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 /* Local (CPU and cgroup) page state & events */ 80 long state[MEMCG_NR_STAT]; 81 unsigned long events[NR_VM_EVENT_ITEMS]; 82 83 /* Delta calculation for lockless upward propagation */ 84 long state_prev[MEMCG_NR_STAT]; 85 unsigned long events_prev[NR_VM_EVENT_ITEMS]; 86 87 /* Cgroup1: threshold notifications & softlimit tree updates */ 88 unsigned long nr_page_events; 89 unsigned long targets[MEM_CGROUP_NTARGETS]; 90 }; 91 92 struct memcg_vmstats { 93 /* Aggregated (CPU and subtree) page state & events */ 94 long state[MEMCG_NR_STAT]; 95 unsigned long events[NR_VM_EVENT_ITEMS]; 96 97 /* Pending child counts during tree propagation */ 98 long state_pending[MEMCG_NR_STAT]; 99 unsigned long events_pending[NR_VM_EVENT_ITEMS]; 100 }; 101 102 struct mem_cgroup_reclaim_iter { 103 struct mem_cgroup *position; 104 /* scan generation, increased every round-trip */ 105 unsigned int generation; 106 }; 107 108 /* 109 * Bitmap and deferred work of shrinker::id corresponding to memcg-aware 110 * shrinkers, which have elements charged to this memcg. 111 */ 112 struct shrinker_info { 113 struct rcu_head rcu; 114 atomic_long_t *nr_deferred; 115 unsigned long *map; 116 }; 117 118 struct lruvec_stats_percpu { 119 /* Local (CPU and cgroup) state */ 120 long state[NR_VM_NODE_STAT_ITEMS]; 121 122 /* Delta calculation for lockless upward propagation */ 123 long state_prev[NR_VM_NODE_STAT_ITEMS]; 124 }; 125 126 struct lruvec_stats { 127 /* Aggregated (CPU and subtree) state */ 128 long state[NR_VM_NODE_STAT_ITEMS]; 129 130 /* Pending child counts during tree propagation */ 131 long state_pending[NR_VM_NODE_STAT_ITEMS]; 132 }; 133 134 /* 135 * per-node information in memory controller. 136 */ 137 struct mem_cgroup_per_node { 138 struct lruvec lruvec; 139 140 struct lruvec_stats_percpu __percpu *lruvec_stats_percpu; 141 struct lruvec_stats lruvec_stats; 142 143 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS]; 144 145 struct mem_cgroup_reclaim_iter iter; 146 147 struct shrinker_info __rcu *shrinker_info; 148 149 struct rb_node tree_node; /* RB tree node */ 150 unsigned long usage_in_excess;/* Set to the value by which */ 151 /* the soft limit is exceeded*/ 152 bool on_tree; 153 struct mem_cgroup *memcg; /* Back pointer, we cannot */ 154 /* use container_of */ 155 }; 156 157 struct mem_cgroup_threshold { 158 struct eventfd_ctx *eventfd; 159 unsigned long threshold; 160 }; 161 162 /* For threshold */ 163 struct mem_cgroup_threshold_ary { 164 /* An array index points to threshold just below or equal to usage. */ 165 int current_threshold; 166 /* Size of entries[] */ 167 unsigned int size; 168 /* Array of thresholds */ 169 struct mem_cgroup_threshold entries[]; 170 }; 171 172 struct mem_cgroup_thresholds { 173 /* Primary thresholds array */ 174 struct mem_cgroup_threshold_ary *primary; 175 /* 176 * Spare threshold array. 177 * This is needed to make mem_cgroup_unregister_event() "never fail". 178 * It must be able to store at least primary->size - 1 entries. 179 */ 180 struct mem_cgroup_threshold_ary *spare; 181 }; 182 183 #if defined(CONFIG_SMP) 184 struct memcg_padding { 185 char x[0]; 186 } ____cacheline_internodealigned_in_smp; 187 #define MEMCG_PADDING(name) struct memcg_padding name 188 #else 189 #define MEMCG_PADDING(name) 190 #endif 191 192 /* 193 * Remember four most recent foreign writebacks with dirty pages in this 194 * cgroup. Inode sharing is expected to be uncommon and, even if we miss 195 * one in a given round, we're likely to catch it later if it keeps 196 * foreign-dirtying, so a fairly low count should be enough. 197 * 198 * See mem_cgroup_track_foreign_dirty_slowpath() for details. 199 */ 200 #define MEMCG_CGWB_FRN_CNT 4 201 202 struct memcg_cgwb_frn { 203 u64 bdi_id; /* bdi->id of the foreign inode */ 204 int memcg_id; /* memcg->css.id of foreign inode */ 205 u64 at; /* jiffies_64 at the time of dirtying */ 206 struct wb_completion done; /* tracks in-flight foreign writebacks */ 207 }; 208 209 /* 210 * Bucket for arbitrarily byte-sized objects charged to a memory 211 * cgroup. The bucket can be reparented in one piece when the cgroup 212 * is destroyed, without having to round up the individual references 213 * of all live memory objects in the wild. 214 */ 215 struct obj_cgroup { 216 struct percpu_ref refcnt; 217 struct mem_cgroup *memcg; 218 atomic_t nr_charged_bytes; 219 union { 220 struct list_head list; 221 struct rcu_head rcu; 222 }; 223 }; 224 225 /* 226 * The memory controller data structure. The memory controller controls both 227 * page cache and RSS per cgroup. We would eventually like to provide 228 * statistics based on the statistics developed by Rik Van Riel for clock-pro, 229 * to help the administrator determine what knobs to tune. 230 */ 231 struct mem_cgroup { 232 struct cgroup_subsys_state css; 233 234 /* Private memcg ID. Used to ID objects that outlive the cgroup */ 235 struct mem_cgroup_id id; 236 237 /* Accounted resources */ 238 struct page_counter memory; /* Both v1 & v2 */ 239 240 union { 241 struct page_counter swap; /* v2 only */ 242 struct page_counter memsw; /* v1 only */ 243 }; 244 245 /* Legacy consumer-oriented counters */ 246 struct page_counter kmem; /* v1 only */ 247 struct page_counter tcpmem; /* v1 only */ 248 249 /* Range enforcement for interrupt charges */ 250 struct work_struct high_work; 251 252 unsigned long soft_limit; 253 254 /* vmpressure notifications */ 255 struct vmpressure vmpressure; 256 257 /* 258 * Should the OOM killer kill all belonging tasks, had it kill one? 259 */ 260 bool oom_group; 261 262 /* protected by memcg_oom_lock */ 263 bool oom_lock; 264 int under_oom; 265 266 int swappiness; 267 /* OOM-Killer disable */ 268 int oom_kill_disable; 269 270 /* memory.events and memory.events.local */ 271 struct cgroup_file events_file; 272 struct cgroup_file events_local_file; 273 274 /* handle for "memory.swap.events" */ 275 struct cgroup_file swap_events_file; 276 277 /* protect arrays of thresholds */ 278 struct mutex thresholds_lock; 279 280 /* thresholds for memory usage. RCU-protected */ 281 struct mem_cgroup_thresholds thresholds; 282 283 /* thresholds for mem+swap usage. RCU-protected */ 284 struct mem_cgroup_thresholds memsw_thresholds; 285 286 /* For oom notifier event fd */ 287 struct list_head oom_notify; 288 289 /* 290 * Should we move charges of a task when a task is moved into this 291 * mem_cgroup ? And what type of charges should we move ? 292 */ 293 unsigned long move_charge_at_immigrate; 294 /* taken only while moving_account > 0 */ 295 spinlock_t move_lock; 296 unsigned long move_lock_flags; 297 298 MEMCG_PADDING(_pad1_); 299 300 /* memory.stat */ 301 struct memcg_vmstats vmstats; 302 303 /* memory.events */ 304 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS]; 305 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS]; 306 307 unsigned long socket_pressure; 308 309 /* Legacy tcp memory accounting */ 310 bool tcpmem_active; 311 int tcpmem_pressure; 312 313 #ifdef CONFIG_MEMCG_KMEM 314 int kmemcg_id; 315 struct obj_cgroup __rcu *objcg; 316 struct list_head objcg_list; /* list of inherited objcgs */ 317 #endif 318 319 MEMCG_PADDING(_pad2_); 320 321 /* 322 * set > 0 if pages under this cgroup are moving to other cgroup. 323 */ 324 atomic_t moving_account; 325 struct task_struct *move_lock_task; 326 327 struct memcg_vmstats_percpu __percpu *vmstats_percpu; 328 329 #ifdef CONFIG_CGROUP_WRITEBACK 330 struct list_head cgwb_list; 331 struct wb_domain cgwb_domain; 332 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT]; 333 #endif 334 335 /* List of events which userspace want to receive */ 336 struct list_head event_list; 337 spinlock_t event_list_lock; 338 339 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 340 struct deferred_split deferred_split_queue; 341 #endif 342 343 struct mem_cgroup_per_node *nodeinfo[]; 344 }; 345 346 /* 347 * size of first charge trial. "32" comes from vmscan.c's magic value. 348 * TODO: maybe necessary to use big numbers in big irons. 349 */ 350 #define MEMCG_CHARGE_BATCH 32U 351 352 extern struct mem_cgroup *root_mem_cgroup; 353 354 enum page_memcg_data_flags { 355 /* page->memcg_data is a pointer to an objcgs vector */ 356 MEMCG_DATA_OBJCGS = (1UL << 0), 357 /* page has been accounted as a non-slab kernel page */ 358 MEMCG_DATA_KMEM = (1UL << 1), 359 /* the next bit after the last actual flag */ 360 __NR_MEMCG_DATA_FLAGS = (1UL << 2), 361 }; 362 363 #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1) 364 365 static inline bool PageMemcgKmem(struct page *page); 366 367 /* 368 * After the initialization objcg->memcg is always pointing at 369 * a valid memcg, but can be atomically swapped to the parent memcg. 370 * 371 * The caller must ensure that the returned memcg won't be released: 372 * e.g. acquire the rcu_read_lock or css_set_lock. 373 */ 374 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg) 375 { 376 return READ_ONCE(objcg->memcg); 377 } 378 379 /* 380 * __page_memcg - get the memory cgroup associated with a non-kmem page 381 * @page: a pointer to the page struct 382 * 383 * Returns a pointer to the memory cgroup associated with the page, 384 * or NULL. This function assumes that the page is known to have a 385 * proper memory cgroup pointer. It's not safe to call this function 386 * against some type of pages, e.g. slab pages or ex-slab pages or 387 * kmem pages. 388 */ 389 static inline struct mem_cgroup *__page_memcg(struct page *page) 390 { 391 unsigned long memcg_data = page->memcg_data; 392 393 VM_BUG_ON_PAGE(PageSlab(page), page); 394 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); 395 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); 396 397 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 398 } 399 400 /* 401 * __page_objcg - get the object cgroup associated with a kmem page 402 * @page: a pointer to the page struct 403 * 404 * Returns a pointer to the object cgroup associated with the page, 405 * or NULL. This function assumes that the page is known to have a 406 * proper object cgroup pointer. It's not safe to call this function 407 * against some type of pages, e.g. slab pages or ex-slab pages or 408 * LRU pages. 409 */ 410 static inline struct obj_cgroup *__page_objcg(struct page *page) 411 { 412 unsigned long memcg_data = page->memcg_data; 413 414 VM_BUG_ON_PAGE(PageSlab(page), page); 415 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); 416 VM_BUG_ON_PAGE(!(memcg_data & MEMCG_DATA_KMEM), page); 417 418 return (struct obj_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 419 } 420 421 /* 422 * page_memcg - get the memory cgroup associated with a page 423 * @page: a pointer to the page struct 424 * 425 * Returns a pointer to the memory cgroup associated with the page, 426 * or NULL. This function assumes that the page is known to have a 427 * proper memory cgroup pointer. It's not safe to call this function 428 * against some type of pages, e.g. slab pages or ex-slab pages. 429 * 430 * For a non-kmem page any of the following ensures page and memcg binding 431 * stability: 432 * 433 * - the page lock 434 * - LRU isolation 435 * - lock_page_memcg() 436 * - exclusive reference 437 * 438 * For a kmem page a caller should hold an rcu read lock to protect memcg 439 * associated with a kmem page from being released. 440 */ 441 static inline struct mem_cgroup *page_memcg(struct page *page) 442 { 443 if (PageMemcgKmem(page)) 444 return obj_cgroup_memcg(__page_objcg(page)); 445 else 446 return __page_memcg(page); 447 } 448 449 /* 450 * page_memcg_rcu - locklessly get the memory cgroup associated with a page 451 * @page: a pointer to the page struct 452 * 453 * Returns a pointer to the memory cgroup associated with the page, 454 * or NULL. This function assumes that the page is known to have a 455 * proper memory cgroup pointer. It's not safe to call this function 456 * against some type of pages, e.g. slab pages or ex-slab pages. 457 */ 458 static inline struct mem_cgroup *page_memcg_rcu(struct page *page) 459 { 460 unsigned long memcg_data = READ_ONCE(page->memcg_data); 461 462 VM_BUG_ON_PAGE(PageSlab(page), page); 463 WARN_ON_ONCE(!rcu_read_lock_held()); 464 465 if (memcg_data & MEMCG_DATA_KMEM) { 466 struct obj_cgroup *objcg; 467 468 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 469 return obj_cgroup_memcg(objcg); 470 } 471 472 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 473 } 474 475 /* 476 * page_memcg_check - get the memory cgroup associated with a page 477 * @page: a pointer to the page struct 478 * 479 * Returns a pointer to the memory cgroup associated with the page, 480 * or NULL. This function unlike page_memcg() can take any page 481 * as an argument. It has to be used in cases when it's not known if a page 482 * has an associated memory cgroup pointer or an object cgroups vector or 483 * an object cgroup. 484 * 485 * For a non-kmem page any of the following ensures page and memcg binding 486 * stability: 487 * 488 * - the page lock 489 * - LRU isolation 490 * - lock_page_memcg() 491 * - exclusive reference 492 * 493 * For a kmem page a caller should hold an rcu read lock to protect memcg 494 * associated with a kmem page from being released. 495 */ 496 static inline struct mem_cgroup *page_memcg_check(struct page *page) 497 { 498 /* 499 * Because page->memcg_data might be changed asynchronously 500 * for slab pages, READ_ONCE() should be used here. 501 */ 502 unsigned long memcg_data = READ_ONCE(page->memcg_data); 503 504 if (memcg_data & MEMCG_DATA_OBJCGS) 505 return NULL; 506 507 if (memcg_data & MEMCG_DATA_KMEM) { 508 struct obj_cgroup *objcg; 509 510 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 511 return obj_cgroup_memcg(objcg); 512 } 513 514 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 515 } 516 517 #ifdef CONFIG_MEMCG_KMEM 518 /* 519 * PageMemcgKmem - check if the page has MemcgKmem flag set 520 * @page: a pointer to the page struct 521 * 522 * Checks if the page has MemcgKmem flag set. The caller must ensure that 523 * the page has an associated memory cgroup. It's not safe to call this function 524 * against some types of pages, e.g. slab pages. 525 */ 526 static inline bool PageMemcgKmem(struct page *page) 527 { 528 VM_BUG_ON_PAGE(page->memcg_data & MEMCG_DATA_OBJCGS, page); 529 return page->memcg_data & MEMCG_DATA_KMEM; 530 } 531 532 /* 533 * page_objcgs - get the object cgroups vector associated with a page 534 * @page: a pointer to the page struct 535 * 536 * Returns a pointer to the object cgroups vector associated with the page, 537 * or NULL. This function assumes that the page is known to have an 538 * associated object cgroups vector. It's not safe to call this function 539 * against pages, which might have an associated memory cgroup: e.g. 540 * kernel stack pages. 541 */ 542 static inline struct obj_cgroup **page_objcgs(struct page *page) 543 { 544 unsigned long memcg_data = READ_ONCE(page->memcg_data); 545 546 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page); 547 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); 548 549 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 550 } 551 552 /* 553 * page_objcgs_check - get the object cgroups vector associated with a page 554 * @page: a pointer to the page struct 555 * 556 * Returns a pointer to the object cgroups vector associated with the page, 557 * or NULL. This function is safe to use if the page can be directly associated 558 * with a memory cgroup. 559 */ 560 static inline struct obj_cgroup **page_objcgs_check(struct page *page) 561 { 562 unsigned long memcg_data = READ_ONCE(page->memcg_data); 563 564 if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS)) 565 return NULL; 566 567 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); 568 569 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 570 } 571 572 #else 573 static inline bool PageMemcgKmem(struct page *page) 574 { 575 return false; 576 } 577 578 static inline struct obj_cgroup **page_objcgs(struct page *page) 579 { 580 return NULL; 581 } 582 583 static inline struct obj_cgroup **page_objcgs_check(struct page *page) 584 { 585 return NULL; 586 } 587 #endif 588 589 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 590 { 591 return (memcg == root_mem_cgroup); 592 } 593 594 static inline bool mem_cgroup_disabled(void) 595 { 596 return !cgroup_subsys_enabled(memory_cgrp_subsys); 597 } 598 599 static inline void mem_cgroup_protection(struct mem_cgroup *root, 600 struct mem_cgroup *memcg, 601 unsigned long *min, 602 unsigned long *low) 603 { 604 *min = *low = 0; 605 606 if (mem_cgroup_disabled()) 607 return; 608 609 /* 610 * There is no reclaim protection applied to a targeted reclaim. 611 * We are special casing this specific case here because 612 * mem_cgroup_protected calculation is not robust enough to keep 613 * the protection invariant for calculated effective values for 614 * parallel reclaimers with different reclaim target. This is 615 * especially a problem for tail memcgs (as they have pages on LRU) 616 * which would want to have effective values 0 for targeted reclaim 617 * but a different value for external reclaim. 618 * 619 * Example 620 * Let's have global and A's reclaim in parallel: 621 * | 622 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G) 623 * |\ 624 * | C (low = 1G, usage = 2.5G) 625 * B (low = 1G, usage = 0.5G) 626 * 627 * For the global reclaim 628 * A.elow = A.low 629 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow 630 * C.elow = min(C.usage, C.low) 631 * 632 * With the effective values resetting we have A reclaim 633 * A.elow = 0 634 * B.elow = B.low 635 * C.elow = C.low 636 * 637 * If the global reclaim races with A's reclaim then 638 * B.elow = C.elow = 0 because children_low_usage > A.elow) 639 * is possible and reclaiming B would be violating the protection. 640 * 641 */ 642 if (root == memcg) 643 return; 644 645 *min = READ_ONCE(memcg->memory.emin); 646 *low = READ_ONCE(memcg->memory.elow); 647 } 648 649 void mem_cgroup_calculate_protection(struct mem_cgroup *root, 650 struct mem_cgroup *memcg); 651 652 static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg) 653 { 654 /* 655 * The root memcg doesn't account charges, and doesn't support 656 * protection. 657 */ 658 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg); 659 660 } 661 662 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg) 663 { 664 if (!mem_cgroup_supports_protection(memcg)) 665 return false; 666 667 return READ_ONCE(memcg->memory.elow) >= 668 page_counter_read(&memcg->memory); 669 } 670 671 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg) 672 { 673 if (!mem_cgroup_supports_protection(memcg)) 674 return false; 675 676 return READ_ONCE(memcg->memory.emin) >= 677 page_counter_read(&memcg->memory); 678 } 679 680 int __mem_cgroup_charge(struct page *page, struct mm_struct *mm, 681 gfp_t gfp_mask); 682 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm, 683 gfp_t gfp_mask) 684 { 685 if (mem_cgroup_disabled()) 686 return 0; 687 return __mem_cgroup_charge(page, mm, gfp_mask); 688 } 689 690 int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm, 691 gfp_t gfp, swp_entry_t entry); 692 void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry); 693 694 void __mem_cgroup_uncharge(struct page *page); 695 static inline void mem_cgroup_uncharge(struct page *page) 696 { 697 if (mem_cgroup_disabled()) 698 return; 699 __mem_cgroup_uncharge(page); 700 } 701 702 void __mem_cgroup_uncharge_list(struct list_head *page_list); 703 static inline void mem_cgroup_uncharge_list(struct list_head *page_list) 704 { 705 if (mem_cgroup_disabled()) 706 return; 707 __mem_cgroup_uncharge_list(page_list); 708 } 709 710 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage); 711 712 /** 713 * mem_cgroup_lruvec - get the lru list vector for a memcg & node 714 * @memcg: memcg of the wanted lruvec 715 * @pgdat: pglist_data 716 * 717 * Returns the lru list vector holding pages for a given @memcg & 718 * @pgdat combination. This can be the node lruvec, if the memory 719 * controller is disabled. 720 */ 721 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg, 722 struct pglist_data *pgdat) 723 { 724 struct mem_cgroup_per_node *mz; 725 struct lruvec *lruvec; 726 727 if (mem_cgroup_disabled()) { 728 lruvec = &pgdat->__lruvec; 729 goto out; 730 } 731 732 if (!memcg) 733 memcg = root_mem_cgroup; 734 735 mz = memcg->nodeinfo[pgdat->node_id]; 736 lruvec = &mz->lruvec; 737 out: 738 /* 739 * Since a node can be onlined after the mem_cgroup was created, 740 * we have to be prepared to initialize lruvec->pgdat here; 741 * and if offlined then reonlined, we need to reinitialize it. 742 */ 743 if (unlikely(lruvec->pgdat != pgdat)) 744 lruvec->pgdat = pgdat; 745 return lruvec; 746 } 747 748 /** 749 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page 750 * @page: the page 751 * 752 * This function relies on page->mem_cgroup being stable. 753 */ 754 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page) 755 { 756 pg_data_t *pgdat = page_pgdat(page); 757 struct mem_cgroup *memcg = page_memcg(page); 758 759 VM_WARN_ON_ONCE_PAGE(!memcg && !mem_cgroup_disabled(), page); 760 return mem_cgroup_lruvec(memcg, pgdat); 761 } 762 763 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p); 764 765 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm); 766 767 struct lruvec *lock_page_lruvec(struct page *page); 768 struct lruvec *lock_page_lruvec_irq(struct page *page); 769 struct lruvec *lock_page_lruvec_irqsave(struct page *page, 770 unsigned long *flags); 771 772 #ifdef CONFIG_DEBUG_VM 773 void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page); 774 #else 775 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) 776 { 777 } 778 #endif 779 780 static inline 781 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){ 782 return css ? container_of(css, struct mem_cgroup, css) : NULL; 783 } 784 785 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg) 786 { 787 return percpu_ref_tryget(&objcg->refcnt); 788 } 789 790 static inline void obj_cgroup_get(struct obj_cgroup *objcg) 791 { 792 percpu_ref_get(&objcg->refcnt); 793 } 794 795 static inline void obj_cgroup_get_many(struct obj_cgroup *objcg, 796 unsigned long nr) 797 { 798 percpu_ref_get_many(&objcg->refcnt, nr); 799 } 800 801 static inline void obj_cgroup_put(struct obj_cgroup *objcg) 802 { 803 percpu_ref_put(&objcg->refcnt); 804 } 805 806 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 807 { 808 if (memcg) 809 css_put(&memcg->css); 810 } 811 812 #define mem_cgroup_from_counter(counter, member) \ 813 container_of(counter, struct mem_cgroup, member) 814 815 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *, 816 struct mem_cgroup *, 817 struct mem_cgroup_reclaim_cookie *); 818 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *); 819 int mem_cgroup_scan_tasks(struct mem_cgroup *, 820 int (*)(struct task_struct *, void *), void *); 821 822 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 823 { 824 if (mem_cgroup_disabled()) 825 return 0; 826 827 return memcg->id.id; 828 } 829 struct mem_cgroup *mem_cgroup_from_id(unsigned short id); 830 831 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 832 { 833 return mem_cgroup_from_css(seq_css(m)); 834 } 835 836 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 837 { 838 struct mem_cgroup_per_node *mz; 839 840 if (mem_cgroup_disabled()) 841 return NULL; 842 843 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 844 return mz->memcg; 845 } 846 847 /** 848 * parent_mem_cgroup - find the accounting parent of a memcg 849 * @memcg: memcg whose parent to find 850 * 851 * Returns the parent memcg, or NULL if this is the root or the memory 852 * controller is in legacy no-hierarchy mode. 853 */ 854 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 855 { 856 if (!memcg->memory.parent) 857 return NULL; 858 return mem_cgroup_from_counter(memcg->memory.parent, memory); 859 } 860 861 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, 862 struct mem_cgroup *root) 863 { 864 if (root == memcg) 865 return true; 866 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); 867 } 868 869 static inline bool mm_match_cgroup(struct mm_struct *mm, 870 struct mem_cgroup *memcg) 871 { 872 struct mem_cgroup *task_memcg; 873 bool match = false; 874 875 rcu_read_lock(); 876 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 877 if (task_memcg) 878 match = mem_cgroup_is_descendant(task_memcg, memcg); 879 rcu_read_unlock(); 880 return match; 881 } 882 883 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page); 884 ino_t page_cgroup_ino(struct page *page); 885 886 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 887 { 888 if (mem_cgroup_disabled()) 889 return true; 890 return !!(memcg->css.flags & CSS_ONLINE); 891 } 892 893 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, 894 int zid, int nr_pages); 895 896 static inline 897 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 898 enum lru_list lru, int zone_idx) 899 { 900 struct mem_cgroup_per_node *mz; 901 902 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 903 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]); 904 } 905 906 void mem_cgroup_handle_over_high(void); 907 908 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg); 909 910 unsigned long mem_cgroup_size(struct mem_cgroup *memcg); 911 912 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, 913 struct task_struct *p); 914 915 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg); 916 917 static inline void mem_cgroup_enter_user_fault(void) 918 { 919 WARN_ON(current->in_user_fault); 920 current->in_user_fault = 1; 921 } 922 923 static inline void mem_cgroup_exit_user_fault(void) 924 { 925 WARN_ON(!current->in_user_fault); 926 current->in_user_fault = 0; 927 } 928 929 static inline bool task_in_memcg_oom(struct task_struct *p) 930 { 931 return p->memcg_in_oom; 932 } 933 934 bool mem_cgroup_oom_synchronize(bool wait); 935 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, 936 struct mem_cgroup *oom_domain); 937 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg); 938 939 #ifdef CONFIG_MEMCG_SWAP 940 extern bool cgroup_memory_noswap; 941 #endif 942 943 void lock_page_memcg(struct page *page); 944 void unlock_page_memcg(struct page *page); 945 946 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val); 947 948 /* idx can be of type enum memcg_stat_item or node_stat_item */ 949 static inline void mod_memcg_state(struct mem_cgroup *memcg, 950 int idx, int val) 951 { 952 unsigned long flags; 953 954 local_irq_save(flags); 955 __mod_memcg_state(memcg, idx, val); 956 local_irq_restore(flags); 957 } 958 959 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 960 { 961 return READ_ONCE(memcg->vmstats.state[idx]); 962 } 963 964 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 965 enum node_stat_item idx) 966 { 967 struct mem_cgroup_per_node *pn; 968 969 if (mem_cgroup_disabled()) 970 return node_page_state(lruvec_pgdat(lruvec), idx); 971 972 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 973 return READ_ONCE(pn->lruvec_stats.state[idx]); 974 } 975 976 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 977 enum node_stat_item idx) 978 { 979 struct mem_cgroup_per_node *pn; 980 long x = 0; 981 int cpu; 982 983 if (mem_cgroup_disabled()) 984 return node_page_state(lruvec_pgdat(lruvec), idx); 985 986 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 987 for_each_possible_cpu(cpu) 988 x += per_cpu(pn->lruvec_stats_percpu->state[idx], cpu); 989 #ifdef CONFIG_SMP 990 if (x < 0) 991 x = 0; 992 #endif 993 return x; 994 } 995 996 void mem_cgroup_flush_stats(void); 997 998 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, 999 int val); 1000 void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val); 1001 1002 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1003 int val) 1004 { 1005 unsigned long flags; 1006 1007 local_irq_save(flags); 1008 __mod_lruvec_kmem_state(p, idx, val); 1009 local_irq_restore(flags); 1010 } 1011 1012 static inline void mod_memcg_lruvec_state(struct lruvec *lruvec, 1013 enum node_stat_item idx, int val) 1014 { 1015 unsigned long flags; 1016 1017 local_irq_save(flags); 1018 __mod_memcg_lruvec_state(lruvec, idx, val); 1019 local_irq_restore(flags); 1020 } 1021 1022 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, 1023 unsigned long count); 1024 1025 static inline void count_memcg_events(struct mem_cgroup *memcg, 1026 enum vm_event_item idx, 1027 unsigned long count) 1028 { 1029 unsigned long flags; 1030 1031 local_irq_save(flags); 1032 __count_memcg_events(memcg, idx, count); 1033 local_irq_restore(flags); 1034 } 1035 1036 static inline void count_memcg_page_event(struct page *page, 1037 enum vm_event_item idx) 1038 { 1039 struct mem_cgroup *memcg = page_memcg(page); 1040 1041 if (memcg) 1042 count_memcg_events(memcg, idx, 1); 1043 } 1044 1045 static inline void count_memcg_event_mm(struct mm_struct *mm, 1046 enum vm_event_item idx) 1047 { 1048 struct mem_cgroup *memcg; 1049 1050 if (mem_cgroup_disabled()) 1051 return; 1052 1053 rcu_read_lock(); 1054 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 1055 if (likely(memcg)) 1056 count_memcg_events(memcg, idx, 1); 1057 rcu_read_unlock(); 1058 } 1059 1060 static inline void memcg_memory_event(struct mem_cgroup *memcg, 1061 enum memcg_memory_event event) 1062 { 1063 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX || 1064 event == MEMCG_SWAP_FAIL; 1065 1066 atomic_long_inc(&memcg->memory_events_local[event]); 1067 if (!swap_event) 1068 cgroup_file_notify(&memcg->events_local_file); 1069 1070 do { 1071 atomic_long_inc(&memcg->memory_events[event]); 1072 if (swap_event) 1073 cgroup_file_notify(&memcg->swap_events_file); 1074 else 1075 cgroup_file_notify(&memcg->events_file); 1076 1077 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) 1078 break; 1079 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) 1080 break; 1081 } while ((memcg = parent_mem_cgroup(memcg)) && 1082 !mem_cgroup_is_root(memcg)); 1083 } 1084 1085 static inline void memcg_memory_event_mm(struct mm_struct *mm, 1086 enum memcg_memory_event event) 1087 { 1088 struct mem_cgroup *memcg; 1089 1090 if (mem_cgroup_disabled()) 1091 return; 1092 1093 rcu_read_lock(); 1094 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 1095 if (likely(memcg)) 1096 memcg_memory_event(memcg, event); 1097 rcu_read_unlock(); 1098 } 1099 1100 void split_page_memcg(struct page *head, unsigned int nr); 1101 1102 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 1103 gfp_t gfp_mask, 1104 unsigned long *total_scanned); 1105 1106 #else /* CONFIG_MEMCG */ 1107 1108 #define MEM_CGROUP_ID_SHIFT 0 1109 #define MEM_CGROUP_ID_MAX 0 1110 1111 static inline struct mem_cgroup *page_memcg(struct page *page) 1112 { 1113 return NULL; 1114 } 1115 1116 static inline struct mem_cgroup *page_memcg_rcu(struct page *page) 1117 { 1118 WARN_ON_ONCE(!rcu_read_lock_held()); 1119 return NULL; 1120 } 1121 1122 static inline struct mem_cgroup *page_memcg_check(struct page *page) 1123 { 1124 return NULL; 1125 } 1126 1127 static inline bool PageMemcgKmem(struct page *page) 1128 { 1129 return false; 1130 } 1131 1132 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 1133 { 1134 return true; 1135 } 1136 1137 static inline bool mem_cgroup_disabled(void) 1138 { 1139 return true; 1140 } 1141 1142 static inline void memcg_memory_event(struct mem_cgroup *memcg, 1143 enum memcg_memory_event event) 1144 { 1145 } 1146 1147 static inline void memcg_memory_event_mm(struct mm_struct *mm, 1148 enum memcg_memory_event event) 1149 { 1150 } 1151 1152 static inline void mem_cgroup_protection(struct mem_cgroup *root, 1153 struct mem_cgroup *memcg, 1154 unsigned long *min, 1155 unsigned long *low) 1156 { 1157 *min = *low = 0; 1158 } 1159 1160 static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root, 1161 struct mem_cgroup *memcg) 1162 { 1163 } 1164 1165 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg) 1166 { 1167 return false; 1168 } 1169 1170 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg) 1171 { 1172 return false; 1173 } 1174 1175 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm, 1176 gfp_t gfp_mask) 1177 { 1178 return 0; 1179 } 1180 1181 static inline int mem_cgroup_swapin_charge_page(struct page *page, 1182 struct mm_struct *mm, gfp_t gfp, swp_entry_t entry) 1183 { 1184 return 0; 1185 } 1186 1187 static inline void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) 1188 { 1189 } 1190 1191 static inline void mem_cgroup_uncharge(struct page *page) 1192 { 1193 } 1194 1195 static inline void mem_cgroup_uncharge_list(struct list_head *page_list) 1196 { 1197 } 1198 1199 static inline void mem_cgroup_migrate(struct page *old, struct page *new) 1200 { 1201 } 1202 1203 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg, 1204 struct pglist_data *pgdat) 1205 { 1206 return &pgdat->__lruvec; 1207 } 1208 1209 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page) 1210 { 1211 pg_data_t *pgdat = page_pgdat(page); 1212 1213 return &pgdat->__lruvec; 1214 } 1215 1216 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) 1217 { 1218 } 1219 1220 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 1221 { 1222 return NULL; 1223 } 1224 1225 static inline bool mm_match_cgroup(struct mm_struct *mm, 1226 struct mem_cgroup *memcg) 1227 { 1228 return true; 1229 } 1230 1231 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) 1232 { 1233 return NULL; 1234 } 1235 1236 static inline 1237 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css) 1238 { 1239 return NULL; 1240 } 1241 1242 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 1243 { 1244 } 1245 1246 static inline struct lruvec *lock_page_lruvec(struct page *page) 1247 { 1248 struct pglist_data *pgdat = page_pgdat(page); 1249 1250 spin_lock(&pgdat->__lruvec.lru_lock); 1251 return &pgdat->__lruvec; 1252 } 1253 1254 static inline struct lruvec *lock_page_lruvec_irq(struct page *page) 1255 { 1256 struct pglist_data *pgdat = page_pgdat(page); 1257 1258 spin_lock_irq(&pgdat->__lruvec.lru_lock); 1259 return &pgdat->__lruvec; 1260 } 1261 1262 static inline struct lruvec *lock_page_lruvec_irqsave(struct page *page, 1263 unsigned long *flagsp) 1264 { 1265 struct pglist_data *pgdat = page_pgdat(page); 1266 1267 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp); 1268 return &pgdat->__lruvec; 1269 } 1270 1271 static inline struct mem_cgroup * 1272 mem_cgroup_iter(struct mem_cgroup *root, 1273 struct mem_cgroup *prev, 1274 struct mem_cgroup_reclaim_cookie *reclaim) 1275 { 1276 return NULL; 1277 } 1278 1279 static inline void mem_cgroup_iter_break(struct mem_cgroup *root, 1280 struct mem_cgroup *prev) 1281 { 1282 } 1283 1284 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, 1285 int (*fn)(struct task_struct *, void *), void *arg) 1286 { 1287 return 0; 1288 } 1289 1290 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 1291 { 1292 return 0; 1293 } 1294 1295 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) 1296 { 1297 WARN_ON_ONCE(id); 1298 /* XXX: This should always return root_mem_cgroup */ 1299 return NULL; 1300 } 1301 1302 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 1303 { 1304 return NULL; 1305 } 1306 1307 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 1308 { 1309 return NULL; 1310 } 1311 1312 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 1313 { 1314 return true; 1315 } 1316 1317 static inline 1318 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 1319 enum lru_list lru, int zone_idx) 1320 { 1321 return 0; 1322 } 1323 1324 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) 1325 { 1326 return 0; 1327 } 1328 1329 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg) 1330 { 1331 return 0; 1332 } 1333 1334 static inline void 1335 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) 1336 { 1337 } 1338 1339 static inline void 1340 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) 1341 { 1342 } 1343 1344 static inline void lock_page_memcg(struct page *page) 1345 { 1346 } 1347 1348 static inline void unlock_page_memcg(struct page *page) 1349 { 1350 } 1351 1352 static inline void mem_cgroup_handle_over_high(void) 1353 { 1354 } 1355 1356 static inline void mem_cgroup_enter_user_fault(void) 1357 { 1358 } 1359 1360 static inline void mem_cgroup_exit_user_fault(void) 1361 { 1362 } 1363 1364 static inline bool task_in_memcg_oom(struct task_struct *p) 1365 { 1366 return false; 1367 } 1368 1369 static inline bool mem_cgroup_oom_synchronize(bool wait) 1370 { 1371 return false; 1372 } 1373 1374 static inline struct mem_cgroup *mem_cgroup_get_oom_group( 1375 struct task_struct *victim, struct mem_cgroup *oom_domain) 1376 { 1377 return NULL; 1378 } 1379 1380 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) 1381 { 1382 } 1383 1384 static inline void __mod_memcg_state(struct mem_cgroup *memcg, 1385 int idx, 1386 int nr) 1387 { 1388 } 1389 1390 static inline void mod_memcg_state(struct mem_cgroup *memcg, 1391 int idx, 1392 int nr) 1393 { 1394 } 1395 1396 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 1397 { 1398 return 0; 1399 } 1400 1401 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 1402 enum node_stat_item idx) 1403 { 1404 return node_page_state(lruvec_pgdat(lruvec), idx); 1405 } 1406 1407 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 1408 enum node_stat_item idx) 1409 { 1410 return node_page_state(lruvec_pgdat(lruvec), idx); 1411 } 1412 1413 static inline void mem_cgroup_flush_stats(void) 1414 { 1415 } 1416 1417 static inline void __mod_memcg_lruvec_state(struct lruvec *lruvec, 1418 enum node_stat_item idx, int val) 1419 { 1420 } 1421 1422 static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1423 int val) 1424 { 1425 struct page *page = virt_to_head_page(p); 1426 1427 __mod_node_page_state(page_pgdat(page), idx, val); 1428 } 1429 1430 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1431 int val) 1432 { 1433 struct page *page = virt_to_head_page(p); 1434 1435 mod_node_page_state(page_pgdat(page), idx, val); 1436 } 1437 1438 static inline void count_memcg_events(struct mem_cgroup *memcg, 1439 enum vm_event_item idx, 1440 unsigned long count) 1441 { 1442 } 1443 1444 static inline void __count_memcg_events(struct mem_cgroup *memcg, 1445 enum vm_event_item idx, 1446 unsigned long count) 1447 { 1448 } 1449 1450 static inline void count_memcg_page_event(struct page *page, 1451 int idx) 1452 { 1453 } 1454 1455 static inline 1456 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx) 1457 { 1458 } 1459 1460 static inline void split_page_memcg(struct page *head, unsigned int nr) 1461 { 1462 } 1463 1464 static inline 1465 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 1466 gfp_t gfp_mask, 1467 unsigned long *total_scanned) 1468 { 1469 return 0; 1470 } 1471 #endif /* CONFIG_MEMCG */ 1472 1473 static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx) 1474 { 1475 __mod_lruvec_kmem_state(p, idx, 1); 1476 } 1477 1478 static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx) 1479 { 1480 __mod_lruvec_kmem_state(p, idx, -1); 1481 } 1482 1483 static inline struct lruvec *parent_lruvec(struct lruvec *lruvec) 1484 { 1485 struct mem_cgroup *memcg; 1486 1487 memcg = lruvec_memcg(lruvec); 1488 if (!memcg) 1489 return NULL; 1490 memcg = parent_mem_cgroup(memcg); 1491 if (!memcg) 1492 return NULL; 1493 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec)); 1494 } 1495 1496 static inline void unlock_page_lruvec(struct lruvec *lruvec) 1497 { 1498 spin_unlock(&lruvec->lru_lock); 1499 } 1500 1501 static inline void unlock_page_lruvec_irq(struct lruvec *lruvec) 1502 { 1503 spin_unlock_irq(&lruvec->lru_lock); 1504 } 1505 1506 static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec, 1507 unsigned long flags) 1508 { 1509 spin_unlock_irqrestore(&lruvec->lru_lock, flags); 1510 } 1511 1512 /* Test requires a stable page->memcg binding, see page_memcg() */ 1513 static inline bool page_matches_lruvec(struct page *page, struct lruvec *lruvec) 1514 { 1515 return lruvec_pgdat(lruvec) == page_pgdat(page) && 1516 lruvec_memcg(lruvec) == page_memcg(page); 1517 } 1518 1519 /* Don't lock again iff page's lruvec locked */ 1520 static inline struct lruvec *relock_page_lruvec_irq(struct page *page, 1521 struct lruvec *locked_lruvec) 1522 { 1523 if (locked_lruvec) { 1524 if (page_matches_lruvec(page, locked_lruvec)) 1525 return locked_lruvec; 1526 1527 unlock_page_lruvec_irq(locked_lruvec); 1528 } 1529 1530 return lock_page_lruvec_irq(page); 1531 } 1532 1533 /* Don't lock again iff page's lruvec locked */ 1534 static inline struct lruvec *relock_page_lruvec_irqsave(struct page *page, 1535 struct lruvec *locked_lruvec, unsigned long *flags) 1536 { 1537 if (locked_lruvec) { 1538 if (page_matches_lruvec(page, locked_lruvec)) 1539 return locked_lruvec; 1540 1541 unlock_page_lruvec_irqrestore(locked_lruvec, *flags); 1542 } 1543 1544 return lock_page_lruvec_irqsave(page, flags); 1545 } 1546 1547 #ifdef CONFIG_CGROUP_WRITEBACK 1548 1549 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb); 1550 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, 1551 unsigned long *pheadroom, unsigned long *pdirty, 1552 unsigned long *pwriteback); 1553 1554 void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, 1555 struct bdi_writeback *wb); 1556 1557 static inline void mem_cgroup_track_foreign_dirty(struct page *page, 1558 struct bdi_writeback *wb) 1559 { 1560 if (mem_cgroup_disabled()) 1561 return; 1562 1563 if (unlikely(&page_memcg(page)->css != wb->memcg_css)) 1564 mem_cgroup_track_foreign_dirty_slowpath(page, wb); 1565 } 1566 1567 void mem_cgroup_flush_foreign(struct bdi_writeback *wb); 1568 1569 #else /* CONFIG_CGROUP_WRITEBACK */ 1570 1571 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) 1572 { 1573 return NULL; 1574 } 1575 1576 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb, 1577 unsigned long *pfilepages, 1578 unsigned long *pheadroom, 1579 unsigned long *pdirty, 1580 unsigned long *pwriteback) 1581 { 1582 } 1583 1584 static inline void mem_cgroup_track_foreign_dirty(struct page *page, 1585 struct bdi_writeback *wb) 1586 { 1587 } 1588 1589 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb) 1590 { 1591 } 1592 1593 #endif /* CONFIG_CGROUP_WRITEBACK */ 1594 1595 struct sock; 1596 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, 1597 gfp_t gfp_mask); 1598 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1599 #ifdef CONFIG_MEMCG 1600 extern struct static_key_false memcg_sockets_enabled_key; 1601 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key) 1602 void mem_cgroup_sk_alloc(struct sock *sk); 1603 void mem_cgroup_sk_free(struct sock *sk); 1604 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1605 { 1606 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure) 1607 return true; 1608 do { 1609 if (time_before(jiffies, READ_ONCE(memcg->socket_pressure))) 1610 return true; 1611 } while ((memcg = parent_mem_cgroup(memcg))); 1612 return false; 1613 } 1614 1615 int alloc_shrinker_info(struct mem_cgroup *memcg); 1616 void free_shrinker_info(struct mem_cgroup *memcg); 1617 void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id); 1618 void reparent_shrinker_deferred(struct mem_cgroup *memcg); 1619 #else 1620 #define mem_cgroup_sockets_enabled 0 1621 static inline void mem_cgroup_sk_alloc(struct sock *sk) { }; 1622 static inline void mem_cgroup_sk_free(struct sock *sk) { }; 1623 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1624 { 1625 return false; 1626 } 1627 1628 static inline void set_shrinker_bit(struct mem_cgroup *memcg, 1629 int nid, int shrinker_id) 1630 { 1631 } 1632 #endif 1633 1634 #ifdef CONFIG_MEMCG_KMEM 1635 bool mem_cgroup_kmem_disabled(void); 1636 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order); 1637 void __memcg_kmem_uncharge_page(struct page *page, int order); 1638 1639 struct obj_cgroup *get_obj_cgroup_from_current(void); 1640 1641 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size); 1642 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size); 1643 1644 extern struct static_key_false memcg_kmem_enabled_key; 1645 1646 extern int memcg_nr_cache_ids; 1647 void memcg_get_cache_ids(void); 1648 void memcg_put_cache_ids(void); 1649 1650 /* 1651 * Helper macro to loop through all memcg-specific caches. Callers must still 1652 * check if the cache is valid (it is either valid or NULL). 1653 * the slab_mutex must be held when looping through those caches 1654 */ 1655 #define for_each_memcg_cache_index(_idx) \ 1656 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++) 1657 1658 static inline bool memcg_kmem_enabled(void) 1659 { 1660 return static_branch_likely(&memcg_kmem_enabled_key); 1661 } 1662 1663 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1664 int order) 1665 { 1666 if (memcg_kmem_enabled()) 1667 return __memcg_kmem_charge_page(page, gfp, order); 1668 return 0; 1669 } 1670 1671 static inline void memcg_kmem_uncharge_page(struct page *page, int order) 1672 { 1673 if (memcg_kmem_enabled()) 1674 __memcg_kmem_uncharge_page(page, order); 1675 } 1676 1677 /* 1678 * A helper for accessing memcg's kmem_id, used for getting 1679 * corresponding LRU lists. 1680 */ 1681 static inline int memcg_cache_id(struct mem_cgroup *memcg) 1682 { 1683 return memcg ? memcg->kmemcg_id : -1; 1684 } 1685 1686 struct mem_cgroup *mem_cgroup_from_obj(void *p); 1687 1688 #else 1689 static inline bool mem_cgroup_kmem_disabled(void) 1690 { 1691 return true; 1692 } 1693 1694 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1695 int order) 1696 { 1697 return 0; 1698 } 1699 1700 static inline void memcg_kmem_uncharge_page(struct page *page, int order) 1701 { 1702 } 1703 1704 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1705 int order) 1706 { 1707 return 0; 1708 } 1709 1710 static inline void __memcg_kmem_uncharge_page(struct page *page, int order) 1711 { 1712 } 1713 1714 #define for_each_memcg_cache_index(_idx) \ 1715 for (; NULL; ) 1716 1717 static inline bool memcg_kmem_enabled(void) 1718 { 1719 return false; 1720 } 1721 1722 static inline int memcg_cache_id(struct mem_cgroup *memcg) 1723 { 1724 return -1; 1725 } 1726 1727 static inline void memcg_get_cache_ids(void) 1728 { 1729 } 1730 1731 static inline void memcg_put_cache_ids(void) 1732 { 1733 } 1734 1735 static inline struct mem_cgroup *mem_cgroup_from_obj(void *p) 1736 { 1737 return NULL; 1738 } 1739 1740 #endif /* CONFIG_MEMCG_KMEM */ 1741 1742 #endif /* _LINUX_MEMCONTROL_H */ 1743