1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* memcontrol.h - Memory Controller 3 * 4 * Copyright IBM Corporation, 2007 5 * Author Balbir Singh <[email protected]> 6 * 7 * Copyright 2007 OpenVZ SWsoft Inc 8 * Author: Pavel Emelianov <[email protected]> 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 25 struct mem_cgroup; 26 struct page; 27 struct mm_struct; 28 struct kmem_cache; 29 30 /* Cgroup-specific page state, on top of universal node page state */ 31 enum 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 42 enum 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 53 enum mem_cgroup_protection { 54 MEMCG_PROT_NONE, 55 MEMCG_PROT_LOW, 56 MEMCG_PROT_MIN, 57 }; 58 59 struct 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 70 struct 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 */ 81 enum 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 88 struct 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 95 struct mem_cgroup_reclaim_iter { 96 struct mem_cgroup *position; 97 /* scan generation, increased every round-trip */ 98 unsigned int generation; 99 }; 100 101 struct 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 */ 109 struct memcg_shrinker_map { 110 struct rcu_head rcu; 111 unsigned long map[0]; 112 }; 113 114 /* 115 * per-zone information in memory controller. 116 */ 117 struct mem_cgroup_per_node { 118 struct lruvec lruvec; 119 120 /* Legacy local VM stats */ 121 struct lruvec_stat __percpu *lruvec_stat_local; 122 123 /* Subtree VM stats (batched updates) */ 124 struct lruvec_stat __percpu *lruvec_stat_cpu; 125 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS]; 126 127 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS]; 128 129 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1]; 130 131 struct memcg_shrinker_map __rcu *shrinker_map; 132 133 struct rb_node tree_node; /* RB tree node */ 134 unsigned long usage_in_excess;/* Set to the value by which */ 135 /* the soft limit is exceeded*/ 136 bool on_tree; 137 bool congested; /* memcg has many dirty pages */ 138 /* backed by a congested BDI */ 139 140 struct mem_cgroup *memcg; /* Back pointer, we cannot */ 141 /* use container_of */ 142 }; 143 144 struct mem_cgroup_threshold { 145 struct eventfd_ctx *eventfd; 146 unsigned long threshold; 147 }; 148 149 /* For threshold */ 150 struct mem_cgroup_threshold_ary { 151 /* An array index points to threshold just below or equal to usage. */ 152 int current_threshold; 153 /* Size of entries[] */ 154 unsigned int size; 155 /* Array of thresholds */ 156 struct mem_cgroup_threshold entries[0]; 157 }; 158 159 struct mem_cgroup_thresholds { 160 /* Primary thresholds array */ 161 struct mem_cgroup_threshold_ary *primary; 162 /* 163 * Spare threshold array. 164 * This is needed to make mem_cgroup_unregister_event() "never fail". 165 * It must be able to store at least primary->size - 1 entries. 166 */ 167 struct mem_cgroup_threshold_ary *spare; 168 }; 169 170 enum memcg_kmem_state { 171 KMEM_NONE, 172 KMEM_ALLOCATED, 173 KMEM_ONLINE, 174 }; 175 176 #if defined(CONFIG_SMP) 177 struct memcg_padding { 178 char x[0]; 179 } ____cacheline_internodealigned_in_smp; 180 #define MEMCG_PADDING(name) struct memcg_padding name; 181 #else 182 #define MEMCG_PADDING(name) 183 #endif 184 185 /* 186 * Remember four most recent foreign writebacks with dirty pages in this 187 * cgroup. Inode sharing is expected to be uncommon and, even if we miss 188 * one in a given round, we're likely to catch it later if it keeps 189 * foreign-dirtying, so a fairly low count should be enough. 190 * 191 * See mem_cgroup_track_foreign_dirty_slowpath() for details. 192 */ 193 #define MEMCG_CGWB_FRN_CNT 4 194 195 struct memcg_cgwb_frn { 196 u64 bdi_id; /* bdi->id of the foreign inode */ 197 int memcg_id; /* memcg->css.id of foreign inode */ 198 u64 at; /* jiffies_64 at the time of dirtying */ 199 struct wb_completion done; /* tracks in-flight foreign writebacks */ 200 }; 201 202 /* 203 * The memory controller data structure. The memory controller controls both 204 * page cache and RSS per cgroup. We would eventually like to provide 205 * statistics based on the statistics developed by Rik Van Riel for clock-pro, 206 * to help the administrator determine what knobs to tune. 207 */ 208 struct mem_cgroup { 209 struct cgroup_subsys_state css; 210 211 /* Private memcg ID. Used to ID objects that outlive the cgroup */ 212 struct mem_cgroup_id id; 213 214 /* Accounted resources */ 215 struct page_counter memory; 216 struct page_counter swap; 217 218 /* Legacy consumer-oriented counters */ 219 struct page_counter memsw; 220 struct page_counter kmem; 221 struct page_counter tcpmem; 222 223 /* Upper bound of normal memory consumption range */ 224 unsigned long high; 225 226 /* Range enforcement for interrupt charges */ 227 struct work_struct high_work; 228 229 unsigned long soft_limit; 230 231 /* vmpressure notifications */ 232 struct vmpressure vmpressure; 233 234 /* 235 * Should the accounting and control be hierarchical, per subtree? 236 */ 237 bool use_hierarchy; 238 239 /* 240 * Should the OOM killer kill all belonging tasks, had it kill one? 241 */ 242 bool oom_group; 243 244 /* protected by memcg_oom_lock */ 245 bool oom_lock; 246 int under_oom; 247 248 int swappiness; 249 /* OOM-Killer disable */ 250 int oom_kill_disable; 251 252 /* memory.events and memory.events.local */ 253 struct cgroup_file events_file; 254 struct cgroup_file events_local_file; 255 256 /* handle for "memory.swap.events" */ 257 struct cgroup_file swap_events_file; 258 259 /* protect arrays of thresholds */ 260 struct mutex thresholds_lock; 261 262 /* thresholds for memory usage. RCU-protected */ 263 struct mem_cgroup_thresholds thresholds; 264 265 /* thresholds for mem+swap usage. RCU-protected */ 266 struct mem_cgroup_thresholds memsw_thresholds; 267 268 /* For oom notifier event fd */ 269 struct list_head oom_notify; 270 271 /* 272 * Should we move charges of a task when a task is moved into this 273 * mem_cgroup ? And what type of charges should we move ? 274 */ 275 unsigned long move_charge_at_immigrate; 276 /* taken only while moving_account > 0 */ 277 spinlock_t move_lock; 278 unsigned long move_lock_flags; 279 280 MEMCG_PADDING(_pad1_); 281 282 /* 283 * set > 0 if pages under this cgroup are moving to other cgroup. 284 */ 285 atomic_t moving_account; 286 struct task_struct *move_lock_task; 287 288 /* Legacy local VM stats and events */ 289 struct memcg_vmstats_percpu __percpu *vmstats_local; 290 291 /* Subtree VM stats and events (batched updates) */ 292 struct memcg_vmstats_percpu __percpu *vmstats_percpu; 293 294 MEMCG_PADDING(_pad2_); 295 296 atomic_long_t vmstats[MEMCG_NR_STAT]; 297 atomic_long_t vmevents[NR_VM_EVENT_ITEMS]; 298 299 /* memory.events */ 300 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS]; 301 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS]; 302 303 unsigned long socket_pressure; 304 305 /* Legacy tcp memory accounting */ 306 bool tcpmem_active; 307 int tcpmem_pressure; 308 309 #ifdef CONFIG_MEMCG_KMEM 310 /* Index in the kmem_cache->memcg_params.memcg_caches array */ 311 int kmemcg_id; 312 enum memcg_kmem_state kmem_state; 313 struct list_head kmem_caches; 314 #endif 315 316 int last_scanned_node; 317 #if MAX_NUMNODES > 1 318 nodemask_t scan_nodes; 319 atomic_t numainfo_events; 320 atomic_t numainfo_updating; 321 #endif 322 323 #ifdef CONFIG_CGROUP_WRITEBACK 324 struct list_head cgwb_list; 325 struct wb_domain cgwb_domain; 326 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT]; 327 #endif 328 329 /* List of events which userspace want to receive */ 330 struct list_head event_list; 331 spinlock_t event_list_lock; 332 333 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 334 struct deferred_split deferred_split_queue; 335 #endif 336 337 struct mem_cgroup_per_node *nodeinfo[0]; 338 /* WARNING: nodeinfo must be the last member here */ 339 }; 340 341 /* 342 * size of first charge trial. "32" comes from vmscan.c's magic value. 343 * TODO: maybe necessary to use big numbers in big irons. 344 */ 345 #define MEMCG_CHARGE_BATCH 32U 346 347 extern struct mem_cgroup *root_mem_cgroup; 348 349 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 350 { 351 return (memcg == root_mem_cgroup); 352 } 353 354 static inline bool mem_cgroup_disabled(void) 355 { 356 return !cgroup_subsys_enabled(memory_cgrp_subsys); 357 } 358 359 enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, 360 struct mem_cgroup *memcg); 361 362 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, 363 gfp_t gfp_mask, struct mem_cgroup **memcgp, 364 bool compound); 365 int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, 366 gfp_t gfp_mask, struct mem_cgroup **memcgp, 367 bool compound); 368 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, 369 bool lrucare, bool compound); 370 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, 371 bool compound); 372 void mem_cgroup_uncharge(struct page *page); 373 void mem_cgroup_uncharge_list(struct list_head *page_list); 374 375 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage); 376 377 static struct mem_cgroup_per_node * 378 mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid) 379 { 380 return memcg->nodeinfo[nid]; 381 } 382 383 /** 384 * mem_cgroup_lruvec - get the lru list vector for a node or a memcg zone 385 * @node: node of the wanted lruvec 386 * @memcg: memcg of the wanted lruvec 387 * 388 * Returns the lru list vector holding pages for a given @node or a given 389 * @memcg and @zone. This can be the node lruvec, if the memory controller 390 * is disabled. 391 */ 392 static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat, 393 struct mem_cgroup *memcg) 394 { 395 struct mem_cgroup_per_node *mz; 396 struct lruvec *lruvec; 397 398 if (mem_cgroup_disabled()) { 399 lruvec = node_lruvec(pgdat); 400 goto out; 401 } 402 403 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); 404 lruvec = &mz->lruvec; 405 out: 406 /* 407 * Since a node can be onlined after the mem_cgroup was created, 408 * we have to be prepared to initialize lruvec->pgdat here; 409 * and if offlined then reonlined, we need to reinitialize it. 410 */ 411 if (unlikely(lruvec->pgdat != pgdat)) 412 lruvec->pgdat = pgdat; 413 return lruvec; 414 } 415 416 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct pglist_data *); 417 418 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p); 419 420 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm); 421 422 struct mem_cgroup *get_mem_cgroup_from_page(struct page *page); 423 424 static inline 425 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){ 426 return css ? container_of(css, struct mem_cgroup, css) : NULL; 427 } 428 429 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 430 { 431 if (memcg) 432 css_put(&memcg->css); 433 } 434 435 #define mem_cgroup_from_counter(counter, member) \ 436 container_of(counter, struct mem_cgroup, member) 437 438 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *, 439 struct mem_cgroup *, 440 struct mem_cgroup_reclaim_cookie *); 441 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *); 442 int mem_cgroup_scan_tasks(struct mem_cgroup *, 443 int (*)(struct task_struct *, void *), void *); 444 445 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 446 { 447 if (mem_cgroup_disabled()) 448 return 0; 449 450 return memcg->id.id; 451 } 452 struct mem_cgroup *mem_cgroup_from_id(unsigned short id); 453 454 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 455 { 456 return mem_cgroup_from_css(seq_css(m)); 457 } 458 459 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 460 { 461 struct mem_cgroup_per_node *mz; 462 463 if (mem_cgroup_disabled()) 464 return NULL; 465 466 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 467 return mz->memcg; 468 } 469 470 /** 471 * parent_mem_cgroup - find the accounting parent of a memcg 472 * @memcg: memcg whose parent to find 473 * 474 * Returns the parent memcg, or NULL if this is the root or the memory 475 * controller is in legacy no-hierarchy mode. 476 */ 477 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 478 { 479 if (!memcg->memory.parent) 480 return NULL; 481 return mem_cgroup_from_counter(memcg->memory.parent, memory); 482 } 483 484 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, 485 struct mem_cgroup *root) 486 { 487 if (root == memcg) 488 return true; 489 if (!root->use_hierarchy) 490 return false; 491 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); 492 } 493 494 static inline bool mm_match_cgroup(struct mm_struct *mm, 495 struct mem_cgroup *memcg) 496 { 497 struct mem_cgroup *task_memcg; 498 bool match = false; 499 500 rcu_read_lock(); 501 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 502 if (task_memcg) 503 match = mem_cgroup_is_descendant(task_memcg, memcg); 504 rcu_read_unlock(); 505 return match; 506 } 507 508 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page); 509 ino_t page_cgroup_ino(struct page *page); 510 511 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 512 { 513 if (mem_cgroup_disabled()) 514 return true; 515 return !!(memcg->css.flags & CSS_ONLINE); 516 } 517 518 /* 519 * For memory reclaim. 520 */ 521 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg); 522 523 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, 524 int zid, int nr_pages); 525 526 static inline 527 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 528 enum lru_list lru, int zone_idx) 529 { 530 struct mem_cgroup_per_node *mz; 531 532 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 533 return mz->lru_zone_size[zone_idx][lru]; 534 } 535 536 void mem_cgroup_handle_over_high(void); 537 538 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg); 539 540 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, 541 struct task_struct *p); 542 543 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg); 544 545 static inline void mem_cgroup_enter_user_fault(void) 546 { 547 WARN_ON(current->in_user_fault); 548 current->in_user_fault = 1; 549 } 550 551 static inline void mem_cgroup_exit_user_fault(void) 552 { 553 WARN_ON(!current->in_user_fault); 554 current->in_user_fault = 0; 555 } 556 557 static inline bool task_in_memcg_oom(struct task_struct *p) 558 { 559 return p->memcg_in_oom; 560 } 561 562 bool mem_cgroup_oom_synchronize(bool wait); 563 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, 564 struct mem_cgroup *oom_domain); 565 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg); 566 567 #ifdef CONFIG_MEMCG_SWAP 568 extern int do_swap_account; 569 #endif 570 571 struct mem_cgroup *lock_page_memcg(struct page *page); 572 void __unlock_page_memcg(struct mem_cgroup *memcg); 573 void unlock_page_memcg(struct page *page); 574 575 /* 576 * idx can be of type enum memcg_stat_item or node_stat_item. 577 * Keep in sync with memcg_exact_page_state(). 578 */ 579 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 580 { 581 long x = atomic_long_read(&memcg->vmstats[idx]); 582 #ifdef CONFIG_SMP 583 if (x < 0) 584 x = 0; 585 #endif 586 return x; 587 } 588 589 /* 590 * idx can be of type enum memcg_stat_item or node_stat_item. 591 * Keep in sync with memcg_exact_page_state(). 592 */ 593 static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg, 594 int idx) 595 { 596 long x = 0; 597 int cpu; 598 599 for_each_possible_cpu(cpu) 600 x += per_cpu(memcg->vmstats_local->stat[idx], cpu); 601 #ifdef CONFIG_SMP 602 if (x < 0) 603 x = 0; 604 #endif 605 return x; 606 } 607 608 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val); 609 610 /* idx can be of type enum memcg_stat_item or node_stat_item */ 611 static inline void mod_memcg_state(struct mem_cgroup *memcg, 612 int idx, int val) 613 { 614 unsigned long flags; 615 616 local_irq_save(flags); 617 __mod_memcg_state(memcg, idx, val); 618 local_irq_restore(flags); 619 } 620 621 /** 622 * mod_memcg_page_state - update page state statistics 623 * @page: the page 624 * @idx: page state item to account 625 * @val: number of pages (positive or negative) 626 * 627 * The @page must be locked or the caller must use lock_page_memcg() 628 * to prevent double accounting when the page is concurrently being 629 * moved to another memcg: 630 * 631 * lock_page(page) or lock_page_memcg(page) 632 * if (TestClearPageState(page)) 633 * mod_memcg_page_state(page, state, -1); 634 * unlock_page(page) or unlock_page_memcg(page) 635 * 636 * Kernel pages are an exception to this, since they'll never move. 637 */ 638 static inline void __mod_memcg_page_state(struct page *page, 639 int idx, int val) 640 { 641 if (page->mem_cgroup) 642 __mod_memcg_state(page->mem_cgroup, idx, val); 643 } 644 645 static inline void mod_memcg_page_state(struct page *page, 646 int idx, int val) 647 { 648 if (page->mem_cgroup) 649 mod_memcg_state(page->mem_cgroup, idx, val); 650 } 651 652 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 653 enum node_stat_item idx) 654 { 655 struct mem_cgroup_per_node *pn; 656 long x; 657 658 if (mem_cgroup_disabled()) 659 return node_page_state(lruvec_pgdat(lruvec), idx); 660 661 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 662 x = atomic_long_read(&pn->lruvec_stat[idx]); 663 #ifdef CONFIG_SMP 664 if (x < 0) 665 x = 0; 666 #endif 667 return x; 668 } 669 670 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 671 enum node_stat_item idx) 672 { 673 struct mem_cgroup_per_node *pn; 674 long x = 0; 675 int cpu; 676 677 if (mem_cgroup_disabled()) 678 return node_page_state(lruvec_pgdat(lruvec), idx); 679 680 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 681 for_each_possible_cpu(cpu) 682 x += per_cpu(pn->lruvec_stat_local->count[idx], cpu); 683 #ifdef CONFIG_SMP 684 if (x < 0) 685 x = 0; 686 #endif 687 return x; 688 } 689 690 void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, 691 int val); 692 void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val); 693 694 static inline void mod_lruvec_state(struct lruvec *lruvec, 695 enum node_stat_item idx, int val) 696 { 697 unsigned long flags; 698 699 local_irq_save(flags); 700 __mod_lruvec_state(lruvec, idx, val); 701 local_irq_restore(flags); 702 } 703 704 static inline void __mod_lruvec_page_state(struct page *page, 705 enum node_stat_item idx, int val) 706 { 707 pg_data_t *pgdat = page_pgdat(page); 708 struct lruvec *lruvec; 709 710 /* Untracked pages have no memcg, no lruvec. Update only the node */ 711 if (!page->mem_cgroup) { 712 __mod_node_page_state(pgdat, idx, val); 713 return; 714 } 715 716 lruvec = mem_cgroup_lruvec(pgdat, page->mem_cgroup); 717 __mod_lruvec_state(lruvec, idx, val); 718 } 719 720 static inline void mod_lruvec_page_state(struct page *page, 721 enum node_stat_item idx, int val) 722 { 723 unsigned long flags; 724 725 local_irq_save(flags); 726 __mod_lruvec_page_state(page, idx, val); 727 local_irq_restore(flags); 728 } 729 730 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 731 gfp_t gfp_mask, 732 unsigned long *total_scanned); 733 734 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, 735 unsigned long count); 736 737 static inline void count_memcg_events(struct mem_cgroup *memcg, 738 enum vm_event_item idx, 739 unsigned long count) 740 { 741 unsigned long flags; 742 743 local_irq_save(flags); 744 __count_memcg_events(memcg, idx, count); 745 local_irq_restore(flags); 746 } 747 748 static inline void count_memcg_page_event(struct page *page, 749 enum vm_event_item idx) 750 { 751 if (page->mem_cgroup) 752 count_memcg_events(page->mem_cgroup, idx, 1); 753 } 754 755 static inline void count_memcg_event_mm(struct mm_struct *mm, 756 enum vm_event_item idx) 757 { 758 struct mem_cgroup *memcg; 759 760 if (mem_cgroup_disabled()) 761 return; 762 763 rcu_read_lock(); 764 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 765 if (likely(memcg)) 766 count_memcg_events(memcg, idx, 1); 767 rcu_read_unlock(); 768 } 769 770 static inline void memcg_memory_event(struct mem_cgroup *memcg, 771 enum memcg_memory_event event) 772 { 773 atomic_long_inc(&memcg->memory_events_local[event]); 774 cgroup_file_notify(&memcg->events_local_file); 775 776 do { 777 atomic_long_inc(&memcg->memory_events[event]); 778 cgroup_file_notify(&memcg->events_file); 779 780 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) 781 break; 782 } while ((memcg = parent_mem_cgroup(memcg)) && 783 !mem_cgroup_is_root(memcg)); 784 } 785 786 static inline void memcg_memory_event_mm(struct mm_struct *mm, 787 enum memcg_memory_event event) 788 { 789 struct mem_cgroup *memcg; 790 791 if (mem_cgroup_disabled()) 792 return; 793 794 rcu_read_lock(); 795 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 796 if (likely(memcg)) 797 memcg_memory_event(memcg, event); 798 rcu_read_unlock(); 799 } 800 801 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 802 void mem_cgroup_split_huge_fixup(struct page *head); 803 #endif 804 805 #else /* CONFIG_MEMCG */ 806 807 #define MEM_CGROUP_ID_SHIFT 0 808 #define MEM_CGROUP_ID_MAX 0 809 810 struct mem_cgroup; 811 812 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 813 { 814 return true; 815 } 816 817 static inline bool mem_cgroup_disabled(void) 818 { 819 return true; 820 } 821 822 static inline void memcg_memory_event(struct mem_cgroup *memcg, 823 enum memcg_memory_event event) 824 { 825 } 826 827 static inline void memcg_memory_event_mm(struct mm_struct *mm, 828 enum memcg_memory_event event) 829 { 830 } 831 832 static inline enum mem_cgroup_protection mem_cgroup_protected( 833 struct mem_cgroup *root, struct mem_cgroup *memcg) 834 { 835 return MEMCG_PROT_NONE; 836 } 837 838 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, 839 gfp_t gfp_mask, 840 struct mem_cgroup **memcgp, 841 bool compound) 842 { 843 *memcgp = NULL; 844 return 0; 845 } 846 847 static inline int mem_cgroup_try_charge_delay(struct page *page, 848 struct mm_struct *mm, 849 gfp_t gfp_mask, 850 struct mem_cgroup **memcgp, 851 bool compound) 852 { 853 *memcgp = NULL; 854 return 0; 855 } 856 857 static inline void mem_cgroup_commit_charge(struct page *page, 858 struct mem_cgroup *memcg, 859 bool lrucare, bool compound) 860 { 861 } 862 863 static inline void mem_cgroup_cancel_charge(struct page *page, 864 struct mem_cgroup *memcg, 865 bool compound) 866 { 867 } 868 869 static inline void mem_cgroup_uncharge(struct page *page) 870 { 871 } 872 873 static inline void mem_cgroup_uncharge_list(struct list_head *page_list) 874 { 875 } 876 877 static inline void mem_cgroup_migrate(struct page *old, struct page *new) 878 { 879 } 880 881 static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat, 882 struct mem_cgroup *memcg) 883 { 884 return node_lruvec(pgdat); 885 } 886 887 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page, 888 struct pglist_data *pgdat) 889 { 890 return &pgdat->lruvec; 891 } 892 893 static inline bool mm_match_cgroup(struct mm_struct *mm, 894 struct mem_cgroup *memcg) 895 { 896 return true; 897 } 898 899 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) 900 { 901 return NULL; 902 } 903 904 static inline struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) 905 { 906 return NULL; 907 } 908 909 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 910 { 911 } 912 913 static inline struct mem_cgroup * 914 mem_cgroup_iter(struct mem_cgroup *root, 915 struct mem_cgroup *prev, 916 struct mem_cgroup_reclaim_cookie *reclaim) 917 { 918 return NULL; 919 } 920 921 static inline void mem_cgroup_iter_break(struct mem_cgroup *root, 922 struct mem_cgroup *prev) 923 { 924 } 925 926 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, 927 int (*fn)(struct task_struct *, void *), void *arg) 928 { 929 return 0; 930 } 931 932 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 933 { 934 return 0; 935 } 936 937 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) 938 { 939 WARN_ON_ONCE(id); 940 /* XXX: This should always return root_mem_cgroup */ 941 return NULL; 942 } 943 944 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 945 { 946 return NULL; 947 } 948 949 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 950 { 951 return NULL; 952 } 953 954 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 955 { 956 return true; 957 } 958 959 static inline 960 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 961 enum lru_list lru, int zone_idx) 962 { 963 return 0; 964 } 965 966 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) 967 { 968 return 0; 969 } 970 971 static inline void 972 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) 973 { 974 } 975 976 static inline void 977 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) 978 { 979 } 980 981 static inline struct mem_cgroup *lock_page_memcg(struct page *page) 982 { 983 return NULL; 984 } 985 986 static inline void __unlock_page_memcg(struct mem_cgroup *memcg) 987 { 988 } 989 990 static inline void unlock_page_memcg(struct page *page) 991 { 992 } 993 994 static inline void mem_cgroup_handle_over_high(void) 995 { 996 } 997 998 static inline void mem_cgroup_enter_user_fault(void) 999 { 1000 } 1001 1002 static inline void mem_cgroup_exit_user_fault(void) 1003 { 1004 } 1005 1006 static inline bool task_in_memcg_oom(struct task_struct *p) 1007 { 1008 return false; 1009 } 1010 1011 static inline bool mem_cgroup_oom_synchronize(bool wait) 1012 { 1013 return false; 1014 } 1015 1016 static inline struct mem_cgroup *mem_cgroup_get_oom_group( 1017 struct task_struct *victim, struct mem_cgroup *oom_domain) 1018 { 1019 return NULL; 1020 } 1021 1022 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) 1023 { 1024 } 1025 1026 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 1027 { 1028 return 0; 1029 } 1030 1031 static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg, 1032 int idx) 1033 { 1034 return 0; 1035 } 1036 1037 static inline void __mod_memcg_state(struct mem_cgroup *memcg, 1038 int idx, 1039 int nr) 1040 { 1041 } 1042 1043 static inline void mod_memcg_state(struct mem_cgroup *memcg, 1044 int idx, 1045 int nr) 1046 { 1047 } 1048 1049 static inline void __mod_memcg_page_state(struct page *page, 1050 int idx, 1051 int nr) 1052 { 1053 } 1054 1055 static inline void mod_memcg_page_state(struct page *page, 1056 int idx, 1057 int nr) 1058 { 1059 } 1060 1061 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 1062 enum node_stat_item idx) 1063 { 1064 return node_page_state(lruvec_pgdat(lruvec), idx); 1065 } 1066 1067 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 1068 enum node_stat_item idx) 1069 { 1070 return node_page_state(lruvec_pgdat(lruvec), idx); 1071 } 1072 1073 static inline void __mod_lruvec_state(struct lruvec *lruvec, 1074 enum node_stat_item idx, int val) 1075 { 1076 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); 1077 } 1078 1079 static inline void mod_lruvec_state(struct lruvec *lruvec, 1080 enum node_stat_item idx, int val) 1081 { 1082 mod_node_page_state(lruvec_pgdat(lruvec), idx, val); 1083 } 1084 1085 static inline void __mod_lruvec_page_state(struct page *page, 1086 enum node_stat_item idx, int val) 1087 { 1088 __mod_node_page_state(page_pgdat(page), idx, val); 1089 } 1090 1091 static inline void mod_lruvec_page_state(struct page *page, 1092 enum node_stat_item idx, int val) 1093 { 1094 mod_node_page_state(page_pgdat(page), idx, val); 1095 } 1096 1097 static inline void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, 1098 int val) 1099 { 1100 struct page *page = virt_to_head_page(p); 1101 1102 __mod_node_page_state(page_pgdat(page), idx, val); 1103 } 1104 1105 static inline 1106 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 1107 gfp_t gfp_mask, 1108 unsigned long *total_scanned) 1109 { 1110 return 0; 1111 } 1112 1113 static inline void mem_cgroup_split_huge_fixup(struct page *head) 1114 { 1115 } 1116 1117 static inline void count_memcg_events(struct mem_cgroup *memcg, 1118 enum vm_event_item idx, 1119 unsigned long count) 1120 { 1121 } 1122 1123 static inline void __count_memcg_events(struct mem_cgroup *memcg, 1124 enum vm_event_item idx, 1125 unsigned long count) 1126 { 1127 } 1128 1129 static inline void count_memcg_page_event(struct page *page, 1130 int idx) 1131 { 1132 } 1133 1134 static inline 1135 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx) 1136 { 1137 } 1138 #endif /* CONFIG_MEMCG */ 1139 1140 /* idx can be of type enum memcg_stat_item or node_stat_item */ 1141 static inline void __inc_memcg_state(struct mem_cgroup *memcg, 1142 int idx) 1143 { 1144 __mod_memcg_state(memcg, idx, 1); 1145 } 1146 1147 /* idx can be of type enum memcg_stat_item or node_stat_item */ 1148 static inline void __dec_memcg_state(struct mem_cgroup *memcg, 1149 int idx) 1150 { 1151 __mod_memcg_state(memcg, idx, -1); 1152 } 1153 1154 /* idx can be of type enum memcg_stat_item or node_stat_item */ 1155 static inline void __inc_memcg_page_state(struct page *page, 1156 int idx) 1157 { 1158 __mod_memcg_page_state(page, idx, 1); 1159 } 1160 1161 /* idx can be of type enum memcg_stat_item or node_stat_item */ 1162 static inline void __dec_memcg_page_state(struct page *page, 1163 int idx) 1164 { 1165 __mod_memcg_page_state(page, idx, -1); 1166 } 1167 1168 static inline void __inc_lruvec_state(struct lruvec *lruvec, 1169 enum node_stat_item idx) 1170 { 1171 __mod_lruvec_state(lruvec, idx, 1); 1172 } 1173 1174 static inline void __dec_lruvec_state(struct lruvec *lruvec, 1175 enum node_stat_item idx) 1176 { 1177 __mod_lruvec_state(lruvec, idx, -1); 1178 } 1179 1180 static inline void __inc_lruvec_page_state(struct page *page, 1181 enum node_stat_item idx) 1182 { 1183 __mod_lruvec_page_state(page, idx, 1); 1184 } 1185 1186 static inline void __dec_lruvec_page_state(struct page *page, 1187 enum node_stat_item idx) 1188 { 1189 __mod_lruvec_page_state(page, idx, -1); 1190 } 1191 1192 static inline void __inc_lruvec_slab_state(void *p, enum node_stat_item idx) 1193 { 1194 __mod_lruvec_slab_state(p, idx, 1); 1195 } 1196 1197 static inline void __dec_lruvec_slab_state(void *p, enum node_stat_item idx) 1198 { 1199 __mod_lruvec_slab_state(p, idx, -1); 1200 } 1201 1202 /* idx can be of type enum memcg_stat_item or node_stat_item */ 1203 static inline void inc_memcg_state(struct mem_cgroup *memcg, 1204 int idx) 1205 { 1206 mod_memcg_state(memcg, idx, 1); 1207 } 1208 1209 /* idx can be of type enum memcg_stat_item or node_stat_item */ 1210 static inline void dec_memcg_state(struct mem_cgroup *memcg, 1211 int idx) 1212 { 1213 mod_memcg_state(memcg, idx, -1); 1214 } 1215 1216 /* idx can be of type enum memcg_stat_item or node_stat_item */ 1217 static inline void inc_memcg_page_state(struct page *page, 1218 int idx) 1219 { 1220 mod_memcg_page_state(page, idx, 1); 1221 } 1222 1223 /* idx can be of type enum memcg_stat_item or node_stat_item */ 1224 static inline void dec_memcg_page_state(struct page *page, 1225 int idx) 1226 { 1227 mod_memcg_page_state(page, idx, -1); 1228 } 1229 1230 static inline void inc_lruvec_state(struct lruvec *lruvec, 1231 enum node_stat_item idx) 1232 { 1233 mod_lruvec_state(lruvec, idx, 1); 1234 } 1235 1236 static inline void dec_lruvec_state(struct lruvec *lruvec, 1237 enum node_stat_item idx) 1238 { 1239 mod_lruvec_state(lruvec, idx, -1); 1240 } 1241 1242 static inline void inc_lruvec_page_state(struct page *page, 1243 enum node_stat_item idx) 1244 { 1245 mod_lruvec_page_state(page, idx, 1); 1246 } 1247 1248 static inline void dec_lruvec_page_state(struct page *page, 1249 enum node_stat_item idx) 1250 { 1251 mod_lruvec_page_state(page, idx, -1); 1252 } 1253 1254 #ifdef CONFIG_CGROUP_WRITEBACK 1255 1256 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb); 1257 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, 1258 unsigned long *pheadroom, unsigned long *pdirty, 1259 unsigned long *pwriteback); 1260 1261 void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, 1262 struct bdi_writeback *wb); 1263 1264 static inline void mem_cgroup_track_foreign_dirty(struct page *page, 1265 struct bdi_writeback *wb) 1266 { 1267 if (unlikely(&page->mem_cgroup->css != wb->memcg_css)) 1268 mem_cgroup_track_foreign_dirty_slowpath(page, wb); 1269 } 1270 1271 void mem_cgroup_flush_foreign(struct bdi_writeback *wb); 1272 1273 #else /* CONFIG_CGROUP_WRITEBACK */ 1274 1275 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) 1276 { 1277 return NULL; 1278 } 1279 1280 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb, 1281 unsigned long *pfilepages, 1282 unsigned long *pheadroom, 1283 unsigned long *pdirty, 1284 unsigned long *pwriteback) 1285 { 1286 } 1287 1288 static inline void mem_cgroup_track_foreign_dirty(struct page *page, 1289 struct bdi_writeback *wb) 1290 { 1291 } 1292 1293 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb) 1294 { 1295 } 1296 1297 #endif /* CONFIG_CGROUP_WRITEBACK */ 1298 1299 struct sock; 1300 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1301 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1302 #ifdef CONFIG_MEMCG 1303 extern struct static_key_false memcg_sockets_enabled_key; 1304 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key) 1305 void mem_cgroup_sk_alloc(struct sock *sk); 1306 void mem_cgroup_sk_free(struct sock *sk); 1307 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1308 { 1309 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure) 1310 return true; 1311 do { 1312 if (time_before(jiffies, memcg->socket_pressure)) 1313 return true; 1314 } while ((memcg = parent_mem_cgroup(memcg))); 1315 return false; 1316 } 1317 1318 extern int memcg_expand_shrinker_maps(int new_id); 1319 1320 extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg, 1321 int nid, int shrinker_id); 1322 #else 1323 #define mem_cgroup_sockets_enabled 0 1324 static inline void mem_cgroup_sk_alloc(struct sock *sk) { }; 1325 static inline void mem_cgroup_sk_free(struct sock *sk) { }; 1326 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1327 { 1328 return false; 1329 } 1330 1331 static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg, 1332 int nid, int shrinker_id) 1333 { 1334 } 1335 #endif 1336 1337 struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep); 1338 void memcg_kmem_put_cache(struct kmem_cache *cachep); 1339 1340 #ifdef CONFIG_MEMCG_KMEM 1341 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order); 1342 void __memcg_kmem_uncharge(struct page *page, int order); 1343 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, 1344 struct mem_cgroup *memcg); 1345 void __memcg_kmem_uncharge_memcg(struct mem_cgroup *memcg, 1346 unsigned int nr_pages); 1347 1348 extern struct static_key_false memcg_kmem_enabled_key; 1349 extern struct workqueue_struct *memcg_kmem_cache_wq; 1350 1351 extern int memcg_nr_cache_ids; 1352 void memcg_get_cache_ids(void); 1353 void memcg_put_cache_ids(void); 1354 1355 /* 1356 * Helper macro to loop through all memcg-specific caches. Callers must still 1357 * check if the cache is valid (it is either valid or NULL). 1358 * the slab_mutex must be held when looping through those caches 1359 */ 1360 #define for_each_memcg_cache_index(_idx) \ 1361 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++) 1362 1363 static inline bool memcg_kmem_enabled(void) 1364 { 1365 return static_branch_unlikely(&memcg_kmem_enabled_key); 1366 } 1367 1368 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) 1369 { 1370 if (memcg_kmem_enabled()) 1371 return __memcg_kmem_charge(page, gfp, order); 1372 return 0; 1373 } 1374 1375 static inline void memcg_kmem_uncharge(struct page *page, int order) 1376 { 1377 if (memcg_kmem_enabled()) 1378 __memcg_kmem_uncharge(page, order); 1379 } 1380 1381 static inline int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, 1382 int order, struct mem_cgroup *memcg) 1383 { 1384 if (memcg_kmem_enabled()) 1385 return __memcg_kmem_charge_memcg(page, gfp, order, memcg); 1386 return 0; 1387 } 1388 1389 static inline void memcg_kmem_uncharge_memcg(struct page *page, int order, 1390 struct mem_cgroup *memcg) 1391 { 1392 if (memcg_kmem_enabled()) 1393 __memcg_kmem_uncharge_memcg(memcg, 1 << order); 1394 } 1395 1396 /* 1397 * helper for accessing a memcg's index. It will be used as an index in the 1398 * child cache array in kmem_cache, and also to derive its name. This function 1399 * will return -1 when this is not a kmem-limited memcg. 1400 */ 1401 static inline int memcg_cache_id(struct mem_cgroup *memcg) 1402 { 1403 return memcg ? memcg->kmemcg_id : -1; 1404 } 1405 1406 #else 1407 1408 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) 1409 { 1410 return 0; 1411 } 1412 1413 static inline void memcg_kmem_uncharge(struct page *page, int order) 1414 { 1415 } 1416 1417 static inline int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) 1418 { 1419 return 0; 1420 } 1421 1422 static inline void __memcg_kmem_uncharge(struct page *page, int order) 1423 { 1424 } 1425 1426 #define for_each_memcg_cache_index(_idx) \ 1427 for (; NULL; ) 1428 1429 static inline bool memcg_kmem_enabled(void) 1430 { 1431 return false; 1432 } 1433 1434 static inline int memcg_cache_id(struct mem_cgroup *memcg) 1435 { 1436 return -1; 1437 } 1438 1439 static inline void memcg_get_cache_ids(void) 1440 { 1441 } 1442 1443 static inline void memcg_put_cache_ids(void) 1444 { 1445 } 1446 1447 #endif /* CONFIG_MEMCG_KMEM */ 1448 1449 #endif /* _LINUX_MEMCONTROL_H */ 1450