ebf | 7087 | /* drop reference from uncharge_folio */ |
f1796544 | 7088 | css_put(&ug->memcg->css); |
a9d5adee JG |
7089 | } |
7090 | ||
c4ed6ebf | 7091 | static void uncharge_folio(struct folio *folio, struct uncharge_gather *ug) |
a9d5adee | 7092 | { |
c4ed6ebf | 7093 | long nr_pages; |
b4e0b68f MS |
7094 | struct mem_cgroup *memcg; |
7095 | struct obj_cgroup *objcg; | |
9f762dbe | 7096 | |
c4ed6ebf | 7097 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
a9d5adee | 7098 | |
a9d5adee JG |
7099 | /* |
7100 | * Nobody should be changing or seriously looking at | |
c4ed6ebf MWO |
7101 | * folio memcg or objcg at this point, we have fully |
7102 | * exclusive access to the folio. | |
a9d5adee | 7103 | */ |
fead2b86 | 7104 | if (folio_memcg_kmem(folio)) { |
1b7e4464 | 7105 | objcg = __folio_objcg(folio); |
b4e0b68f MS |
7106 | /* |
7107 | * This get matches the put at the end of the function and | |
7108 | * kmem pages do not hold memcg references anymore. | |
7109 | */ | |
7110 | memcg = get_mem_cgroup_from_objcg(objcg); | |
7111 | } else { | |
1b7e4464 | 7112 | memcg = __folio_memcg(folio); |
b4e0b68f | 7113 | } |
a9d5adee | 7114 | |
b4e0b68f MS |
7115 | if (!memcg) |
7116 | return; | |
7117 | ||
7118 | if (ug->memcg != memcg) { | |
a9d5adee JG |
7119 | if (ug->memcg) { |
7120 | uncharge_batch(ug); | |
7121 | uncharge_gather_clear(ug); | |
7122 | } | |
b4e0b68f | 7123 | ug->memcg = memcg; |
c4ed6ebf | 7124 | ug->nid = folio_nid(folio); |
f1796544 MH |
7125 | |
7126 | /* pairs with css_put in uncharge_batch */ | |
b4e0b68f | 7127 | css_get(&memcg->css); |
a9d5adee JG |
7128 | } |
7129 | ||
c4ed6ebf | 7130 | nr_pages = folio_nr_pages(folio); |
a9d5adee | 7131 | |
fead2b86 | 7132 | if (folio_memcg_kmem(folio)) { |
b4e0b68f | 7133 | ug->nr_memory += nr_pages; |
9f762dbe | 7134 | ug->nr_kmem += nr_pages; |
b4e0b68f | 7135 | |
c4ed6ebf | 7136 | folio->memcg_data = 0; |
b4e0b68f MS |
7137 | obj_cgroup_put(objcg); |
7138 | } else { | |
7139 | /* LRU pages aren't accounted at the root level */ | |
7140 | if (!mem_cgroup_is_root(memcg)) | |
7141 | ug->nr_memory += nr_pages; | |
18b2db3b | 7142 | ug->pgpgout++; |
a9d5adee | 7143 | |
c4ed6ebf | 7144 | folio->memcg_data = 0; |
b4e0b68f MS |
7145 | } |
7146 | ||
7147 | css_put(&memcg->css); | |
747db954 JW |
7148 | } |
7149 | ||
bbc6b703 | 7150 | void __mem_cgroup_uncharge(struct folio *folio) |
0a31bc97 | 7151 | { |
a9d5adee JG |
7152 | struct uncharge_gather ug; |
7153 | ||
bbc6b703 MWO |
7154 | /* Don't touch folio->lru of any random page, pre-check: */ |
7155 | if (!folio_memcg(folio)) | |
0a31bc97 JW |
7156 | return; |
7157 | ||
a9d5adee | 7158 | uncharge_gather_clear(&ug); |
bbc6b703 | 7159 | uncharge_folio(folio, &ug); |
a9d5adee | 7160 | uncharge_batch(&ug); |
747db954 | 7161 | } |
0a31bc97 | 7162 | |
747db954 | 7163 | /** |
2c8d8f97 | 7164 | * __mem_cgroup_uncharge_list - uncharge a list of page |
747db954 JW |
7165 | * @page_list: list of pages to uncharge |
7166 | * | |
7167 | * Uncharge a list of pages previously charged with | |
2c8d8f97 | 7168 | * __mem_cgroup_charge(). |
747db954 | 7169 | */ |
2c8d8f97 | 7170 | void __mem_cgroup_uncharge_list(struct list_head *page_list) |
747db954 | 7171 | { |
c41a40b6 | 7172 | struct uncharge_gather ug; |
c4ed6ebf | 7173 | struct folio *folio; |
c41a40b6 | 7174 | |
c41a40b6 | 7175 | uncharge_gather_clear(&ug); |
c4ed6ebf MWO |
7176 | list_for_each_entry(folio, page_list, lru) |
7177 | uncharge_folio(folio, &ug); | |
c41a40b6 MS |
7178 | if (ug.memcg) |
7179 | uncharge_batch(&ug); | |
0a31bc97 JW |
7180 | } |
7181 | ||
7182 | /** | |
d21bba2b MWO |
7183 | * mem_cgroup_migrate - Charge a folio's replacement. |
7184 | * @old: Currently circulating folio. | |
7185 | * @new: Replacement folio. | |
0a31bc97 | 7186 | * |
d21bba2b | 7187 | * Charge @new as a replacement folio for @old. @old will |
6a93ca8f | 7188 | * be uncharged upon free. |
0a31bc97 | 7189 | * |
d21bba2b | 7190 | * Both folios must be locked, @new->mapping must be set up. |
0a31bc97 | 7191 | */ |
d21bba2b | 7192 | void mem_cgroup_migrate(struct folio *old, struct folio *new) |
0a31bc97 | 7193 | { |
29833315 | 7194 | struct mem_cgroup *memcg; |
d21bba2b | 7195 | long nr_pages = folio_nr_pages(new); |
d93c4130 | 7196 | unsigned long flags; |
0a31bc97 | 7197 | |
d21bba2b MWO |
7198 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); |
7199 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7200 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7201 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != nr_pages, new); | |
0a31bc97 JW |
7202 | |
7203 | if (mem_cgroup_disabled()) | |
7204 | return; | |
7205 | ||
d21bba2b MWO |
7206 | /* Page cache replacement: new folio already charged? */ |
7207 | if (folio_memcg(new)) | |
0a31bc97 JW |
7208 | return; |
7209 | ||
d21bba2b MWO |
7210 | memcg = folio_memcg(old); |
7211 | VM_WARN_ON_ONCE_FOLIO(!memcg, old); | |
29833315 | 7212 | if (!memcg) |
0a31bc97 JW |
7213 | return; |
7214 | ||
44b7a8d3 | 7215 | /* Force-charge the new page. The old one will be freed soon */ |
8dc87c7d MS |
7216 | if (!mem_cgroup_is_root(memcg)) { |
7217 | page_counter_charge(&memcg->memory, nr_pages); | |
7218 | if (do_memsw_account()) | |
7219 | page_counter_charge(&memcg->memsw, nr_pages); | |
7220 | } | |
0a31bc97 | 7221 | |
1a3e1f40 | 7222 | css_get(&memcg->css); |
d21bba2b | 7223 | commit_charge(new, memcg); |
44b7a8d3 | 7224 | |
d93c4130 | 7225 | local_irq_save(flags); |
6e0110c2 | 7226 | mem_cgroup_charge_statistics(memcg, nr_pages); |
d21bba2b | 7227 | memcg_check_events(memcg, folio_nid(new)); |
d93c4130 | 7228 | local_irq_restore(flags); |
0a31bc97 JW |
7229 | } |
7230 | ||
ef12947c | 7231 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
7232 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
7233 | ||
2d758073 | 7234 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
7235 | { |
7236 | struct mem_cgroup *memcg; | |
7237 | ||
2d758073 JW |
7238 | if (!mem_cgroup_sockets_enabled) |
7239 | return; | |
7240 | ||
e876ecc6 | 7241 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
086f694a | 7242 | if (!in_task()) |
e876ecc6 SB |
7243 | return; |
7244 | ||
11092087 JW |
7245 | rcu_read_lock(); |
7246 | memcg = mem_cgroup_from_task(current); | |
7848ed62 | 7247 | if (mem_cgroup_is_root(memcg)) |
f7e1cb6e | 7248 | goto out; |
0db15298 | 7249 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 7250 | goto out; |
8965aa28 | 7251 | if (css_tryget(&memcg->css)) |
11092087 | 7252 | sk->sk_memcg = memcg; |
f7e1cb6e | 7253 | out: |
11092087 JW |
7254 | rcu_read_unlock(); |
7255 | } | |
11092087 | 7256 | |
2d758073 | 7257 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 7258 | { |
2d758073 JW |
7259 | if (sk->sk_memcg) |
7260 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7261 | } |
7262 | ||
7263 | /** | |
7264 | * mem_cgroup_charge_skmem - charge socket memory | |
7265 | * @memcg: memcg to charge | |
7266 | * @nr_pages: number of pages to charge | |
4b1327be | 7267 | * @gfp_mask: reclaim mode |
11092087 JW |
7268 | * |
7269 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
4b1327be | 7270 | * @memcg's configured limit, %false if it doesn't. |
11092087 | 7271 | */ |
4b1327be WW |
7272 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, |
7273 | gfp_t gfp_mask) | |
11092087 | 7274 | { |
f7e1cb6e | 7275 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7276 | struct page_counter *fail; |
f7e1cb6e | 7277 | |
0db15298 JW |
7278 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7279 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7280 | return true; |
7281 | } | |
0db15298 | 7282 | memcg->tcpmem_pressure = 1; |
4b1327be WW |
7283 | if (gfp_mask & __GFP_NOFAIL) { |
7284 | page_counter_charge(&memcg->tcpmem, nr_pages); | |
7285 | return true; | |
7286 | } | |
f7e1cb6e | 7287 | return false; |
11092087 | 7288 | } |
d886f4e4 | 7289 | |
4b1327be WW |
7290 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) { |
7291 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); | |
f7e1cb6e | 7292 | return true; |
4b1327be | 7293 | } |
f7e1cb6e | 7294 | |
11092087 JW |
7295 | return false; |
7296 | } | |
7297 | ||
7298 | /** | |
7299 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7300 | * @memcg: memcg to uncharge |
7301 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7302 | */ |
7303 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7304 | { | |
f7e1cb6e | 7305 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7306 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7307 | return; |
7308 | } | |
d886f4e4 | 7309 | |
c9019e9b | 7310 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7311 | |
475d0487 | 7312 | refill_stock(memcg, nr_pages); |
11092087 JW |
7313 | } |
7314 | ||
f7e1cb6e JW |
7315 | static int __init cgroup_memory(char *s) |
7316 | { | |
7317 | char *token; | |
7318 | ||
7319 | while ((token = strsep(&s, ",")) != NULL) { | |
7320 | if (!*token) | |
7321 | continue; | |
7322 | if (!strcmp(token, "nosocket")) | |
7323 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7324 | if (!strcmp(token, "nokmem")) |
7325 | cgroup_memory_nokmem = true; | |
b6c1a8af YS |
7326 | if (!strcmp(token, "nobpf")) |
7327 | cgroup_memory_nobpf = true; | |
f7e1cb6e | 7328 | } |
460a79e1 | 7329 | return 1; |
f7e1cb6e JW |
7330 | } |
7331 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7332 | |
2d11085e | 7333 | /* |
1081312f MH |
7334 | * subsys_initcall() for memory controller. |
7335 | * | |
308167fc SAS |
7336 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7337 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7338 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7339 | * should be initialized from here. | |
2d11085e MH |
7340 | */ |
7341 | static int __init mem_cgroup_init(void) | |
7342 | { | |
95a045f6 JW |
7343 | int cpu, node; |
7344 | ||
f3344adf MS |
7345 | /* |
7346 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7347 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7348 | * to work fine, we should make sure that the overfill threshold can't | |
7349 | * exceed S32_MAX / PAGE_SIZE. | |
7350 | */ | |
7351 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7352 | ||
308167fc SAS |
7353 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7354 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7355 | |
7356 | for_each_possible_cpu(cpu) | |
7357 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7358 | drain_local_stock); | |
7359 | ||
7360 | for_each_node(node) { | |
7361 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7362 | |
7363 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7364 | node_online(node) ? node : NUMA_NO_NODE); | |
7365 | ||
ef8f2327 | 7366 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7367 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7368 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7369 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7370 | } | |
7371 | ||
2d11085e MH |
7372 | return 0; |
7373 | } | |
7374 | subsys_initcall(mem_cgroup_init); | |
21afa38e | 7375 | |
e55b9f96 | 7376 | #ifdef CONFIG_SWAP |
358c07fc AB |
7377 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7378 | { | |
1c2d479a | 7379 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7380 | /* |
7381 | * The root cgroup cannot be destroyed, so it's refcount must | |
7382 | * always be >= 1. | |
7383 | */ | |
7848ed62 | 7384 | if (WARN_ON_ONCE(mem_cgroup_is_root(memcg))) { |
358c07fc AB |
7385 | VM_BUG_ON(1); |
7386 | break; | |
7387 | } | |
7388 | memcg = parent_mem_cgroup(memcg); | |
7389 | if (!memcg) | |
7390 | memcg = root_mem_cgroup; | |
7391 | } | |
7392 | return memcg; | |
7393 | } | |
7394 | ||
21afa38e JW |
7395 | /** |
7396 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
3ecb0087 | 7397 | * @folio: folio whose memsw charge to transfer |
21afa38e JW |
7398 | * @entry: swap entry to move the charge to |
7399 | * | |
3ecb0087 | 7400 | * Transfer the memsw charge of @folio to @entry. |
21afa38e | 7401 | */ |
3ecb0087 | 7402 | void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry) |
21afa38e | 7403 | { |
1f47b61f | 7404 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7405 | unsigned int nr_entries; |
21afa38e JW |
7406 | unsigned short oldid; |
7407 | ||
3ecb0087 MWO |
7408 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
7409 | VM_BUG_ON_FOLIO(folio_ref_count(folio), folio); | |
21afa38e | 7410 | |
76358ab5 AS |
7411 | if (mem_cgroup_disabled()) |
7412 | return; | |
7413 | ||
b94c4e94 | 7414 | if (!do_memsw_account()) |
21afa38e JW |
7415 | return; |
7416 | ||
3ecb0087 | 7417 | memcg = folio_memcg(folio); |
21afa38e | 7418 | |
3ecb0087 | 7419 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
21afa38e JW |
7420 | if (!memcg) |
7421 | return; | |
7422 | ||
1f47b61f VD |
7423 | /* |
7424 | * In case the memcg owning these pages has been offlined and doesn't | |
7425 | * have an ID allocated to it anymore, charge the closest online | |
7426 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7427 | */ | |
7428 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
3ecb0087 | 7429 | nr_entries = folio_nr_pages(folio); |
d6810d73 HY |
7430 | /* Get references for the tail pages, too */ |
7431 | if (nr_entries > 1) | |
7432 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7433 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7434 | nr_entries); | |
3ecb0087 | 7435 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 7436 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 7437 | |
3ecb0087 | 7438 | folio->memcg_data = 0; |
21afa38e JW |
7439 | |
7440 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7441 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7442 | |
b25806dc | 7443 | if (memcg != swap_memcg) { |
1f47b61f | 7444 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7445 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7446 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7447 | } |
7448 | ||
ce9ce665 SAS |
7449 | /* |
7450 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7451 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7452 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7453 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 | 7454 | */ |
be3e67b5 | 7455 | memcg_stats_lock(); |
6e0110c2 | 7456 | mem_cgroup_charge_statistics(memcg, -nr_entries); |
be3e67b5 | 7457 | memcg_stats_unlock(); |
3ecb0087 | 7458 | memcg_check_events(memcg, folio_nid(folio)); |
73f576c0 | 7459 | |
1a3e1f40 | 7460 | css_put(&memcg->css); |
21afa38e JW |
7461 | } |
7462 | ||
38d8b4e6 | 7463 | /** |
e2e3fdc7 MWO |
7464 | * __mem_cgroup_try_charge_swap - try charging swap space for a folio |
7465 | * @folio: folio being added to swap | |
37e84351 VD |
7466 | * @entry: swap entry to charge |
7467 | * | |
e2e3fdc7 | 7468 | * Try to charge @folio's memcg for the swap space at @entry. |
37e84351 VD |
7469 | * |
7470 | * Returns 0 on success, -ENOMEM on failure. | |
7471 | */ | |
e2e3fdc7 | 7472 | int __mem_cgroup_try_charge_swap(struct folio *folio, swp_entry_t entry) |
37e84351 | 7473 | { |
e2e3fdc7 | 7474 | unsigned int nr_pages = folio_nr_pages(folio); |
37e84351 | 7475 | struct page_counter *counter; |
38d8b4e6 | 7476 | struct mem_cgroup *memcg; |
37e84351 VD |
7477 | unsigned short oldid; |
7478 | ||
b94c4e94 | 7479 | if (do_memsw_account()) |
37e84351 VD |
7480 | return 0; |
7481 | ||
e2e3fdc7 | 7482 | memcg = folio_memcg(folio); |
37e84351 | 7483 | |
e2e3fdc7 | 7484 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
37e84351 VD |
7485 | if (!memcg) |
7486 | return 0; | |
7487 | ||
f3a53a3a TH |
7488 | if (!entry.val) { |
7489 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7490 | return 0; |
f3a53a3a | 7491 | } |
bb98f2c5 | 7492 | |
1f47b61f VD |
7493 | memcg = mem_cgroup_id_get_online(memcg); |
7494 | ||
b25806dc | 7495 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7496 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7497 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7498 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7499 | mem_cgroup_id_put(memcg); |
37e84351 | 7500 | return -ENOMEM; |
1f47b61f | 7501 | } |
37e84351 | 7502 | |
38d8b4e6 HY |
7503 | /* Get references for the tail pages, too */ |
7504 | if (nr_pages > 1) | |
7505 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7506 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
e2e3fdc7 | 7507 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 7508 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7509 | |
37e84351 VD |
7510 | return 0; |
7511 | } | |
7512 | ||
21afa38e | 7513 | /** |
01c4b28c | 7514 | * __mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7515 | * @entry: swap entry to uncharge |
38d8b4e6 | 7516 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7517 | */ |
01c4b28c | 7518 | void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7519 | { |
7520 | struct mem_cgroup *memcg; | |
7521 | unsigned short id; | |
7522 | ||
c91bdc93 JW |
7523 | if (mem_cgroup_disabled()) |
7524 | return; | |
7525 | ||
38d8b4e6 | 7526 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7527 | rcu_read_lock(); |
adbe427b | 7528 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7529 | if (memcg) { |
b25806dc | 7530 | if (!mem_cgroup_is_root(memcg)) { |
b94c4e94 | 7531 | if (do_memsw_account()) |
38d8b4e6 | 7532 | page_counter_uncharge(&memcg->memsw, nr_pages); |
b94c4e94 JW |
7533 | else |
7534 | page_counter_uncharge(&memcg->swap, nr_pages); | |
37e84351 | 7535 | } |
c9019e9b | 7536 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7537 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7538 | } |
7539 | rcu_read_unlock(); | |
7540 | } | |
7541 | ||
d8b38438 VD |
7542 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7543 | { | |
7544 | long nr_swap_pages = get_nr_swap_pages(); | |
7545 | ||
b25806dc | 7546 | if (mem_cgroup_disabled() || do_memsw_account()) |
d8b38438 | 7547 | return nr_swap_pages; |
7848ed62 | 7548 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) |
d8b38438 | 7549 | nr_swap_pages = min_t(long, nr_swap_pages, |
bbec2e15 | 7550 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7551 | page_counter_read(&memcg->swap)); |
7552 | return nr_swap_pages; | |
7553 | } | |
7554 | ||
9202d527 | 7555 | bool mem_cgroup_swap_full(struct folio *folio) |
5ccc5aba VD |
7556 | { |
7557 | struct mem_cgroup *memcg; | |
7558 | ||
9202d527 | 7559 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
5ccc5aba VD |
7560 | |
7561 | if (vm_swap_full()) | |
7562 | return true; | |
b25806dc | 7563 | if (do_memsw_account()) |
5ccc5aba VD |
7564 | return false; |
7565 | ||
9202d527 | 7566 | memcg = folio_memcg(folio); |
5ccc5aba VD |
7567 | if (!memcg) |
7568 | return false; | |
7569 | ||
7848ed62 | 7570 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
4b82ab4f JK |
7571 | unsigned long usage = page_counter_read(&memcg->swap); |
7572 | ||
7573 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7574 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7575 | return true; |
4b82ab4f | 7576 | } |
5ccc5aba VD |
7577 | |
7578 | return false; | |
7579 | } | |
7580 | ||
eccb52e7 | 7581 | static int __init setup_swap_account(char *s) |
21afa38e | 7582 | { |
b25806dc JW |
7583 | pr_warn_once("The swapaccount= commandline option is deprecated. " |
7584 | "Please report your usecase to linux-mm@kvack.org if you " | |
7585 | "depend on this functionality.\n"); | |
21afa38e JW |
7586 | return 1; |
7587 | } | |
eccb52e7 | 7588 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7589 | |
37e84351 VD |
7590 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7591 | struct cftype *cft) | |
7592 | { | |
7593 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7594 | ||
7595 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7596 | } | |
7597 | ||
4b82ab4f JK |
7598 | static int swap_high_show(struct seq_file *m, void *v) |
7599 | { | |
7600 | return seq_puts_memcg_tunable(m, | |
7601 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7602 | } | |
7603 | ||
7604 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7605 | char *buf, size_t nbytes, loff_t off) | |
7606 | { | |
7607 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7608 | unsigned long high; | |
7609 | int err; | |
7610 | ||
7611 | buf = strstrip(buf); | |
7612 | err = page_counter_memparse(buf, "max", &high); | |
7613 | if (err) | |
7614 | return err; | |
7615 | ||
7616 | page_counter_set_high(&memcg->swap, high); | |
7617 | ||
7618 | return nbytes; | |
7619 | } | |
7620 | ||
37e84351 VD |
7621 | static int swap_max_show(struct seq_file *m, void *v) |
7622 | { | |
677dc973 CD |
7623 | return seq_puts_memcg_tunable(m, |
7624 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7625 | } |
7626 | ||
7627 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7628 | char *buf, size_t nbytes, loff_t off) | |
7629 | { | |
7630 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7631 | unsigned long max; | |
7632 | int err; | |
7633 | ||
7634 | buf = strstrip(buf); | |
7635 | err = page_counter_memparse(buf, "max", &max); | |
7636 | if (err) | |
7637 | return err; | |
7638 | ||
be09102b | 7639 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7640 | |
7641 | return nbytes; | |
7642 | } | |
7643 | ||
f3a53a3a TH |
7644 | static int swap_events_show(struct seq_file *m, void *v) |
7645 | { | |
aa9694bb | 7646 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7647 | |
4b82ab4f JK |
7648 | seq_printf(m, "high %lu\n", |
7649 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7650 | seq_printf(m, "max %lu\n", |
7651 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7652 | seq_printf(m, "fail %lu\n", | |
7653 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7654 | ||
7655 | return 0; | |
7656 | } | |
7657 | ||
37e84351 VD |
7658 | static struct cftype swap_files[] = { |
7659 | { | |
7660 | .name = "swap.current", | |
7661 | .flags = CFTYPE_NOT_ON_ROOT, | |
7662 | .read_u64 = swap_current_read, | |
7663 | }, | |
4b82ab4f JK |
7664 | { |
7665 | .name = "swap.high", | |
7666 | .flags = CFTYPE_NOT_ON_ROOT, | |
7667 | .seq_show = swap_high_show, | |
7668 | .write = swap_high_write, | |
7669 | }, | |
37e84351 VD |
7670 | { |
7671 | .name = "swap.max", | |
7672 | .flags = CFTYPE_NOT_ON_ROOT, | |
7673 | .seq_show = swap_max_show, | |
7674 | .write = swap_max_write, | |
7675 | }, | |
f3a53a3a TH |
7676 | { |
7677 | .name = "swap.events", | |
7678 | .flags = CFTYPE_NOT_ON_ROOT, | |
7679 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7680 | .seq_show = swap_events_show, | |
7681 | }, | |
37e84351 VD |
7682 | { } /* terminate */ |
7683 | }; | |
7684 | ||
eccb52e7 | 7685 | static struct cftype memsw_files[] = { |
21afa38e JW |
7686 | { |
7687 | .name = "memsw.usage_in_bytes", | |
7688 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7689 | .read_u64 = mem_cgroup_read_u64, | |
7690 | }, | |
7691 | { | |
7692 | .name = "memsw.max_usage_in_bytes", | |
7693 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7694 | .write = mem_cgroup_reset, | |
7695 | .read_u64 = mem_cgroup_read_u64, | |
7696 | }, | |
7697 | { | |
7698 | .name = "memsw.limit_in_bytes", | |
7699 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7700 | .write = mem_cgroup_write, | |
7701 | .read_u64 = mem_cgroup_read_u64, | |
7702 | }, | |
7703 | { | |
7704 | .name = "memsw.failcnt", | |
7705 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7706 | .write = mem_cgroup_reset, | |
7707 | .read_u64 = mem_cgroup_read_u64, | |
7708 | }, | |
7709 | { }, /* terminate */ | |
7710 | }; | |
7711 | ||
f4840ccf JW |
7712 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
7713 | /** | |
7714 | * obj_cgroup_may_zswap - check if this cgroup can zswap | |
7715 | * @objcg: the object cgroup | |
7716 | * | |
7717 | * Check if the hierarchical zswap limit has been reached. | |
7718 | * | |
7719 | * This doesn't check for specific headroom, and it is not atomic | |
7720 | * either. But with zswap, the size of the allocation is only known | |
7721 | * once compression has occured, and this optimistic pre-check avoids | |
7722 | * spending cycles on compression when there is already no room left | |
7723 | * or zswap is disabled altogether somewhere in the hierarchy. | |
7724 | */ | |
7725 | bool obj_cgroup_may_zswap(struct obj_cgroup *objcg) | |
7726 | { | |
7727 | struct mem_cgroup *memcg, *original_memcg; | |
7728 | bool ret = true; | |
7729 | ||
7730 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7731 | return true; | |
7732 | ||
7733 | original_memcg = get_mem_cgroup_from_objcg(objcg); | |
7848ed62 | 7734 | for (memcg = original_memcg; !mem_cgroup_is_root(memcg); |
f4840ccf JW |
7735 | memcg = parent_mem_cgroup(memcg)) { |
7736 | unsigned long max = READ_ONCE(memcg->zswap_max); | |
7737 | unsigned long pages; | |
7738 | ||
7739 | if (max == PAGE_COUNTER_MAX) | |
7740 | continue; | |
7741 | if (max == 0) { | |
7742 | ret = false; | |
7743 | break; | |
7744 | } | |
7745 | ||
7746 | cgroup_rstat_flush(memcg->css.cgroup); | |
7747 | pages = memcg_page_state(memcg, MEMCG_ZSWAP_B) / PAGE_SIZE; | |
7748 | if (pages < max) | |
7749 | continue; | |
7750 | ret = false; | |
7751 | break; | |
7752 | } | |
7753 | mem_cgroup_put(original_memcg); | |
7754 | return ret; | |
7755 | } | |
7756 | ||
7757 | /** | |
7758 | * obj_cgroup_charge_zswap - charge compression backend memory | |
7759 | * @objcg: the object cgroup | |
7760 | * @size: size of compressed object | |
7761 | * | |
7762 | * This forces the charge after obj_cgroup_may_swap() allowed | |
7763 | * compression and storage in zwap for this cgroup to go ahead. | |
7764 | */ | |
7765 | void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size) | |
7766 | { | |
7767 | struct mem_cgroup *memcg; | |
7768 | ||
7769 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7770 | return; | |
7771 | ||
7772 | VM_WARN_ON_ONCE(!(current->flags & PF_MEMALLOC)); | |
7773 | ||
7774 | /* PF_MEMALLOC context, charging must succeed */ | |
7775 | if (obj_cgroup_charge(objcg, GFP_KERNEL, size)) | |
7776 | VM_WARN_ON_ONCE(1); | |
7777 | ||
7778 | rcu_read_lock(); | |
7779 | memcg = obj_cgroup_memcg(objcg); | |
7780 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, size); | |
7781 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, 1); | |
7782 | rcu_read_unlock(); | |
7783 | } | |
7784 | ||
7785 | /** | |
7786 | * obj_cgroup_uncharge_zswap - uncharge compression backend memory | |
7787 | * @objcg: the object cgroup | |
7788 | * @size: size of compressed object | |
7789 | * | |
7790 | * Uncharges zswap memory on page in. | |
7791 | */ | |
7792 | void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size) | |
7793 | { | |
7794 | struct mem_cgroup *memcg; | |
7795 | ||
7796 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7797 | return; | |
7798 | ||
7799 | obj_cgroup_uncharge(objcg, size); | |
7800 | ||
7801 | rcu_read_lock(); | |
7802 | memcg = obj_cgroup_memcg(objcg); | |
7803 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, -size); | |
7804 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, -1); | |
7805 | rcu_read_unlock(); | |
7806 | } | |
7807 | ||
7808 | static u64 zswap_current_read(struct cgroup_subsys_state *css, | |
7809 | struct cftype *cft) | |
7810 | { | |
7811 | cgroup_rstat_flush(css->cgroup); | |
7812 | return memcg_page_state(mem_cgroup_from_css(css), MEMCG_ZSWAP_B); | |
7813 | } | |
7814 | ||
7815 | static int zswap_max_show(struct seq_file *m, void *v) | |
7816 | { | |
7817 | return seq_puts_memcg_tunable(m, | |
7818 | READ_ONCE(mem_cgroup_from_seq(m)->zswap_max)); | |
7819 | } | |
7820 | ||
7821 | static ssize_t zswap_max_write(struct kernfs_open_file *of, | |
7822 | char *buf, size_t nbytes, loff_t off) | |
7823 | { | |
7824 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7825 | unsigned long max; | |
7826 | int err; | |
7827 | ||
7828 | buf = strstrip(buf); | |
7829 | err = page_counter_memparse(buf, "max", &max); | |
7830 | if (err) | |
7831 | return err; | |
7832 | ||
7833 | xchg(&memcg->zswap_max, max); | |
7834 | ||
7835 | return nbytes; | |
7836 | } | |
7837 | ||
7838 | static struct cftype zswap_files[] = { | |
7839 | { | |
7840 | .name = "zswap.current", | |
7841 | .flags = CFTYPE_NOT_ON_ROOT, | |
7842 | .read_u64 = zswap_current_read, | |
7843 | }, | |
7844 | { | |
7845 | .name = "zswap.max", | |
7846 | .flags = CFTYPE_NOT_ON_ROOT, | |
7847 | .seq_show = zswap_max_show, | |
7848 | .write = zswap_max_write, | |
7849 | }, | |
7850 | { } /* terminate */ | |
7851 | }; | |
7852 | #endif /* CONFIG_MEMCG_KMEM && CONFIG_ZSWAP */ | |
7853 | ||
21afa38e JW |
7854 | static int __init mem_cgroup_swap_init(void) |
7855 | { | |
2d1c4980 | 7856 | if (mem_cgroup_disabled()) |
eccb52e7 JW |
7857 | return 0; |
7858 | ||
7859 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7860 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
f4840ccf JW |
7861 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
7862 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, zswap_files)); | |
7863 | #endif | |
21afa38e JW |
7864 | return 0; |
7865 | } | |
b25806dc | 7866 | subsys_initcall(mem_cgroup_swap_init); |
21afa38e | 7867 | |
e55b9f96 | 7868 | #endif /* CONFIG_SWAP */ |