at v5.2-rc3 33 kB view raw
1/* SPDX-License-Identifier: GPL-2.0-or-later */ 2/* memcontrol.h - Memory Controller 3 * 4 * Copyright IBM Corporation, 2007 5 * Author Balbir Singh <balbir@linux.vnet.ibm.com> 6 * 7 * Copyright 2007 OpenVZ SWsoft Inc 8 * Author: Pavel Emelianov <xemul@openvz.org> 9 */ 10 11#ifndef _LINUX_MEMCONTROL_H 12#define _LINUX_MEMCONTROL_H 13#include <linux/cgroup.h> 14#include <linux/vm_event_item.h> 15#include <linux/hardirq.h> 16#include <linux/jump_label.h> 17#include <linux/page_counter.h> 18#include <linux/vmpressure.h> 19#include <linux/eventfd.h> 20#include <linux/mm.h> 21#include <linux/vmstat.h> 22#include <linux/writeback.h> 23#include <linux/page-flags.h> 24 25struct mem_cgroup; 26struct page; 27struct mm_struct; 28struct kmem_cache; 29 30/* Cgroup-specific page state, on top of universal node page state */ 31enum memcg_stat_item { 32 MEMCG_CACHE = NR_VM_NODE_STAT_ITEMS, 33 MEMCG_RSS, 34 MEMCG_RSS_HUGE, 35 MEMCG_SWAP, 36 MEMCG_SOCK, 37 /* XXX: why are these zone and not node counters? */ 38 MEMCG_KERNEL_STACK_KB, 39 MEMCG_NR_STAT, 40}; 41 42enum memcg_memory_event { 43 MEMCG_LOW, 44 MEMCG_HIGH, 45 MEMCG_MAX, 46 MEMCG_OOM, 47 MEMCG_OOM_KILL, 48 MEMCG_SWAP_MAX, 49 MEMCG_SWAP_FAIL, 50 MEMCG_NR_MEMORY_EVENTS, 51}; 52 53enum mem_cgroup_protection { 54 MEMCG_PROT_NONE, 55 MEMCG_PROT_LOW, 56 MEMCG_PROT_MIN, 57}; 58 59struct mem_cgroup_reclaim_cookie { 60 pg_data_t *pgdat; 61 int priority; 62 unsigned int generation; 63}; 64 65#ifdef CONFIG_MEMCG 66 67#define MEM_CGROUP_ID_SHIFT 16 68#define MEM_CGROUP_ID_MAX USHRT_MAX 69 70struct mem_cgroup_id { 71 int id; 72 refcount_t ref; 73}; 74 75/* 76 * Per memcg event counter is incremented at every pagein/pageout. With THP, 77 * it will be incremated by the number of pages. This counter is used for 78 * for trigger some periodic events. This is straightforward and better 79 * than using jiffies etc. to handle periodic memcg event. 80 */ 81enum mem_cgroup_events_target { 82 MEM_CGROUP_TARGET_THRESH, 83 MEM_CGROUP_TARGET_SOFTLIMIT, 84 MEM_CGROUP_TARGET_NUMAINFO, 85 MEM_CGROUP_NTARGETS, 86}; 87 88struct memcg_vmstats_percpu { 89 long stat[MEMCG_NR_STAT]; 90 unsigned long events[NR_VM_EVENT_ITEMS]; 91 unsigned long nr_page_events; 92 unsigned long targets[MEM_CGROUP_NTARGETS]; 93}; 94 95struct mem_cgroup_reclaim_iter { 96 struct mem_cgroup *position; 97 /* scan generation, increased every round-trip */ 98 unsigned int generation; 99}; 100 101struct lruvec_stat { 102 long count[NR_VM_NODE_STAT_ITEMS]; 103}; 104 105/* 106 * Bitmap of shrinker::id corresponding to memcg-aware shrinkers, 107 * which have elements charged to this memcg. 108 */ 109struct memcg_shrinker_map { 110 struct rcu_head rcu; 111 unsigned long map[0]; 112}; 113 114/* 115 * per-zone information in memory controller. 116 */ 117struct mem_cgroup_per_node { 118 struct lruvec lruvec; 119 120 struct lruvec_stat __percpu *lruvec_stat_cpu; 121 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS]; 122 atomic_long_t lruvec_stat_local[NR_VM_NODE_STAT_ITEMS]; 123 124 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS]; 125 126 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1]; 127 128#ifdef CONFIG_MEMCG_KMEM 129 struct memcg_shrinker_map __rcu *shrinker_map; 130#endif 131 struct rb_node tree_node; /* RB tree node */ 132 unsigned long usage_in_excess;/* Set to the value by which */ 133 /* the soft limit is exceeded*/ 134 bool on_tree; 135 bool congested; /* memcg has many dirty pages */ 136 /* backed by a congested BDI */ 137 138 struct mem_cgroup *memcg; /* Back pointer, we cannot */ 139 /* use container_of */ 140}; 141 142struct mem_cgroup_threshold { 143 struct eventfd_ctx *eventfd; 144 unsigned long threshold; 145}; 146 147/* For threshold */ 148struct mem_cgroup_threshold_ary { 149 /* An array index points to threshold just below or equal to usage. */ 150 int current_threshold; 151 /* Size of entries[] */ 152 unsigned int size; 153 /* Array of thresholds */ 154 struct mem_cgroup_threshold entries[0]; 155}; 156 157struct mem_cgroup_thresholds { 158 /* Primary thresholds array */ 159 struct mem_cgroup_threshold_ary *primary; 160 /* 161 * Spare threshold array. 162 * This is needed to make mem_cgroup_unregister_event() "never fail". 163 * It must be able to store at least primary->size - 1 entries. 164 */ 165 struct mem_cgroup_threshold_ary *spare; 166}; 167 168enum memcg_kmem_state { 169 KMEM_NONE, 170 KMEM_ALLOCATED, 171 KMEM_ONLINE, 172}; 173 174#if defined(CONFIG_SMP) 175struct memcg_padding { 176 char x[0]; 177} ____cacheline_internodealigned_in_smp; 178#define MEMCG_PADDING(name) struct memcg_padding name; 179#else 180#define MEMCG_PADDING(name) 181#endif 182 183/* 184 * The memory controller data structure. The memory controller controls both 185 * page cache and RSS per cgroup. We would eventually like to provide 186 * statistics based on the statistics developed by Rik Van Riel for clock-pro, 187 * to help the administrator determine what knobs to tune. 188 */ 189struct mem_cgroup { 190 struct cgroup_subsys_state css; 191 192 /* Private memcg ID. Used to ID objects that outlive the cgroup */ 193 struct mem_cgroup_id id; 194 195 /* Accounted resources */ 196 struct page_counter memory; 197 struct page_counter swap; 198 199 /* Legacy consumer-oriented counters */ 200 struct page_counter memsw; 201 struct page_counter kmem; 202 struct page_counter tcpmem; 203 204 /* Upper bound of normal memory consumption range */ 205 unsigned long high; 206 207 /* Range enforcement for interrupt charges */ 208 struct work_struct high_work; 209 210 unsigned long soft_limit; 211 212 /* vmpressure notifications */ 213 struct vmpressure vmpressure; 214 215 /* 216 * Should the accounting and control be hierarchical, per subtree? 217 */ 218 bool use_hierarchy; 219 220 /* 221 * Should the OOM killer kill all belonging tasks, had it kill one? 222 */ 223 bool oom_group; 224 225 /* protected by memcg_oom_lock */ 226 bool oom_lock; 227 int under_oom; 228 229 int swappiness; 230 /* OOM-Killer disable */ 231 int oom_kill_disable; 232 233 /* memory.events */ 234 struct cgroup_file events_file; 235 236 /* handle for "memory.swap.events" */ 237 struct cgroup_file swap_events_file; 238 239 /* protect arrays of thresholds */ 240 struct mutex thresholds_lock; 241 242 /* thresholds for memory usage. RCU-protected */ 243 struct mem_cgroup_thresholds thresholds; 244 245 /* thresholds for mem+swap usage. RCU-protected */ 246 struct mem_cgroup_thresholds memsw_thresholds; 247 248 /* For oom notifier event fd */ 249 struct list_head oom_notify; 250 251 /* 252 * Should we move charges of a task when a task is moved into this 253 * mem_cgroup ? And what type of charges should we move ? 254 */ 255 unsigned long move_charge_at_immigrate; 256 /* taken only while moving_account > 0 */ 257 spinlock_t move_lock; 258 unsigned long move_lock_flags; 259 260 MEMCG_PADDING(_pad1_); 261 262 /* 263 * set > 0 if pages under this cgroup are moving to other cgroup. 264 */ 265 atomic_t moving_account; 266 struct task_struct *move_lock_task; 267 268 /* memory.stat */ 269 struct memcg_vmstats_percpu __percpu *vmstats_percpu; 270 271 MEMCG_PADDING(_pad2_); 272 273 atomic_long_t vmstats[MEMCG_NR_STAT]; 274 atomic_long_t vmstats_local[MEMCG_NR_STAT]; 275 276 atomic_long_t vmevents[NR_VM_EVENT_ITEMS]; 277 atomic_long_t vmevents_local[NR_VM_EVENT_ITEMS]; 278 279 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS]; 280 281 unsigned long socket_pressure; 282 283 /* Legacy tcp memory accounting */ 284 bool tcpmem_active; 285 int tcpmem_pressure; 286 287#ifdef CONFIG_MEMCG_KMEM 288 /* Index in the kmem_cache->memcg_params.memcg_caches array */ 289 int kmemcg_id; 290 enum memcg_kmem_state kmem_state; 291 struct list_head kmem_caches; 292#endif 293 294 int last_scanned_node; 295#if MAX_NUMNODES > 1 296 nodemask_t scan_nodes; 297 atomic_t numainfo_events; 298 atomic_t numainfo_updating; 299#endif 300 301#ifdef CONFIG_CGROUP_WRITEBACK 302 struct list_head cgwb_list; 303 struct wb_domain cgwb_domain; 304#endif 305 306 /* List of events which userspace want to receive */ 307 struct list_head event_list; 308 spinlock_t event_list_lock; 309 310 struct mem_cgroup_per_node *nodeinfo[0]; 311 /* WARNING: nodeinfo must be the last member here */ 312}; 313 314/* 315 * size of first charge trial. "32" comes from vmscan.c's magic value. 316 * TODO: maybe necessary to use big numbers in big irons. 317 */ 318#define MEMCG_CHARGE_BATCH 32U 319 320extern struct mem_cgroup *root_mem_cgroup; 321 322static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 323{ 324 return (memcg == root_mem_cgroup); 325} 326 327static inline bool mem_cgroup_disabled(void) 328{ 329 return !cgroup_subsys_enabled(memory_cgrp_subsys); 330} 331 332enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, 333 struct mem_cgroup *memcg); 334 335int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, 336 gfp_t gfp_mask, struct mem_cgroup **memcgp, 337 bool compound); 338int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, 339 gfp_t gfp_mask, struct mem_cgroup **memcgp, 340 bool compound); 341void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, 342 bool lrucare, bool compound); 343void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, 344 bool compound); 345void mem_cgroup_uncharge(struct page *page); 346void mem_cgroup_uncharge_list(struct list_head *page_list); 347 348void mem_cgroup_migrate(struct page *oldpage, struct page *newpage); 349 350static struct mem_cgroup_per_node * 351mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid) 352{ 353 return memcg->nodeinfo[nid]; 354} 355 356/** 357 * mem_cgroup_lruvec - get the lru list vector for a node or a memcg zone 358 * @node: node of the wanted lruvec 359 * @memcg: memcg of the wanted lruvec 360 * 361 * Returns the lru list vector holding pages for a given @node or a given 362 * @memcg and @zone. This can be the node lruvec, if the memory controller 363 * is disabled. 364 */ 365static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat, 366 struct mem_cgroup *memcg) 367{ 368 struct mem_cgroup_per_node *mz; 369 struct lruvec *lruvec; 370 371 if (mem_cgroup_disabled()) { 372 lruvec = node_lruvec(pgdat); 373 goto out; 374 } 375 376 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); 377 lruvec = &mz->lruvec; 378out: 379 /* 380 * Since a node can be onlined after the mem_cgroup was created, 381 * we have to be prepared to initialize lruvec->pgdat here; 382 * and if offlined then reonlined, we need to reinitialize it. 383 */ 384 if (unlikely(lruvec->pgdat != pgdat)) 385 lruvec->pgdat = pgdat; 386 return lruvec; 387} 388 389struct lruvec *mem_cgroup_page_lruvec(struct page *, struct pglist_data *); 390 391bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg); 392struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p); 393 394struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm); 395 396struct mem_cgroup *get_mem_cgroup_from_page(struct page *page); 397 398static inline 399struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){ 400 return css ? container_of(css, struct mem_cgroup, css) : NULL; 401} 402 403static inline void mem_cgroup_put(struct mem_cgroup *memcg) 404{ 405 if (memcg) 406 css_put(&memcg->css); 407} 408 409#define mem_cgroup_from_counter(counter, member) \ 410 container_of(counter, struct mem_cgroup, member) 411 412struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *, 413 struct mem_cgroup *, 414 struct mem_cgroup_reclaim_cookie *); 415void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *); 416int mem_cgroup_scan_tasks(struct mem_cgroup *, 417 int (*)(struct task_struct *, void *), void *); 418 419static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 420{ 421 if (mem_cgroup_disabled()) 422 return 0; 423 424 return memcg->id.id; 425} 426struct mem_cgroup *mem_cgroup_from_id(unsigned short id); 427 428static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 429{ 430 return mem_cgroup_from_css(seq_css(m)); 431} 432 433static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 434{ 435 struct mem_cgroup_per_node *mz; 436 437 if (mem_cgroup_disabled()) 438 return NULL; 439 440 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 441 return mz->memcg; 442} 443 444/** 445 * parent_mem_cgroup - find the accounting parent of a memcg 446 * @memcg: memcg whose parent to find 447 * 448 * Returns the parent memcg, or NULL if this is the root or the memory 449 * controller is in legacy no-hierarchy mode. 450 */ 451static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 452{ 453 if (!memcg->memory.parent) 454 return NULL; 455 return mem_cgroup_from_counter(memcg->memory.parent, memory); 456} 457 458static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, 459 struct mem_cgroup *root) 460{ 461 if (root == memcg) 462 return true; 463 if (!root->use_hierarchy) 464 return false; 465 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); 466} 467 468static inline bool mm_match_cgroup(struct mm_struct *mm, 469 struct mem_cgroup *memcg) 470{ 471 struct mem_cgroup *task_memcg; 472 bool match = false; 473 474 rcu_read_lock(); 475 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 476 if (task_memcg) 477 match = mem_cgroup_is_descendant(task_memcg, memcg); 478 rcu_read_unlock(); 479 return match; 480} 481 482struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page); 483ino_t page_cgroup_ino(struct page *page); 484 485static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 486{ 487 if (mem_cgroup_disabled()) 488 return true; 489 return !!(memcg->css.flags & CSS_ONLINE); 490} 491 492/* 493 * For memory reclaim. 494 */ 495int mem_cgroup_select_victim_node(struct mem_cgroup *memcg); 496 497void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, 498 int zid, int nr_pages); 499 500static inline 501unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 502 enum lru_list lru, int zone_idx) 503{ 504 struct mem_cgroup_per_node *mz; 505 506 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 507 return mz->lru_zone_size[zone_idx][lru]; 508} 509 510void mem_cgroup_handle_over_high(void); 511 512unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg); 513 514void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, 515 struct task_struct *p); 516 517void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg); 518 519static inline void mem_cgroup_enter_user_fault(void) 520{ 521 WARN_ON(current->in_user_fault); 522 current->in_user_fault = 1; 523} 524 525static inline void mem_cgroup_exit_user_fault(void) 526{ 527 WARN_ON(!current->in_user_fault); 528 current->in_user_fault = 0; 529} 530 531static inline bool task_in_memcg_oom(struct task_struct *p) 532{ 533 return p->memcg_in_oom; 534} 535 536bool mem_cgroup_oom_synchronize(bool wait); 537struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, 538 struct mem_cgroup *oom_domain); 539void mem_cgroup_print_oom_group(struct mem_cgroup *memcg); 540 541#ifdef CONFIG_MEMCG_SWAP 542extern int do_swap_account; 543#endif 544 545struct mem_cgroup *lock_page_memcg(struct page *page); 546void __unlock_page_memcg(struct mem_cgroup *memcg); 547void unlock_page_memcg(struct page *page); 548 549/* 550 * idx can be of type enum memcg_stat_item or node_stat_item. 551 * Keep in sync with memcg_exact_page_state(). 552 */ 553static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 554{ 555 long x = atomic_long_read(&memcg->vmstats[idx]); 556#ifdef CONFIG_SMP 557 if (x < 0) 558 x = 0; 559#endif 560 return x; 561} 562 563/* 564 * idx can be of type enum memcg_stat_item or node_stat_item. 565 * Keep in sync with memcg_exact_page_state(). 566 */ 567static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg, 568 int idx) 569{ 570 long x = atomic_long_read(&memcg->vmstats_local[idx]); 571#ifdef CONFIG_SMP 572 if (x < 0) 573 x = 0; 574#endif 575 return x; 576} 577 578void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val); 579 580/* idx can be of type enum memcg_stat_item or node_stat_item */ 581static inline void mod_memcg_state(struct mem_cgroup *memcg, 582 int idx, int val) 583{ 584 unsigned long flags; 585 586 local_irq_save(flags); 587 __mod_memcg_state(memcg, idx, val); 588 local_irq_restore(flags); 589} 590 591/** 592 * mod_memcg_page_state - update page state statistics 593 * @page: the page 594 * @idx: page state item to account 595 * @val: number of pages (positive or negative) 596 * 597 * The @page must be locked or the caller must use lock_page_memcg() 598 * to prevent double accounting when the page is concurrently being 599 * moved to another memcg: 600 * 601 * lock_page(page) or lock_page_memcg(page) 602 * if (TestClearPageState(page)) 603 * mod_memcg_page_state(page, state, -1); 604 * unlock_page(page) or unlock_page_memcg(page) 605 * 606 * Kernel pages are an exception to this, since they'll never move. 607 */ 608static inline void __mod_memcg_page_state(struct page *page, 609 int idx, int val) 610{ 611 if (page->mem_cgroup) 612 __mod_memcg_state(page->mem_cgroup, idx, val); 613} 614 615static inline void mod_memcg_page_state(struct page *page, 616 int idx, int val) 617{ 618 if (page->mem_cgroup) 619 mod_memcg_state(page->mem_cgroup, idx, val); 620} 621 622static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 623 enum node_stat_item idx) 624{ 625 struct mem_cgroup_per_node *pn; 626 long x; 627 628 if (mem_cgroup_disabled()) 629 return node_page_state(lruvec_pgdat(lruvec), idx); 630 631 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 632 x = atomic_long_read(&pn->lruvec_stat[idx]); 633#ifdef CONFIG_SMP 634 if (x < 0) 635 x = 0; 636#endif 637 return x; 638} 639 640static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 641 enum node_stat_item idx) 642{ 643 struct mem_cgroup_per_node *pn; 644 long x; 645 646 if (mem_cgroup_disabled()) 647 return node_page_state(lruvec_pgdat(lruvec), idx); 648 649 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 650 x = atomic_long_read(&pn->lruvec_stat_local[idx]); 651#ifdef CONFIG_SMP 652 if (x < 0) 653 x = 0; 654#endif 655 return x; 656} 657 658void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, 659 int val); 660 661static inline void mod_lruvec_state(struct lruvec *lruvec, 662 enum node_stat_item idx, int val) 663{ 664 unsigned long flags; 665 666 local_irq_save(flags); 667 __mod_lruvec_state(lruvec, idx, val); 668 local_irq_restore(flags); 669} 670 671static inline void __mod_lruvec_page_state(struct page *page, 672 enum node_stat_item idx, int val) 673{ 674 pg_data_t *pgdat = page_pgdat(page); 675 struct lruvec *lruvec; 676 677 /* Untracked pages have no memcg, no lruvec. Update only the node */ 678 if (!page->mem_cgroup) { 679 __mod_node_page_state(pgdat, idx, val); 680 return; 681 } 682 683 lruvec = mem_cgroup_lruvec(pgdat, page->mem_cgroup); 684 __mod_lruvec_state(lruvec, idx, val); 685} 686 687static inline void mod_lruvec_page_state(struct page *page, 688 enum node_stat_item idx, int val) 689{ 690 unsigned long flags; 691 692 local_irq_save(flags); 693 __mod_lruvec_page_state(page, idx, val); 694 local_irq_restore(flags); 695} 696 697unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 698 gfp_t gfp_mask, 699 unsigned long *total_scanned); 700 701void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, 702 unsigned long count); 703 704static inline void count_memcg_events(struct mem_cgroup *memcg, 705 enum vm_event_item idx, 706 unsigned long count) 707{ 708 unsigned long flags; 709 710 local_irq_save(flags); 711 __count_memcg_events(memcg, idx, count); 712 local_irq_restore(flags); 713} 714 715static inline void count_memcg_page_event(struct page *page, 716 enum vm_event_item idx) 717{ 718 if (page->mem_cgroup) 719 count_memcg_events(page->mem_cgroup, idx, 1); 720} 721 722static inline void count_memcg_event_mm(struct mm_struct *mm, 723 enum vm_event_item idx) 724{ 725 struct mem_cgroup *memcg; 726 727 if (mem_cgroup_disabled()) 728 return; 729 730 rcu_read_lock(); 731 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 732 if (likely(memcg)) 733 count_memcg_events(memcg, idx, 1); 734 rcu_read_unlock(); 735} 736 737static inline void memcg_memory_event(struct mem_cgroup *memcg, 738 enum memcg_memory_event event) 739{ 740 do { 741 atomic_long_inc(&memcg->memory_events[event]); 742 cgroup_file_notify(&memcg->events_file); 743 744 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) 745 break; 746 } while ((memcg = parent_mem_cgroup(memcg)) && 747 !mem_cgroup_is_root(memcg)); 748} 749 750static inline void memcg_memory_event_mm(struct mm_struct *mm, 751 enum memcg_memory_event event) 752{ 753 struct mem_cgroup *memcg; 754 755 if (mem_cgroup_disabled()) 756 return; 757 758 rcu_read_lock(); 759 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 760 if (likely(memcg)) 761 memcg_memory_event(memcg, event); 762 rcu_read_unlock(); 763} 764 765#ifdef CONFIG_TRANSPARENT_HUGEPAGE 766void mem_cgroup_split_huge_fixup(struct page *head); 767#endif 768 769#else /* CONFIG_MEMCG */ 770 771#define MEM_CGROUP_ID_SHIFT 0 772#define MEM_CGROUP_ID_MAX 0 773 774struct mem_cgroup; 775 776static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 777{ 778 return true; 779} 780 781static inline bool mem_cgroup_disabled(void) 782{ 783 return true; 784} 785 786static inline void memcg_memory_event(struct mem_cgroup *memcg, 787 enum memcg_memory_event event) 788{ 789} 790 791static inline void memcg_memory_event_mm(struct mm_struct *mm, 792 enum memcg_memory_event event) 793{ 794} 795 796static inline enum mem_cgroup_protection mem_cgroup_protected( 797 struct mem_cgroup *root, struct mem_cgroup *memcg) 798{ 799 return MEMCG_PROT_NONE; 800} 801 802static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, 803 gfp_t gfp_mask, 804 struct mem_cgroup **memcgp, 805 bool compound) 806{ 807 *memcgp = NULL; 808 return 0; 809} 810 811static inline int mem_cgroup_try_charge_delay(struct page *page, 812 struct mm_struct *mm, 813 gfp_t gfp_mask, 814 struct mem_cgroup **memcgp, 815 bool compound) 816{ 817 *memcgp = NULL; 818 return 0; 819} 820 821static inline void mem_cgroup_commit_charge(struct page *page, 822 struct mem_cgroup *memcg, 823 bool lrucare, bool compound) 824{ 825} 826 827static inline void mem_cgroup_cancel_charge(struct page *page, 828 struct mem_cgroup *memcg, 829 bool compound) 830{ 831} 832 833static inline void mem_cgroup_uncharge(struct page *page) 834{ 835} 836 837static inline void mem_cgroup_uncharge_list(struct list_head *page_list) 838{ 839} 840 841static inline void mem_cgroup_migrate(struct page *old, struct page *new) 842{ 843} 844 845static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat, 846 struct mem_cgroup *memcg) 847{ 848 return node_lruvec(pgdat); 849} 850 851static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page, 852 struct pglist_data *pgdat) 853{ 854 return &pgdat->lruvec; 855} 856 857static inline bool mm_match_cgroup(struct mm_struct *mm, 858 struct mem_cgroup *memcg) 859{ 860 return true; 861} 862 863static inline bool task_in_mem_cgroup(struct task_struct *task, 864 const struct mem_cgroup *memcg) 865{ 866 return true; 867} 868 869static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) 870{ 871 return NULL; 872} 873 874static inline struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) 875{ 876 return NULL; 877} 878 879static inline void mem_cgroup_put(struct mem_cgroup *memcg) 880{ 881} 882 883static inline struct mem_cgroup * 884mem_cgroup_iter(struct mem_cgroup *root, 885 struct mem_cgroup *prev, 886 struct mem_cgroup_reclaim_cookie *reclaim) 887{ 888 return NULL; 889} 890 891static inline void mem_cgroup_iter_break(struct mem_cgroup *root, 892 struct mem_cgroup *prev) 893{ 894} 895 896static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, 897 int (*fn)(struct task_struct *, void *), void *arg) 898{ 899 return 0; 900} 901 902static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 903{ 904 return 0; 905} 906 907static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) 908{ 909 WARN_ON_ONCE(id); 910 /* XXX: This should always return root_mem_cgroup */ 911 return NULL; 912} 913 914static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 915{ 916 return NULL; 917} 918 919static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 920{ 921 return NULL; 922} 923 924static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 925{ 926 return true; 927} 928 929static inline 930unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 931 enum lru_list lru, int zone_idx) 932{ 933 return 0; 934} 935 936static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) 937{ 938 return 0; 939} 940 941static inline void 942mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) 943{ 944} 945 946static inline void 947mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) 948{ 949} 950 951static inline struct mem_cgroup *lock_page_memcg(struct page *page) 952{ 953 return NULL; 954} 955 956static inline void __unlock_page_memcg(struct mem_cgroup *memcg) 957{ 958} 959 960static inline void unlock_page_memcg(struct page *page) 961{ 962} 963 964static inline void mem_cgroup_handle_over_high(void) 965{ 966} 967 968static inline void mem_cgroup_enter_user_fault(void) 969{ 970} 971 972static inline void mem_cgroup_exit_user_fault(void) 973{ 974} 975 976static inline bool task_in_memcg_oom(struct task_struct *p) 977{ 978 return false; 979} 980 981static inline bool mem_cgroup_oom_synchronize(bool wait) 982{ 983 return false; 984} 985 986static inline struct mem_cgroup *mem_cgroup_get_oom_group( 987 struct task_struct *victim, struct mem_cgroup *oom_domain) 988{ 989 return NULL; 990} 991 992static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) 993{ 994} 995 996static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 997{ 998 return 0; 999} 1000 1001static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg, 1002 int idx) 1003{ 1004 return 0; 1005} 1006 1007static inline void __mod_memcg_state(struct mem_cgroup *memcg, 1008 int idx, 1009 int nr) 1010{ 1011} 1012 1013static inline void mod_memcg_state(struct mem_cgroup *memcg, 1014 int idx, 1015 int nr) 1016{ 1017} 1018 1019static inline void __mod_memcg_page_state(struct page *page, 1020 int idx, 1021 int nr) 1022{ 1023} 1024 1025static inline void mod_memcg_page_state(struct page *page, 1026 int idx, 1027 int nr) 1028{ 1029} 1030 1031static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 1032 enum node_stat_item idx) 1033{ 1034 return node_page_state(lruvec_pgdat(lruvec), idx); 1035} 1036 1037static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 1038 enum node_stat_item idx) 1039{ 1040 return node_page_state(lruvec_pgdat(lruvec), idx); 1041} 1042 1043static inline void __mod_lruvec_state(struct lruvec *lruvec, 1044 enum node_stat_item idx, int val) 1045{ 1046 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); 1047} 1048 1049static inline void mod_lruvec_state(struct lruvec *lruvec, 1050 enum node_stat_item idx, int val) 1051{ 1052 mod_node_page_state(lruvec_pgdat(lruvec), idx, val); 1053} 1054 1055static inline void __mod_lruvec_page_state(struct page *page, 1056 enum node_stat_item idx, int val) 1057{ 1058 __mod_node_page_state(page_pgdat(page), idx, val); 1059} 1060 1061static inline void mod_lruvec_page_state(struct page *page, 1062 enum node_stat_item idx, int val) 1063{ 1064 mod_node_page_state(page_pgdat(page), idx, val); 1065} 1066 1067static inline 1068unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 1069 gfp_t gfp_mask, 1070 unsigned long *total_scanned) 1071{ 1072 return 0; 1073} 1074 1075static inline void mem_cgroup_split_huge_fixup(struct page *head) 1076{ 1077} 1078 1079static inline void count_memcg_events(struct mem_cgroup *memcg, 1080 enum vm_event_item idx, 1081 unsigned long count) 1082{ 1083} 1084 1085static inline void __count_memcg_events(struct mem_cgroup *memcg, 1086 enum vm_event_item idx, 1087 unsigned long count) 1088{ 1089} 1090 1091static inline void count_memcg_page_event(struct page *page, 1092 int idx) 1093{ 1094} 1095 1096static inline 1097void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx) 1098{ 1099} 1100#endif /* CONFIG_MEMCG */ 1101 1102/* idx can be of type enum memcg_stat_item or node_stat_item */ 1103static inline void __inc_memcg_state(struct mem_cgroup *memcg, 1104 int idx) 1105{ 1106 __mod_memcg_state(memcg, idx, 1); 1107} 1108 1109/* idx can be of type enum memcg_stat_item or node_stat_item */ 1110static inline void __dec_memcg_state(struct mem_cgroup *memcg, 1111 int idx) 1112{ 1113 __mod_memcg_state(memcg, idx, -1); 1114} 1115 1116/* idx can be of type enum memcg_stat_item or node_stat_item */ 1117static inline void __inc_memcg_page_state(struct page *page, 1118 int idx) 1119{ 1120 __mod_memcg_page_state(page, idx, 1); 1121} 1122 1123/* idx can be of type enum memcg_stat_item or node_stat_item */ 1124static inline void __dec_memcg_page_state(struct page *page, 1125 int idx) 1126{ 1127 __mod_memcg_page_state(page, idx, -1); 1128} 1129 1130static inline void __inc_lruvec_state(struct lruvec *lruvec, 1131 enum node_stat_item idx) 1132{ 1133 __mod_lruvec_state(lruvec, idx, 1); 1134} 1135 1136static inline void __dec_lruvec_state(struct lruvec *lruvec, 1137 enum node_stat_item idx) 1138{ 1139 __mod_lruvec_state(lruvec, idx, -1); 1140} 1141 1142static inline void __inc_lruvec_page_state(struct page *page, 1143 enum node_stat_item idx) 1144{ 1145 __mod_lruvec_page_state(page, idx, 1); 1146} 1147 1148static inline void __dec_lruvec_page_state(struct page *page, 1149 enum node_stat_item idx) 1150{ 1151 __mod_lruvec_page_state(page, idx, -1); 1152} 1153 1154/* idx can be of type enum memcg_stat_item or node_stat_item */ 1155static inline void inc_memcg_state(struct mem_cgroup *memcg, 1156 int idx) 1157{ 1158 mod_memcg_state(memcg, idx, 1); 1159} 1160 1161/* idx can be of type enum memcg_stat_item or node_stat_item */ 1162static inline void dec_memcg_state(struct mem_cgroup *memcg, 1163 int idx) 1164{ 1165 mod_memcg_state(memcg, idx, -1); 1166} 1167 1168/* idx can be of type enum memcg_stat_item or node_stat_item */ 1169static inline void inc_memcg_page_state(struct page *page, 1170 int idx) 1171{ 1172 mod_memcg_page_state(page, idx, 1); 1173} 1174 1175/* idx can be of type enum memcg_stat_item or node_stat_item */ 1176static inline void dec_memcg_page_state(struct page *page, 1177 int idx) 1178{ 1179 mod_memcg_page_state(page, idx, -1); 1180} 1181 1182static inline void inc_lruvec_state(struct lruvec *lruvec, 1183 enum node_stat_item idx) 1184{ 1185 mod_lruvec_state(lruvec, idx, 1); 1186} 1187 1188static inline void dec_lruvec_state(struct lruvec *lruvec, 1189 enum node_stat_item idx) 1190{ 1191 mod_lruvec_state(lruvec, idx, -1); 1192} 1193 1194static inline void inc_lruvec_page_state(struct page *page, 1195 enum node_stat_item idx) 1196{ 1197 mod_lruvec_page_state(page, idx, 1); 1198} 1199 1200static inline void dec_lruvec_page_state(struct page *page, 1201 enum node_stat_item idx) 1202{ 1203 mod_lruvec_page_state(page, idx, -1); 1204} 1205 1206#ifdef CONFIG_CGROUP_WRITEBACK 1207 1208struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb); 1209void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, 1210 unsigned long *pheadroom, unsigned long *pdirty, 1211 unsigned long *pwriteback); 1212 1213#else /* CONFIG_CGROUP_WRITEBACK */ 1214 1215static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) 1216{ 1217 return NULL; 1218} 1219 1220static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb, 1221 unsigned long *pfilepages, 1222 unsigned long *pheadroom, 1223 unsigned long *pdirty, 1224 unsigned long *pwriteback) 1225{ 1226} 1227 1228#endif /* CONFIG_CGROUP_WRITEBACK */ 1229 1230struct sock; 1231bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1232void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1233#ifdef CONFIG_MEMCG 1234extern struct static_key_false memcg_sockets_enabled_key; 1235#define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key) 1236void mem_cgroup_sk_alloc(struct sock *sk); 1237void mem_cgroup_sk_free(struct sock *sk); 1238static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1239{ 1240 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure) 1241 return true; 1242 do { 1243 if (time_before(jiffies, memcg->socket_pressure)) 1244 return true; 1245 } while ((memcg = parent_mem_cgroup(memcg))); 1246 return false; 1247} 1248#else 1249#define mem_cgroup_sockets_enabled 0 1250static inline void mem_cgroup_sk_alloc(struct sock *sk) { }; 1251static inline void mem_cgroup_sk_free(struct sock *sk) { }; 1252static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1253{ 1254 return false; 1255} 1256#endif 1257 1258struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep); 1259void memcg_kmem_put_cache(struct kmem_cache *cachep); 1260 1261#ifdef CONFIG_MEMCG_KMEM 1262int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order); 1263void __memcg_kmem_uncharge(struct page *page, int order); 1264int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, 1265 struct mem_cgroup *memcg); 1266 1267extern struct static_key_false memcg_kmem_enabled_key; 1268extern struct workqueue_struct *memcg_kmem_cache_wq; 1269 1270extern int memcg_nr_cache_ids; 1271void memcg_get_cache_ids(void); 1272void memcg_put_cache_ids(void); 1273 1274/* 1275 * Helper macro to loop through all memcg-specific caches. Callers must still 1276 * check if the cache is valid (it is either valid or NULL). 1277 * the slab_mutex must be held when looping through those caches 1278 */ 1279#define for_each_memcg_cache_index(_idx) \ 1280 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++) 1281 1282static inline bool memcg_kmem_enabled(void) 1283{ 1284 return static_branch_unlikely(&memcg_kmem_enabled_key); 1285} 1286 1287static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) 1288{ 1289 if (memcg_kmem_enabled()) 1290 return __memcg_kmem_charge(page, gfp, order); 1291 return 0; 1292} 1293 1294static inline void memcg_kmem_uncharge(struct page *page, int order) 1295{ 1296 if (memcg_kmem_enabled()) 1297 __memcg_kmem_uncharge(page, order); 1298} 1299 1300static inline int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, 1301 int order, struct mem_cgroup *memcg) 1302{ 1303 if (memcg_kmem_enabled()) 1304 return __memcg_kmem_charge_memcg(page, gfp, order, memcg); 1305 return 0; 1306} 1307/* 1308 * helper for accessing a memcg's index. It will be used as an index in the 1309 * child cache array in kmem_cache, and also to derive its name. This function 1310 * will return -1 when this is not a kmem-limited memcg. 1311 */ 1312static inline int memcg_cache_id(struct mem_cgroup *memcg) 1313{ 1314 return memcg ? memcg->kmemcg_id : -1; 1315} 1316 1317extern int memcg_expand_shrinker_maps(int new_id); 1318 1319extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg, 1320 int nid, int shrinker_id); 1321#else 1322 1323static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) 1324{ 1325 return 0; 1326} 1327 1328static inline void memcg_kmem_uncharge(struct page *page, int order) 1329{ 1330} 1331 1332static inline int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) 1333{ 1334 return 0; 1335} 1336 1337static inline void __memcg_kmem_uncharge(struct page *page, int order) 1338{ 1339} 1340 1341#define for_each_memcg_cache_index(_idx) \ 1342 for (; NULL; ) 1343 1344static inline bool memcg_kmem_enabled(void) 1345{ 1346 return false; 1347} 1348 1349static inline int memcg_cache_id(struct mem_cgroup *memcg) 1350{ 1351 return -1; 1352} 1353 1354static inline void memcg_get_cache_ids(void) 1355{ 1356} 1357 1358static inline void memcg_put_cache_ids(void) 1359{ 1360} 1361 1362static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg, 1363 int nid, int shrinker_id) { } 1364#endif /* CONFIG_MEMCG_KMEM */ 1365 1366#endif /* _LINUX_MEMCONTROL_H */