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 <asm/page.h> 26 27 /* Free memory management - zoned buddy allocator. */ 28 #ifndef CONFIG_ARCH_FORCE_MAX_ORDER 29 #define MAX_ORDER 10 30 #else 31 #define MAX_ORDER CONFIG_ARCH_FORCE_MAX_ORDER 32 #endif 33 #define MAX_ORDER_NR_PAGES (1 << MAX_ORDER) 34 35 #define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES) 36 37 /* 38 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed 39 * costly to service. That is between allocation orders which should 40 * coalesce naturally under reasonable reclaim pressure and those which 41 * will not. 42 */ 43 #define PAGE_ALLOC_COSTLY_ORDER 3 44 45 enum migratetype { 46 MIGRATE_UNMOVABLE, 47 MIGRATE_MOVABLE, 48 MIGRATE_RECLAIMABLE, 49 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ 50 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, 51 #ifdef CONFIG_CMA 52 /* 53 * MIGRATE_CMA migration type is designed to mimic the way 54 * ZONE_MOVABLE works. Only movable pages can be allocated 55 * from MIGRATE_CMA pageblocks and page allocator never 56 * implicitly change migration type of MIGRATE_CMA pageblock. 57 * 58 * The way to use it is to change migratetype of a range of 59 * pageblocks to MIGRATE_CMA which can be done by 60 * __free_pageblock_cma() function. 61 */ 62 MIGRATE_CMA, 63 #endif 64 #ifdef CONFIG_MEMORY_ISOLATION 65 MIGRATE_ISOLATE, /* can't allocate from here */ 66 #endif 67 MIGRATE_TYPES 68 }; 69 70 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */ 71 extern const char * const migratetype_names[MIGRATE_TYPES]; 72 73 #ifdef CONFIG_CMA 74 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) 75 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA) 76 #else 77 # define is_migrate_cma(migratetype) false 78 # define is_migrate_cma_page(_page) false 79 #endif 80 81 static inline bool is_migrate_movable(int mt) 82 { 83 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE; 84 } 85 86 /* 87 * Check whether a migratetype can be merged with another migratetype. 88 * 89 * It is only mergeable when it can fall back to other migratetypes for 90 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c. 91 */ 92 static inline bool migratetype_is_mergeable(int mt) 93 { 94 return mt < MIGRATE_PCPTYPES; 95 } 96 97 #define for_each_migratetype_order(order, type) \ 98 for (order = 0; order <= MAX_ORDER; order++) \ 99 for (type = 0; type < MIGRATE_TYPES; type++) 100 101 extern int page_group_by_mobility_disabled; 102 103 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1) 104 105 #define get_pageblock_migratetype(page) \ 106 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK) 107 108 struct free_area { 109 struct list_head free_list[MIGRATE_TYPES]; 110 unsigned long nr_free; 111 }; 112 113 struct pglist_data; 114 115 #ifdef CONFIG_NUMA 116 enum numa_stat_item { 117 NUMA_HIT, /* allocated in intended node */ 118 NUMA_MISS, /* allocated in non intended node */ 119 NUMA_FOREIGN, /* was intended here, hit elsewhere */ 120 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ 121 NUMA_LOCAL, /* allocation from local node */ 122 NUMA_OTHER, /* allocation from other node */ 123 NR_VM_NUMA_EVENT_ITEMS 124 }; 125 #else 126 #define NR_VM_NUMA_EVENT_ITEMS 0 127 #endif 128 129 enum zone_stat_item { 130 /* First 128 byte cacheline (assuming 64 bit words) */ 131 NR_FREE_PAGES, 132 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ 133 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, 134 NR_ZONE_ACTIVE_ANON, 135 NR_ZONE_INACTIVE_FILE, 136 NR_ZONE_ACTIVE_FILE, 137 NR_ZONE_UNEVICTABLE, 138 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */ 139 NR_MLOCK, /* mlock()ed pages found and moved off LRU */ 140 /* Second 128 byte cacheline */ 141 NR_BOUNCE, 142 #if IS_ENABLED(CONFIG_ZSMALLOC) 143 NR_ZSPAGES, /* allocated in zsmalloc */ 144 #endif 145 NR_FREE_CMA_PAGES, 146 #ifdef CONFIG_UNACCEPTED_MEMORY 147 NR_UNACCEPTED, 148 #endif 149 NR_VM_ZONE_STAT_ITEMS }; 150 151 enum node_stat_item { 152 NR_LRU_BASE, 153 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ 154 NR_ACTIVE_ANON, /* " " " " " */ 155 NR_INACTIVE_FILE, /* " " " " " */ 156 NR_ACTIVE_FILE, /* " " " " " */ 157 NR_UNEVICTABLE, /* " " " " " */ 158 NR_SLAB_RECLAIMABLE_B, 159 NR_SLAB_UNRECLAIMABLE_B, 160 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ 161 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ 162 WORKINGSET_NODES, 163 WORKINGSET_REFAULT_BASE, 164 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE, 165 WORKINGSET_REFAULT_FILE, 166 WORKINGSET_ACTIVATE_BASE, 167 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE, 168 WORKINGSET_ACTIVATE_FILE, 169 WORKINGSET_RESTORE_BASE, 170 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE, 171 WORKINGSET_RESTORE_FILE, 172 WORKINGSET_NODERECLAIM, 173 NR_ANON_MAPPED, /* Mapped anonymous pages */ 174 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. 175 only modified from process context */ 176 NR_FILE_PAGES, 177 NR_FILE_DIRTY, 178 NR_WRITEBACK, 179 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ 180 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ 181 NR_SHMEM_THPS, 182 NR_SHMEM_PMDMAPPED, 183 NR_FILE_THPS, 184 NR_FILE_PMDMAPPED, 185 NR_ANON_THPS, 186 NR_VMSCAN_WRITE, 187 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ 188 NR_DIRTIED, /* page dirtyings since bootup */ 189 NR_WRITTEN, /* page writings since bootup */ 190 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */ 191 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */ 192 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */ 193 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */ 194 NR_KERNEL_STACK_KB, /* measured in KiB */ 195 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK) 196 NR_KERNEL_SCS_KB, /* measured in KiB */ 197 #endif 198 NR_PAGETABLE, /* used for pagetables */ 199 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */ 200 #ifdef CONFIG_SWAP 201 NR_SWAPCACHE, 202 #endif 203 #ifdef CONFIG_NUMA_BALANCING 204 PGPROMOTE_SUCCESS, /* promote successfully */ 205 PGPROMOTE_CANDIDATE, /* candidate pages to promote */ 206 #endif 207 NR_VM_NODE_STAT_ITEMS 208 }; 209 210 /* 211 * Returns true if the item should be printed in THPs (/proc/vmstat 212 * currently prints number of anon, file and shmem THPs. But the item 213 * is charged in pages). 214 */ 215 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item) 216 { 217 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) 218 return false; 219 220 return item == NR_ANON_THPS || 221 item == NR_FILE_THPS || 222 item == NR_SHMEM_THPS || 223 item == NR_SHMEM_PMDMAPPED || 224 item == NR_FILE_PMDMAPPED; 225 } 226 227 /* 228 * Returns true if the value is measured in bytes (most vmstat values are 229 * measured in pages). This defines the API part, the internal representation 230 * might be different. 231 */ 232 static __always_inline bool vmstat_item_in_bytes(int idx) 233 { 234 /* 235 * Global and per-node slab counters track slab pages. 236 * It's expected that changes are multiples of PAGE_SIZE. 237 * Internally values are stored in pages. 238 * 239 * Per-memcg and per-lruvec counters track memory, consumed 240 * by individual slab objects. These counters are actually 241 * byte-precise. 242 */ 243 return (idx == NR_SLAB_RECLAIMABLE_B || 244 idx == NR_SLAB_UNRECLAIMABLE_B); 245 } 246 247 /* 248 * We do arithmetic on the LRU lists in various places in the code, 249 * so it is important to keep the active lists LRU_ACTIVE higher in 250 * the array than the corresponding inactive lists, and to keep 251 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. 252 * 253 * This has to be kept in sync with the statistics in zone_stat_item 254 * above and the descriptions in vmstat_text in mm/vmstat.c 255 */ 256 #define LRU_BASE 0 257 #define LRU_ACTIVE 1 258 #define LRU_FILE 2 259 260 enum lru_list { 261 LRU_INACTIVE_ANON = LRU_BASE, 262 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, 263 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, 264 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, 265 LRU_UNEVICTABLE, 266 NR_LRU_LISTS 267 }; 268 269 enum vmscan_throttle_state { 270 VMSCAN_THROTTLE_WRITEBACK, 271 VMSCAN_THROTTLE_ISOLATED, 272 VMSCAN_THROTTLE_NOPROGRESS, 273 VMSCAN_THROTTLE_CONGESTED, 274 NR_VMSCAN_THROTTLE, 275 }; 276 277 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) 278 279 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) 280 281 static inline bool is_file_lru(enum lru_list lru) 282 { 283 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); 284 } 285 286 static inline bool is_active_lru(enum lru_list lru) 287 { 288 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); 289 } 290 291 #define WORKINGSET_ANON 0 292 #define WORKINGSET_FILE 1 293 #define ANON_AND_FILE 2 294 295 enum lruvec_flags { 296 LRUVEC_CONGESTED, /* lruvec has many dirty pages 297 * backed by a congested BDI 298 */ 299 }; 300 301 #endif /* !__GENERATING_BOUNDS_H */ 302 303 /* 304 * Evictable pages are divided into multiple generations. The youngest and the 305 * oldest generation numbers, max_seq and min_seq, are monotonically increasing. 306 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An 307 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the 308 * corresponding generation. The gen counter in folio->flags stores gen+1 while 309 * a page is on one of lrugen->folios[]. Otherwise it stores 0. 310 * 311 * A page is added to the youngest generation on faulting. The aging needs to 312 * check the accessed bit at least twice before handing this page over to the 313 * eviction. The first check takes care of the accessed bit set on the initial 314 * fault; the second check makes sure this page hasn't been used since then. 315 * This process, AKA second chance, requires a minimum of two generations, 316 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive 317 * LRU, e.g., /proc/vmstat, these two generations are considered active; the 318 * rest of generations, if they exist, are considered inactive. See 319 * lru_gen_is_active(). 320 * 321 * PG_active is always cleared while a page is on one of lrugen->folios[] so 322 * that the aging needs not to worry about it. And it's set again when a page 323 * considered active is isolated for non-reclaiming purposes, e.g., migration. 324 * See lru_gen_add_folio() and lru_gen_del_folio(). 325 * 326 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the 327 * number of categories of the active/inactive LRU when keeping track of 328 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits 329 * in folio->flags. 330 */ 331 #define MIN_NR_GENS 2U 332 #define MAX_NR_GENS 4U 333 334 /* 335 * Each generation is divided into multiple tiers. A page accessed N times 336 * through file descriptors is in tier order_base_2(N). A page in the first tier 337 * (N=0,1) is marked by PG_referenced unless it was faulted in through page 338 * tables or read ahead. A page in any other tier (N>1) is marked by 339 * PG_referenced and PG_workingset. This implies a minimum of two tiers is 340 * supported without using additional bits in folio->flags. 341 * 342 * In contrast to moving across generations which requires the LRU lock, moving 343 * across tiers only involves atomic operations on folio->flags and therefore 344 * has a negligible cost in the buffered access path. In the eviction path, 345 * comparisons of refaulted/(evicted+protected) from the first tier and the 346 * rest infer whether pages accessed multiple times through file descriptors 347 * are statistically hot and thus worth protecting. 348 * 349 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the 350 * number of categories of the active/inactive LRU when keeping track of 351 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in 352 * folio->flags. 353 */ 354 #define MAX_NR_TIERS 4U 355 356 #ifndef __GENERATING_BOUNDS_H 357 358 struct lruvec; 359 struct page_vma_mapped_walk; 360 361 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF) 362 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF) 363 364 #ifdef CONFIG_LRU_GEN 365 366 enum { 367 LRU_GEN_ANON, 368 LRU_GEN_FILE, 369 }; 370 371 enum { 372 LRU_GEN_CORE, 373 LRU_GEN_MM_WALK, 374 LRU_GEN_NONLEAF_YOUNG, 375 NR_LRU_GEN_CAPS 376 }; 377 378 #define MIN_LRU_BATCH BITS_PER_LONG 379 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64) 380 381 /* whether to keep historical stats from evicted generations */ 382 #ifdef CONFIG_LRU_GEN_STATS 383 #define NR_HIST_GENS MAX_NR_GENS 384 #else 385 #define NR_HIST_GENS 1U 386 #endif 387 388 /* 389 * The youngest generation number is stored in max_seq for both anon and file 390 * types as they are aged on an equal footing. The oldest generation numbers are 391 * stored in min_seq[] separately for anon and file types as clean file pages 392 * can be evicted regardless of swap constraints. 393 * 394 * Normally anon and file min_seq are in sync. But if swapping is constrained, 395 * e.g., out of swap space, file min_seq is allowed to advance and leave anon 396 * min_seq behind. 397 * 398 * The number of pages in each generation is eventually consistent and therefore 399 * can be transiently negative when reset_batch_size() is pending. 400 */ 401 struct lru_gen_folio { 402 /* the aging increments the youngest generation number */ 403 unsigned long max_seq; 404 /* the eviction increments the oldest generation numbers */ 405 unsigned long min_seq[ANON_AND_FILE]; 406 /* the birth time of each generation in jiffies */ 407 unsigned long timestamps[MAX_NR_GENS]; 408 /* the multi-gen LRU lists, lazily sorted on eviction */ 409 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 410 /* the multi-gen LRU sizes, eventually consistent */ 411 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 412 /* the exponential moving average of refaulted */ 413 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS]; 414 /* the exponential moving average of evicted+protected */ 415 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS]; 416 /* the first tier doesn't need protection, hence the minus one */ 417 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1]; 418 /* can be modified without holding the LRU lock */ 419 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 420 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 421 /* whether the multi-gen LRU is enabled */ 422 bool enabled; 423 #ifdef CONFIG_MEMCG 424 /* the memcg generation this lru_gen_folio belongs to */ 425 u8 gen; 426 /* the list segment this lru_gen_folio belongs to */ 427 u8 seg; 428 /* per-node lru_gen_folio list for global reclaim */ 429 struct hlist_nulls_node list; 430 #endif 431 }; 432 433 enum { 434 MM_LEAF_TOTAL, /* total leaf entries */ 435 MM_LEAF_OLD, /* old leaf entries */ 436 MM_LEAF_YOUNG, /* young leaf entries */ 437 MM_NONLEAF_TOTAL, /* total non-leaf entries */ 438 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */ 439 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */ 440 NR_MM_STATS 441 }; 442 443 /* double-buffering Bloom filters */ 444 #define NR_BLOOM_FILTERS 2 445 446 struct lru_gen_mm_state { 447 /* set to max_seq after each iteration */ 448 unsigned long seq; 449 /* where the current iteration continues after */ 450 struct list_head *head; 451 /* where the last iteration ended before */ 452 struct list_head *tail; 453 /* Bloom filters flip after each iteration */ 454 unsigned long *filters[NR_BLOOM_FILTERS]; 455 /* the mm stats for debugging */ 456 unsigned long stats[NR_HIST_GENS][NR_MM_STATS]; 457 }; 458 459 struct lru_gen_mm_walk { 460 /* the lruvec under reclaim */ 461 struct lruvec *lruvec; 462 /* unstable max_seq from lru_gen_folio */ 463 unsigned long max_seq; 464 /* the next address within an mm to scan */ 465 unsigned long next_addr; 466 /* to batch promoted pages */ 467 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 468 /* to batch the mm stats */ 469 int mm_stats[NR_MM_STATS]; 470 /* total batched items */ 471 int batched; 472 bool can_swap; 473 bool force_scan; 474 }; 475 476 void lru_gen_init_lruvec(struct lruvec *lruvec); 477 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw); 478 479 #ifdef CONFIG_MEMCG 480 481 /* 482 * For each node, memcgs are divided into two generations: the old and the 483 * young. For each generation, memcgs are randomly sharded into multiple bins 484 * to improve scalability. For each bin, the hlist_nulls is virtually divided 485 * into three segments: the head, the tail and the default. 486 * 487 * An onlining memcg is added to the tail of a random bin in the old generation. 488 * The eviction starts at the head of a random bin in the old generation. The 489 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes 490 * the old generation, is incremented when all its bins become empty. 491 * 492 * There are four operations: 493 * 1. MEMCG_LRU_HEAD, which moves an memcg to the head of a random bin in its 494 * current generation (old or young) and updates its "seg" to "head"; 495 * 2. MEMCG_LRU_TAIL, which moves an memcg to the tail of a random bin in its 496 * current generation (old or young) and updates its "seg" to "tail"; 497 * 3. MEMCG_LRU_OLD, which moves an memcg to the head of a random bin in the old 498 * generation, updates its "gen" to "old" and resets its "seg" to "default"; 499 * 4. MEMCG_LRU_YOUNG, which moves an memcg to the tail of a random bin in the 500 * young generation, updates its "gen" to "young" and resets its "seg" to 501 * "default". 502 * 503 * The events that trigger the above operations are: 504 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD; 505 * 2. The first attempt to reclaim an memcg below low, which triggers 506 * MEMCG_LRU_TAIL; 507 * 3. The first attempt to reclaim an memcg below reclaimable size threshold, 508 * which triggers MEMCG_LRU_TAIL; 509 * 4. The second attempt to reclaim an memcg below reclaimable size threshold, 510 * which triggers MEMCG_LRU_YOUNG; 511 * 5. Attempting to reclaim an memcg below min, which triggers MEMCG_LRU_YOUNG; 512 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG; 513 * 7. Offlining an memcg, which triggers MEMCG_LRU_OLD. 514 * 515 * Note that memcg LRU only applies to global reclaim, and the round-robin 516 * incrementing of their max_seq counters ensures the eventual fairness to all 517 * eligible memcgs. For memcg reclaim, it still relies on mem_cgroup_iter(). 518 */ 519 #define MEMCG_NR_GENS 2 520 #define MEMCG_NR_BINS 8 521 522 struct lru_gen_memcg { 523 /* the per-node memcg generation counter */ 524 unsigned long seq; 525 /* each memcg has one lru_gen_folio per node */ 526 unsigned long nr_memcgs[MEMCG_NR_GENS]; 527 /* per-node lru_gen_folio list for global reclaim */ 528 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS]; 529 /* protects the above */ 530 spinlock_t lock; 531 }; 532 533 void lru_gen_init_pgdat(struct pglist_data *pgdat); 534 535 void lru_gen_init_memcg(struct mem_cgroup *memcg); 536 void lru_gen_exit_memcg(struct mem_cgroup *memcg); 537 void lru_gen_online_memcg(struct mem_cgroup *memcg); 538 void lru_gen_offline_memcg(struct mem_cgroup *memcg); 539 void lru_gen_release_memcg(struct mem_cgroup *memcg); 540 void lru_gen_soft_reclaim(struct lruvec *lruvec); 541 542 #else /* !CONFIG_MEMCG */ 543 544 #define MEMCG_NR_GENS 1 545 546 struct lru_gen_memcg { 547 }; 548 549 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat) 550 { 551 } 552 553 #endif /* CONFIG_MEMCG */ 554 555 #else /* !CONFIG_LRU_GEN */ 556 557 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat) 558 { 559 } 560 561 static inline void lru_gen_init_lruvec(struct lruvec *lruvec) 562 { 563 } 564 565 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw) 566 { 567 } 568 569 #ifdef CONFIG_MEMCG 570 571 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg) 572 { 573 } 574 575 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg) 576 { 577 } 578 579 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg) 580 { 581 } 582 583 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg) 584 { 585 } 586 587 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg) 588 { 589 } 590 591 static inline void lru_gen_soft_reclaim(struct lruvec *lruvec) 592 { 593 } 594 595 #endif /* CONFIG_MEMCG */ 596 597 #endif /* CONFIG_LRU_GEN */ 598 599 struct lruvec { 600 struct list_head lists[NR_LRU_LISTS]; 601 /* per lruvec lru_lock for memcg */ 602 spinlock_t lru_lock; 603 /* 604 * These track the cost of reclaiming one LRU - file or anon - 605 * over the other. As the observed cost of reclaiming one LRU 606 * increases, the reclaim scan balance tips toward the other. 607 */ 608 unsigned long anon_cost; 609 unsigned long file_cost; 610 /* Non-resident age, driven by LRU movement */ 611 atomic_long_t nonresident_age; 612 /* Refaults at the time of last reclaim cycle */ 613 unsigned long refaults[ANON_AND_FILE]; 614 /* Various lruvec state flags (enum lruvec_flags) */ 615 unsigned long flags; 616 #ifdef CONFIG_LRU_GEN 617 /* evictable pages divided into generations */ 618 struct lru_gen_folio lrugen; 619 /* to concurrently iterate lru_gen_mm_list */ 620 struct lru_gen_mm_state mm_state; 621 #endif 622 #ifdef CONFIG_MEMCG 623 struct pglist_data *pgdat; 624 #endif 625 }; 626 627 /* Isolate unmapped pages */ 628 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) 629 /* Isolate for asynchronous migration */ 630 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) 631 /* Isolate unevictable pages */ 632 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) 633 634 /* LRU Isolation modes. */ 635 typedef unsigned __bitwise isolate_mode_t; 636 637 enum zone_watermarks { 638 WMARK_MIN, 639 WMARK_LOW, 640 WMARK_HIGH, 641 WMARK_PROMO, 642 NR_WMARK 643 }; 644 645 /* 646 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list 647 * for THP which will usually be GFP_MOVABLE. Even if it is another type, 648 * it should not contribute to serious fragmentation causing THP allocation 649 * failures. 650 */ 651 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 652 #define NR_PCP_THP 1 653 #else 654 #define NR_PCP_THP 0 655 #endif 656 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1)) 657 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP) 658 659 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost) 660 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost) 661 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost) 662 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost) 663 664 /* Fields and list protected by pagesets local_lock in page_alloc.c */ 665 struct per_cpu_pages { 666 spinlock_t lock; /* Protects lists field */ 667 int count; /* number of pages in the list */ 668 int high; /* high watermark, emptying needed */ 669 int batch; /* chunk size for buddy add/remove */ 670 short free_factor; /* batch scaling factor during free */ 671 #ifdef CONFIG_NUMA 672 short expire; /* When 0, remote pagesets are drained */ 673 #endif 674 675 /* Lists of pages, one per migrate type stored on the pcp-lists */ 676 struct list_head lists[NR_PCP_LISTS]; 677 } ____cacheline_aligned_in_smp; 678 679 struct per_cpu_zonestat { 680 #ifdef CONFIG_SMP 681 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; 682 s8 stat_threshold; 683 #endif 684 #ifdef CONFIG_NUMA 685 /* 686 * Low priority inaccurate counters that are only folded 687 * on demand. Use a large type to avoid the overhead of 688 * folding during refresh_cpu_vm_stats. 689 */ 690 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; 691 #endif 692 }; 693 694 struct per_cpu_nodestat { 695 s8 stat_threshold; 696 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS]; 697 }; 698 699 #endif /* !__GENERATING_BOUNDS.H */ 700 701 enum zone_type { 702 /* 703 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able 704 * to DMA to all of the addressable memory (ZONE_NORMAL). 705 * On architectures where this area covers the whole 32 bit address 706 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller 707 * DMA addressing constraints. This distinction is important as a 32bit 708 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit 709 * platforms may need both zones as they support peripherals with 710 * different DMA addressing limitations. 711 */ 712 #ifdef CONFIG_ZONE_DMA 713 ZONE_DMA, 714 #endif 715 #ifdef CONFIG_ZONE_DMA32 716 ZONE_DMA32, 717 #endif 718 /* 719 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be 720 * performed on pages in ZONE_NORMAL if the DMA devices support 721 * transfers to all addressable memory. 722 */ 723 ZONE_NORMAL, 724 #ifdef CONFIG_HIGHMEM 725 /* 726 * A memory area that is only addressable by the kernel through 727 * mapping portions into its own address space. This is for example 728 * used by i386 to allow the kernel to address the memory beyond 729 * 900MB. The kernel will set up special mappings (page 730 * table entries on i386) for each page that the kernel needs to 731 * access. 732 */ 733 ZONE_HIGHMEM, 734 #endif 735 /* 736 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains 737 * movable pages with few exceptional cases described below. Main use 738 * cases for ZONE_MOVABLE are to make memory offlining/unplug more 739 * likely to succeed, and to locally limit unmovable allocations - e.g., 740 * to increase the number of THP/huge pages. Notable special cases are: 741 * 742 * 1. Pinned pages: (long-term) pinning of movable pages might 743 * essentially turn such pages unmovable. Therefore, we do not allow 744 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and 745 * faulted, they come from the right zone right away. However, it is 746 * still possible that address space already has pages in 747 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has 748 * touches that memory before pinning). In such case we migrate them 749 * to a different zone. When migration fails - pinning fails. 750 * 2. memblock allocations: kernelcore/movablecore setups might create 751 * situations where ZONE_MOVABLE contains unmovable allocations 752 * after boot. Memory offlining and allocations fail early. 753 * 3. Memory holes: kernelcore/movablecore setups might create very rare 754 * situations where ZONE_MOVABLE contains memory holes after boot, 755 * for example, if we have sections that are only partially 756 * populated. Memory offlining and allocations fail early. 757 * 4. PG_hwpoison pages: while poisoned pages can be skipped during 758 * memory offlining, such pages cannot be allocated. 759 * 5. Unmovable PG_offline pages: in paravirtualized environments, 760 * hotplugged memory blocks might only partially be managed by the 761 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The 762 * parts not manged by the buddy are unmovable PG_offline pages. In 763 * some cases (virtio-mem), such pages can be skipped during 764 * memory offlining, however, cannot be moved/allocated. These 765 * techniques might use alloc_contig_range() to hide previously 766 * exposed pages from the buddy again (e.g., to implement some sort 767 * of memory unplug in virtio-mem). 768 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create 769 * situations where ZERO_PAGE(0) which is allocated differently 770 * on different platforms may end up in a movable zone. ZERO_PAGE(0) 771 * cannot be migrated. 772 * 7. Memory-hotplug: when using memmap_on_memory and onlining the 773 * memory to the MOVABLE zone, the vmemmap pages are also placed in 774 * such zone. Such pages cannot be really moved around as they are 775 * self-stored in the range, but they are treated as movable when 776 * the range they describe is about to be offlined. 777 * 778 * In general, no unmovable allocations that degrade memory offlining 779 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range()) 780 * have to expect that migrating pages in ZONE_MOVABLE can fail (even 781 * if has_unmovable_pages() states that there are no unmovable pages, 782 * there can be false negatives). 783 */ 784 ZONE_MOVABLE, 785 #ifdef CONFIG_ZONE_DEVICE 786 ZONE_DEVICE, 787 #endif 788 __MAX_NR_ZONES 789 790 }; 791 792 #ifndef __GENERATING_BOUNDS_H 793 794 #define ASYNC_AND_SYNC 2 795 796 struct zone { 797 /* Read-mostly fields */ 798 799 /* zone watermarks, access with *_wmark_pages(zone) macros */ 800 unsigned long _watermark[NR_WMARK]; 801 unsigned long watermark_boost; 802 803 unsigned long nr_reserved_highatomic; 804 805 /* 806 * We don't know if the memory that we're going to allocate will be 807 * freeable or/and it will be released eventually, so to avoid totally 808 * wasting several GB of ram we must reserve some of the lower zone 809 * memory (otherwise we risk to run OOM on the lower zones despite 810 * there being tons of freeable ram on the higher zones). This array is 811 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl 812 * changes. 813 */ 814 long lowmem_reserve[MAX_NR_ZONES]; 815 816 #ifdef CONFIG_NUMA 817 int node; 818 #endif 819 struct pglist_data *zone_pgdat; 820 struct per_cpu_pages __percpu *per_cpu_pageset; 821 struct per_cpu_zonestat __percpu *per_cpu_zonestats; 822 /* 823 * the high and batch values are copied to individual pagesets for 824 * faster access 825 */ 826 int pageset_high; 827 int pageset_batch; 828 829 #ifndef CONFIG_SPARSEMEM 830 /* 831 * Flags for a pageblock_nr_pages block. See pageblock-flags.h. 832 * In SPARSEMEM, this map is stored in struct mem_section 833 */ 834 unsigned long *pageblock_flags; 835 #endif /* CONFIG_SPARSEMEM */ 836 837 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ 838 unsigned long zone_start_pfn; 839 840 /* 841 * spanned_pages is the total pages spanned by the zone, including 842 * holes, which is calculated as: 843 * spanned_pages = zone_end_pfn - zone_start_pfn; 844 * 845 * present_pages is physical pages existing within the zone, which 846 * is calculated as: 847 * present_pages = spanned_pages - absent_pages(pages in holes); 848 * 849 * present_early_pages is present pages existing within the zone 850 * located on memory available since early boot, excluding hotplugged 851 * memory. 852 * 853 * managed_pages is present pages managed by the buddy system, which 854 * is calculated as (reserved_pages includes pages allocated by the 855 * bootmem allocator): 856 * managed_pages = present_pages - reserved_pages; 857 * 858 * cma pages is present pages that are assigned for CMA use 859 * (MIGRATE_CMA). 860 * 861 * So present_pages may be used by memory hotplug or memory power 862 * management logic to figure out unmanaged pages by checking 863 * (present_pages - managed_pages). And managed_pages should be used 864 * by page allocator and vm scanner to calculate all kinds of watermarks 865 * and thresholds. 866 * 867 * Locking rules: 868 * 869 * zone_start_pfn and spanned_pages are protected by span_seqlock. 870 * It is a seqlock because it has to be read outside of zone->lock, 871 * and it is done in the main allocator path. But, it is written 872 * quite infrequently. 873 * 874 * The span_seq lock is declared along with zone->lock because it is 875 * frequently read in proximity to zone->lock. It's good to 876 * give them a chance of being in the same cacheline. 877 * 878 * Write access to present_pages at runtime should be protected by 879 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of 880 * present_pages should use get_online_mems() to get a stable value. 881 */ 882 atomic_long_t managed_pages; 883 unsigned long spanned_pages; 884 unsigned long present_pages; 885 #if defined(CONFIG_MEMORY_HOTPLUG) 886 unsigned long present_early_pages; 887 #endif 888 #ifdef CONFIG_CMA 889 unsigned long cma_pages; 890 #endif 891 892 const char *name; 893 894 #ifdef CONFIG_MEMORY_ISOLATION 895 /* 896 * Number of isolated pageblock. It is used to solve incorrect 897 * freepage counting problem due to racy retrieving migratetype 898 * of pageblock. Protected by zone->lock. 899 */ 900 unsigned long nr_isolate_pageblock; 901 #endif 902 903 #ifdef CONFIG_MEMORY_HOTPLUG 904 /* see spanned/present_pages for more description */ 905 seqlock_t span_seqlock; 906 #endif 907 908 int initialized; 909 910 /* Write-intensive fields used from the page allocator */ 911 CACHELINE_PADDING(_pad1_); 912 913 /* free areas of different sizes */ 914 struct free_area free_area[MAX_ORDER + 1]; 915 916 #ifdef CONFIG_UNACCEPTED_MEMORY 917 /* Pages to be accepted. All pages on the list are MAX_ORDER */ 918 struct list_head unaccepted_pages; 919 #endif 920 921 /* zone flags, see below */ 922 unsigned long flags; 923 924 /* Primarily protects free_area */ 925 spinlock_t lock; 926 927 /* Write-intensive fields used by compaction and vmstats. */ 928 CACHELINE_PADDING(_pad2_); 929 930 /* 931 * When free pages are below this point, additional steps are taken 932 * when reading the number of free pages to avoid per-cpu counter 933 * drift allowing watermarks to be breached 934 */ 935 unsigned long percpu_drift_mark; 936 937 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 938 /* pfn where compaction free scanner should start */ 939 unsigned long compact_cached_free_pfn; 940 /* pfn where compaction migration scanner should start */ 941 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC]; 942 unsigned long compact_init_migrate_pfn; 943 unsigned long compact_init_free_pfn; 944 #endif 945 946 #ifdef CONFIG_COMPACTION 947 /* 948 * On compaction failure, 1<<compact_defer_shift compactions 949 * are skipped before trying again. The number attempted since 950 * last failure is tracked with compact_considered. 951 * compact_order_failed is the minimum compaction failed order. 952 */ 953 unsigned int compact_considered; 954 unsigned int compact_defer_shift; 955 int compact_order_failed; 956 #endif 957 958 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 959 /* Set to true when the PG_migrate_skip bits should be cleared */ 960 bool compact_blockskip_flush; 961 #endif 962 963 bool contiguous; 964 965 CACHELINE_PADDING(_pad3_); 966 /* Zone statistics */ 967 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; 968 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; 969 } ____cacheline_internodealigned_in_smp; 970 971 enum pgdat_flags { 972 PGDAT_DIRTY, /* reclaim scanning has recently found 973 * many dirty file pages at the tail 974 * of the LRU. 975 */ 976 PGDAT_WRITEBACK, /* reclaim scanning has recently found 977 * many pages under writeback 978 */ 979 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */ 980 }; 981 982 enum zone_flags { 983 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks. 984 * Cleared when kswapd is woken. 985 */ 986 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */ 987 }; 988 989 static inline unsigned long zone_managed_pages(struct zone *zone) 990 { 991 return (unsigned long)atomic_long_read(&zone->managed_pages); 992 } 993 994 static inline unsigned long zone_cma_pages(struct zone *zone) 995 { 996 #ifdef CONFIG_CMA 997 return zone->cma_pages; 998 #else 999 return 0; 1000 #endif 1001 } 1002 1003 static inline unsigned long zone_end_pfn(const struct zone *zone) 1004 { 1005 return zone->zone_start_pfn + zone->spanned_pages; 1006 } 1007 1008 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) 1009 { 1010 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); 1011 } 1012 1013 static inline bool zone_is_initialized(struct zone *zone) 1014 { 1015 return zone->initialized; 1016 } 1017 1018 static inline bool zone_is_empty(struct zone *zone) 1019 { 1020 return zone->spanned_pages == 0; 1021 } 1022 1023 #ifndef BUILD_VDSO32_64 1024 /* 1025 * The zone field is never updated after free_area_init_core() 1026 * sets it, so none of the operations on it need to be atomic. 1027 */ 1028 1029 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */ 1030 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH) 1031 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH) 1032 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH) 1033 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH) 1034 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH) 1035 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH) 1036 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH) 1037 1038 /* 1039 * Define the bit shifts to access each section. For non-existent 1040 * sections we define the shift as 0; that plus a 0 mask ensures 1041 * the compiler will optimise away reference to them. 1042 */ 1043 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0)) 1044 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0)) 1045 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0)) 1046 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0)) 1047 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0)) 1048 1049 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */ 1050 #ifdef NODE_NOT_IN_PAGE_FLAGS 1051 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT) 1052 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \ 1053 SECTIONS_PGOFF : ZONES_PGOFF) 1054 #else 1055 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT) 1056 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \ 1057 NODES_PGOFF : ZONES_PGOFF) 1058 #endif 1059 1060 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0)) 1061 1062 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1) 1063 #define NODES_MASK ((1UL << NODES_WIDTH) - 1) 1064 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1) 1065 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1) 1066 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1) 1067 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1) 1068 1069 static inline enum zone_type page_zonenum(const struct page *page) 1070 { 1071 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT); 1072 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK; 1073 } 1074 1075 static inline enum zone_type folio_zonenum(const struct folio *folio) 1076 { 1077 return page_zonenum(&folio->page); 1078 } 1079 1080 #ifdef CONFIG_ZONE_DEVICE 1081 static inline bool is_zone_device_page(const struct page *page) 1082 { 1083 return page_zonenum(page) == ZONE_DEVICE; 1084 } 1085 1086 /* 1087 * Consecutive zone device pages should not be merged into the same sgl 1088 * or bvec segment with other types of pages or if they belong to different 1089 * pgmaps. Otherwise getting the pgmap of a given segment is not possible 1090 * without scanning the entire segment. This helper returns true either if 1091 * both pages are not zone device pages or both pages are zone device pages 1092 * with the same pgmap. 1093 */ 1094 static inline bool zone_device_pages_have_same_pgmap(const struct page *a, 1095 const struct page *b) 1096 { 1097 if (is_zone_device_page(a) != is_zone_device_page(b)) 1098 return false; 1099 if (!is_zone_device_page(a)) 1100 return true; 1101 return a->pgmap == b->pgmap; 1102 } 1103 1104 extern void memmap_init_zone_device(struct zone *, unsigned long, 1105 unsigned long, struct dev_pagemap *); 1106 #else 1107 static inline bool is_zone_device_page(const struct page *page) 1108 { 1109 return false; 1110 } 1111 static inline bool zone_device_pages_have_same_pgmap(const struct page *a, 1112 const struct page *b) 1113 { 1114 return true; 1115 } 1116 #endif 1117 1118 static inline bool folio_is_zone_device(const struct folio *folio) 1119 { 1120 return is_zone_device_page(&folio->page); 1121 } 1122 1123 static inline bool is_zone_movable_page(const struct page *page) 1124 { 1125 return page_zonenum(page) == ZONE_MOVABLE; 1126 } 1127 #endif 1128 1129 /* 1130 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty 1131 * intersection with the given zone 1132 */ 1133 static inline bool zone_intersects(struct zone *zone, 1134 unsigned long start_pfn, unsigned long nr_pages) 1135 { 1136 if (zone_is_empty(zone)) 1137 return false; 1138 if (start_pfn >= zone_end_pfn(zone) || 1139 start_pfn + nr_pages <= zone->zone_start_pfn) 1140 return false; 1141 1142 return true; 1143 } 1144 1145 /* 1146 * The "priority" of VM scanning is how much of the queues we will scan in one 1147 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the 1148 * queues ("queue_length >> 12") during an aging round. 1149 */ 1150 #define DEF_PRIORITY 12 1151 1152 /* Maximum number of zones on a zonelist */ 1153 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) 1154 1155 enum { 1156 ZONELIST_FALLBACK, /* zonelist with fallback */ 1157 #ifdef CONFIG_NUMA 1158 /* 1159 * The NUMA zonelists are doubled because we need zonelists that 1160 * restrict the allocations to a single node for __GFP_THISNODE. 1161 */ 1162 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */ 1163 #endif 1164 MAX_ZONELISTS 1165 }; 1166 1167 /* 1168 * This struct contains information about a zone in a zonelist. It is stored 1169 * here to avoid dereferences into large structures and lookups of tables 1170 */ 1171 struct zoneref { 1172 struct zone *zone; /* Pointer to actual zone */ 1173 int zone_idx; /* zone_idx(zoneref->zone) */ 1174 }; 1175 1176 /* 1177 * One allocation request operates on a zonelist. A zonelist 1178 * is a list of zones, the first one is the 'goal' of the 1179 * allocation, the other zones are fallback zones, in decreasing 1180 * priority. 1181 * 1182 * To speed the reading of the zonelist, the zonerefs contain the zone index 1183 * of the entry being read. Helper functions to access information given 1184 * a struct zoneref are 1185 * 1186 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs 1187 * zonelist_zone_idx() - Return the index of the zone for an entry 1188 * zonelist_node_idx() - Return the index of the node for an entry 1189 */ 1190 struct zonelist { 1191 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; 1192 }; 1193 1194 /* 1195 * The array of struct pages for flatmem. 1196 * It must be declared for SPARSEMEM as well because there are configurations 1197 * that rely on that. 1198 */ 1199 extern struct page *mem_map; 1200 1201 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1202 struct deferred_split { 1203 spinlock_t split_queue_lock; 1204 struct list_head split_queue; 1205 unsigned long split_queue_len; 1206 }; 1207 #endif 1208 1209 #ifdef CONFIG_MEMORY_FAILURE 1210 /* 1211 * Per NUMA node memory failure handling statistics. 1212 */ 1213 struct memory_failure_stats { 1214 /* 1215 * Number of raw pages poisoned. 1216 * Cases not accounted: memory outside kernel control, offline page, 1217 * arch-specific memory_failure (SGX), hwpoison_filter() filtered 1218 * error events, and unpoison actions from hwpoison_unpoison. 1219 */ 1220 unsigned long total; 1221 /* 1222 * Recovery results of poisoned raw pages handled by memory_failure, 1223 * in sync with mf_result. 1224 * total = ignored + failed + delayed + recovered. 1225 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted. 1226 */ 1227 unsigned long ignored; 1228 unsigned long failed; 1229 unsigned long delayed; 1230 unsigned long recovered; 1231 }; 1232 #endif 1233 1234 /* 1235 * On NUMA machines, each NUMA node would have a pg_data_t to describe 1236 * it's memory layout. On UMA machines there is a single pglist_data which 1237 * describes the whole memory. 1238 * 1239 * Memory statistics and page replacement data structures are maintained on a 1240 * per-zone basis. 1241 */ 1242 typedef struct pglist_data { 1243 /* 1244 * node_zones contains just the zones for THIS node. Not all of the 1245 * zones may be populated, but it is the full list. It is referenced by 1246 * this node's node_zonelists as well as other node's node_zonelists. 1247 */ 1248 struct zone node_zones[MAX_NR_ZONES]; 1249 1250 /* 1251 * node_zonelists contains references to all zones in all nodes. 1252 * Generally the first zones will be references to this node's 1253 * node_zones. 1254 */ 1255 struct zonelist node_zonelists[MAX_ZONELISTS]; 1256 1257 int nr_zones; /* number of populated zones in this node */ 1258 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */ 1259 struct page *node_mem_map; 1260 #ifdef CONFIG_PAGE_EXTENSION 1261 struct page_ext *node_page_ext; 1262 #endif 1263 #endif 1264 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT) 1265 /* 1266 * Must be held any time you expect node_start_pfn, 1267 * node_present_pages, node_spanned_pages or nr_zones to stay constant. 1268 * Also synchronizes pgdat->first_deferred_pfn during deferred page 1269 * init. 1270 * 1271 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to 1272 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG 1273 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT. 1274 * 1275 * Nests above zone->lock and zone->span_seqlock 1276 */ 1277 spinlock_t node_size_lock; 1278 #endif 1279 unsigned long node_start_pfn; 1280 unsigned long node_present_pages; /* total number of physical pages */ 1281 unsigned long node_spanned_pages; /* total size of physical page 1282 range, including holes */ 1283 int node_id; 1284 wait_queue_head_t kswapd_wait; 1285 wait_queue_head_t pfmemalloc_wait; 1286 1287 /* workqueues for throttling reclaim for different reasons. */ 1288 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE]; 1289 1290 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */ 1291 unsigned long nr_reclaim_start; /* nr pages written while throttled 1292 * when throttling started. */ 1293 #ifdef CONFIG_MEMORY_HOTPLUG 1294 struct mutex kswapd_lock; 1295 #endif 1296 struct task_struct *kswapd; /* Protected by kswapd_lock */ 1297 int kswapd_order; 1298 enum zone_type kswapd_highest_zoneidx; 1299 1300 int kswapd_failures; /* Number of 'reclaimed == 0' runs */ 1301 1302 #ifdef CONFIG_COMPACTION 1303 int kcompactd_max_order; 1304 enum zone_type kcompactd_highest_zoneidx; 1305 wait_queue_head_t kcompactd_wait; 1306 struct task_struct *kcompactd; 1307 bool proactive_compact_trigger; 1308 #endif 1309 /* 1310 * This is a per-node reserve of pages that are not available 1311 * to userspace allocations. 1312 */ 1313 unsigned long totalreserve_pages; 1314 1315 #ifdef CONFIG_NUMA 1316 /* 1317 * node reclaim becomes active if more unmapped pages exist. 1318 */ 1319 unsigned long min_unmapped_pages; 1320 unsigned long min_slab_pages; 1321 #endif /* CONFIG_NUMA */ 1322 1323 /* Write-intensive fields used by page reclaim */ 1324 CACHELINE_PADDING(_pad1_); 1325 1326 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1327 /* 1328 * If memory initialisation on large machines is deferred then this 1329 * is the first PFN that needs to be initialised. 1330 */ 1331 unsigned long first_deferred_pfn; 1332 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1333 1334 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1335 struct deferred_split deferred_split_queue; 1336 #endif 1337 1338 #ifdef CONFIG_NUMA_BALANCING 1339 /* start time in ms of current promote rate limit period */ 1340 unsigned int nbp_rl_start; 1341 /* number of promote candidate pages at start time of current rate limit period */ 1342 unsigned long nbp_rl_nr_cand; 1343 /* promote threshold in ms */ 1344 unsigned int nbp_threshold; 1345 /* start time in ms of current promote threshold adjustment period */ 1346 unsigned int nbp_th_start; 1347 /* 1348 * number of promote candidate pages at start time of current promote 1349 * threshold adjustment period 1350 */ 1351 unsigned long nbp_th_nr_cand; 1352 #endif 1353 /* Fields commonly accessed by the page reclaim scanner */ 1354 1355 /* 1356 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED. 1357 * 1358 * Use mem_cgroup_lruvec() to look up lruvecs. 1359 */ 1360 struct lruvec __lruvec; 1361 1362 unsigned long flags; 1363 1364 #ifdef CONFIG_LRU_GEN 1365 /* kswap mm walk data */ 1366 struct lru_gen_mm_walk mm_walk; 1367 /* lru_gen_folio list */ 1368 struct lru_gen_memcg memcg_lru; 1369 #endif 1370 1371 CACHELINE_PADDING(_pad2_); 1372 1373 /* Per-node vmstats */ 1374 struct per_cpu_nodestat __percpu *per_cpu_nodestats; 1375 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS]; 1376 #ifdef CONFIG_NUMA 1377 struct memory_tier __rcu *memtier; 1378 #endif 1379 #ifdef CONFIG_MEMORY_FAILURE 1380 struct memory_failure_stats mf_stats; 1381 #endif 1382 } pg_data_t; 1383 1384 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) 1385 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) 1386 1387 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) 1388 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) 1389 1390 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) 1391 { 1392 return pgdat->node_start_pfn + pgdat->node_spanned_pages; 1393 } 1394 1395 #include <linux/memory_hotplug.h> 1396 1397 void build_all_zonelists(pg_data_t *pgdat); 1398 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order, 1399 enum zone_type highest_zoneidx); 1400 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, 1401 int highest_zoneidx, unsigned int alloc_flags, 1402 long free_pages); 1403 bool zone_watermark_ok(struct zone *z, unsigned int order, 1404 unsigned long mark, int highest_zoneidx, 1405 unsigned int alloc_flags); 1406 bool zone_watermark_ok_safe(struct zone *z, unsigned int order, 1407 unsigned long mark, int highest_zoneidx); 1408 /* 1409 * Memory initialization context, use to differentiate memory added by 1410 * the platform statically or via memory hotplug interface. 1411 */ 1412 enum meminit_context { 1413 MEMINIT_EARLY, 1414 MEMINIT_HOTPLUG, 1415 }; 1416 1417 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, 1418 unsigned long size); 1419 1420 extern void lruvec_init(struct lruvec *lruvec); 1421 1422 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec) 1423 { 1424 #ifdef CONFIG_MEMCG 1425 return lruvec->pgdat; 1426 #else 1427 return container_of(lruvec, struct pglist_data, __lruvec); 1428 #endif 1429 } 1430 1431 #ifdef CONFIG_HAVE_MEMORYLESS_NODES 1432 int local_memory_node(int node_id); 1433 #else 1434 static inline int local_memory_node(int node_id) { return node_id; }; 1435 #endif 1436 1437 /* 1438 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. 1439 */ 1440 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) 1441 1442 #ifdef CONFIG_ZONE_DEVICE 1443 static inline bool zone_is_zone_device(struct zone *zone) 1444 { 1445 return zone_idx(zone) == ZONE_DEVICE; 1446 } 1447 #else 1448 static inline bool zone_is_zone_device(struct zone *zone) 1449 { 1450 return false; 1451 } 1452 #endif 1453 1454 /* 1455 * Returns true if a zone has pages managed by the buddy allocator. 1456 * All the reclaim decisions have to use this function rather than 1457 * populated_zone(). If the whole zone is reserved then we can easily 1458 * end up with populated_zone() && !managed_zone(). 1459 */ 1460 static inline bool managed_zone(struct zone *zone) 1461 { 1462 return zone_managed_pages(zone); 1463 } 1464 1465 /* Returns true if a zone has memory */ 1466 static inline bool populated_zone(struct zone *zone) 1467 { 1468 return zone->present_pages; 1469 } 1470 1471 #ifdef CONFIG_NUMA 1472 static inline int zone_to_nid(struct zone *zone) 1473 { 1474 return zone->node; 1475 } 1476 1477 static inline void zone_set_nid(struct zone *zone, int nid) 1478 { 1479 zone->node = nid; 1480 } 1481 #else 1482 static inline int zone_to_nid(struct zone *zone) 1483 { 1484 return 0; 1485 } 1486 1487 static inline void zone_set_nid(struct zone *zone, int nid) {} 1488 #endif 1489 1490 extern int movable_zone; 1491 1492 static inline int is_highmem_idx(enum zone_type idx) 1493 { 1494 #ifdef CONFIG_HIGHMEM 1495 return (idx == ZONE_HIGHMEM || 1496 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM)); 1497 #else 1498 return 0; 1499 #endif 1500 } 1501 1502 /** 1503 * is_highmem - helper function to quickly check if a struct zone is a 1504 * highmem zone or not. This is an attempt to keep references 1505 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. 1506 * @zone: pointer to struct zone variable 1507 * Return: 1 for a highmem zone, 0 otherwise 1508 */ 1509 static inline int is_highmem(struct zone *zone) 1510 { 1511 return is_highmem_idx(zone_idx(zone)); 1512 } 1513 1514 #ifdef CONFIG_ZONE_DMA 1515 bool has_managed_dma(void); 1516 #else 1517 static inline bool has_managed_dma(void) 1518 { 1519 return false; 1520 } 1521 #endif 1522 1523 /* These two functions are used to setup the per zone pages min values */ 1524 struct ctl_table; 1525 1526 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *, 1527 loff_t *); 1528 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *, 1529 size_t *, loff_t *); 1530 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES]; 1531 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *, 1532 size_t *, loff_t *); 1533 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int, 1534 void *, size_t *, loff_t *); 1535 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, 1536 void *, size_t *, loff_t *); 1537 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, 1538 void *, size_t *, loff_t *); 1539 int numa_zonelist_order_handler(struct ctl_table *, int, 1540 void *, size_t *, loff_t *); 1541 extern int percpu_pagelist_high_fraction; 1542 extern char numa_zonelist_order[]; 1543 #define NUMA_ZONELIST_ORDER_LEN 16 1544 1545 #ifndef CONFIG_NUMA 1546 1547 extern struct pglist_data contig_page_data; 1548 static inline struct pglist_data *NODE_DATA(int nid) 1549 { 1550 return &contig_page_data; 1551 } 1552 1553 #else /* CONFIG_NUMA */ 1554 1555 #include <asm/mmzone.h> 1556 1557 #endif /* !CONFIG_NUMA */ 1558 1559 extern struct pglist_data *first_online_pgdat(void); 1560 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); 1561 extern struct zone *next_zone(struct zone *zone); 1562 1563 /** 1564 * for_each_online_pgdat - helper macro to iterate over all online nodes 1565 * @pgdat: pointer to a pg_data_t variable 1566 */ 1567 #define for_each_online_pgdat(pgdat) \ 1568 for (pgdat = first_online_pgdat(); \ 1569 pgdat; \ 1570 pgdat = next_online_pgdat(pgdat)) 1571 /** 1572 * for_each_zone - helper macro to iterate over all memory zones 1573 * @zone: pointer to struct zone variable 1574 * 1575 * The user only needs to declare the zone variable, for_each_zone 1576 * fills it in. 1577 */ 1578 #define for_each_zone(zone) \ 1579 for (zone = (first_online_pgdat())->node_zones; \ 1580 zone; \ 1581 zone = next_zone(zone)) 1582 1583 #define for_each_populated_zone(zone) \ 1584 for (zone = (first_online_pgdat())->node_zones; \ 1585 zone; \ 1586 zone = next_zone(zone)) \ 1587 if (!populated_zone(zone)) \ 1588 ; /* do nothing */ \ 1589 else 1590 1591 static inline struct zone *zonelist_zone(struct zoneref *zoneref) 1592 { 1593 return zoneref->zone; 1594 } 1595 1596 static inline int zonelist_zone_idx(struct zoneref *zoneref) 1597 { 1598 return zoneref->zone_idx; 1599 } 1600 1601 static inline int zonelist_node_idx(struct zoneref *zoneref) 1602 { 1603 return zone_to_nid(zoneref->zone); 1604 } 1605 1606 struct zoneref *__next_zones_zonelist(struct zoneref *z, 1607 enum zone_type highest_zoneidx, 1608 nodemask_t *nodes); 1609 1610 /** 1611 * 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 1612 * @z: The cursor used as a starting point for the search 1613 * @highest_zoneidx: The zone index of the highest zone to return 1614 * @nodes: An optional nodemask to filter the zonelist with 1615 * 1616 * This function returns the next zone at or below a given zone index that is 1617 * within the allowed nodemask using a cursor as the starting point for the 1618 * search. The zoneref returned is a cursor that represents the current zone 1619 * being examined. It should be advanced by one before calling 1620 * next_zones_zonelist again. 1621 * 1622 * Return: the next zone at or below highest_zoneidx within the allowed 1623 * nodemask using a cursor within a zonelist as a starting point 1624 */ 1625 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z, 1626 enum zone_type highest_zoneidx, 1627 nodemask_t *nodes) 1628 { 1629 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx)) 1630 return z; 1631 return __next_zones_zonelist(z, highest_zoneidx, nodes); 1632 } 1633 1634 /** 1635 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist 1636 * @zonelist: The zonelist to search for a suitable zone 1637 * @highest_zoneidx: The zone index of the highest zone to return 1638 * @nodes: An optional nodemask to filter the zonelist with 1639 * 1640 * This function returns the first zone at or below a given zone index that is 1641 * within the allowed nodemask. The zoneref returned is a cursor that can be 1642 * used to iterate the zonelist with next_zones_zonelist by advancing it by 1643 * one before calling. 1644 * 1645 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is 1646 * never NULL). This may happen either genuinely, or due to concurrent nodemask 1647 * update due to cpuset modification. 1648 * 1649 * Return: Zoneref pointer for the first suitable zone found 1650 */ 1651 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, 1652 enum zone_type highest_zoneidx, 1653 nodemask_t *nodes) 1654 { 1655 return next_zones_zonelist(zonelist->_zonerefs, 1656 highest_zoneidx, nodes); 1657 } 1658 1659 /** 1660 * 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 1661 * @zone: The current zone in the iterator 1662 * @z: The current pointer within zonelist->_zonerefs being iterated 1663 * @zlist: The zonelist being iterated 1664 * @highidx: The zone index of the highest zone to return 1665 * @nodemask: Nodemask allowed by the allocator 1666 * 1667 * This iterator iterates though all zones at or below a given zone index and 1668 * within a given nodemask 1669 */ 1670 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ 1671 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \ 1672 zone; \ 1673 z = next_zones_zonelist(++z, highidx, nodemask), \ 1674 zone = zonelist_zone(z)) 1675 1676 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \ 1677 for (zone = z->zone; \ 1678 zone; \ 1679 z = next_zones_zonelist(++z, highidx, nodemask), \ 1680 zone = zonelist_zone(z)) 1681 1682 1683 /** 1684 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index 1685 * @zone: The current zone in the iterator 1686 * @z: The current pointer within zonelist->zones being iterated 1687 * @zlist: The zonelist being iterated 1688 * @highidx: The zone index of the highest zone to return 1689 * 1690 * This iterator iterates though all zones at or below a given zone index. 1691 */ 1692 #define for_each_zone_zonelist(zone, z, zlist, highidx) \ 1693 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) 1694 1695 /* Whether the 'nodes' are all movable nodes */ 1696 static inline bool movable_only_nodes(nodemask_t *nodes) 1697 { 1698 struct zonelist *zonelist; 1699 struct zoneref *z; 1700 int nid; 1701 1702 if (nodes_empty(*nodes)) 1703 return false; 1704 1705 /* 1706 * We can chose arbitrary node from the nodemask to get a 1707 * zonelist as they are interlinked. We just need to find 1708 * at least one zone that can satisfy kernel allocations. 1709 */ 1710 nid = first_node(*nodes); 1711 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK]; 1712 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes); 1713 return (!z->zone) ? true : false; 1714 } 1715 1716 1717 #ifdef CONFIG_SPARSEMEM 1718 #include <asm/sparsemem.h> 1719 #endif 1720 1721 #ifdef CONFIG_FLATMEM 1722 #define pfn_to_nid(pfn) (0) 1723 #endif 1724 1725 #ifdef CONFIG_SPARSEMEM 1726 1727 /* 1728 * PA_SECTION_SHIFT physical address to/from section number 1729 * PFN_SECTION_SHIFT pfn to/from section number 1730 */ 1731 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) 1732 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) 1733 1734 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) 1735 1736 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) 1737 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) 1738 1739 #define SECTION_BLOCKFLAGS_BITS \ 1740 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) 1741 1742 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS 1743 #error Allocator MAX_ORDER exceeds SECTION_SIZE 1744 #endif 1745 1746 static inline unsigned long pfn_to_section_nr(unsigned long pfn) 1747 { 1748 return pfn >> PFN_SECTION_SHIFT; 1749 } 1750 static inline unsigned long section_nr_to_pfn(unsigned long sec) 1751 { 1752 return sec << PFN_SECTION_SHIFT; 1753 } 1754 1755 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) 1756 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) 1757 1758 #define SUBSECTION_SHIFT 21 1759 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT) 1760 1761 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT) 1762 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT) 1763 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1)) 1764 1765 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS 1766 #error Subsection size exceeds section size 1767 #else 1768 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT)) 1769 #endif 1770 1771 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION) 1772 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK) 1773 1774 struct mem_section_usage { 1775 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1776 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION); 1777 #endif 1778 /* See declaration of similar field in struct zone */ 1779 unsigned long pageblock_flags[0]; 1780 }; 1781 1782 void subsection_map_init(unsigned long pfn, unsigned long nr_pages); 1783 1784 struct page; 1785 struct page_ext; 1786 struct mem_section { 1787 /* 1788 * This is, logically, a pointer to an array of struct 1789 * pages. However, it is stored with some other magic. 1790 * (see sparse.c::sparse_init_one_section()) 1791 * 1792 * Additionally during early boot we encode node id of 1793 * the location of the section here to guide allocation. 1794 * (see sparse.c::memory_present()) 1795 * 1796 * Making it a UL at least makes someone do a cast 1797 * before using it wrong. 1798 */ 1799 unsigned long section_mem_map; 1800 1801 struct mem_section_usage *usage; 1802 #ifdef CONFIG_PAGE_EXTENSION 1803 /* 1804 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use 1805 * section. (see page_ext.h about this.) 1806 */ 1807 struct page_ext *page_ext; 1808 unsigned long pad; 1809 #endif 1810 /* 1811 * WARNING: mem_section must be a power-of-2 in size for the 1812 * calculation and use of SECTION_ROOT_MASK to make sense. 1813 */ 1814 }; 1815 1816 #ifdef CONFIG_SPARSEMEM_EXTREME 1817 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) 1818 #else 1819 #define SECTIONS_PER_ROOT 1 1820 #endif 1821 1822 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) 1823 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) 1824 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) 1825 1826 #ifdef CONFIG_SPARSEMEM_EXTREME 1827 extern struct mem_section **mem_section; 1828 #else 1829 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; 1830 #endif 1831 1832 static inline unsigned long *section_to_usemap(struct mem_section *ms) 1833 { 1834 return ms->usage->pageblock_flags; 1835 } 1836 1837 static inline struct mem_section *__nr_to_section(unsigned long nr) 1838 { 1839 unsigned long root = SECTION_NR_TO_ROOT(nr); 1840 1841 if (unlikely(root >= NR_SECTION_ROOTS)) 1842 return NULL; 1843 1844 #ifdef CONFIG_SPARSEMEM_EXTREME 1845 if (!mem_section || !mem_section[root]) 1846 return NULL; 1847 #endif 1848 return &mem_section[root][nr & SECTION_ROOT_MASK]; 1849 } 1850 extern size_t mem_section_usage_size(void); 1851 1852 /* 1853 * We use the lower bits of the mem_map pointer to store 1854 * a little bit of information. The pointer is calculated 1855 * as mem_map - section_nr_to_pfn(pnum). The result is 1856 * aligned to the minimum alignment of the two values: 1857 * 1. All mem_map arrays are page-aligned. 1858 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT 1859 * lowest bits. PFN_SECTION_SHIFT is arch-specific 1860 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the 1861 * worst combination is powerpc with 256k pages, 1862 * which results in PFN_SECTION_SHIFT equal 6. 1863 * To sum it up, at least 6 bits are available on all architectures. 1864 * However, we can exceed 6 bits on some other architectures except 1865 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available 1866 * with the worst case of 64K pages on arm64) if we make sure the 1867 * exceeded bit is not applicable to powerpc. 1868 */ 1869 enum { 1870 SECTION_MARKED_PRESENT_BIT, 1871 SECTION_HAS_MEM_MAP_BIT, 1872 SECTION_IS_ONLINE_BIT, 1873 SECTION_IS_EARLY_BIT, 1874 #ifdef CONFIG_ZONE_DEVICE 1875 SECTION_TAINT_ZONE_DEVICE_BIT, 1876 #endif 1877 SECTION_MAP_LAST_BIT, 1878 }; 1879 1880 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT) 1881 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT) 1882 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT) 1883 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT) 1884 #ifdef CONFIG_ZONE_DEVICE 1885 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT) 1886 #endif 1887 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1)) 1888 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT 1889 1890 static inline struct page *__section_mem_map_addr(struct mem_section *section) 1891 { 1892 unsigned long map = section->section_mem_map; 1893 map &= SECTION_MAP_MASK; 1894 return (struct page *)map; 1895 } 1896 1897 static inline int present_section(struct mem_section *section) 1898 { 1899 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); 1900 } 1901 1902 static inline int present_section_nr(unsigned long nr) 1903 { 1904 return present_section(__nr_to_section(nr)); 1905 } 1906 1907 static inline int valid_section(struct mem_section *section) 1908 { 1909 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); 1910 } 1911 1912 static inline int early_section(struct mem_section *section) 1913 { 1914 return (section && (section->section_mem_map & SECTION_IS_EARLY)); 1915 } 1916 1917 static inline int valid_section_nr(unsigned long nr) 1918 { 1919 return valid_section(__nr_to_section(nr)); 1920 } 1921 1922 static inline int online_section(struct mem_section *section) 1923 { 1924 return (section && (section->section_mem_map & SECTION_IS_ONLINE)); 1925 } 1926 1927 #ifdef CONFIG_ZONE_DEVICE 1928 static inline int online_device_section(struct mem_section *section) 1929 { 1930 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE; 1931 1932 return section && ((section->section_mem_map & flags) == flags); 1933 } 1934 #else 1935 static inline int online_device_section(struct mem_section *section) 1936 { 1937 return 0; 1938 } 1939 #endif 1940 1941 static inline int online_section_nr(unsigned long nr) 1942 { 1943 return online_section(__nr_to_section(nr)); 1944 } 1945 1946 #ifdef CONFIG_MEMORY_HOTPLUG 1947 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1948 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1949 #endif 1950 1951 static inline struct mem_section *__pfn_to_section(unsigned long pfn) 1952 { 1953 return __nr_to_section(pfn_to_section_nr(pfn)); 1954 } 1955 1956 extern unsigned long __highest_present_section_nr; 1957 1958 static inline int subsection_map_index(unsigned long pfn) 1959 { 1960 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION; 1961 } 1962 1963 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1964 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1965 { 1966 int idx = subsection_map_index(pfn); 1967 1968 return test_bit(idx, ms->usage->subsection_map); 1969 } 1970 #else 1971 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1972 { 1973 return 1; 1974 } 1975 #endif 1976 1977 #ifndef CONFIG_HAVE_ARCH_PFN_VALID 1978 /** 1979 * pfn_valid - check if there is a valid memory map entry for a PFN 1980 * @pfn: the page frame number to check 1981 * 1982 * Check if there is a valid memory map entry aka struct page for the @pfn. 1983 * Note, that availability of the memory map entry does not imply that 1984 * there is actual usable memory at that @pfn. The struct page may 1985 * represent a hole or an unusable page frame. 1986 * 1987 * Return: 1 for PFNs that have memory map entries and 0 otherwise 1988 */ 1989 static inline int pfn_valid(unsigned long pfn) 1990 { 1991 struct mem_section *ms; 1992 1993 /* 1994 * Ensure the upper PAGE_SHIFT bits are clear in the 1995 * pfn. Else it might lead to false positives when 1996 * some of the upper bits are set, but the lower bits 1997 * match a valid pfn. 1998 */ 1999 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn) 2000 return 0; 2001 2002 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 2003 return 0; 2004 ms = __pfn_to_section(pfn); 2005 if (!valid_section(ms)) 2006 return 0; 2007 /* 2008 * Traditionally early sections always returned pfn_valid() for 2009 * the entire section-sized span. 2010 */ 2011 return early_section(ms) || pfn_section_valid(ms, pfn); 2012 } 2013 #endif 2014 2015 static inline int pfn_in_present_section(unsigned long pfn) 2016 { 2017 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 2018 return 0; 2019 return present_section(__pfn_to_section(pfn)); 2020 } 2021 2022 static inline unsigned long next_present_section_nr(unsigned long section_nr) 2023 { 2024 while (++section_nr <= __highest_present_section_nr) { 2025 if (present_section_nr(section_nr)) 2026 return section_nr; 2027 } 2028 2029 return -1; 2030 } 2031 2032 /* 2033 * These are _only_ used during initialisation, therefore they 2034 * can use __initdata ... They could have names to indicate 2035 * this restriction. 2036 */ 2037 #ifdef CONFIG_NUMA 2038 #define pfn_to_nid(pfn) \ 2039 ({ \ 2040 unsigned long __pfn_to_nid_pfn = (pfn); \ 2041 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ 2042 }) 2043 #else 2044 #define pfn_to_nid(pfn) (0) 2045 #endif 2046 2047 void sparse_init(void); 2048 #else 2049 #define sparse_init() do {} while (0) 2050 #define sparse_index_init(_sec, _nid) do {} while (0) 2051 #define pfn_in_present_section pfn_valid 2052 #define subsection_map_init(_pfn, _nr_pages) do {} while (0) 2053 #endif /* CONFIG_SPARSEMEM */ 2054 2055 #endif /* !__GENERATING_BOUNDS.H */ 2056 #endif /* !__ASSEMBLY__ */ 2057 #endif /* _LINUX_MMZONE_H */ 2058