1/* SPDX-License-Identifier: GPL-2.0 */ 2#ifndef _LINUX_MMZONE_H 3#define _LINUX_MMZONE_H 4 5#ifndef __ASSEMBLY__ 6#ifndef __GENERATING_BOUNDS_H 7 8#include <linux/spinlock.h> 9#include <linux/list.h> 10#include <linux/list_nulls.h> 11#include <linux/wait.h> 12#include <linux/bitops.h> 13#include <linux/cache.h> 14#include <linux/threads.h> 15#include <linux/numa.h> 16#include <linux/init.h> 17#include <linux/seqlock.h> 18#include <linux/nodemask.h> 19#include <linux/pageblock-flags.h> 20#include <linux/page-flags-layout.h> 21#include <linux/atomic.h> 22#include <linux/mm_types.h> 23#include <linux/page-flags.h> 24#include <linux/local_lock.h> 25#include <linux/zswap.h> 26#include <asm/page.h> 27 28/* Free memory management - zoned buddy allocator. */ 29#ifndef CONFIG_ARCH_FORCE_MAX_ORDER 30#define MAX_PAGE_ORDER 10 31#else 32#define MAX_PAGE_ORDER CONFIG_ARCH_FORCE_MAX_ORDER 33#endif 34#define MAX_ORDER_NR_PAGES (1 << MAX_PAGE_ORDER) 35 36#define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES) 37 38#define NR_PAGE_ORDERS (MAX_PAGE_ORDER + 1) 39 40/* 41 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed 42 * costly to service. That is between allocation orders which should 43 * coalesce naturally under reasonable reclaim pressure and those which 44 * will not. 45 */ 46#define PAGE_ALLOC_COSTLY_ORDER 3 47 48enum migratetype { 49 MIGRATE_UNMOVABLE, 50 MIGRATE_MOVABLE, 51 MIGRATE_RECLAIMABLE, 52 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ 53 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, 54#ifdef CONFIG_CMA 55 /* 56 * MIGRATE_CMA migration type is designed to mimic the way 57 * ZONE_MOVABLE works. Only movable pages can be allocated 58 * from MIGRATE_CMA pageblocks and page allocator never 59 * implicitly change migration type of MIGRATE_CMA pageblock. 60 * 61 * The way to use it is to change migratetype of a range of 62 * pageblocks to MIGRATE_CMA which can be done by 63 * __free_pageblock_cma() function. 64 */ 65 MIGRATE_CMA, 66#endif 67#ifdef CONFIG_MEMORY_ISOLATION 68 MIGRATE_ISOLATE, /* can't allocate from here */ 69#endif 70 MIGRATE_TYPES 71}; 72 73/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */ 74extern const char * const migratetype_names[MIGRATE_TYPES]; 75 76#ifdef CONFIG_CMA 77# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) 78# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA) 79# define is_migrate_cma_folio(folio, pfn) (MIGRATE_CMA == \ 80 get_pfnblock_flags_mask(&folio->page, pfn, MIGRATETYPE_MASK)) 81#else 82# define is_migrate_cma(migratetype) false 83# define is_migrate_cma_page(_page) false 84# define is_migrate_cma_folio(folio, pfn) false 85#endif 86 87static inline bool is_migrate_movable(int mt) 88{ 89 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE; 90} 91 92/* 93 * Check whether a migratetype can be merged with another migratetype. 94 * 95 * It is only mergeable when it can fall back to other migratetypes for 96 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c. 97 */ 98static inline bool migratetype_is_mergeable(int mt) 99{ 100 return mt < MIGRATE_PCPTYPES; 101} 102 103#define for_each_migratetype_order(order, type) \ 104 for (order = 0; order < NR_PAGE_ORDERS; order++) \ 105 for (type = 0; type < MIGRATE_TYPES; type++) 106 107extern int page_group_by_mobility_disabled; 108 109#define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1) 110 111#define get_pageblock_migratetype(page) \ 112 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK) 113 114#define folio_migratetype(folio) \ 115 get_pfnblock_flags_mask(&folio->page, folio_pfn(folio), \ 116 MIGRATETYPE_MASK) 117struct free_area { 118 struct list_head free_list[MIGRATE_TYPES]; 119 unsigned long nr_free; 120}; 121 122struct pglist_data; 123 124#ifdef CONFIG_NUMA 125enum numa_stat_item { 126 NUMA_HIT, /* allocated in intended node */ 127 NUMA_MISS, /* allocated in non intended node */ 128 NUMA_FOREIGN, /* was intended here, hit elsewhere */ 129 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ 130 NUMA_LOCAL, /* allocation from local node */ 131 NUMA_OTHER, /* allocation from other node */ 132 NR_VM_NUMA_EVENT_ITEMS 133}; 134#else 135#define NR_VM_NUMA_EVENT_ITEMS 0 136#endif 137 138enum zone_stat_item { 139 /* First 128 byte cacheline (assuming 64 bit words) */ 140 NR_FREE_PAGES, 141 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ 142 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, 143 NR_ZONE_ACTIVE_ANON, 144 NR_ZONE_INACTIVE_FILE, 145 NR_ZONE_ACTIVE_FILE, 146 NR_ZONE_UNEVICTABLE, 147 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */ 148 NR_MLOCK, /* mlock()ed pages found and moved off LRU */ 149 /* Second 128 byte cacheline */ 150 NR_BOUNCE, 151#if IS_ENABLED(CONFIG_ZSMALLOC) 152 NR_ZSPAGES, /* allocated in zsmalloc */ 153#endif 154 NR_FREE_CMA_PAGES, 155#ifdef CONFIG_UNACCEPTED_MEMORY 156 NR_UNACCEPTED, 157#endif 158 NR_VM_ZONE_STAT_ITEMS }; 159 160enum node_stat_item { 161 NR_LRU_BASE, 162 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ 163 NR_ACTIVE_ANON, /* " " " " " */ 164 NR_INACTIVE_FILE, /* " " " " " */ 165 NR_ACTIVE_FILE, /* " " " " " */ 166 NR_UNEVICTABLE, /* " " " " " */ 167 NR_SLAB_RECLAIMABLE_B, 168 NR_SLAB_UNRECLAIMABLE_B, 169 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ 170 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ 171 WORKINGSET_NODES, 172 WORKINGSET_REFAULT_BASE, 173 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE, 174 WORKINGSET_REFAULT_FILE, 175 WORKINGSET_ACTIVATE_BASE, 176 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE, 177 WORKINGSET_ACTIVATE_FILE, 178 WORKINGSET_RESTORE_BASE, 179 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE, 180 WORKINGSET_RESTORE_FILE, 181 WORKINGSET_NODERECLAIM, 182 NR_ANON_MAPPED, /* Mapped anonymous pages */ 183 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. 184 only modified from process context */ 185 NR_FILE_PAGES, 186 NR_FILE_DIRTY, 187 NR_WRITEBACK, 188 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ 189 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ 190 NR_SHMEM_THPS, 191 NR_SHMEM_PMDMAPPED, 192 NR_FILE_THPS, 193 NR_FILE_PMDMAPPED, 194 NR_ANON_THPS, 195 NR_VMSCAN_WRITE, 196 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ 197 NR_DIRTIED, /* page dirtyings since bootup */ 198 NR_WRITTEN, /* page writings since bootup */ 199 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */ 200 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */ 201 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */ 202 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */ 203 NR_KERNEL_STACK_KB, /* measured in KiB */ 204#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK) 205 NR_KERNEL_SCS_KB, /* measured in KiB */ 206#endif 207 NR_PAGETABLE, /* used for pagetables */ 208 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */ 209#ifdef CONFIG_SWAP 210 NR_SWAPCACHE, 211#endif 212#ifdef CONFIG_NUMA_BALANCING 213 PGPROMOTE_SUCCESS, /* promote successfully */ 214 PGPROMOTE_CANDIDATE, /* candidate pages to promote */ 215#endif 216 /* PGDEMOTE_*: pages demoted */ 217 PGDEMOTE_KSWAPD, 218 PGDEMOTE_DIRECT, 219 PGDEMOTE_KHUGEPAGED, 220 NR_VM_NODE_STAT_ITEMS 221}; 222 223/* 224 * Returns true if the item should be printed in THPs (/proc/vmstat 225 * currently prints number of anon, file and shmem THPs. But the item 226 * is charged in pages). 227 */ 228static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item) 229{ 230 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) 231 return false; 232 233 return item == NR_ANON_THPS || 234 item == NR_FILE_THPS || 235 item == NR_SHMEM_THPS || 236 item == NR_SHMEM_PMDMAPPED || 237 item == NR_FILE_PMDMAPPED; 238} 239 240/* 241 * Returns true if the value is measured in bytes (most vmstat values are 242 * measured in pages). This defines the API part, the internal representation 243 * might be different. 244 */ 245static __always_inline bool vmstat_item_in_bytes(int idx) 246{ 247 /* 248 * Global and per-node slab counters track slab pages. 249 * It's expected that changes are multiples of PAGE_SIZE. 250 * Internally values are stored in pages. 251 * 252 * Per-memcg and per-lruvec counters track memory, consumed 253 * by individual slab objects. These counters are actually 254 * byte-precise. 255 */ 256 return (idx == NR_SLAB_RECLAIMABLE_B || 257 idx == NR_SLAB_UNRECLAIMABLE_B); 258} 259 260/* 261 * We do arithmetic on the LRU lists in various places in the code, 262 * so it is important to keep the active lists LRU_ACTIVE higher in 263 * the array than the corresponding inactive lists, and to keep 264 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. 265 * 266 * This has to be kept in sync with the statistics in zone_stat_item 267 * above and the descriptions in vmstat_text in mm/vmstat.c 268 */ 269#define LRU_BASE 0 270#define LRU_ACTIVE 1 271#define LRU_FILE 2 272 273enum lru_list { 274 LRU_INACTIVE_ANON = LRU_BASE, 275 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, 276 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, 277 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, 278 LRU_UNEVICTABLE, 279 NR_LRU_LISTS 280}; 281 282enum vmscan_throttle_state { 283 VMSCAN_THROTTLE_WRITEBACK, 284 VMSCAN_THROTTLE_ISOLATED, 285 VMSCAN_THROTTLE_NOPROGRESS, 286 VMSCAN_THROTTLE_CONGESTED, 287 NR_VMSCAN_THROTTLE, 288}; 289 290#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) 291 292#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) 293 294static inline bool is_file_lru(enum lru_list lru) 295{ 296 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); 297} 298 299static inline bool is_active_lru(enum lru_list lru) 300{ 301 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); 302} 303 304#define WORKINGSET_ANON 0 305#define WORKINGSET_FILE 1 306#define ANON_AND_FILE 2 307 308enum lruvec_flags { 309 /* 310 * An lruvec has many dirty pages backed by a congested BDI: 311 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim. 312 * It can be cleared by cgroup reclaim or kswapd. 313 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim. 314 * It can only be cleared by kswapd. 315 * 316 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup 317 * reclaim, but not vice versa. This only applies to the root cgroup. 318 * The goal is to prevent cgroup reclaim on the root cgroup (e.g. 319 * memory.reclaim) to unthrottle an unbalanced node (that was throttled 320 * by kswapd). 321 */ 322 LRUVEC_CGROUP_CONGESTED, 323 LRUVEC_NODE_CONGESTED, 324}; 325 326#endif /* !__GENERATING_BOUNDS_H */ 327 328/* 329 * Evictable pages are divided into multiple generations. The youngest and the 330 * oldest generation numbers, max_seq and min_seq, are monotonically increasing. 331 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An 332 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the 333 * corresponding generation. The gen counter in folio->flags stores gen+1 while 334 * a page is on one of lrugen->folios[]. Otherwise it stores 0. 335 * 336 * A page is added to the youngest generation on faulting. The aging needs to 337 * check the accessed bit at least twice before handing this page over to the 338 * eviction. The first check takes care of the accessed bit set on the initial 339 * fault; the second check makes sure this page hasn't been used since then. 340 * This process, AKA second chance, requires a minimum of two generations, 341 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive 342 * LRU, e.g., /proc/vmstat, these two generations are considered active; the 343 * rest of generations, if they exist, are considered inactive. See 344 * lru_gen_is_active(). 345 * 346 * PG_active is always cleared while a page is on one of lrugen->folios[] so 347 * that the aging needs not to worry about it. And it's set again when a page 348 * considered active is isolated for non-reclaiming purposes, e.g., migration. 349 * See lru_gen_add_folio() and lru_gen_del_folio(). 350 * 351 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the 352 * number of categories of the active/inactive LRU when keeping track of 353 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits 354 * in folio->flags. 355 */ 356#define MIN_NR_GENS 2U 357#define MAX_NR_GENS 4U 358 359/* 360 * Each generation is divided into multiple tiers. A page accessed N times 361 * through file descriptors is in tier order_base_2(N). A page in the first tier 362 * (N=0,1) is marked by PG_referenced unless it was faulted in through page 363 * tables or read ahead. A page in any other tier (N>1) is marked by 364 * PG_referenced and PG_workingset. This implies a minimum of two tiers is 365 * supported without using additional bits in folio->flags. 366 * 367 * In contrast to moving across generations which requires the LRU lock, moving 368 * across tiers only involves atomic operations on folio->flags and therefore 369 * has a negligible cost in the buffered access path. In the eviction path, 370 * comparisons of refaulted/(evicted+protected) from the first tier and the 371 * rest infer whether pages accessed multiple times through file descriptors 372 * are statistically hot and thus worth protecting. 373 * 374 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the 375 * number of categories of the active/inactive LRU when keeping track of 376 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in 377 * folio->flags. 378 */ 379#define MAX_NR_TIERS 4U 380 381#ifndef __GENERATING_BOUNDS_H 382 383struct lruvec; 384struct page_vma_mapped_walk; 385 386#define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF) 387#define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF) 388 389#ifdef CONFIG_LRU_GEN 390 391enum { 392 LRU_GEN_ANON, 393 LRU_GEN_FILE, 394}; 395 396enum { 397 LRU_GEN_CORE, 398 LRU_GEN_MM_WALK, 399 LRU_GEN_NONLEAF_YOUNG, 400 NR_LRU_GEN_CAPS 401}; 402 403#define MIN_LRU_BATCH BITS_PER_LONG 404#define MAX_LRU_BATCH (MIN_LRU_BATCH * 64) 405 406/* whether to keep historical stats from evicted generations */ 407#ifdef CONFIG_LRU_GEN_STATS 408#define NR_HIST_GENS MAX_NR_GENS 409#else 410#define NR_HIST_GENS 1U 411#endif 412 413/* 414 * The youngest generation number is stored in max_seq for both anon and file 415 * types as they are aged on an equal footing. The oldest generation numbers are 416 * stored in min_seq[] separately for anon and file types as clean file pages 417 * can be evicted regardless of swap constraints. 418 * 419 * Normally anon and file min_seq are in sync. But if swapping is constrained, 420 * e.g., out of swap space, file min_seq is allowed to advance and leave anon 421 * min_seq behind. 422 * 423 * The number of pages in each generation is eventually consistent and therefore 424 * can be transiently negative when reset_batch_size() is pending. 425 */ 426struct lru_gen_folio { 427 /* the aging increments the youngest generation number */ 428 unsigned long max_seq; 429 /* the eviction increments the oldest generation numbers */ 430 unsigned long min_seq[ANON_AND_FILE]; 431 /* the birth time of each generation in jiffies */ 432 unsigned long timestamps[MAX_NR_GENS]; 433 /* the multi-gen LRU lists, lazily sorted on eviction */ 434 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 435 /* the multi-gen LRU sizes, eventually consistent */ 436 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 437 /* the exponential moving average of refaulted */ 438 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS]; 439 /* the exponential moving average of evicted+protected */ 440 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS]; 441 /* the first tier doesn't need protection, hence the minus one */ 442 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1]; 443 /* can be modified without holding the LRU lock */ 444 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 445 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 446 /* whether the multi-gen LRU is enabled */ 447 bool enabled; 448 /* the memcg generation this lru_gen_folio belongs to */ 449 u8 gen; 450 /* the list segment this lru_gen_folio belongs to */ 451 u8 seg; 452 /* per-node lru_gen_folio list for global reclaim */ 453 struct hlist_nulls_node list; 454}; 455 456enum { 457 MM_LEAF_TOTAL, /* total leaf entries */ 458 MM_LEAF_OLD, /* old leaf entries */ 459 MM_LEAF_YOUNG, /* young leaf entries */ 460 MM_NONLEAF_TOTAL, /* total non-leaf entries */ 461 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */ 462 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */ 463 NR_MM_STATS 464}; 465 466/* double-buffering Bloom filters */ 467#define NR_BLOOM_FILTERS 2 468 469struct lru_gen_mm_state { 470 /* synced with max_seq after each iteration */ 471 unsigned long seq; 472 /* where the current iteration continues after */ 473 struct list_head *head; 474 /* where the last iteration ended before */ 475 struct list_head *tail; 476 /* Bloom filters flip after each iteration */ 477 unsigned long *filters[NR_BLOOM_FILTERS]; 478 /* the mm stats for debugging */ 479 unsigned long stats[NR_HIST_GENS][NR_MM_STATS]; 480}; 481 482struct lru_gen_mm_walk { 483 /* the lruvec under reclaim */ 484 struct lruvec *lruvec; 485 /* max_seq from lru_gen_folio: can be out of date */ 486 unsigned long seq; 487 /* the next address within an mm to scan */ 488 unsigned long next_addr; 489 /* to batch promoted pages */ 490 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 491 /* to batch the mm stats */ 492 int mm_stats[NR_MM_STATS]; 493 /* total batched items */ 494 int batched; 495 bool can_swap; 496 bool force_scan; 497}; 498 499/* 500 * For each node, memcgs are divided into two generations: the old and the 501 * young. For each generation, memcgs are randomly sharded into multiple bins 502 * to improve scalability. For each bin, the hlist_nulls is virtually divided 503 * into three segments: the head, the tail and the default. 504 * 505 * An onlining memcg is added to the tail of a random bin in the old generation. 506 * The eviction starts at the head of a random bin in the old generation. The 507 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes 508 * the old generation, is incremented when all its bins become empty. 509 * 510 * There are four operations: 511 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its 512 * current generation (old or young) and updates its "seg" to "head"; 513 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its 514 * current generation (old or young) and updates its "seg" to "tail"; 515 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old 516 * generation, updates its "gen" to "old" and resets its "seg" to "default"; 517 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the 518 * young generation, updates its "gen" to "young" and resets its "seg" to 519 * "default". 520 * 521 * The events that trigger the above operations are: 522 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD; 523 * 2. The first attempt to reclaim a memcg below low, which triggers 524 * MEMCG_LRU_TAIL; 525 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size 526 * threshold, which triggers MEMCG_LRU_TAIL; 527 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size 528 * threshold, which triggers MEMCG_LRU_YOUNG; 529 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG; 530 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG; 531 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD. 532 * 533 * Notes: 534 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing 535 * of their max_seq counters ensures the eventual fairness to all eligible 536 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter(). 537 * 2. There are only two valid generations: old (seq) and young (seq+1). 538 * MEMCG_NR_GENS is set to three so that when reading the generation counter 539 * locklessly, a stale value (seq-1) does not wraparound to young. 540 */ 541#define MEMCG_NR_GENS 3 542#define MEMCG_NR_BINS 8 543 544struct lru_gen_memcg { 545 /* the per-node memcg generation counter */ 546 unsigned long seq; 547 /* each memcg has one lru_gen_folio per node */ 548 unsigned long nr_memcgs[MEMCG_NR_GENS]; 549 /* per-node lru_gen_folio list for global reclaim */ 550 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS]; 551 /* protects the above */ 552 spinlock_t lock; 553}; 554 555void lru_gen_init_pgdat(struct pglist_data *pgdat); 556void lru_gen_init_lruvec(struct lruvec *lruvec); 557void lru_gen_look_around(struct page_vma_mapped_walk *pvmw); 558 559void lru_gen_init_memcg(struct mem_cgroup *memcg); 560void lru_gen_exit_memcg(struct mem_cgroup *memcg); 561void lru_gen_online_memcg(struct mem_cgroup *memcg); 562void lru_gen_offline_memcg(struct mem_cgroup *memcg); 563void lru_gen_release_memcg(struct mem_cgroup *memcg); 564void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid); 565 566#else /* !CONFIG_LRU_GEN */ 567 568static inline void lru_gen_init_pgdat(struct pglist_data *pgdat) 569{ 570} 571 572static inline void lru_gen_init_lruvec(struct lruvec *lruvec) 573{ 574} 575 576static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw) 577{ 578} 579 580static inline void lru_gen_init_memcg(struct mem_cgroup *memcg) 581{ 582} 583 584static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg) 585{ 586} 587 588static inline void lru_gen_online_memcg(struct mem_cgroup *memcg) 589{ 590} 591 592static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg) 593{ 594} 595 596static inline void lru_gen_release_memcg(struct mem_cgroup *memcg) 597{ 598} 599 600static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid) 601{ 602} 603 604#endif /* CONFIG_LRU_GEN */ 605 606struct lruvec { 607 struct list_head lists[NR_LRU_LISTS]; 608 /* per lruvec lru_lock for memcg */ 609 spinlock_t lru_lock; 610 /* 611 * These track the cost of reclaiming one LRU - file or anon - 612 * over the other. As the observed cost of reclaiming one LRU 613 * increases, the reclaim scan balance tips toward the other. 614 */ 615 unsigned long anon_cost; 616 unsigned long file_cost; 617 /* Non-resident age, driven by LRU movement */ 618 atomic_long_t nonresident_age; 619 /* Refaults at the time of last reclaim cycle */ 620 unsigned long refaults[ANON_AND_FILE]; 621 /* Various lruvec state flags (enum lruvec_flags) */ 622 unsigned long flags; 623#ifdef CONFIG_LRU_GEN 624 /* evictable pages divided into generations */ 625 struct lru_gen_folio lrugen; 626#ifdef CONFIG_LRU_GEN_WALKS_MMU 627 /* to concurrently iterate lru_gen_mm_list */ 628 struct lru_gen_mm_state mm_state; 629#endif 630#endif /* CONFIG_LRU_GEN */ 631#ifdef CONFIG_MEMCG 632 struct pglist_data *pgdat; 633#endif 634 struct zswap_lruvec_state zswap_lruvec_state; 635}; 636 637/* Isolate for asynchronous migration */ 638#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) 639/* Isolate unevictable pages */ 640#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) 641 642/* LRU Isolation modes. */ 643typedef unsigned __bitwise isolate_mode_t; 644 645enum zone_watermarks { 646 WMARK_MIN, 647 WMARK_LOW, 648 WMARK_HIGH, 649 WMARK_PROMO, 650 NR_WMARK 651}; 652 653/* 654 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list 655 * for THP which will usually be GFP_MOVABLE. Even if it is another type, 656 * it should not contribute to serious fragmentation causing THP allocation 657 * failures. 658 */ 659#ifdef CONFIG_TRANSPARENT_HUGEPAGE 660#define NR_PCP_THP 1 661#else 662#define NR_PCP_THP 0 663#endif 664#define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1)) 665#define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP) 666 667#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost) 668#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost) 669#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost) 670#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost) 671 672/* 673 * Flags used in pcp->flags field. 674 * 675 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the 676 * previous page freeing. To avoid to drain PCP for an accident 677 * high-order page freeing. 678 * 679 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before 680 * draining PCP for consecutive high-order pages freeing without 681 * allocation if data cache slice of CPU is large enough. To reduce 682 * zone lock contention and keep cache-hot pages reusing. 683 */ 684#define PCPF_PREV_FREE_HIGH_ORDER BIT(0) 685#define PCPF_FREE_HIGH_BATCH BIT(1) 686 687struct per_cpu_pages { 688 spinlock_t lock; /* Protects lists field */ 689 int count; /* number of pages in the list */ 690 int high; /* high watermark, emptying needed */ 691 int high_min; /* min high watermark */ 692 int high_max; /* max high watermark */ 693 int batch; /* chunk size for buddy add/remove */ 694 u8 flags; /* protected by pcp->lock */ 695 u8 alloc_factor; /* batch scaling factor during allocate */ 696#ifdef CONFIG_NUMA 697 u8 expire; /* When 0, remote pagesets are drained */ 698#endif 699 short free_count; /* consecutive free count */ 700 701 /* Lists of pages, one per migrate type stored on the pcp-lists */ 702 struct list_head lists[NR_PCP_LISTS]; 703} ____cacheline_aligned_in_smp; 704 705struct per_cpu_zonestat { 706#ifdef CONFIG_SMP 707 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; 708 s8 stat_threshold; 709#endif 710#ifdef CONFIG_NUMA 711 /* 712 * Low priority inaccurate counters that are only folded 713 * on demand. Use a large type to avoid the overhead of 714 * folding during refresh_cpu_vm_stats. 715 */ 716 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; 717#endif 718}; 719 720struct per_cpu_nodestat { 721 s8 stat_threshold; 722 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS]; 723}; 724 725#endif /* !__GENERATING_BOUNDS.H */ 726 727enum zone_type { 728 /* 729 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able 730 * to DMA to all of the addressable memory (ZONE_NORMAL). 731 * On architectures where this area covers the whole 32 bit address 732 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller 733 * DMA addressing constraints. This distinction is important as a 32bit 734 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit 735 * platforms may need both zones as they support peripherals with 736 * different DMA addressing limitations. 737 */ 738#ifdef CONFIG_ZONE_DMA 739 ZONE_DMA, 740#endif 741#ifdef CONFIG_ZONE_DMA32 742 ZONE_DMA32, 743#endif 744 /* 745 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be 746 * performed on pages in ZONE_NORMAL if the DMA devices support 747 * transfers to all addressable memory. 748 */ 749 ZONE_NORMAL, 750#ifdef CONFIG_HIGHMEM 751 /* 752 * A memory area that is only addressable by the kernel through 753 * mapping portions into its own address space. This is for example 754 * used by i386 to allow the kernel to address the memory beyond 755 * 900MB. The kernel will set up special mappings (page 756 * table entries on i386) for each page that the kernel needs to 757 * access. 758 */ 759 ZONE_HIGHMEM, 760#endif 761 /* 762 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains 763 * movable pages with few exceptional cases described below. Main use 764 * cases for ZONE_MOVABLE are to make memory offlining/unplug more 765 * likely to succeed, and to locally limit unmovable allocations - e.g., 766 * to increase the number of THP/huge pages. Notable special cases are: 767 * 768 * 1. Pinned pages: (long-term) pinning of movable pages might 769 * essentially turn such pages unmovable. Therefore, we do not allow 770 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and 771 * faulted, they come from the right zone right away. However, it is 772 * still possible that address space already has pages in 773 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has 774 * touches that memory before pinning). In such case we migrate them 775 * to a different zone. When migration fails - pinning fails. 776 * 2. memblock allocations: kernelcore/movablecore setups might create 777 * situations where ZONE_MOVABLE contains unmovable allocations 778 * after boot. Memory offlining and allocations fail early. 779 * 3. Memory holes: kernelcore/movablecore setups might create very rare 780 * situations where ZONE_MOVABLE contains memory holes after boot, 781 * for example, if we have sections that are only partially 782 * populated. Memory offlining and allocations fail early. 783 * 4. PG_hwpoison pages: while poisoned pages can be skipped during 784 * memory offlining, such pages cannot be allocated. 785 * 5. Unmovable PG_offline pages: in paravirtualized environments, 786 * hotplugged memory blocks might only partially be managed by the 787 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The 788 * parts not manged by the buddy are unmovable PG_offline pages. In 789 * some cases (virtio-mem), such pages can be skipped during 790 * memory offlining, however, cannot be moved/allocated. These 791 * techniques might use alloc_contig_range() to hide previously 792 * exposed pages from the buddy again (e.g., to implement some sort 793 * of memory unplug in virtio-mem). 794 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create 795 * situations where ZERO_PAGE(0) which is allocated differently 796 * on different platforms may end up in a movable zone. ZERO_PAGE(0) 797 * cannot be migrated. 798 * 7. Memory-hotplug: when using memmap_on_memory and onlining the 799 * memory to the MOVABLE zone, the vmemmap pages are also placed in 800 * such zone. Such pages cannot be really moved around as they are 801 * self-stored in the range, but they are treated as movable when 802 * the range they describe is about to be offlined. 803 * 804 * In general, no unmovable allocations that degrade memory offlining 805 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range()) 806 * have to expect that migrating pages in ZONE_MOVABLE can fail (even 807 * if has_unmovable_pages() states that there are no unmovable pages, 808 * there can be false negatives). 809 */ 810 ZONE_MOVABLE, 811#ifdef CONFIG_ZONE_DEVICE 812 ZONE_DEVICE, 813#endif 814 __MAX_NR_ZONES 815 816}; 817 818#ifndef __GENERATING_BOUNDS_H 819 820#define ASYNC_AND_SYNC 2 821 822struct zone { 823 /* Read-mostly fields */ 824 825 /* zone watermarks, access with *_wmark_pages(zone) macros */ 826 unsigned long _watermark[NR_WMARK]; 827 unsigned long watermark_boost; 828 829 unsigned long nr_reserved_highatomic; 830 831 /* 832 * We don't know if the memory that we're going to allocate will be 833 * freeable or/and it will be released eventually, so to avoid totally 834 * wasting several GB of ram we must reserve some of the lower zone 835 * memory (otherwise we risk to run OOM on the lower zones despite 836 * there being tons of freeable ram on the higher zones). This array is 837 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl 838 * changes. 839 */ 840 long lowmem_reserve[MAX_NR_ZONES]; 841 842#ifdef CONFIG_NUMA 843 int node; 844#endif 845 struct pglist_data *zone_pgdat; 846 struct per_cpu_pages __percpu *per_cpu_pageset; 847 struct per_cpu_zonestat __percpu *per_cpu_zonestats; 848 /* 849 * the high and batch values are copied to individual pagesets for 850 * faster access 851 */ 852 int pageset_high_min; 853 int pageset_high_max; 854 int pageset_batch; 855 856#ifndef CONFIG_SPARSEMEM 857 /* 858 * Flags for a pageblock_nr_pages block. See pageblock-flags.h. 859 * In SPARSEMEM, this map is stored in struct mem_section 860 */ 861 unsigned long *pageblock_flags; 862#endif /* CONFIG_SPARSEMEM */ 863 864 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ 865 unsigned long zone_start_pfn; 866 867 /* 868 * spanned_pages is the total pages spanned by the zone, including 869 * holes, which is calculated as: 870 * spanned_pages = zone_end_pfn - zone_start_pfn; 871 * 872 * present_pages is physical pages existing within the zone, which 873 * is calculated as: 874 * present_pages = spanned_pages - absent_pages(pages in holes); 875 * 876 * present_early_pages is present pages existing within the zone 877 * located on memory available since early boot, excluding hotplugged 878 * memory. 879 * 880 * managed_pages is present pages managed by the buddy system, which 881 * is calculated as (reserved_pages includes pages allocated by the 882 * bootmem allocator): 883 * managed_pages = present_pages - reserved_pages; 884 * 885 * cma pages is present pages that are assigned for CMA use 886 * (MIGRATE_CMA). 887 * 888 * So present_pages may be used by memory hotplug or memory power 889 * management logic to figure out unmanaged pages by checking 890 * (present_pages - managed_pages). And managed_pages should be used 891 * by page allocator and vm scanner to calculate all kinds of watermarks 892 * and thresholds. 893 * 894 * Locking rules: 895 * 896 * zone_start_pfn and spanned_pages are protected by span_seqlock. 897 * It is a seqlock because it has to be read outside of zone->lock, 898 * and it is done in the main allocator path. But, it is written 899 * quite infrequently. 900 * 901 * The span_seq lock is declared along with zone->lock because it is 902 * frequently read in proximity to zone->lock. It's good to 903 * give them a chance of being in the same cacheline. 904 * 905 * Write access to present_pages at runtime should be protected by 906 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of 907 * present_pages should use get_online_mems() to get a stable value. 908 */ 909 atomic_long_t managed_pages; 910 unsigned long spanned_pages; 911 unsigned long present_pages; 912#if defined(CONFIG_MEMORY_HOTPLUG) 913 unsigned long present_early_pages; 914#endif 915#ifdef CONFIG_CMA 916 unsigned long cma_pages; 917#endif 918 919 const char *name; 920 921#ifdef CONFIG_MEMORY_ISOLATION 922 /* 923 * Number of isolated pageblock. It is used to solve incorrect 924 * freepage counting problem due to racy retrieving migratetype 925 * of pageblock. Protected by zone->lock. 926 */ 927 unsigned long nr_isolate_pageblock; 928#endif 929 930#ifdef CONFIG_MEMORY_HOTPLUG 931 /* see spanned/present_pages for more description */ 932 seqlock_t span_seqlock; 933#endif 934 935 int initialized; 936 937 /* Write-intensive fields used from the page allocator */ 938 CACHELINE_PADDING(_pad1_); 939 940 /* free areas of different sizes */ 941 struct free_area free_area[NR_PAGE_ORDERS]; 942 943#ifdef CONFIG_UNACCEPTED_MEMORY 944 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */ 945 struct list_head unaccepted_pages; 946#endif 947 948 /* zone flags, see below */ 949 unsigned long flags; 950 951 /* Primarily protects free_area */ 952 spinlock_t lock; 953 954 /* Write-intensive fields used by compaction and vmstats. */ 955 CACHELINE_PADDING(_pad2_); 956 957 /* 958 * When free pages are below this point, additional steps are taken 959 * when reading the number of free pages to avoid per-cpu counter 960 * drift allowing watermarks to be breached 961 */ 962 unsigned long percpu_drift_mark; 963 964#if defined CONFIG_COMPACTION || defined CONFIG_CMA 965 /* pfn where compaction free scanner should start */ 966 unsigned long compact_cached_free_pfn; 967 /* pfn where compaction migration scanner should start */ 968 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC]; 969 unsigned long compact_init_migrate_pfn; 970 unsigned long compact_init_free_pfn; 971#endif 972 973#ifdef CONFIG_COMPACTION 974 /* 975 * On compaction failure, 1<<compact_defer_shift compactions 976 * are skipped before trying again. The number attempted since 977 * last failure is tracked with compact_considered. 978 * compact_order_failed is the minimum compaction failed order. 979 */ 980 unsigned int compact_considered; 981 unsigned int compact_defer_shift; 982 int compact_order_failed; 983#endif 984 985#if defined CONFIG_COMPACTION || defined CONFIG_CMA 986 /* Set to true when the PG_migrate_skip bits should be cleared */ 987 bool compact_blockskip_flush; 988#endif 989 990 bool contiguous; 991 992 CACHELINE_PADDING(_pad3_); 993 /* Zone statistics */ 994 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; 995 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; 996} ____cacheline_internodealigned_in_smp; 997 998enum pgdat_flags { 999 PGDAT_DIRTY, /* reclaim scanning has recently found 1000 * many dirty file pages at the tail 1001 * of the LRU. 1002 */ 1003 PGDAT_WRITEBACK, /* reclaim scanning has recently found 1004 * many pages under writeback 1005 */ 1006 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */ 1007}; 1008 1009enum zone_flags { 1010 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks. 1011 * Cleared when kswapd is woken. 1012 */ 1013 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */ 1014 ZONE_BELOW_HIGH, /* zone is below high watermark. */ 1015}; 1016 1017static inline unsigned long zone_managed_pages(struct zone *zone) 1018{ 1019 return (unsigned long)atomic_long_read(&zone->managed_pages); 1020} 1021 1022static inline unsigned long zone_cma_pages(struct zone *zone) 1023{ 1024#ifdef CONFIG_CMA 1025 return zone->cma_pages; 1026#else 1027 return 0; 1028#endif 1029} 1030 1031static inline unsigned long zone_end_pfn(const struct zone *zone) 1032{ 1033 return zone->zone_start_pfn + zone->spanned_pages; 1034} 1035 1036static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) 1037{ 1038 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); 1039} 1040 1041static inline bool zone_is_initialized(struct zone *zone) 1042{ 1043 return zone->initialized; 1044} 1045 1046static inline bool zone_is_empty(struct zone *zone) 1047{ 1048 return zone->spanned_pages == 0; 1049} 1050 1051#ifndef BUILD_VDSO32_64 1052/* 1053 * The zone field is never updated after free_area_init_core() 1054 * sets it, so none of the operations on it need to be atomic. 1055 */ 1056 1057/* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */ 1058#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH) 1059#define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH) 1060#define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH) 1061#define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH) 1062#define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH) 1063#define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH) 1064#define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH) 1065 1066/* 1067 * Define the bit shifts to access each section. For non-existent 1068 * sections we define the shift as 0; that plus a 0 mask ensures 1069 * the compiler will optimise away reference to them. 1070 */ 1071#define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0)) 1072#define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0)) 1073#define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0)) 1074#define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0)) 1075#define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0)) 1076 1077/* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */ 1078#ifdef NODE_NOT_IN_PAGE_FLAGS 1079#define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT) 1080#define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \ 1081 SECTIONS_PGOFF : ZONES_PGOFF) 1082#else 1083#define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT) 1084#define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \ 1085 NODES_PGOFF : ZONES_PGOFF) 1086#endif 1087 1088#define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0)) 1089 1090#define ZONES_MASK ((1UL << ZONES_WIDTH) - 1) 1091#define NODES_MASK ((1UL << NODES_WIDTH) - 1) 1092#define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1) 1093#define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1) 1094#define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1) 1095#define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1) 1096 1097static inline enum zone_type page_zonenum(const struct page *page) 1098{ 1099 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT); 1100 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK; 1101} 1102 1103static inline enum zone_type folio_zonenum(const struct folio *folio) 1104{ 1105 return page_zonenum(&folio->page); 1106} 1107 1108#ifdef CONFIG_ZONE_DEVICE 1109static inline bool is_zone_device_page(const struct page *page) 1110{ 1111 return page_zonenum(page) == ZONE_DEVICE; 1112} 1113 1114/* 1115 * Consecutive zone device pages should not be merged into the same sgl 1116 * or bvec segment with other types of pages or if they belong to different 1117 * pgmaps. Otherwise getting the pgmap of a given segment is not possible 1118 * without scanning the entire segment. This helper returns true either if 1119 * both pages are not zone device pages or both pages are zone device pages 1120 * with the same pgmap. 1121 */ 1122static inline bool zone_device_pages_have_same_pgmap(const struct page *a, 1123 const struct page *b) 1124{ 1125 if (is_zone_device_page(a) != is_zone_device_page(b)) 1126 return false; 1127 if (!is_zone_device_page(a)) 1128 return true; 1129 return a->pgmap == b->pgmap; 1130} 1131 1132extern void memmap_init_zone_device(struct zone *, unsigned long, 1133 unsigned long, struct dev_pagemap *); 1134#else 1135static inline bool is_zone_device_page(const struct page *page) 1136{ 1137 return false; 1138} 1139static inline bool zone_device_pages_have_same_pgmap(const struct page *a, 1140 const struct page *b) 1141{ 1142 return true; 1143} 1144#endif 1145 1146static inline bool folio_is_zone_device(const struct folio *folio) 1147{ 1148 return is_zone_device_page(&folio->page); 1149} 1150 1151static inline bool is_zone_movable_page(const struct page *page) 1152{ 1153 return page_zonenum(page) == ZONE_MOVABLE; 1154} 1155 1156static inline bool folio_is_zone_movable(const struct folio *folio) 1157{ 1158 return folio_zonenum(folio) == ZONE_MOVABLE; 1159} 1160#endif 1161 1162/* 1163 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty 1164 * intersection with the given zone 1165 */ 1166static inline bool zone_intersects(struct zone *zone, 1167 unsigned long start_pfn, unsigned long nr_pages) 1168{ 1169 if (zone_is_empty(zone)) 1170 return false; 1171 if (start_pfn >= zone_end_pfn(zone) || 1172 start_pfn + nr_pages <= zone->zone_start_pfn) 1173 return false; 1174 1175 return true; 1176} 1177 1178/* 1179 * The "priority" of VM scanning is how much of the queues we will scan in one 1180 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the 1181 * queues ("queue_length >> 12") during an aging round. 1182 */ 1183#define DEF_PRIORITY 12 1184 1185/* Maximum number of zones on a zonelist */ 1186#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) 1187 1188enum { 1189 ZONELIST_FALLBACK, /* zonelist with fallback */ 1190#ifdef CONFIG_NUMA 1191 /* 1192 * The NUMA zonelists are doubled because we need zonelists that 1193 * restrict the allocations to a single node for __GFP_THISNODE. 1194 */ 1195 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */ 1196#endif 1197 MAX_ZONELISTS 1198}; 1199 1200/* 1201 * This struct contains information about a zone in a zonelist. It is stored 1202 * here to avoid dereferences into large structures and lookups of tables 1203 */ 1204struct zoneref { 1205 struct zone *zone; /* Pointer to actual zone */ 1206 int zone_idx; /* zone_idx(zoneref->zone) */ 1207}; 1208 1209/* 1210 * One allocation request operates on a zonelist. A zonelist 1211 * is a list of zones, the first one is the 'goal' of the 1212 * allocation, the other zones are fallback zones, in decreasing 1213 * priority. 1214 * 1215 * To speed the reading of the zonelist, the zonerefs contain the zone index 1216 * of the entry being read. Helper functions to access information given 1217 * a struct zoneref are 1218 * 1219 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs 1220 * zonelist_zone_idx() - Return the index of the zone for an entry 1221 * zonelist_node_idx() - Return the index of the node for an entry 1222 */ 1223struct zonelist { 1224 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; 1225}; 1226 1227/* 1228 * The array of struct pages for flatmem. 1229 * It must be declared for SPARSEMEM as well because there are configurations 1230 * that rely on that. 1231 */ 1232extern struct page *mem_map; 1233 1234#ifdef CONFIG_TRANSPARENT_HUGEPAGE 1235struct deferred_split { 1236 spinlock_t split_queue_lock; 1237 struct list_head split_queue; 1238 unsigned long split_queue_len; 1239}; 1240#endif 1241 1242#ifdef CONFIG_MEMORY_FAILURE 1243/* 1244 * Per NUMA node memory failure handling statistics. 1245 */ 1246struct memory_failure_stats { 1247 /* 1248 * Number of raw pages poisoned. 1249 * Cases not accounted: memory outside kernel control, offline page, 1250 * arch-specific memory_failure (SGX), hwpoison_filter() filtered 1251 * error events, and unpoison actions from hwpoison_unpoison. 1252 */ 1253 unsigned long total; 1254 /* 1255 * Recovery results of poisoned raw pages handled by memory_failure, 1256 * in sync with mf_result. 1257 * total = ignored + failed + delayed + recovered. 1258 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted. 1259 */ 1260 unsigned long ignored; 1261 unsigned long failed; 1262 unsigned long delayed; 1263 unsigned long recovered; 1264}; 1265#endif 1266 1267/* 1268 * On NUMA machines, each NUMA node would have a pg_data_t to describe 1269 * it's memory layout. On UMA machines there is a single pglist_data which 1270 * describes the whole memory. 1271 * 1272 * Memory statistics and page replacement data structures are maintained on a 1273 * per-zone basis. 1274 */ 1275typedef struct pglist_data { 1276 /* 1277 * node_zones contains just the zones for THIS node. Not all of the 1278 * zones may be populated, but it is the full list. It is referenced by 1279 * this node's node_zonelists as well as other node's node_zonelists. 1280 */ 1281 struct zone node_zones[MAX_NR_ZONES]; 1282 1283 /* 1284 * node_zonelists contains references to all zones in all nodes. 1285 * Generally the first zones will be references to this node's 1286 * node_zones. 1287 */ 1288 struct zonelist node_zonelists[MAX_ZONELISTS]; 1289 1290 int nr_zones; /* number of populated zones in this node */ 1291#ifdef CONFIG_FLATMEM /* means !SPARSEMEM */ 1292 struct page *node_mem_map; 1293#ifdef CONFIG_PAGE_EXTENSION 1294 struct page_ext *node_page_ext; 1295#endif 1296#endif 1297#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT) 1298 /* 1299 * Must be held any time you expect node_start_pfn, 1300 * node_present_pages, node_spanned_pages or nr_zones to stay constant. 1301 * Also synchronizes pgdat->first_deferred_pfn during deferred page 1302 * init. 1303 * 1304 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to 1305 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG 1306 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT. 1307 * 1308 * Nests above zone->lock and zone->span_seqlock 1309 */ 1310 spinlock_t node_size_lock; 1311#endif 1312 unsigned long node_start_pfn; 1313 unsigned long node_present_pages; /* total number of physical pages */ 1314 unsigned long node_spanned_pages; /* total size of physical page 1315 range, including holes */ 1316 int node_id; 1317 wait_queue_head_t kswapd_wait; 1318 wait_queue_head_t pfmemalloc_wait; 1319 1320 /* workqueues for throttling reclaim for different reasons. */ 1321 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE]; 1322 1323 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */ 1324 unsigned long nr_reclaim_start; /* nr pages written while throttled 1325 * when throttling started. */ 1326#ifdef CONFIG_MEMORY_HOTPLUG 1327 struct mutex kswapd_lock; 1328#endif 1329 struct task_struct *kswapd; /* Protected by kswapd_lock */ 1330 int kswapd_order; 1331 enum zone_type kswapd_highest_zoneidx; 1332 1333 int kswapd_failures; /* Number of 'reclaimed == 0' runs */ 1334 1335#ifdef CONFIG_COMPACTION 1336 int kcompactd_max_order; 1337 enum zone_type kcompactd_highest_zoneidx; 1338 wait_queue_head_t kcompactd_wait; 1339 struct task_struct *kcompactd; 1340 bool proactive_compact_trigger; 1341#endif 1342 /* 1343 * This is a per-node reserve of pages that are not available 1344 * to userspace allocations. 1345 */ 1346 unsigned long totalreserve_pages; 1347 1348#ifdef CONFIG_NUMA 1349 /* 1350 * node reclaim becomes active if more unmapped pages exist. 1351 */ 1352 unsigned long min_unmapped_pages; 1353 unsigned long min_slab_pages; 1354#endif /* CONFIG_NUMA */ 1355 1356 /* Write-intensive fields used by page reclaim */ 1357 CACHELINE_PADDING(_pad1_); 1358 1359#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1360 /* 1361 * If memory initialisation on large machines is deferred then this 1362 * is the first PFN that needs to be initialised. 1363 */ 1364 unsigned long first_deferred_pfn; 1365#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1366 1367#ifdef CONFIG_TRANSPARENT_HUGEPAGE 1368 struct deferred_split deferred_split_queue; 1369#endif 1370 1371#ifdef CONFIG_NUMA_BALANCING 1372 /* start time in ms of current promote rate limit period */ 1373 unsigned int nbp_rl_start; 1374 /* number of promote candidate pages at start time of current rate limit period */ 1375 unsigned long nbp_rl_nr_cand; 1376 /* promote threshold in ms */ 1377 unsigned int nbp_threshold; 1378 /* start time in ms of current promote threshold adjustment period */ 1379 unsigned int nbp_th_start; 1380 /* 1381 * number of promote candidate pages at start time of current promote 1382 * threshold adjustment period 1383 */ 1384 unsigned long nbp_th_nr_cand; 1385#endif 1386 /* Fields commonly accessed by the page reclaim scanner */ 1387 1388 /* 1389 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED. 1390 * 1391 * Use mem_cgroup_lruvec() to look up lruvecs. 1392 */ 1393 struct lruvec __lruvec; 1394 1395 unsigned long flags; 1396 1397#ifdef CONFIG_LRU_GEN 1398 /* kswap mm walk data */ 1399 struct lru_gen_mm_walk mm_walk; 1400 /* lru_gen_folio list */ 1401 struct lru_gen_memcg memcg_lru; 1402#endif 1403 1404 CACHELINE_PADDING(_pad2_); 1405 1406 /* Per-node vmstats */ 1407 struct per_cpu_nodestat __percpu *per_cpu_nodestats; 1408 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS]; 1409#ifdef CONFIG_NUMA 1410 struct memory_tier __rcu *memtier; 1411#endif 1412#ifdef CONFIG_MEMORY_FAILURE 1413 struct memory_failure_stats mf_stats; 1414#endif 1415} pg_data_t; 1416 1417#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) 1418#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) 1419 1420#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) 1421#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) 1422 1423static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) 1424{ 1425 return pgdat->node_start_pfn + pgdat->node_spanned_pages; 1426} 1427 1428#include <linux/memory_hotplug.h> 1429 1430void build_all_zonelists(pg_data_t *pgdat); 1431void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order, 1432 enum zone_type highest_zoneidx); 1433bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, 1434 int highest_zoneidx, unsigned int alloc_flags, 1435 long free_pages); 1436bool zone_watermark_ok(struct zone *z, unsigned int order, 1437 unsigned long mark, int highest_zoneidx, 1438 unsigned int alloc_flags); 1439bool zone_watermark_ok_safe(struct zone *z, unsigned int order, 1440 unsigned long mark, int highest_zoneidx); 1441/* 1442 * Memory initialization context, use to differentiate memory added by 1443 * the platform statically or via memory hotplug interface. 1444 */ 1445enum meminit_context { 1446 MEMINIT_EARLY, 1447 MEMINIT_HOTPLUG, 1448}; 1449 1450extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, 1451 unsigned long size); 1452 1453extern void lruvec_init(struct lruvec *lruvec); 1454 1455static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec) 1456{ 1457#ifdef CONFIG_MEMCG 1458 return lruvec->pgdat; 1459#else 1460 return container_of(lruvec, struct pglist_data, __lruvec); 1461#endif 1462} 1463 1464#ifdef CONFIG_HAVE_MEMORYLESS_NODES 1465int local_memory_node(int node_id); 1466#else 1467static inline int local_memory_node(int node_id) { return node_id; }; 1468#endif 1469 1470/* 1471 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. 1472 */ 1473#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) 1474 1475#ifdef CONFIG_ZONE_DEVICE 1476static inline bool zone_is_zone_device(struct zone *zone) 1477{ 1478 return zone_idx(zone) == ZONE_DEVICE; 1479} 1480#else 1481static inline bool zone_is_zone_device(struct zone *zone) 1482{ 1483 return false; 1484} 1485#endif 1486 1487/* 1488 * Returns true if a zone has pages managed by the buddy allocator. 1489 * All the reclaim decisions have to use this function rather than 1490 * populated_zone(). If the whole zone is reserved then we can easily 1491 * end up with populated_zone() && !managed_zone(). 1492 */ 1493static inline bool managed_zone(struct zone *zone) 1494{ 1495 return zone_managed_pages(zone); 1496} 1497 1498/* Returns true if a zone has memory */ 1499static inline bool populated_zone(struct zone *zone) 1500{ 1501 return zone->present_pages; 1502} 1503 1504#ifdef CONFIG_NUMA 1505static inline int zone_to_nid(struct zone *zone) 1506{ 1507 return zone->node; 1508} 1509 1510static inline void zone_set_nid(struct zone *zone, int nid) 1511{ 1512 zone->node = nid; 1513} 1514#else 1515static inline int zone_to_nid(struct zone *zone) 1516{ 1517 return 0; 1518} 1519 1520static inline void zone_set_nid(struct zone *zone, int nid) {} 1521#endif 1522 1523extern int movable_zone; 1524 1525static inline int is_highmem_idx(enum zone_type idx) 1526{ 1527#ifdef CONFIG_HIGHMEM 1528 return (idx == ZONE_HIGHMEM || 1529 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM)); 1530#else 1531 return 0; 1532#endif 1533} 1534 1535/** 1536 * is_highmem - helper function to quickly check if a struct zone is a 1537 * highmem zone or not. This is an attempt to keep references 1538 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. 1539 * @zone: pointer to struct zone variable 1540 * Return: 1 for a highmem zone, 0 otherwise 1541 */ 1542static inline int is_highmem(struct zone *zone) 1543{ 1544 return is_highmem_idx(zone_idx(zone)); 1545} 1546 1547#ifdef CONFIG_ZONE_DMA 1548bool has_managed_dma(void); 1549#else 1550static inline bool has_managed_dma(void) 1551{ 1552 return false; 1553} 1554#endif 1555 1556 1557#ifndef CONFIG_NUMA 1558 1559extern struct pglist_data contig_page_data; 1560static inline struct pglist_data *NODE_DATA(int nid) 1561{ 1562 return &contig_page_data; 1563} 1564 1565#else /* CONFIG_NUMA */ 1566 1567#include <asm/mmzone.h> 1568 1569#endif /* !CONFIG_NUMA */ 1570 1571extern struct pglist_data *first_online_pgdat(void); 1572extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); 1573extern struct zone *next_zone(struct zone *zone); 1574 1575/** 1576 * for_each_online_pgdat - helper macro to iterate over all online nodes 1577 * @pgdat: pointer to a pg_data_t variable 1578 */ 1579#define for_each_online_pgdat(pgdat) \ 1580 for (pgdat = first_online_pgdat(); \ 1581 pgdat; \ 1582 pgdat = next_online_pgdat(pgdat)) 1583/** 1584 * for_each_zone - helper macro to iterate over all memory zones 1585 * @zone: pointer to struct zone variable 1586 * 1587 * The user only needs to declare the zone variable, for_each_zone 1588 * fills it in. 1589 */ 1590#define for_each_zone(zone) \ 1591 for (zone = (first_online_pgdat())->node_zones; \ 1592 zone; \ 1593 zone = next_zone(zone)) 1594 1595#define for_each_populated_zone(zone) \ 1596 for (zone = (first_online_pgdat())->node_zones; \ 1597 zone; \ 1598 zone = next_zone(zone)) \ 1599 if (!populated_zone(zone)) \ 1600 ; /* do nothing */ \ 1601 else 1602 1603static inline struct zone *zonelist_zone(struct zoneref *zoneref) 1604{ 1605 return zoneref->zone; 1606} 1607 1608static inline int zonelist_zone_idx(struct zoneref *zoneref) 1609{ 1610 return zoneref->zone_idx; 1611} 1612 1613static inline int zonelist_node_idx(struct zoneref *zoneref) 1614{ 1615 return zone_to_nid(zoneref->zone); 1616} 1617 1618struct zoneref *__next_zones_zonelist(struct zoneref *z, 1619 enum zone_type highest_zoneidx, 1620 nodemask_t *nodes); 1621 1622/** 1623 * next_zones_zonelist - Returns the next zone at or below highest_zoneidx within the allowed nodemask using a cursor within a zonelist as a starting point 1624 * @z: The cursor used as a starting point for the search 1625 * @highest_zoneidx: The zone index of the highest zone to return 1626 * @nodes: An optional nodemask to filter the zonelist with 1627 * 1628 * This function returns the next zone at or below a given zone index that is 1629 * within the allowed nodemask using a cursor as the starting point for the 1630 * search. The zoneref returned is a cursor that represents the current zone 1631 * being examined. It should be advanced by one before calling 1632 * next_zones_zonelist again. 1633 * 1634 * Return: the next zone at or below highest_zoneidx within the allowed 1635 * nodemask using a cursor within a zonelist as a starting point 1636 */ 1637static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z, 1638 enum zone_type highest_zoneidx, 1639 nodemask_t *nodes) 1640{ 1641 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx)) 1642 return z; 1643 return __next_zones_zonelist(z, highest_zoneidx, nodes); 1644} 1645 1646/** 1647 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist 1648 * @zonelist: The zonelist to search for a suitable zone 1649 * @highest_zoneidx: The zone index of the highest zone to return 1650 * @nodes: An optional nodemask to filter the zonelist with 1651 * 1652 * This function returns the first zone at or below a given zone index that is 1653 * within the allowed nodemask. The zoneref returned is a cursor that can be 1654 * used to iterate the zonelist with next_zones_zonelist by advancing it by 1655 * one before calling. 1656 * 1657 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is 1658 * never NULL). This may happen either genuinely, or due to concurrent nodemask 1659 * update due to cpuset modification. 1660 * 1661 * Return: Zoneref pointer for the first suitable zone found 1662 */ 1663static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, 1664 enum zone_type highest_zoneidx, 1665 nodemask_t *nodes) 1666{ 1667 return next_zones_zonelist(zonelist->_zonerefs, 1668 highest_zoneidx, nodes); 1669} 1670 1671/** 1672 * for_each_zone_zonelist_nodemask - helper macro to iterate over valid zones in a zonelist at or below a given zone index and within a nodemask 1673 * @zone: The current zone in the iterator 1674 * @z: The current pointer within zonelist->_zonerefs being iterated 1675 * @zlist: The zonelist being iterated 1676 * @highidx: The zone index of the highest zone to return 1677 * @nodemask: Nodemask allowed by the allocator 1678 * 1679 * This iterator iterates though all zones at or below a given zone index and 1680 * within a given nodemask 1681 */ 1682#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ 1683 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \ 1684 zone; \ 1685 z = next_zones_zonelist(++z, highidx, nodemask), \ 1686 zone = zonelist_zone(z)) 1687 1688#define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \ 1689 for (zone = z->zone; \ 1690 zone; \ 1691 z = next_zones_zonelist(++z, highidx, nodemask), \ 1692 zone = zonelist_zone(z)) 1693 1694 1695/** 1696 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index 1697 * @zone: The current zone in the iterator 1698 * @z: The current pointer within zonelist->zones being iterated 1699 * @zlist: The zonelist being iterated 1700 * @highidx: The zone index of the highest zone to return 1701 * 1702 * This iterator iterates though all zones at or below a given zone index. 1703 */ 1704#define for_each_zone_zonelist(zone, z, zlist, highidx) \ 1705 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) 1706 1707/* Whether the 'nodes' are all movable nodes */ 1708static inline bool movable_only_nodes(nodemask_t *nodes) 1709{ 1710 struct zonelist *zonelist; 1711 struct zoneref *z; 1712 int nid; 1713 1714 if (nodes_empty(*nodes)) 1715 return false; 1716 1717 /* 1718 * We can chose arbitrary node from the nodemask to get a 1719 * zonelist as they are interlinked. We just need to find 1720 * at least one zone that can satisfy kernel allocations. 1721 */ 1722 nid = first_node(*nodes); 1723 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK]; 1724 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes); 1725 return (!z->zone) ? true : false; 1726} 1727 1728 1729#ifdef CONFIG_SPARSEMEM 1730#include <asm/sparsemem.h> 1731#endif 1732 1733#ifdef CONFIG_FLATMEM 1734#define pfn_to_nid(pfn) (0) 1735#endif 1736 1737#ifdef CONFIG_SPARSEMEM 1738 1739/* 1740 * PA_SECTION_SHIFT physical address to/from section number 1741 * PFN_SECTION_SHIFT pfn to/from section number 1742 */ 1743#define PA_SECTION_SHIFT (SECTION_SIZE_BITS) 1744#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) 1745 1746#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) 1747 1748#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) 1749#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) 1750 1751#define SECTION_BLOCKFLAGS_BITS \ 1752 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) 1753 1754#if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS 1755#error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE 1756#endif 1757 1758static inline unsigned long pfn_to_section_nr(unsigned long pfn) 1759{ 1760 return pfn >> PFN_SECTION_SHIFT; 1761} 1762static inline unsigned long section_nr_to_pfn(unsigned long sec) 1763{ 1764 return sec << PFN_SECTION_SHIFT; 1765} 1766 1767#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) 1768#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) 1769 1770#define SUBSECTION_SHIFT 21 1771#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT) 1772 1773#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT) 1774#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT) 1775#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1)) 1776 1777#if SUBSECTION_SHIFT > SECTION_SIZE_BITS 1778#error Subsection size exceeds section size 1779#else 1780#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT)) 1781#endif 1782 1783#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION) 1784#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK) 1785 1786struct mem_section_usage { 1787 struct rcu_head rcu; 1788#ifdef CONFIG_SPARSEMEM_VMEMMAP 1789 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION); 1790#endif 1791 /* See declaration of similar field in struct zone */ 1792 unsigned long pageblock_flags[0]; 1793}; 1794 1795void subsection_map_init(unsigned long pfn, unsigned long nr_pages); 1796 1797struct page; 1798struct page_ext; 1799struct mem_section { 1800 /* 1801 * This is, logically, a pointer to an array of struct 1802 * pages. However, it is stored with some other magic. 1803 * (see sparse.c::sparse_init_one_section()) 1804 * 1805 * Additionally during early boot we encode node id of 1806 * the location of the section here to guide allocation. 1807 * (see sparse.c::memory_present()) 1808 * 1809 * Making it a UL at least makes someone do a cast 1810 * before using it wrong. 1811 */ 1812 unsigned long section_mem_map; 1813 1814 struct mem_section_usage *usage; 1815#ifdef CONFIG_PAGE_EXTENSION 1816 /* 1817 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use 1818 * section. (see page_ext.h about this.) 1819 */ 1820 struct page_ext *page_ext; 1821 unsigned long pad; 1822#endif 1823 /* 1824 * WARNING: mem_section must be a power-of-2 in size for the 1825 * calculation and use of SECTION_ROOT_MASK to make sense. 1826 */ 1827}; 1828 1829#ifdef CONFIG_SPARSEMEM_EXTREME 1830#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) 1831#else 1832#define SECTIONS_PER_ROOT 1 1833#endif 1834 1835#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) 1836#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) 1837#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) 1838 1839#ifdef CONFIG_SPARSEMEM_EXTREME 1840extern struct mem_section **mem_section; 1841#else 1842extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; 1843#endif 1844 1845static inline unsigned long *section_to_usemap(struct mem_section *ms) 1846{ 1847 return ms->usage->pageblock_flags; 1848} 1849 1850static inline struct mem_section *__nr_to_section(unsigned long nr) 1851{ 1852 unsigned long root = SECTION_NR_TO_ROOT(nr); 1853 1854 if (unlikely(root >= NR_SECTION_ROOTS)) 1855 return NULL; 1856 1857#ifdef CONFIG_SPARSEMEM_EXTREME 1858 if (!mem_section || !mem_section[root]) 1859 return NULL; 1860#endif 1861 return &mem_section[root][nr & SECTION_ROOT_MASK]; 1862} 1863extern size_t mem_section_usage_size(void); 1864 1865/* 1866 * We use the lower bits of the mem_map pointer to store 1867 * a little bit of information. The pointer is calculated 1868 * as mem_map - section_nr_to_pfn(pnum). The result is 1869 * aligned to the minimum alignment of the two values: 1870 * 1. All mem_map arrays are page-aligned. 1871 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT 1872 * lowest bits. PFN_SECTION_SHIFT is arch-specific 1873 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the 1874 * worst combination is powerpc with 256k pages, 1875 * which results in PFN_SECTION_SHIFT equal 6. 1876 * To sum it up, at least 6 bits are available on all architectures. 1877 * However, we can exceed 6 bits on some other architectures except 1878 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available 1879 * with the worst case of 64K pages on arm64) if we make sure the 1880 * exceeded bit is not applicable to powerpc. 1881 */ 1882enum { 1883 SECTION_MARKED_PRESENT_BIT, 1884 SECTION_HAS_MEM_MAP_BIT, 1885 SECTION_IS_ONLINE_BIT, 1886 SECTION_IS_EARLY_BIT, 1887#ifdef CONFIG_ZONE_DEVICE 1888 SECTION_TAINT_ZONE_DEVICE_BIT, 1889#endif 1890 SECTION_MAP_LAST_BIT, 1891}; 1892 1893#define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT) 1894#define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT) 1895#define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT) 1896#define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT) 1897#ifdef CONFIG_ZONE_DEVICE 1898#define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT) 1899#endif 1900#define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1)) 1901#define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT 1902 1903static inline struct page *__section_mem_map_addr(struct mem_section *section) 1904{ 1905 unsigned long map = section->section_mem_map; 1906 map &= SECTION_MAP_MASK; 1907 return (struct page *)map; 1908} 1909 1910static inline int present_section(struct mem_section *section) 1911{ 1912 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); 1913} 1914 1915static inline int present_section_nr(unsigned long nr) 1916{ 1917 return present_section(__nr_to_section(nr)); 1918} 1919 1920static inline int valid_section(struct mem_section *section) 1921{ 1922 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); 1923} 1924 1925static inline int early_section(struct mem_section *section) 1926{ 1927 return (section && (section->section_mem_map & SECTION_IS_EARLY)); 1928} 1929 1930static inline int valid_section_nr(unsigned long nr) 1931{ 1932 return valid_section(__nr_to_section(nr)); 1933} 1934 1935static inline int online_section(struct mem_section *section) 1936{ 1937 return (section && (section->section_mem_map & SECTION_IS_ONLINE)); 1938} 1939 1940#ifdef CONFIG_ZONE_DEVICE 1941static inline int online_device_section(struct mem_section *section) 1942{ 1943 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE; 1944 1945 return section && ((section->section_mem_map & flags) == flags); 1946} 1947#else 1948static inline int online_device_section(struct mem_section *section) 1949{ 1950 return 0; 1951} 1952#endif 1953 1954static inline int online_section_nr(unsigned long nr) 1955{ 1956 return online_section(__nr_to_section(nr)); 1957} 1958 1959#ifdef CONFIG_MEMORY_HOTPLUG 1960void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1961void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1962#endif 1963 1964static inline struct mem_section *__pfn_to_section(unsigned long pfn) 1965{ 1966 return __nr_to_section(pfn_to_section_nr(pfn)); 1967} 1968 1969extern unsigned long __highest_present_section_nr; 1970 1971static inline int subsection_map_index(unsigned long pfn) 1972{ 1973 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION; 1974} 1975 1976#ifdef CONFIG_SPARSEMEM_VMEMMAP 1977static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1978{ 1979 int idx = subsection_map_index(pfn); 1980 1981 return test_bit(idx, READ_ONCE(ms->usage)->subsection_map); 1982} 1983#else 1984static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1985{ 1986 return 1; 1987} 1988#endif 1989 1990#ifndef CONFIG_HAVE_ARCH_PFN_VALID 1991/** 1992 * pfn_valid - check if there is a valid memory map entry for a PFN 1993 * @pfn: the page frame number to check 1994 * 1995 * Check if there is a valid memory map entry aka struct page for the @pfn. 1996 * Note, that availability of the memory map entry does not imply that 1997 * there is actual usable memory at that @pfn. The struct page may 1998 * represent a hole or an unusable page frame. 1999 * 2000 * Return: 1 for PFNs that have memory map entries and 0 otherwise 2001 */ 2002static inline int pfn_valid(unsigned long pfn) 2003{ 2004 struct mem_section *ms; 2005 int ret; 2006 2007 /* 2008 * Ensure the upper PAGE_SHIFT bits are clear in the 2009 * pfn. Else it might lead to false positives when 2010 * some of the upper bits are set, but the lower bits 2011 * match a valid pfn. 2012 */ 2013 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn) 2014 return 0; 2015 2016 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 2017 return 0; 2018 ms = __pfn_to_section(pfn); 2019 rcu_read_lock_sched(); 2020 if (!valid_section(ms)) { 2021 rcu_read_unlock_sched(); 2022 return 0; 2023 } 2024 /* 2025 * Traditionally early sections always returned pfn_valid() for 2026 * the entire section-sized span. 2027 */ 2028 ret = early_section(ms) || pfn_section_valid(ms, pfn); 2029 rcu_read_unlock_sched(); 2030 2031 return ret; 2032} 2033#endif 2034 2035static inline int pfn_in_present_section(unsigned long pfn) 2036{ 2037 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 2038 return 0; 2039 return present_section(__pfn_to_section(pfn)); 2040} 2041 2042static inline unsigned long next_present_section_nr(unsigned long section_nr) 2043{ 2044 while (++section_nr <= __highest_present_section_nr) { 2045 if (present_section_nr(section_nr)) 2046 return section_nr; 2047 } 2048 2049 return -1; 2050} 2051 2052/* 2053 * These are _only_ used during initialisation, therefore they 2054 * can use __initdata ... They could have names to indicate 2055 * this restriction. 2056 */ 2057#ifdef CONFIG_NUMA 2058#define pfn_to_nid(pfn) \ 2059({ \ 2060 unsigned long __pfn_to_nid_pfn = (pfn); \ 2061 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ 2062}) 2063#else 2064#define pfn_to_nid(pfn) (0) 2065#endif 2066 2067void sparse_init(void); 2068#else 2069#define sparse_init() do {} while (0) 2070#define sparse_index_init(_sec, _nid) do {} while (0) 2071#define pfn_in_present_section pfn_valid 2072#define subsection_map_init(_pfn, _nr_pages) do {} while (0) 2073#endif /* CONFIG_SPARSEMEM */ 2074 2075#endif /* !__GENERATING_BOUNDS.H */ 2076#endif /* !__ASSEMBLY__ */ 2077#endif /* _LINUX_MMZONE_H */ 2078