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/wait.h> 11 #include <linux/bitops.h> 12 #include <linux/cache.h> 13 #include <linux/threads.h> 14 #include <linux/numa.h> 15 #include <linux/init.h> 16 #include <linux/seqlock.h> 17 #include <linux/nodemask.h> 18 #include <linux/pageblock-flags.h> 19 #include <linux/page-flags-layout.h> 20 #include <linux/atomic.h> 21 #include <linux/mm_types.h> 22 #include <linux/page-flags.h> 23 #include <asm/page.h> 24 25 /* Free memory management - zoned buddy allocator. */ 26 #ifndef CONFIG_FORCE_MAX_ZONEORDER 27 #define MAX_ORDER 11 28 #else 29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER 30 #endif 31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1)) 32 33 /* 34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed 35 * costly to service. That is between allocation orders which should 36 * coalesce naturally under reasonable reclaim pressure and those which 37 * will not. 38 */ 39 #define PAGE_ALLOC_COSTLY_ORDER 3 40 41 enum migratetype { 42 MIGRATE_UNMOVABLE, 43 MIGRATE_MOVABLE, 44 MIGRATE_RECLAIMABLE, 45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ 46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, 47 #ifdef CONFIG_CMA 48 /* 49 * MIGRATE_CMA migration type is designed to mimic the way 50 * ZONE_MOVABLE works. Only movable pages can be allocated 51 * from MIGRATE_CMA pageblocks and page allocator never 52 * implicitly change migration type of MIGRATE_CMA pageblock. 53 * 54 * The way to use it is to change migratetype of a range of 55 * pageblocks to MIGRATE_CMA which can be done by 56 * __free_pageblock_cma() function. What is important though 57 * is that a range of pageblocks must be aligned to 58 * MAX_ORDER_NR_PAGES should biggest page be bigger then 59 * a single pageblock. 60 */ 61 MIGRATE_CMA, 62 #endif 63 #ifdef CONFIG_MEMORY_ISOLATION 64 MIGRATE_ISOLATE, /* can't allocate from here */ 65 #endif 66 MIGRATE_TYPES 67 }; 68 69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */ 70 extern const char * const migratetype_names[MIGRATE_TYPES]; 71 72 #ifdef CONFIG_CMA 73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) 74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA) 75 #else 76 # define is_migrate_cma(migratetype) false 77 # define is_migrate_cma_page(_page) false 78 #endif 79 80 static inline bool is_migrate_movable(int mt) 81 { 82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE; 83 } 84 85 #define for_each_migratetype_order(order, type) \ 86 for (order = 0; order < MAX_ORDER; order++) \ 87 for (type = 0; type < MIGRATE_TYPES; type++) 88 89 extern int page_group_by_mobility_disabled; 90 91 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1) 92 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1) 93 94 #define get_pageblock_migratetype(page) \ 95 get_pfnblock_flags_mask(page, page_to_pfn(page), \ 96 PB_migrate_end, MIGRATETYPE_MASK) 97 98 struct free_area { 99 struct list_head free_list[MIGRATE_TYPES]; 100 unsigned long nr_free; 101 }; 102 103 /* Used for pages not on another list */ 104 static inline void add_to_free_area(struct page *page, struct free_area *area, 105 int migratetype) 106 { 107 list_add(&page->lru, &area->free_list[migratetype]); 108 area->nr_free++; 109 } 110 111 /* Used for pages not on another list */ 112 static inline void add_to_free_area_tail(struct page *page, struct free_area *area, 113 int migratetype) 114 { 115 list_add_tail(&page->lru, &area->free_list[migratetype]); 116 area->nr_free++; 117 } 118 119 #ifdef CONFIG_SHUFFLE_PAGE_ALLOCATOR 120 /* Used to preserve page allocation order entropy */ 121 void add_to_free_area_random(struct page *page, struct free_area *area, 122 int migratetype); 123 #else 124 static inline void add_to_free_area_random(struct page *page, 125 struct free_area *area, int migratetype) 126 { 127 add_to_free_area(page, area, migratetype); 128 } 129 #endif 130 131 /* Used for pages which are on another list */ 132 static inline void move_to_free_area(struct page *page, struct free_area *area, 133 int migratetype) 134 { 135 list_move(&page->lru, &area->free_list[migratetype]); 136 } 137 138 static inline struct page *get_page_from_free_area(struct free_area *area, 139 int migratetype) 140 { 141 return list_first_entry_or_null(&area->free_list[migratetype], 142 struct page, lru); 143 } 144 145 static inline void del_page_from_free_area(struct page *page, 146 struct free_area *area) 147 { 148 list_del(&page->lru); 149 __ClearPageBuddy(page); 150 set_page_private(page, 0); 151 area->nr_free--; 152 } 153 154 static inline bool free_area_empty(struct free_area *area, int migratetype) 155 { 156 return list_empty(&area->free_list[migratetype]); 157 } 158 159 struct pglist_data; 160 161 /* 162 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel. 163 * So add a wild amount of padding here to ensure that they fall into separate 164 * cachelines. There are very few zone structures in the machine, so space 165 * consumption is not a concern here. 166 */ 167 #if defined(CONFIG_SMP) 168 struct zone_padding { 169 char x[0]; 170 } ____cacheline_internodealigned_in_smp; 171 #define ZONE_PADDING(name) struct zone_padding name; 172 #else 173 #define ZONE_PADDING(name) 174 #endif 175 176 #ifdef CONFIG_NUMA 177 enum numa_stat_item { 178 NUMA_HIT, /* allocated in intended node */ 179 NUMA_MISS, /* allocated in non intended node */ 180 NUMA_FOREIGN, /* was intended here, hit elsewhere */ 181 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ 182 NUMA_LOCAL, /* allocation from local node */ 183 NUMA_OTHER, /* allocation from other node */ 184 NR_VM_NUMA_STAT_ITEMS 185 }; 186 #else 187 #define NR_VM_NUMA_STAT_ITEMS 0 188 #endif 189 190 enum zone_stat_item { 191 /* First 128 byte cacheline (assuming 64 bit words) */ 192 NR_FREE_PAGES, 193 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ 194 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, 195 NR_ZONE_ACTIVE_ANON, 196 NR_ZONE_INACTIVE_FILE, 197 NR_ZONE_ACTIVE_FILE, 198 NR_ZONE_UNEVICTABLE, 199 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */ 200 NR_MLOCK, /* mlock()ed pages found and moved off LRU */ 201 NR_PAGETABLE, /* used for pagetables */ 202 NR_KERNEL_STACK_KB, /* measured in KiB */ 203 /* Second 128 byte cacheline */ 204 NR_BOUNCE, 205 #if IS_ENABLED(CONFIG_ZSMALLOC) 206 NR_ZSPAGES, /* allocated in zsmalloc */ 207 #endif 208 NR_FREE_CMA_PAGES, 209 NR_VM_ZONE_STAT_ITEMS }; 210 211 enum node_stat_item { 212 NR_LRU_BASE, 213 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ 214 NR_ACTIVE_ANON, /* " " " " " */ 215 NR_INACTIVE_FILE, /* " " " " " */ 216 NR_ACTIVE_FILE, /* " " " " " */ 217 NR_UNEVICTABLE, /* " " " " " */ 218 NR_SLAB_RECLAIMABLE, 219 NR_SLAB_UNRECLAIMABLE, 220 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ 221 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ 222 WORKINGSET_NODES, 223 WORKINGSET_REFAULT, 224 WORKINGSET_ACTIVATE, 225 WORKINGSET_RESTORE, 226 WORKINGSET_NODERECLAIM, 227 NR_ANON_MAPPED, /* Mapped anonymous pages */ 228 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. 229 only modified from process context */ 230 NR_FILE_PAGES, 231 NR_FILE_DIRTY, 232 NR_WRITEBACK, 233 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ 234 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ 235 NR_SHMEM_THPS, 236 NR_SHMEM_PMDMAPPED, 237 NR_FILE_THPS, 238 NR_FILE_PMDMAPPED, 239 NR_ANON_THPS, 240 NR_UNSTABLE_NFS, /* NFS unstable pages */ 241 NR_VMSCAN_WRITE, 242 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ 243 NR_DIRTIED, /* page dirtyings since bootup */ 244 NR_WRITTEN, /* page writings since bootup */ 245 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */ 246 NR_VM_NODE_STAT_ITEMS 247 }; 248 249 /* 250 * We do arithmetic on the LRU lists in various places in the code, 251 * so it is important to keep the active lists LRU_ACTIVE higher in 252 * the array than the corresponding inactive lists, and to keep 253 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. 254 * 255 * This has to be kept in sync with the statistics in zone_stat_item 256 * above and the descriptions in vmstat_text in mm/vmstat.c 257 */ 258 #define LRU_BASE 0 259 #define LRU_ACTIVE 1 260 #define LRU_FILE 2 261 262 enum lru_list { 263 LRU_INACTIVE_ANON = LRU_BASE, 264 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, 265 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, 266 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, 267 LRU_UNEVICTABLE, 268 NR_LRU_LISTS 269 }; 270 271 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) 272 273 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) 274 275 static inline bool is_file_lru(enum lru_list lru) 276 { 277 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); 278 } 279 280 static inline bool is_active_lru(enum lru_list lru) 281 { 282 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); 283 } 284 285 struct zone_reclaim_stat { 286 /* 287 * The pageout code in vmscan.c keeps track of how many of the 288 * mem/swap backed and file backed pages are referenced. 289 * The higher the rotated/scanned ratio, the more valuable 290 * that cache is. 291 * 292 * The anon LRU stats live in [0], file LRU stats in [1] 293 */ 294 unsigned long recent_rotated[2]; 295 unsigned long recent_scanned[2]; 296 }; 297 298 enum lruvec_flags { 299 LRUVEC_CONGESTED, /* lruvec has many dirty pages 300 * backed by a congested BDI 301 */ 302 }; 303 304 struct lruvec { 305 struct list_head lists[NR_LRU_LISTS]; 306 struct zone_reclaim_stat reclaim_stat; 307 /* Evictions & activations on the inactive file list */ 308 atomic_long_t inactive_age; 309 /* Refaults at the time of last reclaim cycle */ 310 unsigned long refaults; 311 /* Various lruvec state flags (enum lruvec_flags) */ 312 unsigned long flags; 313 #ifdef CONFIG_MEMCG 314 struct pglist_data *pgdat; 315 #endif 316 }; 317 318 /* Isolate unmapped pages */ 319 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) 320 /* Isolate for asynchronous migration */ 321 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) 322 /* Isolate unevictable pages */ 323 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) 324 325 /* LRU Isolation modes. */ 326 typedef unsigned __bitwise isolate_mode_t; 327 328 enum zone_watermarks { 329 WMARK_MIN, 330 WMARK_LOW, 331 WMARK_HIGH, 332 NR_WMARK 333 }; 334 335 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost) 336 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost) 337 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost) 338 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost) 339 340 struct per_cpu_pages { 341 int count; /* number of pages in the list */ 342 int high; /* high watermark, emptying needed */ 343 int batch; /* chunk size for buddy add/remove */ 344 345 /* Lists of pages, one per migrate type stored on the pcp-lists */ 346 struct list_head lists[MIGRATE_PCPTYPES]; 347 }; 348 349 struct per_cpu_pageset { 350 struct per_cpu_pages pcp; 351 #ifdef CONFIG_NUMA 352 s8 expire; 353 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS]; 354 #endif 355 #ifdef CONFIG_SMP 356 s8 stat_threshold; 357 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; 358 #endif 359 }; 360 361 struct per_cpu_nodestat { 362 s8 stat_threshold; 363 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS]; 364 }; 365 366 #endif /* !__GENERATING_BOUNDS.H */ 367 368 enum zone_type { 369 /* 370 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able 371 * to DMA to all of the addressable memory (ZONE_NORMAL). 372 * On architectures where this area covers the whole 32 bit address 373 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller 374 * DMA addressing constraints. This distinction is important as a 32bit 375 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit 376 * platforms may need both zones as they support peripherals with 377 * different DMA addressing limitations. 378 * 379 * Some examples: 380 * 381 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the 382 * rest of the lower 4G. 383 * 384 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on 385 * the specific device. 386 * 387 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the 388 * lower 4G. 389 * 390 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary 391 * depending on the specific device. 392 * 393 * - s390 uses ZONE_DMA fixed to the lower 2G. 394 * 395 * - ia64 and riscv only use ZONE_DMA32. 396 * 397 * - parisc uses neither. 398 */ 399 #ifdef CONFIG_ZONE_DMA 400 ZONE_DMA, 401 #endif 402 #ifdef CONFIG_ZONE_DMA32 403 ZONE_DMA32, 404 #endif 405 /* 406 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be 407 * performed on pages in ZONE_NORMAL if the DMA devices support 408 * transfers to all addressable memory. 409 */ 410 ZONE_NORMAL, 411 #ifdef CONFIG_HIGHMEM 412 /* 413 * A memory area that is only addressable by the kernel through 414 * mapping portions into its own address space. This is for example 415 * used by i386 to allow the kernel to address the memory beyond 416 * 900MB. The kernel will set up special mappings (page 417 * table entries on i386) for each page that the kernel needs to 418 * access. 419 */ 420 ZONE_HIGHMEM, 421 #endif 422 ZONE_MOVABLE, 423 #ifdef CONFIG_ZONE_DEVICE 424 ZONE_DEVICE, 425 #endif 426 __MAX_NR_ZONES 427 428 }; 429 430 #ifndef __GENERATING_BOUNDS_H 431 432 struct zone { 433 /* Read-mostly fields */ 434 435 /* zone watermarks, access with *_wmark_pages(zone) macros */ 436 unsigned long _watermark[NR_WMARK]; 437 unsigned long watermark_boost; 438 439 unsigned long nr_reserved_highatomic; 440 441 /* 442 * We don't know if the memory that we're going to allocate will be 443 * freeable or/and it will be released eventually, so to avoid totally 444 * wasting several GB of ram we must reserve some of the lower zone 445 * memory (otherwise we risk to run OOM on the lower zones despite 446 * there being tons of freeable ram on the higher zones). This array is 447 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl 448 * changes. 449 */ 450 long lowmem_reserve[MAX_NR_ZONES]; 451 452 #ifdef CONFIG_NUMA 453 int node; 454 #endif 455 struct pglist_data *zone_pgdat; 456 struct per_cpu_pageset __percpu *pageset; 457 458 #ifndef CONFIG_SPARSEMEM 459 /* 460 * Flags for a pageblock_nr_pages block. See pageblock-flags.h. 461 * In SPARSEMEM, this map is stored in struct mem_section 462 */ 463 unsigned long *pageblock_flags; 464 #endif /* CONFIG_SPARSEMEM */ 465 466 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ 467 unsigned long zone_start_pfn; 468 469 /* 470 * spanned_pages is the total pages spanned by the zone, including 471 * holes, which is calculated as: 472 * spanned_pages = zone_end_pfn - zone_start_pfn; 473 * 474 * present_pages is physical pages existing within the zone, which 475 * is calculated as: 476 * present_pages = spanned_pages - absent_pages(pages in holes); 477 * 478 * managed_pages is present pages managed by the buddy system, which 479 * is calculated as (reserved_pages includes pages allocated by the 480 * bootmem allocator): 481 * managed_pages = present_pages - reserved_pages; 482 * 483 * So present_pages may be used by memory hotplug or memory power 484 * management logic to figure out unmanaged pages by checking 485 * (present_pages - managed_pages). And managed_pages should be used 486 * by page allocator and vm scanner to calculate all kinds of watermarks 487 * and thresholds. 488 * 489 * Locking rules: 490 * 491 * zone_start_pfn and spanned_pages are protected by span_seqlock. 492 * It is a seqlock because it has to be read outside of zone->lock, 493 * and it is done in the main allocator path. But, it is written 494 * quite infrequently. 495 * 496 * The span_seq lock is declared along with zone->lock because it is 497 * frequently read in proximity to zone->lock. It's good to 498 * give them a chance of being in the same cacheline. 499 * 500 * Write access to present_pages at runtime should be protected by 501 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of 502 * present_pages should get_online_mems() to get a stable value. 503 */ 504 atomic_long_t managed_pages; 505 unsigned long spanned_pages; 506 unsigned long present_pages; 507 508 const char *name; 509 510 #ifdef CONFIG_MEMORY_ISOLATION 511 /* 512 * Number of isolated pageblock. It is used to solve incorrect 513 * freepage counting problem due to racy retrieving migratetype 514 * of pageblock. Protected by zone->lock. 515 */ 516 unsigned long nr_isolate_pageblock; 517 #endif 518 519 #ifdef CONFIG_MEMORY_HOTPLUG 520 /* see spanned/present_pages for more description */ 521 seqlock_t span_seqlock; 522 #endif 523 524 int initialized; 525 526 /* Write-intensive fields used from the page allocator */ 527 ZONE_PADDING(_pad1_) 528 529 /* free areas of different sizes */ 530 struct free_area free_area[MAX_ORDER]; 531 532 /* zone flags, see below */ 533 unsigned long flags; 534 535 /* Primarily protects free_area */ 536 spinlock_t lock; 537 538 /* Write-intensive fields used by compaction and vmstats. */ 539 ZONE_PADDING(_pad2_) 540 541 /* 542 * When free pages are below this point, additional steps are taken 543 * when reading the number of free pages to avoid per-cpu counter 544 * drift allowing watermarks to be breached 545 */ 546 unsigned long percpu_drift_mark; 547 548 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 549 /* pfn where compaction free scanner should start */ 550 unsigned long compact_cached_free_pfn; 551 /* pfn where async and sync compaction migration scanner should start */ 552 unsigned long compact_cached_migrate_pfn[2]; 553 unsigned long compact_init_migrate_pfn; 554 unsigned long compact_init_free_pfn; 555 #endif 556 557 #ifdef CONFIG_COMPACTION 558 /* 559 * On compaction failure, 1<<compact_defer_shift compactions 560 * are skipped before trying again. The number attempted since 561 * last failure is tracked with compact_considered. 562 */ 563 unsigned int compact_considered; 564 unsigned int compact_defer_shift; 565 int compact_order_failed; 566 #endif 567 568 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 569 /* Set to true when the PG_migrate_skip bits should be cleared */ 570 bool compact_blockskip_flush; 571 #endif 572 573 bool contiguous; 574 575 ZONE_PADDING(_pad3_) 576 /* Zone statistics */ 577 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; 578 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS]; 579 } ____cacheline_internodealigned_in_smp; 580 581 enum pgdat_flags { 582 PGDAT_DIRTY, /* reclaim scanning has recently found 583 * many dirty file pages at the tail 584 * of the LRU. 585 */ 586 PGDAT_WRITEBACK, /* reclaim scanning has recently found 587 * many pages under writeback 588 */ 589 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */ 590 }; 591 592 enum zone_flags { 593 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks. 594 * Cleared when kswapd is woken. 595 */ 596 }; 597 598 static inline unsigned long zone_managed_pages(struct zone *zone) 599 { 600 return (unsigned long)atomic_long_read(&zone->managed_pages); 601 } 602 603 static inline unsigned long zone_end_pfn(const struct zone *zone) 604 { 605 return zone->zone_start_pfn + zone->spanned_pages; 606 } 607 608 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) 609 { 610 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); 611 } 612 613 static inline bool zone_is_initialized(struct zone *zone) 614 { 615 return zone->initialized; 616 } 617 618 static inline bool zone_is_empty(struct zone *zone) 619 { 620 return zone->spanned_pages == 0; 621 } 622 623 /* 624 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty 625 * intersection with the given zone 626 */ 627 static inline bool zone_intersects(struct zone *zone, 628 unsigned long start_pfn, unsigned long nr_pages) 629 { 630 if (zone_is_empty(zone)) 631 return false; 632 if (start_pfn >= zone_end_pfn(zone) || 633 start_pfn + nr_pages <= zone->zone_start_pfn) 634 return false; 635 636 return true; 637 } 638 639 /* 640 * The "priority" of VM scanning is how much of the queues we will scan in one 641 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the 642 * queues ("queue_length >> 12") during an aging round. 643 */ 644 #define DEF_PRIORITY 12 645 646 /* Maximum number of zones on a zonelist */ 647 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) 648 649 enum { 650 ZONELIST_FALLBACK, /* zonelist with fallback */ 651 #ifdef CONFIG_NUMA 652 /* 653 * The NUMA zonelists are doubled because we need zonelists that 654 * restrict the allocations to a single node for __GFP_THISNODE. 655 */ 656 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */ 657 #endif 658 MAX_ZONELISTS 659 }; 660 661 /* 662 * This struct contains information about a zone in a zonelist. It is stored 663 * here to avoid dereferences into large structures and lookups of tables 664 */ 665 struct zoneref { 666 struct zone *zone; /* Pointer to actual zone */ 667 int zone_idx; /* zone_idx(zoneref->zone) */ 668 }; 669 670 /* 671 * One allocation request operates on a zonelist. A zonelist 672 * is a list of zones, the first one is the 'goal' of the 673 * allocation, the other zones are fallback zones, in decreasing 674 * priority. 675 * 676 * To speed the reading of the zonelist, the zonerefs contain the zone index 677 * of the entry being read. Helper functions to access information given 678 * a struct zoneref are 679 * 680 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs 681 * zonelist_zone_idx() - Return the index of the zone for an entry 682 * zonelist_node_idx() - Return the index of the node for an entry 683 */ 684 struct zonelist { 685 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; 686 }; 687 688 #ifndef CONFIG_DISCONTIGMEM 689 /* The array of struct pages - for discontigmem use pgdat->lmem_map */ 690 extern struct page *mem_map; 691 #endif 692 693 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 694 struct deferred_split { 695 spinlock_t split_queue_lock; 696 struct list_head split_queue; 697 unsigned long split_queue_len; 698 }; 699 #endif 700 701 /* 702 * On NUMA machines, each NUMA node would have a pg_data_t to describe 703 * it's memory layout. On UMA machines there is a single pglist_data which 704 * describes the whole memory. 705 * 706 * Memory statistics and page replacement data structures are maintained on a 707 * per-zone basis. 708 */ 709 struct bootmem_data; 710 typedef struct pglist_data { 711 struct zone node_zones[MAX_NR_ZONES]; 712 struct zonelist node_zonelists[MAX_ZONELISTS]; 713 int nr_zones; 714 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */ 715 struct page *node_mem_map; 716 #ifdef CONFIG_PAGE_EXTENSION 717 struct page_ext *node_page_ext; 718 #endif 719 #endif 720 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT) 721 /* 722 * Must be held any time you expect node_start_pfn, 723 * node_present_pages, node_spanned_pages or nr_zones to stay constant. 724 * 725 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to 726 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG 727 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT. 728 * 729 * Nests above zone->lock and zone->span_seqlock 730 */ 731 spinlock_t node_size_lock; 732 #endif 733 unsigned long node_start_pfn; 734 unsigned long node_present_pages; /* total number of physical pages */ 735 unsigned long node_spanned_pages; /* total size of physical page 736 range, including holes */ 737 int node_id; 738 wait_queue_head_t kswapd_wait; 739 wait_queue_head_t pfmemalloc_wait; 740 struct task_struct *kswapd; /* Protected by 741 mem_hotplug_begin/end() */ 742 int kswapd_order; 743 enum zone_type kswapd_classzone_idx; 744 745 int kswapd_failures; /* Number of 'reclaimed == 0' runs */ 746 747 #ifdef CONFIG_COMPACTION 748 int kcompactd_max_order; 749 enum zone_type kcompactd_classzone_idx; 750 wait_queue_head_t kcompactd_wait; 751 struct task_struct *kcompactd; 752 #endif 753 /* 754 * This is a per-node reserve of pages that are not available 755 * to userspace allocations. 756 */ 757 unsigned long totalreserve_pages; 758 759 #ifdef CONFIG_NUMA 760 /* 761 * node reclaim becomes active if more unmapped pages exist. 762 */ 763 unsigned long min_unmapped_pages; 764 unsigned long min_slab_pages; 765 #endif /* CONFIG_NUMA */ 766 767 /* Write-intensive fields used by page reclaim */ 768 ZONE_PADDING(_pad1_) 769 spinlock_t lru_lock; 770 771 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 772 /* 773 * If memory initialisation on large machines is deferred then this 774 * is the first PFN that needs to be initialised. 775 */ 776 unsigned long first_deferred_pfn; 777 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 778 779 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 780 struct deferred_split deferred_split_queue; 781 #endif 782 783 /* Fields commonly accessed by the page reclaim scanner */ 784 785 /* 786 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED. 787 * 788 * Use mem_cgroup_lruvec() to look up lruvecs. 789 */ 790 struct lruvec __lruvec; 791 792 unsigned long flags; 793 794 ZONE_PADDING(_pad2_) 795 796 /* Per-node vmstats */ 797 struct per_cpu_nodestat __percpu *per_cpu_nodestats; 798 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS]; 799 } pg_data_t; 800 801 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) 802 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) 803 #ifdef CONFIG_FLAT_NODE_MEM_MAP 804 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) 805 #else 806 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) 807 #endif 808 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) 809 810 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) 811 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) 812 813 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) 814 { 815 return pgdat->node_start_pfn + pgdat->node_spanned_pages; 816 } 817 818 static inline bool pgdat_is_empty(pg_data_t *pgdat) 819 { 820 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages; 821 } 822 823 #include <linux/memory_hotplug.h> 824 825 void build_all_zonelists(pg_data_t *pgdat); 826 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order, 827 enum zone_type classzone_idx); 828 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, 829 int classzone_idx, unsigned int alloc_flags, 830 long free_pages); 831 bool zone_watermark_ok(struct zone *z, unsigned int order, 832 unsigned long mark, int classzone_idx, 833 unsigned int alloc_flags); 834 bool zone_watermark_ok_safe(struct zone *z, unsigned int order, 835 unsigned long mark, int classzone_idx); 836 enum memmap_context { 837 MEMMAP_EARLY, 838 MEMMAP_HOTPLUG, 839 }; 840 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, 841 unsigned long size); 842 843 extern void lruvec_init(struct lruvec *lruvec); 844 845 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec) 846 { 847 #ifdef CONFIG_MEMCG 848 return lruvec->pgdat; 849 #else 850 return container_of(lruvec, struct pglist_data, __lruvec); 851 #endif 852 } 853 854 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx); 855 856 #ifdef CONFIG_HAVE_MEMORY_PRESENT 857 void memory_present(int nid, unsigned long start, unsigned long end); 858 #else 859 static inline void memory_present(int nid, unsigned long start, unsigned long end) {} 860 #endif 861 862 #if defined(CONFIG_SPARSEMEM) 863 void memblocks_present(void); 864 #else 865 static inline void memblocks_present(void) {} 866 #endif 867 868 #ifdef CONFIG_HAVE_MEMORYLESS_NODES 869 int local_memory_node(int node_id); 870 #else 871 static inline int local_memory_node(int node_id) { return node_id; }; 872 #endif 873 874 /* 875 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. 876 */ 877 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) 878 879 /* 880 * Returns true if a zone has pages managed by the buddy allocator. 881 * All the reclaim decisions have to use this function rather than 882 * populated_zone(). If the whole zone is reserved then we can easily 883 * end up with populated_zone() && !managed_zone(). 884 */ 885 static inline bool managed_zone(struct zone *zone) 886 { 887 return zone_managed_pages(zone); 888 } 889 890 /* Returns true if a zone has memory */ 891 static inline bool populated_zone(struct zone *zone) 892 { 893 return zone->present_pages; 894 } 895 896 #ifdef CONFIG_NUMA 897 static inline int zone_to_nid(struct zone *zone) 898 { 899 return zone->node; 900 } 901 902 static inline void zone_set_nid(struct zone *zone, int nid) 903 { 904 zone->node = nid; 905 } 906 #else 907 static inline int zone_to_nid(struct zone *zone) 908 { 909 return 0; 910 } 911 912 static inline void zone_set_nid(struct zone *zone, int nid) {} 913 #endif 914 915 extern int movable_zone; 916 917 #ifdef CONFIG_HIGHMEM 918 static inline int zone_movable_is_highmem(void) 919 { 920 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 921 return movable_zone == ZONE_HIGHMEM; 922 #else 923 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM; 924 #endif 925 } 926 #endif 927 928 static inline int is_highmem_idx(enum zone_type idx) 929 { 930 #ifdef CONFIG_HIGHMEM 931 return (idx == ZONE_HIGHMEM || 932 (idx == ZONE_MOVABLE && zone_movable_is_highmem())); 933 #else 934 return 0; 935 #endif 936 } 937 938 /** 939 * is_highmem - helper function to quickly check if a struct zone is a 940 * highmem zone or not. This is an attempt to keep references 941 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. 942 * @zone - pointer to struct zone variable 943 */ 944 static inline int is_highmem(struct zone *zone) 945 { 946 #ifdef CONFIG_HIGHMEM 947 return is_highmem_idx(zone_idx(zone)); 948 #else 949 return 0; 950 #endif 951 } 952 953 /* These two functions are used to setup the per zone pages min values */ 954 struct ctl_table; 955 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, 956 void __user *, size_t *, loff_t *); 957 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int, 958 void __user *, size_t *, loff_t *); 959 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, 960 void __user *, size_t *, loff_t *); 961 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES]; 962 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, 963 void __user *, size_t *, loff_t *); 964 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, 965 void __user *, size_t *, loff_t *); 966 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, 967 void __user *, size_t *, loff_t *); 968 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, 969 void __user *, size_t *, loff_t *); 970 971 extern int numa_zonelist_order_handler(struct ctl_table *, int, 972 void __user *, size_t *, loff_t *); 973 extern char numa_zonelist_order[]; 974 #define NUMA_ZONELIST_ORDER_LEN 16 975 976 #ifndef CONFIG_NEED_MULTIPLE_NODES 977 978 extern struct pglist_data contig_page_data; 979 #define NODE_DATA(nid) (&contig_page_data) 980 #define NODE_MEM_MAP(nid) mem_map 981 982 #else /* CONFIG_NEED_MULTIPLE_NODES */ 983 984 #include <asm/mmzone.h> 985 986 #endif /* !CONFIG_NEED_MULTIPLE_NODES */ 987 988 extern struct pglist_data *first_online_pgdat(void); 989 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); 990 extern struct zone *next_zone(struct zone *zone); 991 992 /** 993 * for_each_online_pgdat - helper macro to iterate over all online nodes 994 * @pgdat - pointer to a pg_data_t variable 995 */ 996 #define for_each_online_pgdat(pgdat) \ 997 for (pgdat = first_online_pgdat(); \ 998 pgdat; \ 999 pgdat = next_online_pgdat(pgdat)) 1000 /** 1001 * for_each_zone - helper macro to iterate over all memory zones 1002 * @zone - pointer to struct zone variable 1003 * 1004 * The user only needs to declare the zone variable, for_each_zone 1005 * fills it in. 1006 */ 1007 #define for_each_zone(zone) \ 1008 for (zone = (first_online_pgdat())->node_zones; \ 1009 zone; \ 1010 zone = next_zone(zone)) 1011 1012 #define for_each_populated_zone(zone) \ 1013 for (zone = (first_online_pgdat())->node_zones; \ 1014 zone; \ 1015 zone = next_zone(zone)) \ 1016 if (!populated_zone(zone)) \ 1017 ; /* do nothing */ \ 1018 else 1019 1020 static inline struct zone *zonelist_zone(struct zoneref *zoneref) 1021 { 1022 return zoneref->zone; 1023 } 1024 1025 static inline int zonelist_zone_idx(struct zoneref *zoneref) 1026 { 1027 return zoneref->zone_idx; 1028 } 1029 1030 static inline int zonelist_node_idx(struct zoneref *zoneref) 1031 { 1032 return zone_to_nid(zoneref->zone); 1033 } 1034 1035 struct zoneref *__next_zones_zonelist(struct zoneref *z, 1036 enum zone_type highest_zoneidx, 1037 nodemask_t *nodes); 1038 1039 /** 1040 * 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 1041 * @z - The cursor used as a starting point for the search 1042 * @highest_zoneidx - The zone index of the highest zone to return 1043 * @nodes - An optional nodemask to filter the zonelist with 1044 * 1045 * This function returns the next zone at or below a given zone index that is 1046 * within the allowed nodemask using a cursor as the starting point for the 1047 * search. The zoneref returned is a cursor that represents the current zone 1048 * being examined. It should be advanced by one before calling 1049 * next_zones_zonelist again. 1050 */ 1051 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z, 1052 enum zone_type highest_zoneidx, 1053 nodemask_t *nodes) 1054 { 1055 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx)) 1056 return z; 1057 return __next_zones_zonelist(z, highest_zoneidx, nodes); 1058 } 1059 1060 /** 1061 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist 1062 * @zonelist - The zonelist to search for a suitable zone 1063 * @highest_zoneidx - The zone index of the highest zone to return 1064 * @nodes - An optional nodemask to filter the zonelist with 1065 * @return - Zoneref pointer for the first suitable zone found (see below) 1066 * 1067 * This function returns the first zone at or below a given zone index that is 1068 * within the allowed nodemask. The zoneref returned is a cursor that can be 1069 * used to iterate the zonelist with next_zones_zonelist by advancing it by 1070 * one before calling. 1071 * 1072 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is 1073 * never NULL). This may happen either genuinely, or due to concurrent nodemask 1074 * update due to cpuset modification. 1075 */ 1076 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, 1077 enum zone_type highest_zoneidx, 1078 nodemask_t *nodes) 1079 { 1080 return next_zones_zonelist(zonelist->_zonerefs, 1081 highest_zoneidx, nodes); 1082 } 1083 1084 /** 1085 * 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 1086 * @zone - The current zone in the iterator 1087 * @z - The current pointer within zonelist->_zonerefs being iterated 1088 * @zlist - The zonelist being iterated 1089 * @highidx - The zone index of the highest zone to return 1090 * @nodemask - Nodemask allowed by the allocator 1091 * 1092 * This iterator iterates though all zones at or below a given zone index and 1093 * within a given nodemask 1094 */ 1095 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ 1096 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \ 1097 zone; \ 1098 z = next_zones_zonelist(++z, highidx, nodemask), \ 1099 zone = zonelist_zone(z)) 1100 1101 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ 1102 for (zone = z->zone; \ 1103 zone; \ 1104 z = next_zones_zonelist(++z, highidx, nodemask), \ 1105 zone = zonelist_zone(z)) 1106 1107 1108 /** 1109 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index 1110 * @zone - The current zone in the iterator 1111 * @z - The current pointer within zonelist->zones being iterated 1112 * @zlist - The zonelist being iterated 1113 * @highidx - The zone index of the highest zone to return 1114 * 1115 * This iterator iterates though all zones at or below a given zone index. 1116 */ 1117 #define for_each_zone_zonelist(zone, z, zlist, highidx) \ 1118 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) 1119 1120 #ifdef CONFIG_SPARSEMEM 1121 #include <asm/sparsemem.h> 1122 #endif 1123 1124 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \ 1125 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) 1126 static inline unsigned long early_pfn_to_nid(unsigned long pfn) 1127 { 1128 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA)); 1129 return 0; 1130 } 1131 #endif 1132 1133 #ifdef CONFIG_FLATMEM 1134 #define pfn_to_nid(pfn) (0) 1135 #endif 1136 1137 #ifdef CONFIG_SPARSEMEM 1138 1139 /* 1140 * SECTION_SHIFT #bits space required to store a section # 1141 * 1142 * PA_SECTION_SHIFT physical address to/from section number 1143 * PFN_SECTION_SHIFT pfn to/from section number 1144 */ 1145 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) 1146 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) 1147 1148 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) 1149 1150 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) 1151 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) 1152 1153 #define SECTION_BLOCKFLAGS_BITS \ 1154 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) 1155 1156 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS 1157 #error Allocator MAX_ORDER exceeds SECTION_SIZE 1158 #endif 1159 1160 static inline unsigned long pfn_to_section_nr(unsigned long pfn) 1161 { 1162 return pfn >> PFN_SECTION_SHIFT; 1163 } 1164 static inline unsigned long section_nr_to_pfn(unsigned long sec) 1165 { 1166 return sec << PFN_SECTION_SHIFT; 1167 } 1168 1169 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) 1170 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) 1171 1172 #define SUBSECTION_SHIFT 21 1173 1174 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT) 1175 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT) 1176 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1)) 1177 1178 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS 1179 #error Subsection size exceeds section size 1180 #else 1181 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT)) 1182 #endif 1183 1184 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION) 1185 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK) 1186 1187 struct mem_section_usage { 1188 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION); 1189 /* See declaration of similar field in struct zone */ 1190 unsigned long pageblock_flags[0]; 1191 }; 1192 1193 void subsection_map_init(unsigned long pfn, unsigned long nr_pages); 1194 1195 struct page; 1196 struct page_ext; 1197 struct mem_section { 1198 /* 1199 * This is, logically, a pointer to an array of struct 1200 * pages. However, it is stored with some other magic. 1201 * (see sparse.c::sparse_init_one_section()) 1202 * 1203 * Additionally during early boot we encode node id of 1204 * the location of the section here to guide allocation. 1205 * (see sparse.c::memory_present()) 1206 * 1207 * Making it a UL at least makes someone do a cast 1208 * before using it wrong. 1209 */ 1210 unsigned long section_mem_map; 1211 1212 struct mem_section_usage *usage; 1213 #ifdef CONFIG_PAGE_EXTENSION 1214 /* 1215 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use 1216 * section. (see page_ext.h about this.) 1217 */ 1218 struct page_ext *page_ext; 1219 unsigned long pad; 1220 #endif 1221 /* 1222 * WARNING: mem_section must be a power-of-2 in size for the 1223 * calculation and use of SECTION_ROOT_MASK to make sense. 1224 */ 1225 }; 1226 1227 #ifdef CONFIG_SPARSEMEM_EXTREME 1228 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) 1229 #else 1230 #define SECTIONS_PER_ROOT 1 1231 #endif 1232 1233 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) 1234 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) 1235 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) 1236 1237 #ifdef CONFIG_SPARSEMEM_EXTREME 1238 extern struct mem_section **mem_section; 1239 #else 1240 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; 1241 #endif 1242 1243 static inline unsigned long *section_to_usemap(struct mem_section *ms) 1244 { 1245 return ms->usage->pageblock_flags; 1246 } 1247 1248 static inline struct mem_section *__nr_to_section(unsigned long nr) 1249 { 1250 #ifdef CONFIG_SPARSEMEM_EXTREME 1251 if (!mem_section) 1252 return NULL; 1253 #endif 1254 if (!mem_section[SECTION_NR_TO_ROOT(nr)]) 1255 return NULL; 1256 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; 1257 } 1258 extern unsigned long __section_nr(struct mem_section *ms); 1259 extern size_t mem_section_usage_size(void); 1260 1261 /* 1262 * We use the lower bits of the mem_map pointer to store 1263 * a little bit of information. The pointer is calculated 1264 * as mem_map - section_nr_to_pfn(pnum). The result is 1265 * aligned to the minimum alignment of the two values: 1266 * 1. All mem_map arrays are page-aligned. 1267 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT 1268 * lowest bits. PFN_SECTION_SHIFT is arch-specific 1269 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the 1270 * worst combination is powerpc with 256k pages, 1271 * which results in PFN_SECTION_SHIFT equal 6. 1272 * To sum it up, at least 6 bits are available. 1273 */ 1274 #define SECTION_MARKED_PRESENT (1UL<<0) 1275 #define SECTION_HAS_MEM_MAP (1UL<<1) 1276 #define SECTION_IS_ONLINE (1UL<<2) 1277 #define SECTION_IS_EARLY (1UL<<3) 1278 #define SECTION_MAP_LAST_BIT (1UL<<4) 1279 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) 1280 #define SECTION_NID_SHIFT 3 1281 1282 static inline struct page *__section_mem_map_addr(struct mem_section *section) 1283 { 1284 unsigned long map = section->section_mem_map; 1285 map &= SECTION_MAP_MASK; 1286 return (struct page *)map; 1287 } 1288 1289 static inline int present_section(struct mem_section *section) 1290 { 1291 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); 1292 } 1293 1294 static inline int present_section_nr(unsigned long nr) 1295 { 1296 return present_section(__nr_to_section(nr)); 1297 } 1298 1299 static inline int valid_section(struct mem_section *section) 1300 { 1301 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); 1302 } 1303 1304 static inline int early_section(struct mem_section *section) 1305 { 1306 return (section && (section->section_mem_map & SECTION_IS_EARLY)); 1307 } 1308 1309 static inline int valid_section_nr(unsigned long nr) 1310 { 1311 return valid_section(__nr_to_section(nr)); 1312 } 1313 1314 static inline int online_section(struct mem_section *section) 1315 { 1316 return (section && (section->section_mem_map & SECTION_IS_ONLINE)); 1317 } 1318 1319 static inline int online_section_nr(unsigned long nr) 1320 { 1321 return online_section(__nr_to_section(nr)); 1322 } 1323 1324 #ifdef CONFIG_MEMORY_HOTPLUG 1325 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1326 #ifdef CONFIG_MEMORY_HOTREMOVE 1327 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1328 #endif 1329 #endif 1330 1331 static inline struct mem_section *__pfn_to_section(unsigned long pfn) 1332 { 1333 return __nr_to_section(pfn_to_section_nr(pfn)); 1334 } 1335 1336 extern unsigned long __highest_present_section_nr; 1337 1338 static inline int subsection_map_index(unsigned long pfn) 1339 { 1340 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION; 1341 } 1342 1343 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1344 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1345 { 1346 int idx = subsection_map_index(pfn); 1347 1348 return test_bit(idx, ms->usage->subsection_map); 1349 } 1350 #else 1351 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1352 { 1353 return 1; 1354 } 1355 #endif 1356 1357 #ifndef CONFIG_HAVE_ARCH_PFN_VALID 1358 static inline int pfn_valid(unsigned long pfn) 1359 { 1360 struct mem_section *ms; 1361 1362 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 1363 return 0; 1364 ms = __nr_to_section(pfn_to_section_nr(pfn)); 1365 if (!valid_section(ms)) 1366 return 0; 1367 /* 1368 * Traditionally early sections always returned pfn_valid() for 1369 * the entire section-sized span. 1370 */ 1371 return early_section(ms) || pfn_section_valid(ms, pfn); 1372 } 1373 #endif 1374 1375 static inline int pfn_present(unsigned long pfn) 1376 { 1377 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 1378 return 0; 1379 return present_section(__nr_to_section(pfn_to_section_nr(pfn))); 1380 } 1381 1382 static inline unsigned long next_present_section_nr(unsigned long section_nr) 1383 { 1384 while (++section_nr <= __highest_present_section_nr) { 1385 if (present_section_nr(section_nr)) 1386 return section_nr; 1387 } 1388 1389 return -1; 1390 } 1391 1392 /* 1393 * These are _only_ used during initialisation, therefore they 1394 * can use __initdata ... They could have names to indicate 1395 * this restriction. 1396 */ 1397 #ifdef CONFIG_NUMA 1398 #define pfn_to_nid(pfn) \ 1399 ({ \ 1400 unsigned long __pfn_to_nid_pfn = (pfn); \ 1401 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ 1402 }) 1403 #else 1404 #define pfn_to_nid(pfn) (0) 1405 #endif 1406 1407 #define early_pfn_valid(pfn) pfn_valid(pfn) 1408 void sparse_init(void); 1409 #else 1410 #define sparse_init() do {} while (0) 1411 #define sparse_index_init(_sec, _nid) do {} while (0) 1412 #define pfn_present pfn_valid 1413 #define subsection_map_init(_pfn, _nr_pages) do {} while (0) 1414 #endif /* CONFIG_SPARSEMEM */ 1415 1416 /* 1417 * During memory init memblocks map pfns to nids. The search is expensive and 1418 * this caches recent lookups. The implementation of __early_pfn_to_nid 1419 * may treat start/end as pfns or sections. 1420 */ 1421 struct mminit_pfnnid_cache { 1422 unsigned long last_start; 1423 unsigned long last_end; 1424 int last_nid; 1425 }; 1426 1427 #ifndef early_pfn_valid 1428 #define early_pfn_valid(pfn) (1) 1429 #endif 1430 1431 void memory_present(int nid, unsigned long start, unsigned long end); 1432 1433 /* 1434 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we 1435 * need to check pfn validity within that MAX_ORDER_NR_PAGES block. 1436 * pfn_valid_within() should be used in this case; we optimise this away 1437 * when we have no holes within a MAX_ORDER_NR_PAGES block. 1438 */ 1439 #ifdef CONFIG_HOLES_IN_ZONE 1440 #define pfn_valid_within(pfn) pfn_valid(pfn) 1441 #else 1442 #define pfn_valid_within(pfn) (1) 1443 #endif 1444 1445 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL 1446 /* 1447 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap 1448 * associated with it or not. This means that a struct page exists for this 1449 * pfn. The caller cannot assume the page is fully initialized in general. 1450 * Hotplugable pages might not have been onlined yet. pfn_to_online_page() 1451 * will ensure the struct page is fully online and initialized. Special pages 1452 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly. 1453 * 1454 * In FLATMEM, it is expected that holes always have valid memmap as long as 1455 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed 1456 * that a valid section has a memmap for the entire section. 1457 * 1458 * However, an ARM, and maybe other embedded architectures in the future 1459 * free memmap backing holes to save memory on the assumption the memmap is 1460 * never used. The page_zone linkages are then broken even though pfn_valid() 1461 * returns true. A walker of the full memmap must then do this additional 1462 * check to ensure the memmap they are looking at is sane by making sure 1463 * the zone and PFN linkages are still valid. This is expensive, but walkers 1464 * of the full memmap are extremely rare. 1465 */ 1466 bool memmap_valid_within(unsigned long pfn, 1467 struct page *page, struct zone *zone); 1468 #else 1469 static inline bool memmap_valid_within(unsigned long pfn, 1470 struct page *page, struct zone *zone) 1471 { 1472 return true; 1473 } 1474 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */ 1475 1476 #endif /* !__GENERATING_BOUNDS.H */ 1477 #endif /* !__ASSEMBLY__ */ 1478 #endif /* _LINUX_MMZONE_H */ 1479