include/linux/mmzone.h

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
is_migrate_movable
add_to_free_area
add_to_free_area_tail
add_to_free_area_random
move_to_free_area
get_page_from_free_area
del_page_from_free_area
free_area_empty
is_file_lru
is_active_lru
zone_managed_pages
zone_end_pfn
zone_spans_pfn
zone_is_initialized
zone_is_empty
zone_intersects
node_lruvec
pgdat_end_pfn
pgdat_is_empty
lruvec_pgdat
memory_present
memblocks_present
local_memory_node
managed_zone
populated_zone
zone_to_nid
zone_set_nid
zone_to_nid
zone_set_nid
zone_movable_is_highmem
is_highmem_idx
is_highmem
zonelist_zone
zonelist_zone_idx
zonelist_node_idx
next_zones_zonelist
first_zones_zonelist
early_pfn_to_nid
pfn_to_section_nr
section_nr_to_pfn
section_to_usemap
__nr_to_section
__section_mem_map_addr
present_section
present_section_nr
valid_section
early_section
valid_section_nr
online_section
online_section_nr
__pfn_to_section
subsection_map_index
pfn_section_valid
pfn_section_valid
pfn_valid
pfn_present
memmap_valid_within
   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 int is_file_lru(enum lru_list lru)
 276 {
 277         return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
 278 }
 279 
 280 static inline int 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 struct lruvec {
 299         struct list_head                lists[NR_LRU_LISTS];
 300         struct zone_reclaim_stat        reclaim_stat;
 301         /* Evictions & activations on the inactive file list */
 302         atomic_long_t                   inactive_age;
 303         /* Refaults at the time of last reclaim cycle */
 304         unsigned long                   refaults;
 305 #ifdef CONFIG_MEMCG
 306         struct pglist_data *pgdat;
 307 #endif
 308 };
 309 
 310 /* Isolate unmapped file */
 311 #define ISOLATE_UNMAPPED        ((__force isolate_mode_t)0x2)
 312 /* Isolate for asynchronous migration */
 313 #define ISOLATE_ASYNC_MIGRATE   ((__force isolate_mode_t)0x4)
 314 /* Isolate unevictable pages */
 315 #define ISOLATE_UNEVICTABLE     ((__force isolate_mode_t)0x8)
 316 
 317 /* LRU Isolation modes. */
 318 typedef unsigned __bitwise isolate_mode_t;
 319 
 320 enum zone_watermarks {
 321         WMARK_MIN,
 322         WMARK_LOW,
 323         WMARK_HIGH,
 324         NR_WMARK
 325 };
 326 
 327 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
 328 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
 329 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
 330 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
 331 
 332 struct per_cpu_pages {
 333         int count;              /* number of pages in the list */
 334         int high;               /* high watermark, emptying needed */
 335         int batch;              /* chunk size for buddy add/remove */
 336 
 337         /* Lists of pages, one per migrate type stored on the pcp-lists */
 338         struct list_head lists[MIGRATE_PCPTYPES];
 339 };
 340 
 341 struct per_cpu_pageset {
 342         struct per_cpu_pages pcp;
 343 #ifdef CONFIG_NUMA
 344         s8 expire;
 345         u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
 346 #endif
 347 #ifdef CONFIG_SMP
 348         s8 stat_threshold;
 349         s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
 350 #endif
 351 };
 352 
 353 struct per_cpu_nodestat {
 354         s8 stat_threshold;
 355         s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
 356 };
 357 
 358 #endif /* !__GENERATING_BOUNDS.H */
 359 
 360 enum zone_type {
 361 #ifdef CONFIG_ZONE_DMA
 362         /*
 363          * ZONE_DMA is used when there are devices that are not able
 364          * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
 365          * carve out the portion of memory that is needed for these devices.
 366          * The range is arch specific.
 367          *
 368          * Some examples
 369          *
 370          * Architecture         Limit
 371          * ---------------------------
 372          * parisc, ia64, sparc  <4G
 373          * s390, powerpc        <2G
 374          * arm                  Various
 375          * alpha                Unlimited or 0-16MB.
 376          *
 377          * i386, x86_64 and multiple other arches
 378          *                      <16M.
 379          */
 380         ZONE_DMA,
 381 #endif
 382 #ifdef CONFIG_ZONE_DMA32
 383         /*
 384          * x86_64 needs two ZONE_DMAs because it supports devices that are
 385          * only able to do DMA to the lower 16M but also 32 bit devices that
 386          * can only do DMA areas below 4G.
 387          */
 388         ZONE_DMA32,
 389 #endif
 390         /*
 391          * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
 392          * performed on pages in ZONE_NORMAL if the DMA devices support
 393          * transfers to all addressable memory.
 394          */
 395         ZONE_NORMAL,
 396 #ifdef CONFIG_HIGHMEM
 397         /*
 398          * A memory area that is only addressable by the kernel through
 399          * mapping portions into its own address space. This is for example
 400          * used by i386 to allow the kernel to address the memory beyond
 401          * 900MB. The kernel will set up special mappings (page
 402          * table entries on i386) for each page that the kernel needs to
 403          * access.
 404          */
 405         ZONE_HIGHMEM,
 406 #endif
 407         ZONE_MOVABLE,
 408 #ifdef CONFIG_ZONE_DEVICE
 409         ZONE_DEVICE,
 410 #endif
 411         __MAX_NR_ZONES
 412 
 413 };
 414 
 415 #ifndef __GENERATING_BOUNDS_H
 416 
 417 struct zone {
 418         /* Read-mostly fields */
 419 
 420         /* zone watermarks, access with *_wmark_pages(zone) macros */
 421         unsigned long _watermark[NR_WMARK];
 422         unsigned long watermark_boost;
 423 
 424         unsigned long nr_reserved_highatomic;
 425 
 426         /*
 427          * We don't know if the memory that we're going to allocate will be
 428          * freeable or/and it will be released eventually, so to avoid totally
 429          * wasting several GB of ram we must reserve some of the lower zone
 430          * memory (otherwise we risk to run OOM on the lower zones despite
 431          * there being tons of freeable ram on the higher zones).  This array is
 432          * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
 433          * changes.
 434          */
 435         long lowmem_reserve[MAX_NR_ZONES];
 436 
 437 #ifdef CONFIG_NUMA
 438         int node;
 439 #endif
 440         struct pglist_data      *zone_pgdat;
 441         struct per_cpu_pageset __percpu *pageset;
 442 
 443 #ifndef CONFIG_SPARSEMEM
 444         /*
 445          * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
 446          * In SPARSEMEM, this map is stored in struct mem_section
 447          */
 448         unsigned long           *pageblock_flags;
 449 #endif /* CONFIG_SPARSEMEM */
 450 
 451         /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
 452         unsigned long           zone_start_pfn;
 453 
 454         /*
 455          * spanned_pages is the total pages spanned by the zone, including
 456          * holes, which is calculated as:
 457          *      spanned_pages = zone_end_pfn - zone_start_pfn;
 458          *
 459          * present_pages is physical pages existing within the zone, which
 460          * is calculated as:
 461          *      present_pages = spanned_pages - absent_pages(pages in holes);
 462          *
 463          * managed_pages is present pages managed by the buddy system, which
 464          * is calculated as (reserved_pages includes pages allocated by the
 465          * bootmem allocator):
 466          *      managed_pages = present_pages - reserved_pages;
 467          *
 468          * So present_pages may be used by memory hotplug or memory power
 469          * management logic to figure out unmanaged pages by checking
 470          * (present_pages - managed_pages). And managed_pages should be used
 471          * by page allocator and vm scanner to calculate all kinds of watermarks
 472          * and thresholds.
 473          *
 474          * Locking rules:
 475          *
 476          * zone_start_pfn and spanned_pages are protected by span_seqlock.
 477          * It is a seqlock because it has to be read outside of zone->lock,
 478          * and it is done in the main allocator path.  But, it is written
 479          * quite infrequently.
 480          *
 481          * The span_seq lock is declared along with zone->lock because it is
 482          * frequently read in proximity to zone->lock.  It's good to
 483          * give them a chance of being in the same cacheline.
 484          *
 485          * Write access to present_pages at runtime should be protected by
 486          * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
 487          * present_pages should get_online_mems() to get a stable value.
 488          */
 489         atomic_long_t           managed_pages;
 490         unsigned long           spanned_pages;
 491         unsigned long           present_pages;
 492 
 493         const char              *name;
 494 
 495 #ifdef CONFIG_MEMORY_ISOLATION
 496         /*
 497          * Number of isolated pageblock. It is used to solve incorrect
 498          * freepage counting problem due to racy retrieving migratetype
 499          * of pageblock. Protected by zone->lock.
 500          */
 501         unsigned long           nr_isolate_pageblock;
 502 #endif
 503 
 504 #ifdef CONFIG_MEMORY_HOTPLUG
 505         /* see spanned/present_pages for more description */
 506         seqlock_t               span_seqlock;
 507 #endif
 508 
 509         int initialized;
 510 
 511         /* Write-intensive fields used from the page allocator */
 512         ZONE_PADDING(_pad1_)
 513 
 514         /* free areas of different sizes */
 515         struct free_area        free_area[MAX_ORDER];
 516 
 517         /* zone flags, see below */
 518         unsigned long           flags;
 519 
 520         /* Primarily protects free_area */
 521         spinlock_t              lock;
 522 
 523         /* Write-intensive fields used by compaction and vmstats. */
 524         ZONE_PADDING(_pad2_)
 525 
 526         /*
 527          * When free pages are below this point, additional steps are taken
 528          * when reading the number of free pages to avoid per-cpu counter
 529          * drift allowing watermarks to be breached
 530          */
 531         unsigned long percpu_drift_mark;
 532 
 533 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
 534         /* pfn where compaction free scanner should start */
 535         unsigned long           compact_cached_free_pfn;
 536         /* pfn where async and sync compaction migration scanner should start */
 537         unsigned long           compact_cached_migrate_pfn[2];
 538         unsigned long           compact_init_migrate_pfn;
 539         unsigned long           compact_init_free_pfn;
 540 #endif
 541 
 542 #ifdef CONFIG_COMPACTION
 543         /*
 544          * On compaction failure, 1<<compact_defer_shift compactions
 545          * are skipped before trying again. The number attempted since
 546          * last failure is tracked with compact_considered.
 547          */
 548         unsigned int            compact_considered;
 549         unsigned int            compact_defer_shift;
 550         int                     compact_order_failed;
 551 #endif
 552 
 553 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
 554         /* Set to true when the PG_migrate_skip bits should be cleared */
 555         bool                    compact_blockskip_flush;
 556 #endif
 557 
 558         bool                    contiguous;
 559 
 560         ZONE_PADDING(_pad3_)
 561         /* Zone statistics */
 562         atomic_long_t           vm_stat[NR_VM_ZONE_STAT_ITEMS];
 563         atomic_long_t           vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
 564 } ____cacheline_internodealigned_in_smp;
 565 
 566 enum pgdat_flags {
 567         PGDAT_CONGESTED,                /* pgdat has many dirty pages backed by
 568                                          * a congested BDI
 569                                          */
 570         PGDAT_DIRTY,                    /* reclaim scanning has recently found
 571                                          * many dirty file pages at the tail
 572                                          * of the LRU.
 573                                          */
 574         PGDAT_WRITEBACK,                /* reclaim scanning has recently found
 575                                          * many pages under writeback
 576                                          */
 577         PGDAT_RECLAIM_LOCKED,           /* prevents concurrent reclaim */
 578 };
 579 
 580 enum zone_flags {
 581         ZONE_BOOSTED_WATERMARK,         /* zone recently boosted watermarks.
 582                                          * Cleared when kswapd is woken.
 583                                          */
 584 };
 585 
 586 static inline unsigned long zone_managed_pages(struct zone *zone)
 587 {
 588         return (unsigned long)atomic_long_read(&zone->managed_pages);
 589 }
 590 
 591 static inline unsigned long zone_end_pfn(const struct zone *zone)
 592 {
 593         return zone->zone_start_pfn + zone->spanned_pages;
 594 }
 595 
 596 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
 597 {
 598         return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
 599 }
 600 
 601 static inline bool zone_is_initialized(struct zone *zone)
 602 {
 603         return zone->initialized;
 604 }
 605 
 606 static inline bool zone_is_empty(struct zone *zone)
 607 {
 608         return zone->spanned_pages == 0;
 609 }
 610 
 611 /*
 612  * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
 613  * intersection with the given zone
 614  */
 615 static inline bool zone_intersects(struct zone *zone,
 616                 unsigned long start_pfn, unsigned long nr_pages)
 617 {
 618         if (zone_is_empty(zone))
 619                 return false;
 620         if (start_pfn >= zone_end_pfn(zone) ||
 621             start_pfn + nr_pages <= zone->zone_start_pfn)
 622                 return false;
 623 
 624         return true;
 625 }
 626 
 627 /*
 628  * The "priority" of VM scanning is how much of the queues we will scan in one
 629  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
 630  * queues ("queue_length >> 12") during an aging round.
 631  */
 632 #define DEF_PRIORITY 12
 633 
 634 /* Maximum number of zones on a zonelist */
 635 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
 636 
 637 enum {
 638         ZONELIST_FALLBACK,      /* zonelist with fallback */
 639 #ifdef CONFIG_NUMA
 640         /*
 641          * The NUMA zonelists are doubled because we need zonelists that
 642          * restrict the allocations to a single node for __GFP_THISNODE.
 643          */
 644         ZONELIST_NOFALLBACK,    /* zonelist without fallback (__GFP_THISNODE) */
 645 #endif
 646         MAX_ZONELISTS
 647 };
 648 
 649 /*
 650  * This struct contains information about a zone in a zonelist. It is stored
 651  * here to avoid dereferences into large structures and lookups of tables
 652  */
 653 struct zoneref {
 654         struct zone *zone;      /* Pointer to actual zone */
 655         int zone_idx;           /* zone_idx(zoneref->zone) */
 656 };
 657 
 658 /*
 659  * One allocation request operates on a zonelist. A zonelist
 660  * is a list of zones, the first one is the 'goal' of the
 661  * allocation, the other zones are fallback zones, in decreasing
 662  * priority.
 663  *
 664  * To speed the reading of the zonelist, the zonerefs contain the zone index
 665  * of the entry being read. Helper functions to access information given
 666  * a struct zoneref are
 667  *
 668  * zonelist_zone()      - Return the struct zone * for an entry in _zonerefs
 669  * zonelist_zone_idx()  - Return the index of the zone for an entry
 670  * zonelist_node_idx()  - Return the index of the node for an entry
 671  */
 672 struct zonelist {
 673         struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
 674 };
 675 
 676 #ifndef CONFIG_DISCONTIGMEM
 677 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
 678 extern struct page *mem_map;
 679 #endif
 680 
 681 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 682 struct deferred_split {
 683         spinlock_t split_queue_lock;
 684         struct list_head split_queue;
 685         unsigned long split_queue_len;
 686 };
 687 #endif
 688 
 689 /*
 690  * On NUMA machines, each NUMA node would have a pg_data_t to describe
 691  * it's memory layout. On UMA machines there is a single pglist_data which
 692  * describes the whole memory.
 693  *
 694  * Memory statistics and page replacement data structures are maintained on a
 695  * per-zone basis.
 696  */
 697 struct bootmem_data;
 698 typedef struct pglist_data {
 699         struct zone node_zones[MAX_NR_ZONES];
 700         struct zonelist node_zonelists[MAX_ZONELISTS];
 701         int nr_zones;
 702 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
 703         struct page *node_mem_map;
 704 #ifdef CONFIG_PAGE_EXTENSION
 705         struct page_ext *node_page_ext;
 706 #endif
 707 #endif
 708 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
 709         /*
 710          * Must be held any time you expect node_start_pfn,
 711          * node_present_pages, node_spanned_pages or nr_zones to stay constant.
 712          *
 713          * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
 714          * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
 715          * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
 716          *
 717          * Nests above zone->lock and zone->span_seqlock
 718          */
 719         spinlock_t node_size_lock;
 720 #endif
 721         unsigned long node_start_pfn;
 722         unsigned long node_present_pages; /* total number of physical pages */
 723         unsigned long node_spanned_pages; /* total size of physical page
 724                                              range, including holes */
 725         int node_id;
 726         wait_queue_head_t kswapd_wait;
 727         wait_queue_head_t pfmemalloc_wait;
 728         struct task_struct *kswapd;     /* Protected by
 729                                            mem_hotplug_begin/end() */
 730         int kswapd_order;
 731         enum zone_type kswapd_classzone_idx;
 732 
 733         int kswapd_failures;            /* Number of 'reclaimed == 0' runs */
 734 
 735 #ifdef CONFIG_COMPACTION
 736         int kcompactd_max_order;
 737         enum zone_type kcompactd_classzone_idx;
 738         wait_queue_head_t kcompactd_wait;
 739         struct task_struct *kcompactd;
 740 #endif
 741         /*
 742          * This is a per-node reserve of pages that are not available
 743          * to userspace allocations.
 744          */
 745         unsigned long           totalreserve_pages;
 746 
 747 #ifdef CONFIG_NUMA
 748         /*
 749          * zone reclaim becomes active if more unmapped pages exist.
 750          */
 751         unsigned long           min_unmapped_pages;
 752         unsigned long           min_slab_pages;
 753 #endif /* CONFIG_NUMA */
 754 
 755         /* Write-intensive fields used by page reclaim */
 756         ZONE_PADDING(_pad1_)
 757         spinlock_t              lru_lock;
 758 
 759 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
 760         /*
 761          * If memory initialisation on large machines is deferred then this
 762          * is the first PFN that needs to be initialised.
 763          */
 764         unsigned long first_deferred_pfn;
 765 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
 766 
 767 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 768         struct deferred_split deferred_split_queue;
 769 #endif
 770 
 771         /* Fields commonly accessed by the page reclaim scanner */
 772         struct lruvec           lruvec;
 773 
 774         unsigned long           flags;
 775 
 776         ZONE_PADDING(_pad2_)
 777 
 778         /* Per-node vmstats */
 779         struct per_cpu_nodestat __percpu *per_cpu_nodestats;
 780         atomic_long_t           vm_stat[NR_VM_NODE_STAT_ITEMS];
 781 } pg_data_t;
 782 
 783 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
 784 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
 785 #ifdef CONFIG_FLAT_NODE_MEM_MAP
 786 #define pgdat_page_nr(pgdat, pagenr)    ((pgdat)->node_mem_map + (pagenr))
 787 #else
 788 #define pgdat_page_nr(pgdat, pagenr)    pfn_to_page((pgdat)->node_start_pfn + (pagenr))
 789 #endif
 790 #define nid_page_nr(nid, pagenr)        pgdat_page_nr(NODE_DATA(nid),(pagenr))
 791 
 792 #define node_start_pfn(nid)     (NODE_DATA(nid)->node_start_pfn)
 793 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
 794 
 795 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
 796 {
 797         return &pgdat->lruvec;
 798 }
 799 
 800 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
 801 {
 802         return pgdat->node_start_pfn + pgdat->node_spanned_pages;
 803 }
 804 
 805 static inline bool pgdat_is_empty(pg_data_t *pgdat)
 806 {
 807         return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
 808 }
 809 
 810 #include <linux/memory_hotplug.h>
 811 
 812 void build_all_zonelists(pg_data_t *pgdat);
 813 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
 814                    enum zone_type classzone_idx);
 815 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
 816                          int classzone_idx, unsigned int alloc_flags,
 817                          long free_pages);
 818 bool zone_watermark_ok(struct zone *z, unsigned int order,
 819                 unsigned long mark, int classzone_idx,
 820                 unsigned int alloc_flags);
 821 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
 822                 unsigned long mark, int classzone_idx);
 823 enum memmap_context {
 824         MEMMAP_EARLY,
 825         MEMMAP_HOTPLUG,
 826 };
 827 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
 828                                      unsigned long size);
 829 
 830 extern void lruvec_init(struct lruvec *lruvec);
 831 
 832 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
 833 {
 834 #ifdef CONFIG_MEMCG
 835         return lruvec->pgdat;
 836 #else
 837         return container_of(lruvec, struct pglist_data, lruvec);
 838 #endif
 839 }
 840 
 841 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
 842 
 843 #ifdef CONFIG_HAVE_MEMORY_PRESENT
 844 void memory_present(int nid, unsigned long start, unsigned long end);
 845 #else
 846 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
 847 #endif
 848 
 849 #if defined(CONFIG_SPARSEMEM)
 850 void memblocks_present(void);
 851 #else
 852 static inline void memblocks_present(void) {}
 853 #endif
 854 
 855 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
 856 int local_memory_node(int node_id);
 857 #else
 858 static inline int local_memory_node(int node_id) { return node_id; };
 859 #endif
 860 
 861 /*
 862  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
 863  */
 864 #define zone_idx(zone)          ((zone) - (zone)->zone_pgdat->node_zones)
 865 
 866 /*
 867  * Returns true if a zone has pages managed by the buddy allocator.
 868  * All the reclaim decisions have to use this function rather than
 869  * populated_zone(). If the whole zone is reserved then we can easily
 870  * end up with populated_zone() && !managed_zone().
 871  */
 872 static inline bool managed_zone(struct zone *zone)
 873 {
 874         return zone_managed_pages(zone);
 875 }
 876 
 877 /* Returns true if a zone has memory */
 878 static inline bool populated_zone(struct zone *zone)
 879 {
 880         return zone->present_pages;
 881 }
 882 
 883 #ifdef CONFIG_NUMA
 884 static inline int zone_to_nid(struct zone *zone)
 885 {
 886         return zone->node;
 887 }
 888 
 889 static inline void zone_set_nid(struct zone *zone, int nid)
 890 {
 891         zone->node = nid;
 892 }
 893 #else
 894 static inline int zone_to_nid(struct zone *zone)
 895 {
 896         return 0;
 897 }
 898 
 899 static inline void zone_set_nid(struct zone *zone, int nid) {}
 900 #endif
 901 
 902 extern int movable_zone;
 903 
 904 #ifdef CONFIG_HIGHMEM
 905 static inline int zone_movable_is_highmem(void)
 906 {
 907 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
 908         return movable_zone == ZONE_HIGHMEM;
 909 #else
 910         return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
 911 #endif
 912 }
 913 #endif
 914 
 915 static inline int is_highmem_idx(enum zone_type idx)
 916 {
 917 #ifdef CONFIG_HIGHMEM
 918         return (idx == ZONE_HIGHMEM ||
 919                 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
 920 #else
 921         return 0;
 922 #endif
 923 }
 924 
 925 /**
 926  * is_highmem - helper function to quickly check if a struct zone is a
 927  *              highmem zone or not.  This is an attempt to keep references
 928  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
 929  * @zone - pointer to struct zone variable
 930  */
 931 static inline int is_highmem(struct zone *zone)
 932 {
 933 #ifdef CONFIG_HIGHMEM
 934         return is_highmem_idx(zone_idx(zone));
 935 #else
 936         return 0;
 937 #endif
 938 }
 939 
 940 /* These two functions are used to setup the per zone pages min values */
 941 struct ctl_table;
 942 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
 943                                         void __user *, size_t *, loff_t *);
 944 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
 945                                         void __user *, size_t *, loff_t *);
 946 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
 947                                         void __user *, size_t *, loff_t *);
 948 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
 949 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
 950                                         void __user *, size_t *, loff_t *);
 951 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
 952                                         void __user *, size_t *, loff_t *);
 953 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
 954                         void __user *, size_t *, loff_t *);
 955 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
 956                         void __user *, size_t *, loff_t *);
 957 
 958 extern int numa_zonelist_order_handler(struct ctl_table *, int,
 959                         void __user *, size_t *, loff_t *);
 960 extern char numa_zonelist_order[];
 961 #define NUMA_ZONELIST_ORDER_LEN 16
 962 
 963 #ifndef CONFIG_NEED_MULTIPLE_NODES
 964 
 965 extern struct pglist_data contig_page_data;
 966 #define NODE_DATA(nid)          (&contig_page_data)
 967 #define NODE_MEM_MAP(nid)       mem_map
 968 
 969 #else /* CONFIG_NEED_MULTIPLE_NODES */
 970 
 971 #include <asm/mmzone.h>
 972 
 973 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 974 
 975 extern struct pglist_data *first_online_pgdat(void);
 976 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
 977 extern struct zone *next_zone(struct zone *zone);
 978 
 979 /**
 980  * for_each_online_pgdat - helper macro to iterate over all online nodes
 981  * @pgdat - pointer to a pg_data_t variable
 982  */
 983 #define for_each_online_pgdat(pgdat)                    \
 984         for (pgdat = first_online_pgdat();              \
 985              pgdat;                                     \
 986              pgdat = next_online_pgdat(pgdat))
 987 /**
 988  * for_each_zone - helper macro to iterate over all memory zones
 989  * @zone - pointer to struct zone variable
 990  *
 991  * The user only needs to declare the zone variable, for_each_zone
 992  * fills it in.
 993  */
 994 #define for_each_zone(zone)                             \
 995         for (zone = (first_online_pgdat())->node_zones; \
 996              zone;                                      \
 997              zone = next_zone(zone))
 998 
 999 #define for_each_populated_zone(zone)                   \
1000         for (zone = (first_online_pgdat())->node_zones; \
1001              zone;                                      \
1002              zone = next_zone(zone))                    \
1003                 if (!populated_zone(zone))              \
1004                         ; /* do nothing */              \
1005                 else
1006 
1007 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1008 {
1009         return zoneref->zone;
1010 }
1011 
1012 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1013 {
1014         return zoneref->zone_idx;
1015 }
1016 
1017 static inline int zonelist_node_idx(struct zoneref *zoneref)
1018 {
1019         return zone_to_nid(zoneref->zone);
1020 }
1021 
1022 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1023                                         enum zone_type highest_zoneidx,
1024                                         nodemask_t *nodes);
1025 
1026 /**
1027  * 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
1028  * @z - The cursor used as a starting point for the search
1029  * @highest_zoneidx - The zone index of the highest zone to return
1030  * @nodes - An optional nodemask to filter the zonelist with
1031  *
1032  * This function returns the next zone at or below a given zone index that is
1033  * within the allowed nodemask using a cursor as the starting point for the
1034  * search. The zoneref returned is a cursor that represents the current zone
1035  * being examined. It should be advanced by one before calling
1036  * next_zones_zonelist again.
1037  */
1038 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1039                                         enum zone_type highest_zoneidx,
1040                                         nodemask_t *nodes)
1041 {
1042         if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1043                 return z;
1044         return __next_zones_zonelist(z, highest_zoneidx, nodes);
1045 }
1046 
1047 /**
1048  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1049  * @zonelist - The zonelist to search for a suitable zone
1050  * @highest_zoneidx - The zone index of the highest zone to return
1051  * @nodes - An optional nodemask to filter the zonelist with
1052  * @return - Zoneref pointer for the first suitable zone found (see below)
1053  *
1054  * This function returns the first zone at or below a given zone index that is
1055  * within the allowed nodemask. The zoneref returned is a cursor that can be
1056  * used to iterate the zonelist with next_zones_zonelist by advancing it by
1057  * one before calling.
1058  *
1059  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1060  * never NULL). This may happen either genuinely, or due to concurrent nodemask
1061  * update due to cpuset modification.
1062  */
1063 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1064                                         enum zone_type highest_zoneidx,
1065                                         nodemask_t *nodes)
1066 {
1067         return next_zones_zonelist(zonelist->_zonerefs,
1068                                                         highest_zoneidx, nodes);
1069 }
1070 
1071 /**
1072  * 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
1073  * @zone - The current zone in the iterator
1074  * @z - The current pointer within zonelist->zones being iterated
1075  * @zlist - The zonelist being iterated
1076  * @highidx - The zone index of the highest zone to return
1077  * @nodemask - Nodemask allowed by the allocator
1078  *
1079  * This iterator iterates though all zones at or below a given zone index and
1080  * within a given nodemask
1081  */
1082 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1083         for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);       \
1084                 zone;                                                   \
1085                 z = next_zones_zonelist(++z, highidx, nodemask),        \
1086                         zone = zonelist_zone(z))
1087 
1088 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1089         for (zone = z->zone;    \
1090                 zone;                                                   \
1091                 z = next_zones_zonelist(++z, highidx, nodemask),        \
1092                         zone = zonelist_zone(z))
1093 
1094 
1095 /**
1096  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1097  * @zone - The current zone in the iterator
1098  * @z - The current pointer within zonelist->zones being iterated
1099  * @zlist - The zonelist being iterated
1100  * @highidx - The zone index of the highest zone to return
1101  *
1102  * This iterator iterates though all zones at or below a given zone index.
1103  */
1104 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1105         for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1106 
1107 #ifdef CONFIG_SPARSEMEM
1108 #include <asm/sparsemem.h>
1109 #endif
1110 
1111 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1112         !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1113 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1114 {
1115         BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1116         return 0;
1117 }
1118 #endif
1119 
1120 #ifdef CONFIG_FLATMEM
1121 #define pfn_to_nid(pfn)         (0)
1122 #endif
1123 
1124 #ifdef CONFIG_SPARSEMEM
1125 
1126 /*
1127  * SECTION_SHIFT                #bits space required to store a section #
1128  *
1129  * PA_SECTION_SHIFT             physical address to/from section number
1130  * PFN_SECTION_SHIFT            pfn to/from section number
1131  */
1132 #define PA_SECTION_SHIFT        (SECTION_SIZE_BITS)
1133 #define PFN_SECTION_SHIFT       (SECTION_SIZE_BITS - PAGE_SHIFT)
1134 
1135 #define NR_MEM_SECTIONS         (1UL << SECTIONS_SHIFT)
1136 
1137 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1138 #define PAGE_SECTION_MASK       (~(PAGES_PER_SECTION-1))
1139 
1140 #define SECTION_BLOCKFLAGS_BITS \
1141         ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1142 
1143 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1144 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1145 #endif
1146 
1147 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1148 {
1149         return pfn >> PFN_SECTION_SHIFT;
1150 }
1151 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1152 {
1153         return sec << PFN_SECTION_SHIFT;
1154 }
1155 
1156 #define SECTION_ALIGN_UP(pfn)   (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1157 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1158 
1159 #define SUBSECTION_SHIFT 21
1160 
1161 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1162 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1163 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1164 
1165 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1166 #error Subsection size exceeds section size
1167 #else
1168 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1169 #endif
1170 
1171 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1172 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1173 
1174 struct mem_section_usage {
1175         DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1176         /* See declaration of similar field in struct zone */
1177         unsigned long pageblock_flags[0];
1178 };
1179 
1180 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1181 
1182 struct page;
1183 struct page_ext;
1184 struct mem_section {
1185         /*
1186          * This is, logically, a pointer to an array of struct
1187          * pages.  However, it is stored with some other magic.
1188          * (see sparse.c::sparse_init_one_section())
1189          *
1190          * Additionally during early boot we encode node id of
1191          * the location of the section here to guide allocation.
1192          * (see sparse.c::memory_present())
1193          *
1194          * Making it a UL at least makes someone do a cast
1195          * before using it wrong.
1196          */
1197         unsigned long section_mem_map;
1198 
1199         struct mem_section_usage *usage;
1200 #ifdef CONFIG_PAGE_EXTENSION
1201         /*
1202          * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1203          * section. (see page_ext.h about this.)
1204          */
1205         struct page_ext *page_ext;
1206         unsigned long pad;
1207 #endif
1208         /*
1209          * WARNING: mem_section must be a power-of-2 in size for the
1210          * calculation and use of SECTION_ROOT_MASK to make sense.
1211          */
1212 };
1213 
1214 #ifdef CONFIG_SPARSEMEM_EXTREME
1215 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1216 #else
1217 #define SECTIONS_PER_ROOT       1
1218 #endif
1219 
1220 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1221 #define NR_SECTION_ROOTS        DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1222 #define SECTION_ROOT_MASK       (SECTIONS_PER_ROOT - 1)
1223 
1224 #ifdef CONFIG_SPARSEMEM_EXTREME
1225 extern struct mem_section **mem_section;
1226 #else
1227 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1228 #endif
1229 
1230 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1231 {
1232         return ms->usage->pageblock_flags;
1233 }
1234 
1235 static inline struct mem_section *__nr_to_section(unsigned long nr)
1236 {
1237 #ifdef CONFIG_SPARSEMEM_EXTREME
1238         if (!mem_section)
1239                 return NULL;
1240 #endif
1241         if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1242                 return NULL;
1243         return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1244 }
1245 extern unsigned long __section_nr(struct mem_section *ms);
1246 extern size_t mem_section_usage_size(void);
1247 
1248 /*
1249  * We use the lower bits of the mem_map pointer to store
1250  * a little bit of information.  The pointer is calculated
1251  * as mem_map - section_nr_to_pfn(pnum).  The result is
1252  * aligned to the minimum alignment of the two values:
1253  *   1. All mem_map arrays are page-aligned.
1254  *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1255  *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1256  *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1257  *      worst combination is powerpc with 256k pages,
1258  *      which results in PFN_SECTION_SHIFT equal 6.
1259  * To sum it up, at least 6 bits are available.
1260  */
1261 #define SECTION_MARKED_PRESENT  (1UL<<0)
1262 #define SECTION_HAS_MEM_MAP     (1UL<<1)
1263 #define SECTION_IS_ONLINE       (1UL<<2)
1264 #define SECTION_IS_EARLY        (1UL<<3)
1265 #define SECTION_MAP_LAST_BIT    (1UL<<4)
1266 #define SECTION_MAP_MASK        (~(SECTION_MAP_LAST_BIT-1))
1267 #define SECTION_NID_SHIFT       3
1268 
1269 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1270 {
1271         unsigned long map = section->section_mem_map;
1272         map &= SECTION_MAP_MASK;
1273         return (struct page *)map;
1274 }
1275 
1276 static inline int present_section(struct mem_section *section)
1277 {
1278         return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1279 }
1280 
1281 static inline int present_section_nr(unsigned long nr)
1282 {
1283         return present_section(__nr_to_section(nr));
1284 }
1285 
1286 static inline int valid_section(struct mem_section *section)
1287 {
1288         return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1289 }
1290 
1291 static inline int early_section(struct mem_section *section)
1292 {
1293         return (section && (section->section_mem_map & SECTION_IS_EARLY));
1294 }
1295 
1296 static inline int valid_section_nr(unsigned long nr)
1297 {
1298         return valid_section(__nr_to_section(nr));
1299 }
1300 
1301 static inline int online_section(struct mem_section *section)
1302 {
1303         return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1304 }
1305 
1306 static inline int online_section_nr(unsigned long nr)
1307 {
1308         return online_section(__nr_to_section(nr));
1309 }
1310 
1311 #ifdef CONFIG_MEMORY_HOTPLUG
1312 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1313 #ifdef CONFIG_MEMORY_HOTREMOVE
1314 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1315 #endif
1316 #endif
1317 
1318 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1319 {
1320         return __nr_to_section(pfn_to_section_nr(pfn));
1321 }
1322 
1323 extern unsigned long __highest_present_section_nr;
1324 
1325 static inline int subsection_map_index(unsigned long pfn)
1326 {
1327         return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1328 }
1329 
1330 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1331 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1332 {
1333         int idx = subsection_map_index(pfn);
1334 
1335         return test_bit(idx, ms->usage->subsection_map);
1336 }
1337 #else
1338 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1339 {
1340         return 1;
1341 }
1342 #endif
1343 
1344 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1345 static inline int pfn_valid(unsigned long pfn)
1346 {
1347         struct mem_section *ms;
1348 
1349         if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1350                 return 0;
1351         ms = __nr_to_section(pfn_to_section_nr(pfn));
1352         if (!valid_section(ms))
1353                 return 0;
1354         /*
1355          * Traditionally early sections always returned pfn_valid() for
1356          * the entire section-sized span.
1357          */
1358         return early_section(ms) || pfn_section_valid(ms, pfn);
1359 }
1360 #endif
1361 
1362 static inline int pfn_present(unsigned long pfn)
1363 {
1364         if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1365                 return 0;
1366         return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1367 }
1368 
1369 /*
1370  * These are _only_ used during initialisation, therefore they
1371  * can use __initdata ...  They could have names to indicate
1372  * this restriction.
1373  */
1374 #ifdef CONFIG_NUMA
1375 #define pfn_to_nid(pfn)                                                 \
1376 ({                                                                      \
1377         unsigned long __pfn_to_nid_pfn = (pfn);                         \
1378         page_to_nid(pfn_to_page(__pfn_to_nid_pfn));                     \
1379 })
1380 #else
1381 #define pfn_to_nid(pfn)         (0)
1382 #endif
1383 
1384 #define early_pfn_valid(pfn)    pfn_valid(pfn)
1385 void sparse_init(void);
1386 #else
1387 #define sparse_init()   do {} while (0)
1388 #define sparse_index_init(_sec, _nid)  do {} while (0)
1389 #define pfn_present pfn_valid
1390 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1391 #endif /* CONFIG_SPARSEMEM */
1392 
1393 /*
1394  * During memory init memblocks map pfns to nids. The search is expensive and
1395  * this caches recent lookups. The implementation of __early_pfn_to_nid
1396  * may treat start/end as pfns or sections.
1397  */
1398 struct mminit_pfnnid_cache {
1399         unsigned long last_start;
1400         unsigned long last_end;
1401         int last_nid;
1402 };
1403 
1404 #ifndef early_pfn_valid
1405 #define early_pfn_valid(pfn)    (1)
1406 #endif
1407 
1408 void memory_present(int nid, unsigned long start, unsigned long end);
1409 
1410 /*
1411  * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1412  * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1413  * pfn_valid_within() should be used in this case; we optimise this away
1414  * when we have no holes within a MAX_ORDER_NR_PAGES block.
1415  */
1416 #ifdef CONFIG_HOLES_IN_ZONE
1417 #define pfn_valid_within(pfn) pfn_valid(pfn)
1418 #else
1419 #define pfn_valid_within(pfn) (1)
1420 #endif
1421 
1422 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1423 /*
1424  * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1425  * associated with it or not. This means that a struct page exists for this
1426  * pfn. The caller cannot assume the page is fully initialized in general.
1427  * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1428  * will ensure the struct page is fully online and initialized. Special pages
1429  * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1430  *
1431  * In FLATMEM, it is expected that holes always have valid memmap as long as
1432  * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1433  * that a valid section has a memmap for the entire section.
1434  *
1435  * However, an ARM, and maybe other embedded architectures in the future
1436  * free memmap backing holes to save memory on the assumption the memmap is
1437  * never used. The page_zone linkages are then broken even though pfn_valid()
1438  * returns true. A walker of the full memmap must then do this additional
1439  * check to ensure the memmap they are looking at is sane by making sure
1440  * the zone and PFN linkages are still valid. This is expensive, but walkers
1441  * of the full memmap are extremely rare.
1442  */
1443 bool memmap_valid_within(unsigned long pfn,
1444                                         struct page *page, struct zone *zone);
1445 #else
1446 static inline bool memmap_valid_within(unsigned long pfn,
1447                                         struct page *page, struct zone *zone)
1448 {
1449         return true;
1450 }
1451 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1452 
1453 #endif /* !__GENERATING_BOUNDS.H */
1454 #endif /* !__ASSEMBLY__ */
1455 #endif /* _LINUX_MMZONE_H */
/* [<][>][^][v][top][bottom][index][help] */
root/include/linux/mmzone.h

INCLUDED FROM

DEFINITIONS
