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 obj_cgroup; 27 struct page; 28 struct mm_struct; 29 struct kmem_cache; 30 31 /* Cgroup-specific page state, on top of universal node page state */ 32 enum memcg_stat_item { 33 MEMCG_SWAP = NR_VM_NODE_STAT_ITEMS, 34 MEMCG_SOCK, 35 MEMCG_PERCPU_B, 36 MEMCG_NR_STAT, 37 }; 38 39 enum memcg_memory_event { 40 MEMCG_LOW, 41 MEMCG_HIGH, 42 MEMCG_MAX, 43 MEMCG_OOM, 44 MEMCG_OOM_KILL, 45 MEMCG_SWAP_HIGH, 46 MEMCG_SWAP_MAX, 47 MEMCG_SWAP_FAIL, 48 MEMCG_NR_MEMORY_EVENTS, 49 }; 50 51 struct mem_cgroup_reclaim_cookie { 52 pg_data_t *pgdat; 53 unsigned int generation; 54 }; 55 56 #ifdef CONFIG_MEMCG 57 58 #define MEM_CGROUP_ID_SHIFT 16 59 #define MEM_CGROUP_ID_MAX USHRT_MAX 60 61 struct mem_cgroup_id { 62 int id; 63 refcount_t ref; 64 }; 65 66 /* 67 * Per memcg event counter is incremented at every pagein/pageout. With THP, 68 * it will be incremented by the number of pages. This counter is used 69 * to trigger some periodic events. This is straightforward and better 70 * than using jiffies etc. to handle periodic memcg event. 71 */ 72 enum mem_cgroup_events_target { 73 MEM_CGROUP_TARGET_THRESH, 74 MEM_CGROUP_TARGET_SOFTLIMIT, 75 MEM_CGROUP_NTARGETS, 76 }; 77 78 struct memcg_vmstats_percpu { 79 long stat[MEMCG_NR_STAT]; 80 unsigned long events[NR_VM_EVENT_ITEMS]; 81 unsigned long nr_page_events; 82 unsigned long targets[MEM_CGROUP_NTARGETS]; 83 }; 84 85 struct mem_cgroup_reclaim_iter { 86 struct mem_cgroup *position; 87 /* scan generation, increased every round-trip */ 88 unsigned int generation; 89 }; 90 91 struct lruvec_stat { 92 long count[NR_VM_NODE_STAT_ITEMS]; 93 }; 94 95 /* 96 * Bitmap of shrinker::id corresponding to memcg-aware shrinkers, 97 * which have elements charged to this memcg. 98 */ 99 struct memcg_shrinker_map { 100 struct rcu_head rcu; 101 unsigned long map[]; 102 }; 103 104 /* 105 * per-node information in memory controller. 106 */ 107 struct mem_cgroup_per_node { 108 struct lruvec lruvec; 109 110 /* Legacy local VM stats */ 111 struct lruvec_stat __percpu *lruvec_stat_local; 112 113 /* Subtree VM stats (batched updates) */ 114 struct lruvec_stat __percpu *lruvec_stat_cpu; 115 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS]; 116 117 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS]; 118 119 struct mem_cgroup_reclaim_iter iter; 120 121 struct memcg_shrinker_map __rcu *shrinker_map; 122 123 struct rb_node tree_node; /* RB tree node */ 124 unsigned long usage_in_excess;/* Set to the value by which */ 125 /* the soft limit is exceeded*/ 126 bool on_tree; 127 struct mem_cgroup *memcg; /* Back pointer, we cannot */ 128 /* use container_of */ 129 }; 130 131 struct mem_cgroup_threshold { 132 struct eventfd_ctx *eventfd; 133 unsigned long threshold; 134 }; 135 136 /* For threshold */ 137 struct mem_cgroup_threshold_ary { 138 /* An array index points to threshold just below or equal to usage. */ 139 int current_threshold; 140 /* Size of entries[] */ 141 unsigned int size; 142 /* Array of thresholds */ 143 struct mem_cgroup_threshold entries[]; 144 }; 145 146 struct mem_cgroup_thresholds { 147 /* Primary thresholds array */ 148 struct mem_cgroup_threshold_ary *primary; 149 /* 150 * Spare threshold array. 151 * This is needed to make mem_cgroup_unregister_event() "never fail". 152 * It must be able to store at least primary->size - 1 entries. 153 */ 154 struct mem_cgroup_threshold_ary *spare; 155 }; 156 157 enum memcg_kmem_state { 158 KMEM_NONE, 159 KMEM_ALLOCATED, 160 KMEM_ONLINE, 161 }; 162 163 #if defined(CONFIG_SMP) 164 struct memcg_padding { 165 char x[0]; 166 } ____cacheline_internodealigned_in_smp; 167 #define MEMCG_PADDING(name) struct memcg_padding name; 168 #else 169 #define MEMCG_PADDING(name) 170 #endif 171 172 /* 173 * Remember four most recent foreign writebacks with dirty pages in this 174 * cgroup. Inode sharing is expected to be uncommon and, even if we miss 175 * one in a given round, we're likely to catch it later if it keeps 176 * foreign-dirtying, so a fairly low count should be enough. 177 * 178 * See mem_cgroup_track_foreign_dirty_slowpath() for details. 179 */ 180 #define MEMCG_CGWB_FRN_CNT 4 181 182 struct memcg_cgwb_frn { 183 u64 bdi_id; /* bdi->id of the foreign inode */ 184 int memcg_id; /* memcg->css.id of foreign inode */ 185 u64 at; /* jiffies_64 at the time of dirtying */ 186 struct wb_completion done; /* tracks in-flight foreign writebacks */ 187 }; 188 189 /* 190 * Bucket for arbitrarily byte-sized objects charged to a memory 191 * cgroup. The bucket can be reparented in one piece when the cgroup 192 * is destroyed, without having to round up the individual references 193 * of all live memory objects in the wild. 194 */ 195 struct obj_cgroup { 196 struct percpu_ref refcnt; 197 struct mem_cgroup *memcg; 198 atomic_t nr_charged_bytes; 199 union { 200 struct list_head list; 201 struct rcu_head rcu; 202 }; 203 }; 204 205 /* 206 * The memory controller data structure. The memory controller controls both 207 * page cache and RSS per cgroup. We would eventually like to provide 208 * statistics based on the statistics developed by Rik Van Riel for clock-pro, 209 * to help the administrator determine what knobs to tune. 210 */ 211 struct mem_cgroup { 212 struct cgroup_subsys_state css; 213 214 /* Private memcg ID. Used to ID objects that outlive the cgroup */ 215 struct mem_cgroup_id id; 216 217 /* Accounted resources */ 218 struct page_counter memory; /* Both v1 & v2 */ 219 220 union { 221 struct page_counter swap; /* v2 only */ 222 struct page_counter memsw; /* v1 only */ 223 }; 224 225 /* Legacy consumer-oriented counters */ 226 struct page_counter kmem; /* v1 only */ 227 struct page_counter tcpmem; /* v1 only */ 228 229 /* Range enforcement for interrupt charges */ 230 struct work_struct high_work; 231 232 unsigned long soft_limit; 233 234 /* vmpressure notifications */ 235 struct vmpressure vmpressure; 236 237 /* 238 * Should the OOM killer kill all belonging tasks, had it kill one? 239 */ 240 bool oom_group; 241 242 /* protected by memcg_oom_lock */ 243 bool oom_lock; 244 int under_oom; 245 246 int swappiness; 247 /* OOM-Killer disable */ 248 int oom_kill_disable; 249 250 /* memory.events and memory.events.local */ 251 struct cgroup_file events_file; 252 struct cgroup_file events_local_file; 253 254 /* handle for "memory.swap.events" */ 255 struct cgroup_file swap_events_file; 256 257 /* protect arrays of thresholds */ 258 struct mutex thresholds_lock; 259 260 /* thresholds for memory usage. RCU-protected */ 261 struct mem_cgroup_thresholds thresholds; 262 263 /* thresholds for mem+swap usage. RCU-protected */ 264 struct mem_cgroup_thresholds memsw_thresholds; 265 266 /* For oom notifier event fd */ 267 struct list_head oom_notify; 268 269 /* 270 * Should we move charges of a task when a task is moved into this 271 * mem_cgroup ? And what type of charges should we move ? 272 */ 273 unsigned long move_charge_at_immigrate; 274 /* taken only while moving_account > 0 */ 275 spinlock_t move_lock; 276 unsigned long move_lock_flags; 277 278 MEMCG_PADDING(_pad1_); 279 280 atomic_long_t vmstats[MEMCG_NR_STAT]; 281 atomic_long_t vmevents[NR_VM_EVENT_ITEMS]; 282 283 /* memory.events */ 284 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS]; 285 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS]; 286 287 unsigned long socket_pressure; 288 289 /* Legacy tcp memory accounting */ 290 bool tcpmem_active; 291 int tcpmem_pressure; 292 293 #ifdef CONFIG_MEMCG_KMEM 294 int kmemcg_id; 295 enum memcg_kmem_state kmem_state; 296 struct obj_cgroup __rcu *objcg; 297 struct list_head objcg_list; /* list of inherited objcgs */ 298 #endif 299 300 MEMCG_PADDING(_pad2_); 301 302 /* 303 * set > 0 if pages under this cgroup are moving to other cgroup. 304 */ 305 atomic_t moving_account; 306 struct task_struct *move_lock_task; 307 308 /* Legacy local VM stats and events */ 309 struct memcg_vmstats_percpu __percpu *vmstats_local; 310 311 /* Subtree VM stats and events (batched updates) */ 312 struct memcg_vmstats_percpu __percpu *vmstats_percpu; 313 314 #ifdef CONFIG_CGROUP_WRITEBACK 315 struct list_head cgwb_list; 316 struct wb_domain cgwb_domain; 317 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT]; 318 #endif 319 320 /* List of events which userspace want to receive */ 321 struct list_head event_list; 322 spinlock_t event_list_lock; 323 324 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 325 struct deferred_split deferred_split_queue; 326 #endif 327 328 struct mem_cgroup_per_node *nodeinfo[0]; 329 /* WARNING: nodeinfo must be the last member here */ 330 }; 331 332 /* 333 * size of first charge trial. "32" comes from vmscan.c's magic value. 334 * TODO: maybe necessary to use big numbers in big irons. 335 */ 336 #define MEMCG_CHARGE_BATCH 32U 337 338 extern struct mem_cgroup *root_mem_cgroup; 339 340 enum page_memcg_data_flags { 341 /* page->memcg_data is a pointer to an objcgs vector */ 342 MEMCG_DATA_OBJCGS = (1UL << 0), 343 /* page has been accounted as a non-slab kernel page */ 344 MEMCG_DATA_KMEM = (1UL << 1), 345 /* the next bit after the last actual flag */ 346 __NR_MEMCG_DATA_FLAGS = (1UL << 2), 347 }; 348 349 #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1) 350 351 /* 352 * page_memcg - get the memory cgroup associated with a page 353 * @page: a pointer to the page struct 354 * 355 * Returns a pointer to the memory cgroup associated with the page, 356 * or NULL. This function assumes that the page is known to have a 357 * proper memory cgroup pointer. It's not safe to call this function 358 * against some type of pages, e.g. slab pages or ex-slab pages. 359 * 360 * Any of the following ensures page and memcg binding stability: 361 * - the page lock 362 * - LRU isolation 363 * - lock_page_memcg() 364 * - exclusive reference 365 */ 366 static inline struct mem_cgroup *page_memcg(struct page *page) 367 { 368 unsigned long memcg_data = page->memcg_data; 369 370 VM_BUG_ON_PAGE(PageSlab(page), page); 371 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); 372 373 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 374 } 375 376 /* 377 * page_memcg_rcu - locklessly get the memory cgroup associated with a page 378 * @page: a pointer to the page struct 379 * 380 * Returns a pointer to the memory cgroup associated with the page, 381 * or NULL. This function assumes that the page is known to have a 382 * proper memory cgroup pointer. It's not safe to call this function 383 * against some type of pages, e.g. slab pages or ex-slab pages. 384 */ 385 static inline struct mem_cgroup *page_memcg_rcu(struct page *page) 386 { 387 VM_BUG_ON_PAGE(PageSlab(page), page); 388 WARN_ON_ONCE(!rcu_read_lock_held()); 389 390 return (struct mem_cgroup *)(READ_ONCE(page->memcg_data) & 391 ~MEMCG_DATA_FLAGS_MASK); 392 } 393 394 /* 395 * page_memcg_check - get the memory cgroup associated with a page 396 * @page: a pointer to the page struct 397 * 398 * Returns a pointer to the memory cgroup associated with the page, 399 * or NULL. This function unlike page_memcg() can take any page 400 * as an argument. It has to be used in cases when it's not known if a page 401 * has an associated memory cgroup pointer or an object cgroups vector. 402 * 403 * Any of the following ensures page and memcg binding stability: 404 * - the page lock 405 * - LRU isolation 406 * - lock_page_memcg() 407 * - exclusive reference 408 */ 409 static inline struct mem_cgroup *page_memcg_check(struct page *page) 410 { 411 /* 412 * Because page->memcg_data might be changed asynchronously 413 * for slab pages, READ_ONCE() should be used here. 414 */ 415 unsigned long memcg_data = READ_ONCE(page->memcg_data); 416 417 if (memcg_data & MEMCG_DATA_OBJCGS) 418 return NULL; 419 420 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 421 } 422 423 /* 424 * PageMemcgKmem - check if the page has MemcgKmem flag set 425 * @page: a pointer to the page struct 426 * 427 * Checks if the page has MemcgKmem flag set. The caller must ensure that 428 * the page has an associated memory cgroup. It's not safe to call this function 429 * against some types of pages, e.g. slab pages. 430 */ 431 static inline bool PageMemcgKmem(struct page *page) 432 { 433 VM_BUG_ON_PAGE(page->memcg_data & MEMCG_DATA_OBJCGS, page); 434 return page->memcg_data & MEMCG_DATA_KMEM; 435 } 436 437 #ifdef CONFIG_MEMCG_KMEM 438 /* 439 * page_objcgs - get the object cgroups vector associated with a page 440 * @page: a pointer to the page struct 441 * 442 * Returns a pointer to the object cgroups vector associated with the page, 443 * or NULL. This function assumes that the page is known to have an 444 * associated object cgroups vector. It's not safe to call this function 445 * against pages, which might have an associated memory cgroup: e.g. 446 * kernel stack pages. 447 */ 448 static inline struct obj_cgroup **page_objcgs(struct page *page) 449 { 450 unsigned long memcg_data = READ_ONCE(page->memcg_data); 451 452 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page); 453 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); 454 455 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 456 } 457 458 /* 459 * page_objcgs_check - get the object cgroups vector associated with a page 460 * @page: a pointer to the page struct 461 * 462 * Returns a pointer to the object cgroups vector associated with the page, 463 * or NULL. This function is safe to use if the page can be directly associated 464 * with a memory cgroup. 465 */ 466 static inline struct obj_cgroup **page_objcgs_check(struct page *page) 467 { 468 unsigned long memcg_data = READ_ONCE(page->memcg_data); 469 470 if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS)) 471 return NULL; 472 473 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); 474 475 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 476 } 477 478 /* 479 * set_page_objcgs - associate a page with a object cgroups vector 480 * @page: a pointer to the page struct 481 * @objcgs: a pointer to the object cgroups vector 482 * 483 * Atomically associates a page with a vector of object cgroups. 484 */ 485 static inline bool set_page_objcgs(struct page *page, 486 struct obj_cgroup **objcgs) 487 { 488 return !cmpxchg(&page->memcg_data, 0, (unsigned long)objcgs | 489 MEMCG_DATA_OBJCGS); 490 } 491 #else 492 static inline struct obj_cgroup **page_objcgs(struct page *page) 493 { 494 return NULL; 495 } 496 497 static inline struct obj_cgroup **page_objcgs_check(struct page *page) 498 { 499 return NULL; 500 } 501 502 static inline bool set_page_objcgs(struct page *page, 503 struct obj_cgroup **objcgs) 504 { 505 return true; 506 } 507 #endif 508 509 static __always_inline bool memcg_stat_item_in_bytes(int idx) 510 { 511 if (idx == MEMCG_PERCPU_B) 512 return true; 513 return vmstat_item_in_bytes(idx); 514 } 515 516 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 517 { 518 return (memcg == root_mem_cgroup); 519 } 520 521 static inline bool mem_cgroup_disabled(void) 522 { 523 return !cgroup_subsys_enabled(memory_cgrp_subsys); 524 } 525 526 static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root, 527 struct mem_cgroup *memcg, 528 bool in_low_reclaim) 529 { 530 if (mem_cgroup_disabled()) 531 return 0; 532 533 /* 534 * There is no reclaim protection applied to a targeted reclaim. 535 * We are special casing this specific case here because 536 * mem_cgroup_protected calculation is not robust enough to keep 537 * the protection invariant for calculated effective values for 538 * parallel reclaimers with different reclaim target. This is 539 * especially a problem for tail memcgs (as they have pages on LRU) 540 * which would want to have effective values 0 for targeted reclaim 541 * but a different value for external reclaim. 542 * 543 * Example 544 * Let's have global and A's reclaim in parallel: 545 * | 546 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G) 547 * |\ 548 * | C (low = 1G, usage = 2.5G) 549 * B (low = 1G, usage = 0.5G) 550 * 551 * For the global reclaim 552 * A.elow = A.low 553 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow 554 * C.elow = min(C.usage, C.low) 555 * 556 * With the effective values resetting we have A reclaim 557 * A.elow = 0 558 * B.elow = B.low 559 * C.elow = C.low 560 * 561 * If the global reclaim races with A's reclaim then 562 * B.elow = C.elow = 0 because children_low_usage > A.elow) 563 * is possible and reclaiming B would be violating the protection. 564 * 565 */ 566 if (root == memcg) 567 return 0; 568 569 if (in_low_reclaim) 570 return READ_ONCE(memcg->memory.emin); 571 572 return max(READ_ONCE(memcg->memory.emin), 573 READ_ONCE(memcg->memory.elow)); 574 } 575 576 void mem_cgroup_calculate_protection(struct mem_cgroup *root, 577 struct mem_cgroup *memcg); 578 579 static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg) 580 { 581 /* 582 * The root memcg doesn't account charges, and doesn't support 583 * protection. 584 */ 585 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg); 586 587 } 588 589 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg) 590 { 591 if (!mem_cgroup_supports_protection(memcg)) 592 return false; 593 594 return READ_ONCE(memcg->memory.elow) >= 595 page_counter_read(&memcg->memory); 596 } 597 598 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg) 599 { 600 if (!mem_cgroup_supports_protection(memcg)) 601 return false; 602 603 return READ_ONCE(memcg->memory.emin) >= 604 page_counter_read(&memcg->memory); 605 } 606 607 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask); 608 609 void mem_cgroup_uncharge(struct page *page); 610 void mem_cgroup_uncharge_list(struct list_head *page_list); 611 612 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage); 613 614 static struct mem_cgroup_per_node * 615 mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid) 616 { 617 return memcg->nodeinfo[nid]; 618 } 619 620 /** 621 * mem_cgroup_lruvec - get the lru list vector for a memcg & node 622 * @memcg: memcg of the wanted lruvec 623 * @pgdat: pglist_data 624 * 625 * Returns the lru list vector holding pages for a given @memcg & 626 * @pgdat combination. This can be the node lruvec, if the memory 627 * controller is disabled. 628 */ 629 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg, 630 struct pglist_data *pgdat) 631 { 632 struct mem_cgroup_per_node *mz; 633 struct lruvec *lruvec; 634 635 if (mem_cgroup_disabled()) { 636 lruvec = &pgdat->__lruvec; 637 goto out; 638 } 639 640 if (!memcg) 641 memcg = root_mem_cgroup; 642 643 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); 644 lruvec = &mz->lruvec; 645 out: 646 /* 647 * Since a node can be onlined after the mem_cgroup was created, 648 * we have to be prepared to initialize lruvec->pgdat here; 649 * and if offlined then reonlined, we need to reinitialize it. 650 */ 651 if (unlikely(lruvec->pgdat != pgdat)) 652 lruvec->pgdat = pgdat; 653 return lruvec; 654 } 655 656 /** 657 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page 658 * @page: the page 659 * @pgdat: pgdat of the page 660 * 661 * This function relies on page->mem_cgroup being stable. 662 */ 663 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page, 664 struct pglist_data *pgdat) 665 { 666 struct mem_cgroup *memcg = page_memcg(page); 667 668 VM_WARN_ON_ONCE_PAGE(!memcg, page); 669 return mem_cgroup_lruvec(memcg, pgdat); 670 } 671 672 static inline bool lruvec_holds_page_lru_lock(struct page *page, 673 struct lruvec *lruvec) 674 { 675 pg_data_t *pgdat = page_pgdat(page); 676 const struct mem_cgroup *memcg; 677 struct mem_cgroup_per_node *mz; 678 679 if (mem_cgroup_disabled()) 680 return lruvec == &pgdat->__lruvec; 681 682 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 683 memcg = page_memcg(page) ? : root_mem_cgroup; 684 685 return lruvec->pgdat == pgdat && mz->memcg == memcg; 686 } 687 688 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p); 689 690 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm); 691 692 struct mem_cgroup *get_mem_cgroup_from_page(struct page *page); 693 694 struct lruvec *lock_page_lruvec(struct page *page); 695 struct lruvec *lock_page_lruvec_irq(struct page *page); 696 struct lruvec *lock_page_lruvec_irqsave(struct page *page, 697 unsigned long *flags); 698 699 #ifdef CONFIG_DEBUG_VM 700 void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page); 701 #else 702 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) 703 { 704 } 705 #endif 706 707 static inline 708 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){ 709 return css ? container_of(css, struct mem_cgroup, css) : NULL; 710 } 711 712 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg) 713 { 714 return percpu_ref_tryget(&objcg->refcnt); 715 } 716 717 static inline void obj_cgroup_get(struct obj_cgroup *objcg) 718 { 719 percpu_ref_get(&objcg->refcnt); 720 } 721 722 static inline void obj_cgroup_put(struct obj_cgroup *objcg) 723 { 724 percpu_ref_put(&objcg->refcnt); 725 } 726 727 /* 728 * After the initialization objcg->memcg is always pointing at 729 * a valid memcg, but can be atomically swapped to the parent memcg. 730 * 731 * The caller must ensure that the returned memcg won't be released: 732 * e.g. acquire the rcu_read_lock or css_set_lock. 733 */ 734 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg) 735 { 736 return READ_ONCE(objcg->memcg); 737 } 738 739 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 740 { 741 if (memcg) 742 css_put(&memcg->css); 743 } 744 745 #define mem_cgroup_from_counter(counter, member) \ 746 container_of(counter, struct mem_cgroup, member) 747 748 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *, 749 struct mem_cgroup *, 750 struct mem_cgroup_reclaim_cookie *); 751 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *); 752 int mem_cgroup_scan_tasks(struct mem_cgroup *, 753 int (*)(struct task_struct *, void *), void *); 754 755 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 756 { 757 if (mem_cgroup_disabled()) 758 return 0; 759 760 return memcg->id.id; 761 } 762 struct mem_cgroup *mem_cgroup_from_id(unsigned short id); 763 764 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 765 { 766 return mem_cgroup_from_css(seq_css(m)); 767 } 768 769 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 770 { 771 struct mem_cgroup_per_node *mz; 772 773 if (mem_cgroup_disabled()) 774 return NULL; 775 776 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 777 return mz->memcg; 778 } 779 780 /** 781 * parent_mem_cgroup - find the accounting parent of a memcg 782 * @memcg: memcg whose parent to find 783 * 784 * Returns the parent memcg, or NULL if this is the root or the memory 785 * controller is in legacy no-hierarchy mode. 786 */ 787 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 788 { 789 if (!memcg->memory.parent) 790 return NULL; 791 return mem_cgroup_from_counter(memcg->memory.parent, memory); 792 } 793 794 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, 795 struct mem_cgroup *root) 796 { 797 if (root == memcg) 798 return true; 799 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); 800 } 801 802 static inline bool mm_match_cgroup(struct mm_struct *mm, 803 struct mem_cgroup *memcg) 804 { 805 struct mem_cgroup *task_memcg; 806 bool match = false; 807 808 rcu_read_lock(); 809 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 810 if (task_memcg) 811 match = mem_cgroup_is_descendant(task_memcg, memcg); 812 rcu_read_unlock(); 813 return match; 814 } 815 816 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page); 817 ino_t page_cgroup_ino(struct page *page); 818 819 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 820 { 821 if (mem_cgroup_disabled()) 822 return true; 823 return !!(memcg->css.flags & CSS_ONLINE); 824 } 825 826 /* 827 * For memory reclaim. 828 */ 829 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg); 830 831 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, 832 int zid, int nr_pages); 833 834 static inline 835 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 836 enum lru_list lru, int zone_idx) 837 { 838 struct mem_cgroup_per_node *mz; 839 840 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 841 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]); 842 } 843 844 void mem_cgroup_handle_over_high(void); 845 846 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg); 847 848 unsigned long mem_cgroup_size(struct mem_cgroup *memcg); 849 850 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, 851 struct task_struct *p); 852 853 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg); 854 855 static inline void mem_cgroup_enter_user_fault(void) 856 { 857 WARN_ON(current->in_user_fault); 858 current->in_user_fault = 1; 859 } 860 861 static inline void mem_cgroup_exit_user_fault(void) 862 { 863 WARN_ON(!current->in_user_fault); 864 current->in_user_fault = 0; 865 } 866 867 static inline bool task_in_memcg_oom(struct task_struct *p) 868 { 869 return p->memcg_in_oom; 870 } 871 872 bool mem_cgroup_oom_synchronize(bool wait); 873 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, 874 struct mem_cgroup *oom_domain); 875 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg); 876 877 #ifdef CONFIG_MEMCG_SWAP 878 extern bool cgroup_memory_noswap; 879 #endif 880 881 struct mem_cgroup *lock_page_memcg(struct page *page); 882 void __unlock_page_memcg(struct mem_cgroup *memcg); 883 void unlock_page_memcg(struct page *page); 884 885 /* 886 * idx can be of type enum memcg_stat_item or node_stat_item. 887 * Keep in sync with memcg_exact_page_state(). 888 */ 889 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 890 { 891 long x = atomic_long_read(&memcg->vmstats[idx]); 892 #ifdef CONFIG_SMP 893 if (x < 0) 894 x = 0; 895 #endif 896 return x; 897 } 898 899 /* 900 * idx can be of type enum memcg_stat_item or node_stat_item. 901 * Keep in sync with memcg_exact_page_state(). 902 */ 903 static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg, 904 int idx) 905 { 906 long x = 0; 907 int cpu; 908 909 for_each_possible_cpu(cpu) 910 x += per_cpu(memcg->vmstats_local->stat[idx], cpu); 911 #ifdef CONFIG_SMP 912 if (x < 0) 913 x = 0; 914 #endif 915 return x; 916 } 917 918 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val); 919 920 /* idx can be of type enum memcg_stat_item or node_stat_item */ 921 static inline void mod_memcg_state(struct mem_cgroup *memcg, 922 int idx, int val) 923 { 924 unsigned long flags; 925 926 local_irq_save(flags); 927 __mod_memcg_state(memcg, idx, val); 928 local_irq_restore(flags); 929 } 930 931 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 932 enum node_stat_item idx) 933 { 934 struct mem_cgroup_per_node *pn; 935 long x; 936 937 if (mem_cgroup_disabled()) 938 return node_page_state(lruvec_pgdat(lruvec), idx); 939 940 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 941 x = atomic_long_read(&pn->lruvec_stat[idx]); 942 #ifdef CONFIG_SMP 943 if (x < 0) 944 x = 0; 945 #endif 946 return x; 947 } 948 949 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 950 enum node_stat_item idx) 951 { 952 struct mem_cgroup_per_node *pn; 953 long x = 0; 954 int cpu; 955 956 if (mem_cgroup_disabled()) 957 return node_page_state(lruvec_pgdat(lruvec), idx); 958 959 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 960 for_each_possible_cpu(cpu) 961 x += per_cpu(pn->lruvec_stat_local->count[idx], cpu); 962 #ifdef CONFIG_SMP 963 if (x < 0) 964 x = 0; 965 #endif 966 return x; 967 } 968 969 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, 970 int val); 971 void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val); 972 973 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 974 int val) 975 { 976 unsigned long flags; 977 978 local_irq_save(flags); 979 __mod_lruvec_kmem_state(p, idx, val); 980 local_irq_restore(flags); 981 } 982 983 static inline void mod_memcg_lruvec_state(struct lruvec *lruvec, 984 enum node_stat_item idx, int val) 985 { 986 unsigned long flags; 987 988 local_irq_save(flags); 989 __mod_memcg_lruvec_state(lruvec, idx, val); 990 local_irq_restore(flags); 991 } 992 993 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 994 gfp_t gfp_mask, 995 unsigned long *total_scanned); 996 997 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, 998 unsigned long count); 999 1000 static inline void count_memcg_events(struct mem_cgroup *memcg, 1001 enum vm_event_item idx, 1002 unsigned long count) 1003 { 1004 unsigned long flags; 1005 1006 local_irq_save(flags); 1007 __count_memcg_events(memcg, idx, count); 1008 local_irq_restore(flags); 1009 } 1010 1011 static inline void count_memcg_page_event(struct page *page, 1012 enum vm_event_item idx) 1013 { 1014 struct mem_cgroup *memcg = page_memcg(page); 1015 1016 if (memcg) 1017 count_memcg_events(memcg, idx, 1); 1018 } 1019 1020 static inline void count_memcg_event_mm(struct mm_struct *mm, 1021 enum vm_event_item idx) 1022 { 1023 struct mem_cgroup *memcg; 1024 1025 if (mem_cgroup_disabled()) 1026 return; 1027 1028 rcu_read_lock(); 1029 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 1030 if (likely(memcg)) 1031 count_memcg_events(memcg, idx, 1); 1032 rcu_read_unlock(); 1033 } 1034 1035 static inline void memcg_memory_event(struct mem_cgroup *memcg, 1036 enum memcg_memory_event event) 1037 { 1038 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX || 1039 event == MEMCG_SWAP_FAIL; 1040 1041 atomic_long_inc(&memcg->memory_events_local[event]); 1042 if (!swap_event) 1043 cgroup_file_notify(&memcg->events_local_file); 1044 1045 do { 1046 atomic_long_inc(&memcg->memory_events[event]); 1047 if (swap_event) 1048 cgroup_file_notify(&memcg->swap_events_file); 1049 else 1050 cgroup_file_notify(&memcg->events_file); 1051 1052 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) 1053 break; 1054 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) 1055 break; 1056 } while ((memcg = parent_mem_cgroup(memcg)) && 1057 !mem_cgroup_is_root(memcg)); 1058 } 1059 1060 static inline void memcg_memory_event_mm(struct mm_struct *mm, 1061 enum memcg_memory_event event) 1062 { 1063 struct mem_cgroup *memcg; 1064 1065 if (mem_cgroup_disabled()) 1066 return; 1067 1068 rcu_read_lock(); 1069 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 1070 if (likely(memcg)) 1071 memcg_memory_event(memcg, event); 1072 rcu_read_unlock(); 1073 } 1074 1075 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1076 void mem_cgroup_split_huge_fixup(struct page *head); 1077 #endif 1078 1079 #else /* CONFIG_MEMCG */ 1080 1081 #define MEM_CGROUP_ID_SHIFT 0 1082 #define MEM_CGROUP_ID_MAX 0 1083 1084 struct mem_cgroup; 1085 1086 static inline struct mem_cgroup *page_memcg(struct page *page) 1087 { 1088 return NULL; 1089 } 1090 1091 static inline struct mem_cgroup *page_memcg_rcu(struct page *page) 1092 { 1093 WARN_ON_ONCE(!rcu_read_lock_held()); 1094 return NULL; 1095 } 1096 1097 static inline struct mem_cgroup *page_memcg_check(struct page *page) 1098 { 1099 return NULL; 1100 } 1101 1102 static inline bool PageMemcgKmem(struct page *page) 1103 { 1104 return false; 1105 } 1106 1107 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 1108 { 1109 return true; 1110 } 1111 1112 static inline bool mem_cgroup_disabled(void) 1113 { 1114 return true; 1115 } 1116 1117 static inline void memcg_memory_event(struct mem_cgroup *memcg, 1118 enum memcg_memory_event event) 1119 { 1120 } 1121 1122 static inline void memcg_memory_event_mm(struct mm_struct *mm, 1123 enum memcg_memory_event event) 1124 { 1125 } 1126 1127 static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root, 1128 struct mem_cgroup *memcg, 1129 bool in_low_reclaim) 1130 { 1131 return 0; 1132 } 1133 1134 static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root, 1135 struct mem_cgroup *memcg) 1136 { 1137 } 1138 1139 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg) 1140 { 1141 return false; 1142 } 1143 1144 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg) 1145 { 1146 return false; 1147 } 1148 1149 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm, 1150 gfp_t gfp_mask) 1151 { 1152 return 0; 1153 } 1154 1155 static inline void mem_cgroup_uncharge(struct page *page) 1156 { 1157 } 1158 1159 static inline void mem_cgroup_uncharge_list(struct list_head *page_list) 1160 { 1161 } 1162 1163 static inline void mem_cgroup_migrate(struct page *old, struct page *new) 1164 { 1165 } 1166 1167 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg, 1168 struct pglist_data *pgdat) 1169 { 1170 return &pgdat->__lruvec; 1171 } 1172 1173 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page, 1174 struct pglist_data *pgdat) 1175 { 1176 return &pgdat->__lruvec; 1177 } 1178 1179 static inline bool lruvec_holds_page_lru_lock(struct page *page, 1180 struct lruvec *lruvec) 1181 { 1182 pg_data_t *pgdat = page_pgdat(page); 1183 1184 return lruvec == &pgdat->__lruvec; 1185 } 1186 1187 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 1188 { 1189 return NULL; 1190 } 1191 1192 static inline bool mm_match_cgroup(struct mm_struct *mm, 1193 struct mem_cgroup *memcg) 1194 { 1195 return true; 1196 } 1197 1198 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) 1199 { 1200 return NULL; 1201 } 1202 1203 static inline struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) 1204 { 1205 return NULL; 1206 } 1207 1208 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 1209 { 1210 } 1211 1212 static inline struct lruvec *lock_page_lruvec(struct page *page) 1213 { 1214 struct pglist_data *pgdat = page_pgdat(page); 1215 1216 spin_lock(&pgdat->__lruvec.lru_lock); 1217 return &pgdat->__lruvec; 1218 } 1219 1220 static inline struct lruvec *lock_page_lruvec_irq(struct page *page) 1221 { 1222 struct pglist_data *pgdat = page_pgdat(page); 1223 1224 spin_lock_irq(&pgdat->__lruvec.lru_lock); 1225 return &pgdat->__lruvec; 1226 } 1227 1228 static inline struct lruvec *lock_page_lruvec_irqsave(struct page *page, 1229 unsigned long *flagsp) 1230 { 1231 struct pglist_data *pgdat = page_pgdat(page); 1232 1233 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp); 1234 return &pgdat->__lruvec; 1235 } 1236 1237 static inline struct mem_cgroup * 1238 mem_cgroup_iter(struct mem_cgroup *root, 1239 struct mem_cgroup *prev, 1240 struct mem_cgroup_reclaim_cookie *reclaim) 1241 { 1242 return NULL; 1243 } 1244 1245 static inline void mem_cgroup_iter_break(struct mem_cgroup *root, 1246 struct mem_cgroup *prev) 1247 { 1248 } 1249 1250 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, 1251 int (*fn)(struct task_struct *, void *), void *arg) 1252 { 1253 return 0; 1254 } 1255 1256 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 1257 { 1258 return 0; 1259 } 1260 1261 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) 1262 { 1263 WARN_ON_ONCE(id); 1264 /* XXX: This should always return root_mem_cgroup */ 1265 return NULL; 1266 } 1267 1268 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 1269 { 1270 return NULL; 1271 } 1272 1273 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 1274 { 1275 return NULL; 1276 } 1277 1278 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 1279 { 1280 return true; 1281 } 1282 1283 static inline 1284 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 1285 enum lru_list lru, int zone_idx) 1286 { 1287 return 0; 1288 } 1289 1290 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) 1291 { 1292 return 0; 1293 } 1294 1295 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg) 1296 { 1297 return 0; 1298 } 1299 1300 static inline void 1301 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) 1302 { 1303 } 1304 1305 static inline void 1306 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) 1307 { 1308 } 1309 1310 static inline struct mem_cgroup *lock_page_memcg(struct page *page) 1311 { 1312 return NULL; 1313 } 1314 1315 static inline void __unlock_page_memcg(struct mem_cgroup *memcg) 1316 { 1317 } 1318 1319 static inline void unlock_page_memcg(struct page *page) 1320 { 1321 } 1322 1323 static inline void mem_cgroup_handle_over_high(void) 1324 { 1325 } 1326 1327 static inline void mem_cgroup_enter_user_fault(void) 1328 { 1329 } 1330 1331 static inline void mem_cgroup_exit_user_fault(void) 1332 { 1333 } 1334 1335 static inline bool task_in_memcg_oom(struct task_struct *p) 1336 { 1337 return false; 1338 } 1339 1340 static inline bool mem_cgroup_oom_synchronize(bool wait) 1341 { 1342 return false; 1343 } 1344 1345 static inline struct mem_cgroup *mem_cgroup_get_oom_group( 1346 struct task_struct *victim, struct mem_cgroup *oom_domain) 1347 { 1348 return NULL; 1349 } 1350 1351 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) 1352 { 1353 } 1354 1355 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 1356 { 1357 return 0; 1358 } 1359 1360 static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg, 1361 int idx) 1362 { 1363 return 0; 1364 } 1365 1366 static inline void __mod_memcg_state(struct mem_cgroup *memcg, 1367 int idx, 1368 int nr) 1369 { 1370 } 1371 1372 static inline void mod_memcg_state(struct mem_cgroup *memcg, 1373 int idx, 1374 int nr) 1375 { 1376 } 1377 1378 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 1379 enum node_stat_item idx) 1380 { 1381 return node_page_state(lruvec_pgdat(lruvec), idx); 1382 } 1383 1384 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 1385 enum node_stat_item idx) 1386 { 1387 return node_page_state(lruvec_pgdat(lruvec), idx); 1388 } 1389 1390 static inline void __mod_memcg_lruvec_state(struct lruvec *lruvec, 1391 enum node_stat_item idx, int val) 1392 { 1393 } 1394 1395 static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1396 int val) 1397 { 1398 struct page *page = virt_to_head_page(p); 1399 1400 __mod_node_page_state(page_pgdat(page), idx, val); 1401 } 1402 1403 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1404 int val) 1405 { 1406 struct page *page = virt_to_head_page(p); 1407 1408 mod_node_page_state(page_pgdat(page), idx, val); 1409 } 1410 1411 static inline 1412 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 1413 gfp_t gfp_mask, 1414 unsigned long *total_scanned) 1415 { 1416 return 0; 1417 } 1418 1419 static inline void mem_cgroup_split_huge_fixup(struct page *head) 1420 { 1421 } 1422 1423 static inline void count_memcg_events(struct mem_cgroup *memcg, 1424 enum vm_event_item idx, 1425 unsigned long count) 1426 { 1427 } 1428 1429 static inline void __count_memcg_events(struct mem_cgroup *memcg, 1430 enum vm_event_item idx, 1431 unsigned long count) 1432 { 1433 } 1434 1435 static inline void count_memcg_page_event(struct page *page, 1436 int idx) 1437 { 1438 } 1439 1440 static inline 1441 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx) 1442 { 1443 } 1444 1445 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) 1446 { 1447 } 1448 #endif /* CONFIG_MEMCG */ 1449 1450 static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx) 1451 { 1452 __mod_lruvec_kmem_state(p, idx, 1); 1453 } 1454 1455 static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx) 1456 { 1457 __mod_lruvec_kmem_state(p, idx, -1); 1458 } 1459 1460 static inline struct lruvec *parent_lruvec(struct lruvec *lruvec) 1461 { 1462 struct mem_cgroup *memcg; 1463 1464 memcg = lruvec_memcg(lruvec); 1465 if (!memcg) 1466 return NULL; 1467 memcg = parent_mem_cgroup(memcg); 1468 if (!memcg) 1469 return NULL; 1470 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec)); 1471 } 1472 1473 static inline void unlock_page_lruvec(struct lruvec *lruvec) 1474 { 1475 spin_unlock(&lruvec->lru_lock); 1476 } 1477 1478 static inline void unlock_page_lruvec_irq(struct lruvec *lruvec) 1479 { 1480 spin_unlock_irq(&lruvec->lru_lock); 1481 } 1482 1483 static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec, 1484 unsigned long flags) 1485 { 1486 spin_unlock_irqrestore(&lruvec->lru_lock, flags); 1487 } 1488 1489 /* Don't lock again iff page's lruvec locked */ 1490 static inline struct lruvec *relock_page_lruvec_irq(struct page *page, 1491 struct lruvec *locked_lruvec) 1492 { 1493 if (locked_lruvec) { 1494 if (lruvec_holds_page_lru_lock(page, locked_lruvec)) 1495 return locked_lruvec; 1496 1497 unlock_page_lruvec_irq(locked_lruvec); 1498 } 1499 1500 return lock_page_lruvec_irq(page); 1501 } 1502 1503 /* Don't lock again iff page's lruvec locked */ 1504 static inline struct lruvec *relock_page_lruvec_irqsave(struct page *page, 1505 struct lruvec *locked_lruvec, unsigned long *flags) 1506 { 1507 if (locked_lruvec) { 1508 if (lruvec_holds_page_lru_lock(page, locked_lruvec)) 1509 return locked_lruvec; 1510 1511 unlock_page_lruvec_irqrestore(locked_lruvec, *flags); 1512 } 1513 1514 return lock_page_lruvec_irqsave(page, flags); 1515 } 1516 1517 #ifdef CONFIG_CGROUP_WRITEBACK 1518 1519 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb); 1520 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, 1521 unsigned long *pheadroom, unsigned long *pdirty, 1522 unsigned long *pwriteback); 1523 1524 void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, 1525 struct bdi_writeback *wb); 1526 1527 static inline void mem_cgroup_track_foreign_dirty(struct page *page, 1528 struct bdi_writeback *wb) 1529 { 1530 if (mem_cgroup_disabled()) 1531 return; 1532 1533 if (unlikely(&page_memcg(page)->css != wb->memcg_css)) 1534 mem_cgroup_track_foreign_dirty_slowpath(page, wb); 1535 } 1536 1537 void mem_cgroup_flush_foreign(struct bdi_writeback *wb); 1538 1539 #else /* CONFIG_CGROUP_WRITEBACK */ 1540 1541 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) 1542 { 1543 return NULL; 1544 } 1545 1546 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb, 1547 unsigned long *pfilepages, 1548 unsigned long *pheadroom, 1549 unsigned long *pdirty, 1550 unsigned long *pwriteback) 1551 { 1552 } 1553 1554 static inline void mem_cgroup_track_foreign_dirty(struct page *page, 1555 struct bdi_writeback *wb) 1556 { 1557 } 1558 1559 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb) 1560 { 1561 } 1562 1563 #endif /* CONFIG_CGROUP_WRITEBACK */ 1564 1565 struct sock; 1566 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1567 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1568 #ifdef CONFIG_MEMCG 1569 extern struct static_key_false memcg_sockets_enabled_key; 1570 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key) 1571 void mem_cgroup_sk_alloc(struct sock *sk); 1572 void mem_cgroup_sk_free(struct sock *sk); 1573 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1574 { 1575 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure) 1576 return true; 1577 do { 1578 if (time_before(jiffies, memcg->socket_pressure)) 1579 return true; 1580 } while ((memcg = parent_mem_cgroup(memcg))); 1581 return false; 1582 } 1583 1584 extern int memcg_expand_shrinker_maps(int new_id); 1585 1586 extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg, 1587 int nid, int shrinker_id); 1588 #else 1589 #define mem_cgroup_sockets_enabled 0 1590 static inline void mem_cgroup_sk_alloc(struct sock *sk) { }; 1591 static inline void mem_cgroup_sk_free(struct sock *sk) { }; 1592 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1593 { 1594 return false; 1595 } 1596 1597 static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg, 1598 int nid, int shrinker_id) 1599 { 1600 } 1601 #endif 1602 1603 #ifdef CONFIG_MEMCG_KMEM 1604 int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, 1605 unsigned int nr_pages); 1606 void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages); 1607 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order); 1608 void __memcg_kmem_uncharge_page(struct page *page, int order); 1609 1610 struct obj_cgroup *get_obj_cgroup_from_current(void); 1611 1612 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size); 1613 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size); 1614 1615 extern struct static_key_false memcg_kmem_enabled_key; 1616 1617 extern int memcg_nr_cache_ids; 1618 void memcg_get_cache_ids(void); 1619 void memcg_put_cache_ids(void); 1620 1621 /* 1622 * Helper macro to loop through all memcg-specific caches. Callers must still 1623 * check if the cache is valid (it is either valid or NULL). 1624 * the slab_mutex must be held when looping through those caches 1625 */ 1626 #define for_each_memcg_cache_index(_idx) \ 1627 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++) 1628 1629 static inline bool memcg_kmem_enabled(void) 1630 { 1631 return static_branch_likely(&memcg_kmem_enabled_key); 1632 } 1633 1634 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1635 int order) 1636 { 1637 if (memcg_kmem_enabled()) 1638 return __memcg_kmem_charge_page(page, gfp, order); 1639 return 0; 1640 } 1641 1642 static inline void memcg_kmem_uncharge_page(struct page *page, int order) 1643 { 1644 if (memcg_kmem_enabled()) 1645 __memcg_kmem_uncharge_page(page, order); 1646 } 1647 1648 /* 1649 * A helper for accessing memcg's kmem_id, used for getting 1650 * corresponding LRU lists. 1651 */ 1652 static inline int memcg_cache_id(struct mem_cgroup *memcg) 1653 { 1654 return memcg ? memcg->kmemcg_id : -1; 1655 } 1656 1657 struct mem_cgroup *mem_cgroup_from_obj(void *p); 1658 1659 #else 1660 1661 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1662 int order) 1663 { 1664 return 0; 1665 } 1666 1667 static inline void memcg_kmem_uncharge_page(struct page *page, int order) 1668 { 1669 } 1670 1671 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1672 int order) 1673 { 1674 return 0; 1675 } 1676 1677 static inline void __memcg_kmem_uncharge_page(struct page *page, int order) 1678 { 1679 } 1680 1681 #define for_each_memcg_cache_index(_idx) \ 1682 for (; NULL; ) 1683 1684 static inline bool memcg_kmem_enabled(void) 1685 { 1686 return false; 1687 } 1688 1689 static inline int memcg_cache_id(struct mem_cgroup *memcg) 1690 { 1691 return -1; 1692 } 1693 1694 static inline void memcg_get_cache_ids(void) 1695 { 1696 } 1697 1698 static inline void memcg_put_cache_ids(void) 1699 { 1700 } 1701 1702 static inline struct mem_cgroup *mem_cgroup_from_obj(void *p) 1703 { 1704 return NULL; 1705 } 1706 1707 #endif /* CONFIG_MEMCG_KMEM */ 1708 1709 #endif /* _LINUX_MEMCONTROL_H */ 1710