include/linux/mmzone.h

   1 #ifndef _LINUX_MMZONE_H
   2 #define _LINUX_MMZONE_H
   3
   4 #ifdef __KERNEL__
   5 #ifndef __ASSEMBLY__
   6
   7 #include <linux/spinlock.h>
   8 #include <linux/list.h>
   9 #include <linux/wait.h>
  10 #include <linux/cache.h>
  11 #include <linux/threads.h>
  12 #include <linux/numa.h>
  13 #include <linux/init.h>
  14 #include <linux/seqlock.h>
  15 #include <linux/nodemask.h>
  16 #include <asm/atomic.h>
  17 #include <asm/page.h>
  18
  19 /* Free memory management - zoned buddy allocator.  */
  20 #ifndef CONFIG_FORCE_MAX_ZONEORDER
  21 #define MAX_ORDER 11
  22 #else
  23 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
  24 #endif
  25 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
  26
  27 /*
  28  * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
  29  * costly to service.  That is between allocation orders which should
  30  * coelesce naturally under reasonable reclaim pressure and those which
  31  * will not.
  32  */
  33 #define PAGE_ALLOC_COSTLY_ORDER 3
  34
  35 struct free_area {
  36         struct list_head        free_list;
  37         unsigned long           nr_free;
  38 };
  39
  40 struct pglist_data;
  41
  42 /*
  43  * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
  44  * So add a wild amount of padding here to ensure that they fall into separate
  45  * cachelines.  There are very few zone structures in the machine, so space
  46  * consumption is not a concern here.
  47  */
  48 #if defined(CONFIG_SMP)
  49 struct zone_padding {
  50         char x[0];
  51 } ____cacheline_internodealigned_in_smp;
  52 #define ZONE_PADDING(name)      struct zone_padding name;
  53 #else
  54 #define ZONE_PADDING(name)
  55 #endif
  56
  57 enum zone_stat_item {
  58         /* First 128 byte cacheline (assuming 64 bit words) */
  59         NR_FREE_PAGES,
  60         NR_INACTIVE,
  61         NR_ACTIVE,
  62         NR_ANON_PAGES,  /* Mapped anonymous pages */
  63         NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
  64                            only modified from process context */
  65         NR_FILE_PAGES,
  66         NR_FILE_DIRTY,
  67         NR_WRITEBACK,
  68         /* Second 128 byte cacheline */
  69         NR_SLAB_RECLAIMABLE,
  70         NR_SLAB_UNRECLAIMABLE,
  71         NR_PAGETABLE,           /* used for pagetables */
  72         NR_UNSTABLE_NFS,        /* NFS unstable pages */
  73         NR_BOUNCE,
  74         NR_VMSCAN_WRITE,
  75 #ifdef CONFIG_NUMA
  76         NUMA_HIT,               /* allocated in intended node */
  77         NUMA_MISS,              /* allocated in non intended node */
  78         NUMA_FOREIGN,           /* was intended here, hit elsewhere */
  79         NUMA_INTERLEAVE_HIT,    /* interleaver preferred this zone */
  80         NUMA_LOCAL,             /* allocation from local node */
  81         NUMA_OTHER,             /* allocation from other node */
  82 #endif
  83         NR_VM_ZONE_STAT_ITEMS };
  84
  85 struct per_cpu_pages {
  86         int count;              /* number of pages in the list */
  87         int high;               /* high watermark, emptying needed */
  88         int batch;              /* chunk size for buddy add/remove */
  89         struct list_head list;  /* the list of pages */
  90 };
  91
  92 struct per_cpu_pageset {
  93         struct per_cpu_pages pcp[2];    /* 0: hot.  1: cold */
  94 #ifdef CONFIG_NUMA
  95         s8 expire;
  96 #endif
  97 #ifdef CONFIG_SMP
  98         s8 stat_threshold;
  99         s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
 100 #endif
 101 } ____cacheline_aligned_in_smp;
 102
 103 #ifdef CONFIG_NUMA
 104 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
 105 #else
 106 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
 107 #endif
 108
 109 enum zone_type {
 110 #ifdef CONFIG_ZONE_DMA
 111         /*
 112          * ZONE_DMA is used when there are devices that are not able
 113          * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
 114          * carve out the portion of memory that is needed for these devices.
 115          * The range is arch specific.
 116          *
 117          * Some examples
 118          *
 119          * Architecture         Limit
 120          * ---------------------------
 121          * parisc, ia64, sparc  <4G
 122          * s390                 <2G
 123          * arm                  Various
 124          * alpha                Unlimited or 0-16MB.
 125          *
 126          * i386, x86_64 and multiple other arches
 127          *                      <16M.
 128          */
 129         ZONE_DMA,
 130 #endif
 131 #ifdef CONFIG_ZONE_DMA32
 132         /*
 133          * x86_64 needs two ZONE_DMAs because it supports devices that are
 134          * only able to do DMA to the lower 16M but also 32 bit devices that
 135          * can only do DMA areas below 4G.
 136          */
 137         ZONE_DMA32,
 138 #endif
 139         /*
 140          * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
 141          * performed on pages in ZONE_NORMAL if the DMA devices support
 142          * transfers to all addressable memory.
 143          */
 144         ZONE_NORMAL,
 145 #ifdef CONFIG_HIGHMEM
 146         /*
 147          * A memory area that is only addressable by the kernel through
 148          * mapping portions into its own address space. This is for example
 149          * used by i386 to allow the kernel to address the memory beyond
 150          * 900MB. The kernel will set up special mappings (page
 151          * table entries on i386) for each page that the kernel needs to
 152          * access.
 153          */
 154         ZONE_HIGHMEM,
 155 #endif
 156         ZONE_MOVABLE,
 157         MAX_NR_ZONES
 158 };
 159
 160 /*
 161  * When a memory allocation must conform to specific limitations (such
 162  * as being suitable for DMA) the caller will pass in hints to the
 163  * allocator in the gfp_mask, in the zone modifier bits.  These bits
 164  * are used to select a priority ordered list of memory zones which
 165  * match the requested limits. See gfp_zone() in include/linux/gfp.h
 166  */
 167
 168 /*
 169  * Count the active zones.  Note that the use of defined(X) outside
 170  * #if and family is not necessarily defined so ensure we cannot use
 171  * it later.  Use __ZONE_COUNT to work out how many shift bits we need.
 172  */
 173 #define __ZONE_COUNT (                  \
 174           defined(CONFIG_ZONE_DMA)      \
 175         + defined(CONFIG_ZONE_DMA32)    \
 176         + 1                             \
 177         + defined(CONFIG_HIGHMEM)       \
 178         + 1                             \
 179 )
 180 #if __ZONE_COUNT < 2
 181 #define ZONES_SHIFT 0
 182 #elif __ZONE_COUNT <= 2
 183 #define ZONES_SHIFT 1
 184 #elif __ZONE_COUNT <= 4
 185 #define ZONES_SHIFT 2
 186 #else
 187 #error ZONES_SHIFT -- too many zones configured adjust calculation
 188 #endif
 189 #undef __ZONE_COUNT
 190
 191 struct zone {
 192         /* Fields commonly accessed by the page allocator */
 193         unsigned long           pages_min, pages_low, pages_high;
 194         /*
 195          * We don't know if the memory that we're going to allocate will be freeable
 196          * or/and it will be released eventually, so to avoid totally wasting several
 197          * GB of ram we must reserve some of the lower zone memory (otherwise we risk
 198          * to run OOM on the lower zones despite there's tons of freeable ram
 199          * on the higher zones). This array is recalculated at runtime if the
 200          * sysctl_lowmem_reserve_ratio sysctl changes.
 201          */
 202         unsigned long           lowmem_reserve[MAX_NR_ZONES];
 203
 204 #ifdef CONFIG_NUMA
 205         int node;
 206         /*
 207          * zone reclaim becomes active if more unmapped pages exist.
 208          */
 209         unsigned long           min_unmapped_pages;
 210         unsigned long           min_slab_pages;
 211         struct per_cpu_pageset  *pageset[NR_CPUS];
 212 #else
 213         struct per_cpu_pageset  pageset[NR_CPUS];
 214 #endif
 215         /*
 216          * free areas of different sizes
 217          */
 218         spinlock_t              lock;
 219 #ifdef CONFIG_MEMORY_HOTPLUG
 220         /* see spanned/present_pages for more description */
 221         seqlock_t               span_seqlock;
 222 #endif
 223         struct free_area        free_area[MAX_ORDER];
 224
 225
 226         ZONE_PADDING(_pad1_)
 227
 228         /* Fields commonly accessed by the page reclaim scanner */
 229         spinlock_t              lru_lock;
 230         struct list_head        active_list;
 231         struct list_head        inactive_list;
 232         unsigned long           nr_scan_active;
 233         unsigned long           nr_scan_inactive;
 234         unsigned long           pages_scanned;     /* since last reclaim */
 235         int                     all_unreclaimable; /* All pages pinned */
 236
 237         /* A count of how many reclaimers are scanning this zone */
 238         atomic_t                reclaim_in_progress;
 239
 240         /* Zone statistics */
 241         atomic_long_t           vm_stat[NR_VM_ZONE_STAT_ITEMS];
 242
 243         /*
 244          * prev_priority holds the scanning priority for this zone.  It is
 245          * defined as the scanning priority at which we achieved our reclaim
 246          * target at the previous try_to_free_pages() or balance_pgdat()
 247          * invokation.
 248          *
 249          * We use prev_priority as a measure of how much stress page reclaim is
 250          * under - it drives the swappiness decision: whether to unmap mapped
 251          * pages.
 252          *
 253          * Access to both this field is quite racy even on uniprocessor.  But
 254          * it is expected to average out OK.
 255          */
 256         int prev_priority;
 257
 258
 259         ZONE_PADDING(_pad2_)
 260         /* Rarely used or read-mostly fields */
 261
 262         /*
 263          * wait_table           -- the array holding the hash table
 264          * wait_table_hash_nr_entries   -- the size of the hash table array
 265          * wait_table_bits      -- wait_table_size == (1 << wait_table_bits)
 266          *
 267          * The purpose of all these is to keep track of the people
 268          * waiting for a page to become available and make them
 269          * runnable again when possible. The trouble is that this
 270          * consumes a lot of space, especially when so few things
 271          * wait on pages at a given time. So instead of using
 272          * per-page waitqueues, we use a waitqueue hash table.
 273          *
 274          * The bucket discipline is to sleep on the same queue when
 275          * colliding and wake all in that wait queue when removing.
 276          * When something wakes, it must check to be sure its page is
 277          * truly available, a la thundering herd. The cost of a
 278          * collision is great, but given the expected load of the
 279          * table, they should be so rare as to be outweighed by the
 280          * benefits from the saved space.
 281          *
 282          * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
 283          * primary users of these fields, and in mm/page_alloc.c
 284          * free_area_init_core() performs the initialization of them.
 285          */
 286         wait_queue_head_t       * wait_table;
 287         unsigned long           wait_table_hash_nr_entries;
 288         unsigned long           wait_table_bits;
 289
 290         /*
 291          * Discontig memory support fields.
 292          */
 293         struct pglist_data      *zone_pgdat;
 294         /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
 295         unsigned long           zone_start_pfn;
 296
 297         /*
 298          * zone_start_pfn, spanned_pages and present_pages are all
 299          * protected by span_seqlock.  It is a seqlock because it has
 300          * to be read outside of zone->lock, and it is done in the main
 301          * allocator path.  But, it is written quite infrequently.
 302          *
 303          * The lock is declared along with zone->lock because it is
 304          * frequently read in proximity to zone->lock.  It's good to
 305          * give them a chance of being in the same cacheline.
 306          */
 307         unsigned long           spanned_pages;  /* total size, including holes */
 308         unsigned long           present_pages;  /* amount of memory (excluding holes) */
 309
 310         /*
 311          * rarely used fields:
 312          */
 313         const char              *name;
 314 } ____cacheline_internodealigned_in_smp;
 315
 316 /*
 317  * The "priority" of VM scanning is how much of the queues we will scan in one
 318  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
 319  * queues ("queue_length >> 12") during an aging round.
 320  */
 321 #define DEF_PRIORITY 12
 322
 323 /* Maximum number of zones on a zonelist */
 324 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
 325
 326 #ifdef CONFIG_NUMA
 327 /*
 328  * We cache key information from each zonelist for smaller cache
 329  * footprint when scanning for free pages in get_page_from_freelist().
 330  *
 331  * 1) The BITMAP fullzones tracks which zones in a zonelist have come
 332  *    up short of free memory since the last time (last_fullzone_zap)
 333  *    we zero'd fullzones.
 334  * 2) The array z_to_n[] maps each zone in the zonelist to its node
 335  *    id, so that we can efficiently evaluate whether that node is
 336  *    set in the current tasks mems_allowed.
 337  *
 338  * Both fullzones and z_to_n[] are one-to-one with the zonelist,
 339  * indexed by a zones offset in the zonelist zones[] array.
 340  *
 341  * The get_page_from_freelist() routine does two scans.  During the
 342  * first scan, we skip zones whose corresponding bit in 'fullzones'
 343  * is set or whose corresponding node in current->mems_allowed (which
 344  * comes from cpusets) is not set.  During the second scan, we bypass
 345  * this zonelist_cache, to ensure we look methodically at each zone.
 346  *
 347  * Once per second, we zero out (zap) fullzones, forcing us to
 348  * reconsider nodes that might have regained more free memory.
 349  * The field last_full_zap is the time we last zapped fullzones.
 350  *
 351  * This mechanism reduces the amount of time we waste repeatedly
 352  * reexaming zones for free memory when they just came up low on
 353  * memory momentarilly ago.
 354  *
 355  * The zonelist_cache struct members logically belong in struct
 356  * zonelist.  However, the mempolicy zonelists constructed for
 357  * MPOL_BIND are intentionally variable length (and usually much
 358  * shorter).  A general purpose mechanism for handling structs with
 359  * multiple variable length members is more mechanism than we want
 360  * here.  We resort to some special case hackery instead.
 361  *
 362  * The MPOL_BIND zonelists don't need this zonelist_cache (in good
 363  * part because they are shorter), so we put the fixed length stuff
 364  * at the front of the zonelist struct, ending in a variable length
 365  * zones[], as is needed by MPOL_BIND.
 366  *
 367  * Then we put the optional zonelist cache on the end of the zonelist
 368  * struct.  This optional stuff is found by a 'zlcache_ptr' pointer in
 369  * the fixed length portion at the front of the struct.  This pointer
 370  * both enables us to find the zonelist cache, and in the case of
 371  * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
 372  * to know that the zonelist cache is not there.
 373  *
 374  * The end result is that struct zonelists come in two flavors:
 375  *  1) The full, fixed length version, shown below, and
 376  *  2) The custom zonelists for MPOL_BIND.
 377  * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
 378  *
 379  * Even though there may be multiple CPU cores on a node modifying
 380  * fullzones or last_full_zap in the same zonelist_cache at the same
 381  * time, we don't lock it.  This is just hint data - if it is wrong now
 382  * and then, the allocator will still function, perhaps a bit slower.
 383  */
 384
 385
 386 struct zonelist_cache {
 387         unsigned short z_to_n[MAX_ZONES_PER_ZONELIST];          /* zone->nid */
 388         DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST);      /* zone full? */
 389         unsigned long last_full_zap;            /* when last zap'd (jiffies) */
 390 };
 391 #else
 392 struct zonelist_cache;
 393 #endif
 394
 395 /*
 396  * One allocation request operates on a zonelist. A zonelist
 397  * is a list of zones, the first one is the 'goal' of the
 398  * allocation, the other zones are fallback zones, in decreasing
 399  * priority.
 400  *
 401  * If zlcache_ptr is not NULL, then it is just the address of zlcache,
 402  * as explained above.  If zlcache_ptr is NULL, there is no zlcache.
 403  */
 404
 405 struct zonelist {
 406         struct zonelist_cache *zlcache_ptr;                  // NULL or &zlcache
 407         struct zone *zones[MAX_ZONES_PER_ZONELIST + 1];      // NULL delimited
 408 #ifdef CONFIG_NUMA
 409         struct zonelist_cache zlcache;                       // optional ...
 410 #endif
 411 };
 412
 413 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
 414 struct node_active_region {
 415         unsigned long start_pfn;
 416         unsigned long end_pfn;
 417         int nid;
 418 };
 419 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
 420
 421 #ifndef CONFIG_DISCONTIGMEM
 422 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
 423 extern struct page *mem_map;
 424 #endif
 425
 426 /*
 427  * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
 428  * (mostly NUMA machines?) to denote a higher-level memory zone than the
 429  * zone denotes.
 430  *
 431  * On NUMA machines, each NUMA node would have a pg_data_t to describe
 432  * it's memory layout.
 433  *
 434  * Memory statistics and page replacement data structures are maintained on a
 435  * per-zone basis.
 436  */
 437 struct bootmem_data;
 438 typedef struct pglist_data {
 439         struct zone node_zones[MAX_NR_ZONES];
 440         struct zonelist node_zonelists[MAX_NR_ZONES];
 441         int nr_zones;
 442 #ifdef CONFIG_FLAT_NODE_MEM_MAP
 443         struct page *node_mem_map;
 444 #endif
 445         struct bootmem_data *bdata;
 446 #ifdef CONFIG_MEMORY_HOTPLUG
 447         /*
 448          * Must be held any time you expect node_start_pfn, node_present_pages
 449          * or node_spanned_pages stay constant.  Holding this will also
 450          * guarantee that any pfn_valid() stays that way.
 451          *
 452          * Nests above zone->lock and zone->size_seqlock.
 453          */
 454         spinlock_t node_size_lock;
 455 #endif
 456         unsigned long node_start_pfn;
 457         unsigned long node_present_pages; /* total number of physical pages */
 458         unsigned long node_spanned_pages; /* total size of physical page
 459                                              range, including holes */
 460         int node_id;
 461         wait_queue_head_t kswapd_wait;
 462         struct task_struct *kswapd;
 463         int kswapd_max_order;
 464 } pg_data_t;
 465
 466 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
 467 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
 468 #ifdef CONFIG_FLAT_NODE_MEM_MAP
 469 #define pgdat_page_nr(pgdat, pagenr)    ((pgdat)->node_mem_map + (pagenr))
 470 #else
 471 #define pgdat_page_nr(pgdat, pagenr)    pfn_to_page((pgdat)->node_start_pfn + (pagenr))
 472 #endif
 473 #define nid_page_nr(nid, pagenr)        pgdat_page_nr(NODE_DATA(nid),(pagenr))
 474
 475 #include <linux/memory_hotplug.h>
 476
 477 void get_zone_counts(unsigned long *active, unsigned long *inactive,
 478                         unsigned long *free);
 479 void build_all_zonelists(void);
 480 void wakeup_kswapd(struct zone *zone, int order);
 481 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
 482                 int classzone_idx, int alloc_flags);
 483 enum memmap_context {
 484         MEMMAP_EARLY,
 485         MEMMAP_HOTPLUG,
 486 };
 487 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
 488                                      unsigned long size,
 489                                      enum memmap_context context);
 490
 491 #ifdef CONFIG_HAVE_MEMORY_PRESENT
 492 void memory_present(int nid, unsigned long start, unsigned long end);
 493 #else
 494 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
 495 #endif
 496
 497 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
 498 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
 499 #endif
 500
 501 /*
 502  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
 503  */
 504 #define zone_idx(zone)          ((zone) - (zone)->zone_pgdat->node_zones)
 505
 506 static inline int populated_zone(struct zone *zone)
 507 {
 508         return (!!zone->present_pages);
 509 }
 510
 511 extern int movable_zone;
 512
 513 static inline int zone_movable_is_highmem(void)
 514 {
 515 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
 516         return movable_zone == ZONE_HIGHMEM;
 517 #else
 518         return 0;
 519 #endif
 520 }
 521
 522 static inline int is_highmem_idx(enum zone_type idx)
 523 {
 524 #ifdef CONFIG_HIGHMEM
 525         return (idx == ZONE_HIGHMEM ||
 526                 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
 527 #else
 528         return 0;
 529 #endif
 530 }
 531
 532 static inline int is_normal_idx(enum zone_type idx)
 533 {
 534         return (idx == ZONE_NORMAL);
 535 }
 536
 537 /**
 538  * is_highmem - helper function to quickly check if a struct zone is a
 539  *              highmem zone or not.  This is an attempt to keep references
 540  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
 541  * @zone - pointer to struct zone variable
 542  */
 543 static inline int is_highmem(struct zone *zone)
 544 {
 545 #ifdef CONFIG_HIGHMEM
 546         int zone_idx = zone - zone->zone_pgdat->node_zones;
 547         return zone_idx == ZONE_HIGHMEM ||
 548                 (zone_idx == ZONE_MOVABLE && zone_movable_is_highmem());
 549 #else
 550         return 0;
 551 #endif
 552 }
 553
 554 static inline int is_normal(struct zone *zone)
 555 {
 556         return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
 557 }
 558
 559 static inline int is_dma32(struct zone *zone)
 560 {
 561 #ifdef CONFIG_ZONE_DMA32
 562         return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
 563 #else
 564         return 0;
 565 #endif
 566 }
 567
 568 static inline int is_dma(struct zone *zone)
 569 {
 570 #ifdef CONFIG_ZONE_DMA
 571         return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
 572 #else
 573         return 0;
 574 #endif
 575 }
 576
 577 /* These two functions are used to setup the per zone pages min values */
 578 struct ctl_table;
 579 struct file;
 580 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
 581                                         void __user *, size_t *, loff_t *);
 582 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
 583 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
 584                                         void __user *, size_t *, loff_t *);
 585 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
 586                                         void __user *, size_t *, loff_t *);
 587 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
 588                         struct file *, void __user *, size_t *, loff_t *);
 589 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
 590                         struct file *, void __user *, size_t *, loff_t *);
 591
 592 extern int numa_zonelist_order_handler(struct ctl_table *, int,
 593                         struct file *, void __user *, size_t *, loff_t *);
 594 extern char numa_zonelist_order[];
 595 #define NUMA_ZONELIST_ORDER_LEN 16      /* string buffer size */
 596
 597 #include <linux/topology.h>
 598 /* Returns the number of the current Node. */
 599 #ifndef numa_node_id
 600 #define numa_node_id()          (cpu_to_node(raw_smp_processor_id()))
 601 #endif
 602
 603 #ifndef CONFIG_NEED_MULTIPLE_NODES
 604
 605 extern struct pglist_data contig_page_data;
 606 #define NODE_DATA(nid)          (&contig_page_data)
 607 #define NODE_MEM_MAP(nid)       mem_map
 608 #define MAX_NODES_SHIFT         1
 609
 610 #else /* CONFIG_NEED_MULTIPLE_NODES */
 611
 612 #include <asm/mmzone.h>
 613
 614 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 615
 616 extern struct pglist_data *first_online_pgdat(void);
 617 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
 618 extern struct zone *next_zone(struct zone *zone);
 619
 620 /**
 621  * for_each_pgdat - helper macro to iterate over all nodes
 622  * @pgdat - pointer to a pg_data_t variable
 623  */
 624 #define for_each_online_pgdat(pgdat)                    \
 625         for (pgdat = first_online_pgdat();              \
 626              pgdat;                                     \
 627              pgdat = next_online_pgdat(pgdat))
 628 /**
 629  * for_each_zone - helper macro to iterate over all memory zones
 630  * @zone - pointer to struct zone variable
 631  *
 632  * The user only needs to declare the zone variable, for_each_zone
 633  * fills it in.
 634  */
 635 #define for_each_zone(zone)                             \
 636         for (zone = (first_online_pgdat())->node_zones; \
 637              zone;                                      \
 638              zone = next_zone(zone))
 639
 640 #ifdef CONFIG_SPARSEMEM
 641 #include <asm/sparsemem.h>
 642 #endif
 643
 644 #if BITS_PER_LONG == 32
 645 /*
 646  * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
 647  * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
 648  */
 649 #define FLAGS_RESERVED          9
 650
 651 #elif BITS_PER_LONG == 64
 652 /*
 653  * with 64 bit flags field, there's plenty of room.
 654  */
 655 #define FLAGS_RESERVED          32
 656
 657 #else
 658
 659 #error BITS_PER_LONG not defined
 660
 661 #endif
 662
 663 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
 664         !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
 665 #define early_pfn_to_nid(nid)  (0UL)
 666 #endif
 667
 668 #ifdef CONFIG_FLATMEM
 669 #define pfn_to_nid(pfn)         (0)
 670 #endif
 671
 672 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
 673 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
 674
 675 #ifdef CONFIG_SPARSEMEM
 676
 677 /*
 678  * SECTION_SHIFT                #bits space required to store a section #
 679  *
 680  * PA_SECTION_SHIFT             physical address to/from section number
 681  * PFN_SECTION_SHIFT            pfn to/from section number
 682  */
 683 #define SECTIONS_SHIFT          (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
 684
 685 #define PA_SECTION_SHIFT        (SECTION_SIZE_BITS)
 686 #define PFN_SECTION_SHIFT       (SECTION_SIZE_BITS - PAGE_SHIFT)
 687
 688 #define NR_MEM_SECTIONS         (1UL << SECTIONS_SHIFT)
 689
 690 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
 691 #define PAGE_SECTION_MASK       (~(PAGES_PER_SECTION-1))
 692
 693 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
 694 #error Allocator MAX_ORDER exceeds SECTION_SIZE
 695 #endif
 696
 697 struct page;
 698 struct mem_section {
 699         /*
 700          * This is, logically, a pointer to an array of struct
 701          * pages.  However, it is stored with some other magic.
 702          * (see sparse.c::sparse_init_one_section())
 703          *
 704          * Additionally during early boot we encode node id of
 705          * the location of the section here to guide allocation.
 706          * (see sparse.c::memory_present())
 707          *
 708          * Making it a UL at least makes someone do a cast
 709          * before using it wrong.
 710          */
 711         unsigned long section_mem_map;
 712 };
 713
 714 #ifdef CONFIG_SPARSEMEM_EXTREME
 715 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
 716 #else
 717 #define SECTIONS_PER_ROOT       1
 718 #endif
 719
 720 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
 721 #define NR_SECTION_ROOTS        (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
 722 #define SECTION_ROOT_MASK       (SECTIONS_PER_ROOT - 1)
 723
 724 #ifdef CONFIG_SPARSEMEM_EXTREME
 725 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
 726 #else
 727 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
 728 #endif
 729
 730 static inline struct mem_section *__nr_to_section(unsigned long nr)
 731 {
 732         if (!mem_section[SECTION_NR_TO_ROOT(nr)])
 733                 return NULL;
 734         return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
 735 }
 736 extern int __section_nr(struct mem_section* ms);
 737
 738 /*
 739  * We use the lower bits of the mem_map pointer to store
 740  * a little bit of information.  There should be at least
 741  * 3 bits here due to 32-bit alignment.
 742  */
 743 #define SECTION_MARKED_PRESENT  (1UL<<0)
 744 #define SECTION_HAS_MEM_MAP     (1UL<<1)
 745 #define SECTION_MAP_LAST_BIT    (1UL<<2)
 746 #define SECTION_MAP_MASK        (~(SECTION_MAP_LAST_BIT-1))
 747 #define SECTION_NID_SHIFT       2
 748
 749 static inline struct page *__section_mem_map_addr(struct mem_section *section)
 750 {
 751         unsigned long map = section->section_mem_map;
 752         map &= SECTION_MAP_MASK;
 753         return (struct page *)map;
 754 }
 755
 756 static inline int valid_section(struct mem_section *section)
 757 {
 758         return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
 759 }
 760
 761 static inline int section_has_mem_map(struct mem_section *section)
 762 {
 763         return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
 764 }
 765
 766 static inline int valid_section_nr(unsigned long nr)
 767 {
 768         return valid_section(__nr_to_section(nr));
 769 }
 770
 771 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
 772 {
 773         return __nr_to_section(pfn_to_section_nr(pfn));
 774 }
 775
 776 static inline int pfn_valid(unsigned long pfn)
 777 {
 778         if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
 779                 return 0;
 780         return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
 781 }
 782
 783 /*
 784  * These are _only_ used during initialisation, therefore they
 785  * can use __initdata ...  They could have names to indicate
 786  * this restriction.
 787  */
 788 #ifdef CONFIG_NUMA
 789 #define pfn_to_nid(pfn)                                                 \
 790 ({                                                                      \
 791         unsigned long __pfn_to_nid_pfn = (pfn);                         \
 792         page_to_nid(pfn_to_page(__pfn_to_nid_pfn));                     \
 793 })
 794 #else
 795 #define pfn_to_nid(pfn)         (0)
 796 #endif
 797
 798 #define early_pfn_valid(pfn)    pfn_valid(pfn)
 799 void sparse_init(void);
 800 #else
 801 #define sparse_init()   do {} while (0)
 802 #define sparse_index_init(_sec, _nid)  do {} while (0)
 803 #endif /* CONFIG_SPARSEMEM */
 804
 805 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
 806 #define early_pfn_in_nid(pfn, nid)      (early_pfn_to_nid(pfn) == (nid))
 807 #else
 808 #define early_pfn_in_nid(pfn, nid)      (1)
 809 #endif
 810
 811 #ifndef early_pfn_valid
 812 #define early_pfn_valid(pfn)    (1)
 813 #endif
 814
 815 void memory_present(int nid, unsigned long start, unsigned long end);
 816 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
 817
 818 /*
 819  * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
 820  * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
 821  * pfn_valid_within() should be used in this case; we optimise this away
 822  * when we have no holes within a MAX_ORDER_NR_PAGES block.
 823  */
 824 #ifdef CONFIG_HOLES_IN_ZONE
 825 #define pfn_valid_within(pfn) pfn_valid(pfn)
 826 #else
 827 #define pfn_valid_within(pfn) (1)
 828 #endif
 829
 830 #endif /* !__ASSEMBLY__ */
 831 #endif /* __KERNEL__ */
 832 #endif /* _LINUX_MMZONE_H */