Home | History | Annotate | Download | only in linux
      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 struct free_area {
     28 	struct list_head	free_list;
     29 	unsigned long		nr_free;
     30 };
     31 
     32 struct pglist_data;
     33 
     34 /*
     35  * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
     36  * So add a wild amount of padding here to ensure that they fall into separate
     37  * cachelines.  There are very few zone structures in the machine, so space
     38  * consumption is not a concern here.
     39  */
     40 #if defined(CONFIG_SMP)
     41 struct zone_padding {
     42 	char x[0];
     43 } ____cacheline_internodealigned_in_smp;
     44 #define ZONE_PADDING(name)	struct zone_padding name;
     45 #else
     46 #define ZONE_PADDING(name)
     47 #endif
     48 
     49 enum zone_stat_item {
     50 	NR_ANON_PAGES,	/* Mapped anonymous pages */
     51 	NR_FILE_MAPPED,	/* pagecache pages mapped into pagetables.
     52 			   only modified from process context */
     53 	NR_FILE_PAGES,
     54 	NR_SLAB,	/* Pages used by slab allocator */
     55 	NR_PAGETABLE,	/* used for pagetables */
     56 	NR_FILE_DIRTY,
     57 	NR_WRITEBACK,
     58 	NR_UNSTABLE_NFS,	/* NFS unstable pages */
     59 	NR_BOUNCE,
     60 #ifdef CONFIG_NUMA
     61 	NUMA_HIT,		/* allocated in intended node */
     62 	NUMA_MISS,		/* allocated in non intended node */
     63 	NUMA_FOREIGN,		/* was intended here, hit elsewhere */
     64 	NUMA_INTERLEAVE_HIT,	/* interleaver preferred this zone */
     65 	NUMA_LOCAL,		/* allocation from local node */
     66 	NUMA_OTHER,		/* allocation from other node */
     67 #endif
     68 	NR_VM_ZONE_STAT_ITEMS };
     69 
     70 struct per_cpu_pages {
     71 	int count;		/* number of pages in the list */
     72 	int high;		/* high watermark, emptying needed */
     73 	int batch;		/* chunk size for buddy add/remove */
     74 	struct list_head list;	/* the list of pages */
     75 };
     76 
     77 struct per_cpu_pageset {
     78 	struct per_cpu_pages pcp[2];	/* 0: hot.  1: cold */
     79 #ifdef CONFIG_SMP
     80 	s8 stat_threshold;
     81 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
     82 #endif
     83 } ____cacheline_aligned_in_smp;
     84 
     85 #ifdef CONFIG_NUMA
     86 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
     87 #else
     88 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
     89 #endif
     90 
     91 #define ZONE_DMA		0
     92 #define ZONE_DMA32		1
     93 #define ZONE_NORMAL		2
     94 #define ZONE_HIGHMEM		3
     95 
     96 #define MAX_NR_ZONES		4	/* Sync this with ZONES_SHIFT */
     97 #define ZONES_SHIFT		2	/* ceil(log2(MAX_NR_ZONES)) */
     98 
     99 
    100 /*
    101  * When a memory allocation must conform to specific limitations (such
    102  * as being suitable for DMA) the caller will pass in hints to the
    103  * allocator in the gfp_mask, in the zone modifier bits.  These bits
    104  * are used to select a priority ordered list of memory zones which
    105  * match the requested limits.  GFP_ZONEMASK defines which bits within
    106  * the gfp_mask should be considered as zone modifiers.  Each valid
    107  * combination of the zone modifier bits has a corresponding list
    108  * of zones (in node_zonelists).  Thus for two zone modifiers there
    109  * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
    110  * be 8 (2 ** 3) zonelists.  GFP_ZONETYPES defines the number of possible
    111  * combinations of zone modifiers in "zone modifier space".
    112  *
    113  * As an optimisation any zone modifier bits which are only valid when
    114  * no other zone modifier bits are set (loners) should be placed in
    115  * the highest order bits of this field.  This allows us to reduce the
    116  * extent of the zonelists thus saving space.  For example in the case
    117  * of three zone modifier bits, we could require up to eight zonelists.
    118  * If the left most zone modifier is a "loner" then the highest valid
    119  * zonelist would be four allowing us to allocate only five zonelists.
    120  * Use the first form for GFP_ZONETYPES when the left most bit is not
    121  * a "loner", otherwise use the second.
    122  *
    123  * NOTE! Make sure this matches the zones in <linux/gfp.h>
    124  */
    125 #define GFP_ZONEMASK	0x07
    126 /* #define GFP_ZONETYPES       (GFP_ZONEMASK + 1) */           /* Non-loner */
    127 #define GFP_ZONETYPES  ((GFP_ZONEMASK + 1) / 2 + 1)            /* Loner */
    128 
    129 /*
    130  * On machines where it is needed (eg PCs) we divide physical memory
    131  * into multiple physical zones. On a 32bit PC we have 4 zones:
    132  *
    133  * ZONE_DMA	  < 16 MB	ISA DMA capable memory
    134  * ZONE_DMA32	     0 MB 	Empty
    135  * ZONE_NORMAL	16-896 MB	direct mapped by the kernel
    136  * ZONE_HIGHMEM	 > 896 MB	only page cache and user processes
    137  */
    138 
    139 struct zone {
    140 	/* Fields commonly accessed by the page allocator */
    141 	unsigned long		free_pages;
    142 	unsigned long		pages_min, pages_low, pages_high;
    143 	/*
    144 	 * We don't know if the memory that we're going to allocate will be freeable
    145 	 * or/and it will be released eventually, so to avoid totally wasting several
    146 	 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
    147 	 * to run OOM on the lower zones despite there's tons of freeable ram
    148 	 * on the higher zones). This array is recalculated at runtime if the
    149 	 * sysctl_lowmem_reserve_ratio sysctl changes.
    150 	 */
    151 	unsigned long		lowmem_reserve[MAX_NR_ZONES];
    152 
    153 #ifdef CONFIG_NUMA
    154 	/*
    155 	 * zone reclaim becomes active if more unmapped pages exist.
    156 	 */
    157 	unsigned long		min_unmapped_ratio;
    158 	struct per_cpu_pageset	*pageset[NR_CPUS];
    159 #else
    160 	struct per_cpu_pageset	pageset[NR_CPUS];
    161 #endif
    162 	/*
    163 	 * free areas of different sizes
    164 	 */
    165 	spinlock_t		lock;
    166 #ifdef CONFIG_MEMORY_HOTPLUG
    167 	/* see spanned/present_pages for more description */
    168 	seqlock_t		span_seqlock;
    169 #endif
    170 	struct free_area	free_area[MAX_ORDER];
    171 
    172 
    173 	ZONE_PADDING(_pad1_)
    174 
    175 	/* Fields commonly accessed by the page reclaim scanner */
    176 	spinlock_t		lru_lock;
    177 	struct list_head	active_list;
    178 	struct list_head	inactive_list;
    179 	unsigned long		nr_scan_active;
    180 	unsigned long		nr_scan_inactive;
    181 	unsigned long		nr_active;
    182 	unsigned long		nr_inactive;
    183 	unsigned long		pages_scanned;	   /* since last reclaim */
    184 	int			all_unreclaimable; /* All pages pinned */
    185 
    186 	/* A count of how many reclaimers are scanning this zone */
    187 	atomic_t		reclaim_in_progress;
    188 
    189 	/* Zone statistics */
    190 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
    191 
    192 	/*
    193 	 * prev_priority holds the scanning priority for this zone.  It is
    194 	 * defined as the scanning priority at which we achieved our reclaim
    195 	 * target at the previous try_to_free_pages() or balance_pgdat()
    196 	 * invokation.
    197 	 *
    198 	 * We use prev_priority as a measure of how much stress page reclaim is
    199 	 * under - it drives the swappiness decision: whether to unmap mapped
    200 	 * pages.
    201 	 *
    202 	 * temp_priority is used to remember the scanning priority at which
    203 	 * this zone was successfully refilled to free_pages == pages_high.
    204 	 *
    205 	 * Access to both these fields is quite racy even on uniprocessor.  But
    206 	 * it is expected to average out OK.
    207 	 */
    208 	int temp_priority;
    209 	int prev_priority;
    210 
    211 
    212 	ZONE_PADDING(_pad2_)
    213 	/* Rarely used or read-mostly fields */
    214 
    215 	/*
    216 	 * wait_table		-- the array holding the hash table
    217 	 * wait_table_hash_nr_entries	-- the size of the hash table array
    218 	 * wait_table_bits	-- wait_table_size == (1 << wait_table_bits)
    219 	 *
    220 	 * The purpose of all these is to keep track of the people
    221 	 * waiting for a page to become available and make them
    222 	 * runnable again when possible. The trouble is that this
    223 	 * consumes a lot of space, especially when so few things
    224 	 * wait on pages at a given time. So instead of using
    225 	 * per-page waitqueues, we use a waitqueue hash table.
    226 	 *
    227 	 * The bucket discipline is to sleep on the same queue when
    228 	 * colliding and wake all in that wait queue when removing.
    229 	 * When something wakes, it must check to be sure its page is
    230 	 * truly available, a la thundering herd. The cost of a
    231 	 * collision is great, but given the expected load of the
    232 	 * table, they should be so rare as to be outweighed by the
    233 	 * benefits from the saved space.
    234 	 *
    235 	 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
    236 	 * primary users of these fields, and in mm/page_alloc.c
    237 	 * free_area_init_core() performs the initialization of them.
    238 	 */
    239 	wait_queue_head_t	* wait_table;
    240 	unsigned long		wait_table_hash_nr_entries;
    241 	unsigned long		wait_table_bits;
    242 
    243 	/*
    244 	 * Discontig memory support fields.
    245 	 */
    246 	struct pglist_data	*zone_pgdat;
    247 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
    248 	unsigned long		zone_start_pfn;
    249 
    250 	/*
    251 	 * zone_start_pfn, spanned_pages and present_pages are all
    252 	 * protected by span_seqlock.  It is a seqlock because it has
    253 	 * to be read outside of zone->lock, and it is done in the main
    254 	 * allocator path.  But, it is written quite infrequently.
    255 	 *
    256 	 * The lock is declared along with zone->lock because it is
    257 	 * frequently read in proximity to zone->lock.  It's good to
    258 	 * give them a chance of being in the same cacheline.
    259 	 */
    260 	unsigned long		spanned_pages;	/* total size, including holes */
    261 	unsigned long		present_pages;	/* amount of memory (excluding holes) */
    262 
    263 	/*
    264 	 * rarely used fields:
    265 	 */
    266 	char			*name;
    267 } ____cacheline_internodealigned_in_smp;
    268 
    269 
    270 /*
    271  * The "priority" of VM scanning is how much of the queues we will scan in one
    272  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
    273  * queues ("queue_length >> 12") during an aging round.
    274  */
    275 #define DEF_PRIORITY 12
    276 
    277 /*
    278  * One allocation request operates on a zonelist. A zonelist
    279  * is a list of zones, the first one is the 'goal' of the
    280  * allocation, the other zones are fallback zones, in decreasing
    281  * priority.
    282  *
    283  * Right now a zonelist takes up less than a cacheline. We never
    284  * modify it apart from boot-up, and only a few indices are used,
    285  * so despite the zonelist table being relatively big, the cache
    286  * footprint of this construct is very small.
    287  */
    288 struct zonelist {
    289 	struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
    290 };
    291 
    292 
    293 /*
    294  * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
    295  * (mostly NUMA machines?) to denote a higher-level memory zone than the
    296  * zone denotes.
    297  *
    298  * On NUMA machines, each NUMA node would have a pg_data_t to describe
    299  * it's memory layout.
    300  *
    301  * Memory statistics and page replacement data structures are maintained on a
    302  * per-zone basis.
    303  */
    304 struct bootmem_data;
    305 typedef struct pglist_data {
    306 	struct zone node_zones[MAX_NR_ZONES];
    307 	struct zonelist node_zonelists[GFP_ZONETYPES];
    308 	int nr_zones;
    309 #ifdef CONFIG_FLAT_NODE_MEM_MAP
    310 	struct page *node_mem_map;
    311 #endif
    312 	struct bootmem_data *bdata;
    313 #ifdef CONFIG_MEMORY_HOTPLUG
    314 	/*
    315 	 * Must be held any time you expect node_start_pfn, node_present_pages
    316 	 * or node_spanned_pages stay constant.  Holding this will also
    317 	 * guarantee that any pfn_valid() stays that way.
    318 	 *
    319 	 * Nests above zone->lock and zone->size_seqlock.
    320 	 */
    321 	spinlock_t node_size_lock;
    322 #endif
    323 	unsigned long node_start_pfn;
    324 	unsigned long node_present_pages; /* total number of physical pages */
    325 	unsigned long node_spanned_pages; /* total size of physical page
    326 					     range, including holes */
    327 	int node_id;
    328 	wait_queue_head_t kswapd_wait;
    329 	struct task_struct *kswapd;
    330 	int kswapd_max_order;
    331 } pg_data_t;
    332 
    333 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
    334 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
    335 #ifdef CONFIG_FLAT_NODE_MEM_MAP
    336 #define pgdat_page_nr(pgdat, pagenr)	((pgdat)->node_mem_map + (pagenr))
    337 #else
    338 #define pgdat_page_nr(pgdat, pagenr)	pfn_to_page((pgdat)->node_start_pfn + (pagenr))
    339 #endif
    340 #define nid_page_nr(nid, pagenr) 	pgdat_page_nr(NODE_DATA(nid),(pagenr))
    341 
    342 #include <linux/memory_hotplug.h>
    343 
    344 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
    345 			unsigned long *free, struct pglist_data *pgdat);
    346 void get_zone_counts(unsigned long *active, unsigned long *inactive,
    347 			unsigned long *free);
    348 void build_all_zonelists(void);
    349 void wakeup_kswapd(struct zone *zone, int order);
    350 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
    351 		int classzone_idx, int alloc_flags);
    352 
    353 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
    354 				     unsigned long size);
    355 
    356 #ifdef CONFIG_HAVE_MEMORY_PRESENT
    357 void memory_present(int nid, unsigned long start, unsigned long end);
    358 #else
    359 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
    360 #endif
    361 
    362 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
    363 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
    364 #endif
    365 
    366 /*
    367  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
    368  */
    369 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
    370 
    371 static inline int populated_zone(struct zone *zone)
    372 {
    373 	return (!!zone->present_pages);
    374 }
    375 
    376 static inline int is_highmem_idx(int idx)
    377 {
    378 	return (idx == ZONE_HIGHMEM);
    379 }
    380 
    381 static inline int is_normal_idx(int idx)
    382 {
    383 	return (idx == ZONE_NORMAL);
    384 }
    385 
    386 /**
    387  * is_highmem - helper function to quickly check if a struct zone is a
    388  *              highmem zone or not.  This is an attempt to keep references
    389  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
    390  * @zone - pointer to struct zone variable
    391  */
    392 static inline int is_highmem(struct zone *zone)
    393 {
    394 	return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
    395 }
    396 
    397 static inline int is_normal(struct zone *zone)
    398 {
    399 	return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
    400 }
    401 
    402 static inline int is_dma32(struct zone *zone)
    403 {
    404 	return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
    405 }
    406 
    407 static inline int is_dma(struct zone *zone)
    408 {
    409 	return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
    410 }
    411 
    412 /* These two functions are used to setup the per zone pages min values */
    413 struct ctl_table;
    414 struct file;
    415 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
    416 					void __user *, size_t *, loff_t *);
    417 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
    418 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
    419 					void __user *, size_t *, loff_t *);
    420 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
    421 					void __user *, size_t *, loff_t *);
    422 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
    423 			struct file *, void __user *, size_t *, loff_t *);
    424 
    425 #include <linux/topology.h>
    426 /* Returns the number of the current Node. */
    427 #ifndef numa_node_id
    428 #define numa_node_id()		(cpu_to_node(raw_smp_processor_id()))
    429 #endif
    430 
    431 #ifndef CONFIG_NEED_MULTIPLE_NODES
    432 
    433 extern struct pglist_data contig_page_data;
    434 #define NODE_DATA(nid)		(&contig_page_data)
    435 #define NODE_MEM_MAP(nid)	mem_map
    436 #define MAX_NODES_SHIFT		1
    437 
    438 #else /* CONFIG_NEED_MULTIPLE_NODES */
    439 
    440 #include <asm/mmzone.h>
    441 
    442 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
    443 
    444 extern struct pglist_data *first_online_pgdat(void);
    445 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
    446 extern struct zone *next_zone(struct zone *zone);
    447 
    448 /**
    449  * for_each_pgdat - helper macro to iterate over all nodes
    450  * @pgdat - pointer to a pg_data_t variable
    451  */
    452 #define for_each_online_pgdat(pgdat)			\
    453 	for (pgdat = first_online_pgdat();		\
    454 	     pgdat;					\
    455 	     pgdat = next_online_pgdat(pgdat))
    456 /**
    457  * for_each_zone - helper macro to iterate over all memory zones
    458  * @zone - pointer to struct zone variable
    459  *
    460  * The user only needs to declare the zone variable, for_each_zone
    461  * fills it in.
    462  */
    463 #define for_each_zone(zone)			        \
    464 	for (zone = (first_online_pgdat())->node_zones; \
    465 	     zone;					\
    466 	     zone = next_zone(zone))
    467 
    468 #ifdef CONFIG_SPARSEMEM
    469 #include <asm/sparsemem.h>
    470 #endif
    471 
    472 #if BITS_PER_LONG == 32
    473 /*
    474  * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
    475  * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
    476  */
    477 #define FLAGS_RESERVED		9
    478 
    479 #elif BITS_PER_LONG == 64
    480 /*
    481  * with 64 bit flags field, there's plenty of room.
    482  */
    483 #define FLAGS_RESERVED		32
    484 
    485 #else
    486 
    487 #error BITS_PER_LONG not defined
    488 
    489 #endif
    490 
    491 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
    492 #define early_pfn_to_nid(nid)  (0UL)
    493 #endif
    494 
    495 #ifdef CONFIG_FLATMEM
    496 #define pfn_to_nid(pfn)		(0)
    497 #endif
    498 
    499 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
    500 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
    501 
    502 #ifdef CONFIG_SPARSEMEM
    503 
    504 /*
    505  * SECTION_SHIFT    		#bits space required to store a section #
    506  *
    507  * PA_SECTION_SHIFT		physical address to/from section number
    508  * PFN_SECTION_SHIFT		pfn to/from section number
    509  */
    510 #define SECTIONS_SHIFT		(MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
    511 
    512 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
    513 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
    514 
    515 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
    516 
    517 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
    518 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
    519 
    520 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
    521 #error Allocator MAX_ORDER exceeds SECTION_SIZE
    522 #endif
    523 
    524 struct page;
    525 struct mem_section {
    526 	/*
    527 	 * This is, logically, a pointer to an array of struct
    528 	 * pages.  However, it is stored with some other magic.
    529 	 * (see sparse.c::sparse_init_one_section())
    530 	 *
    531 	 * Additionally during early boot we encode node id of
    532 	 * the location of the section here to guide allocation.
    533 	 * (see sparse.c::memory_present())
    534 	 *
    535 	 * Making it a UL at least makes someone do a cast
    536 	 * before using it wrong.
    537 	 */
    538 	unsigned long section_mem_map;
    539 };
    540 
    541 #ifdef CONFIG_SPARSEMEM_EXTREME
    542 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
    543 #else
    544 #define SECTIONS_PER_ROOT	1
    545 #endif
    546 
    547 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
    548 #define NR_SECTION_ROOTS	(NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
    549 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
    550 
    551 #ifdef CONFIG_SPARSEMEM_EXTREME
    552 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
    553 #else
    554 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
    555 #endif
    556 
    557 static inline struct mem_section *__nr_to_section(unsigned long nr)
    558 {
    559 	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
    560 		return NULL;
    561 	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
    562 }
    563 extern int __section_nr(struct mem_section* ms);
    564 
    565 /*
    566  * We use the lower bits of the mem_map pointer to store
    567  * a little bit of information.  There should be at least
    568  * 3 bits here due to 32-bit alignment.
    569  */
    570 #define	SECTION_MARKED_PRESENT	(1UL<<0)
    571 #define SECTION_HAS_MEM_MAP	(1UL<<1)
    572 #define SECTION_MAP_LAST_BIT	(1UL<<2)
    573 #define SECTION_MAP_MASK	(~(SECTION_MAP_LAST_BIT-1))
    574 #define SECTION_NID_SHIFT	2
    575 
    576 static inline struct page *__section_mem_map_addr(struct mem_section *section)
    577 {
    578 	unsigned long map = section->section_mem_map;
    579 	map &= SECTION_MAP_MASK;
    580 	return (struct page *)map;
    581 }
    582 
    583 static inline int valid_section(struct mem_section *section)
    584 {
    585 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
    586 }
    587 
    588 static inline int section_has_mem_map(struct mem_section *section)
    589 {
    590 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
    591 }
    592 
    593 static inline int valid_section_nr(unsigned long nr)
    594 {
    595 	return valid_section(__nr_to_section(nr));
    596 }
    597 
    598 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
    599 {
    600 	return __nr_to_section(pfn_to_section_nr(pfn));
    601 }
    602 
    603 static inline int pfn_valid(unsigned long pfn)
    604 {
    605 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
    606 		return 0;
    607 	return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
    608 }
    609 
    610 /*
    611  * These are _only_ used during initialisation, therefore they
    612  * can use __initdata ...  They could have names to indicate
    613  * this restriction.
    614  */
    615 #ifdef CONFIG_NUMA
    616 #define pfn_to_nid(pfn)							\
    617 ({									\
    618 	unsigned long __pfn_to_nid_pfn = (pfn);				\
    619 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
    620 })
    621 #else
    622 #define pfn_to_nid(pfn)		(0)
    623 #endif
    624 
    625 #define early_pfn_valid(pfn)	pfn_valid(pfn)
    626 void sparse_init(void);
    627 #else
    628 #define sparse_init()	do {} while (0)
    629 #define sparse_index_init(_sec, _nid)  do {} while (0)
    630 #endif /* CONFIG_SPARSEMEM */
    631 
    632 #ifndef early_pfn_valid
    633 #define early_pfn_valid(pfn)	(1)
    634 #endif
    635 
    636 void memory_present(int nid, unsigned long start, unsigned long end);
    637 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
    638 
    639 #endif /* !__ASSEMBLY__ */
    640 #endif /* __KERNEL__ */
    641 #endif /* _LINUX_MMZONE_H */
    642