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