1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
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 <linux/pageblock-flags.h>
17 #include <asm/atomic.h>
20 /* Free memory management - zoned buddy allocator. */
21 #ifndef CONFIG_FORCE_MAX_ZONEORDER
24 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
26 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
29 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
30 * costly to service. That is between allocation orders which should
31 * coelesce naturally under reasonable reclaim pressure and those which
34 #define PAGE_ALLOC_COSTLY_ORDER 3
36 #define MIGRATE_UNMOVABLE 0
37 #define MIGRATE_MOVABLE 1
38 #define MIGRATE_TYPES 2
40 #define for_each_migratetype_order(order, type) \
41 for (order = 0; order < MAX_ORDER; order++) \
42 for (type = 0; type < MIGRATE_TYPES; type++)
45 struct list_head free_list[MIGRATE_TYPES];
46 unsigned long nr_free;
52 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
53 * So add a wild amount of padding here to ensure that they fall into separate
54 * cachelines. There are very few zone structures in the machine, so space
55 * consumption is not a concern here.
57 #if defined(CONFIG_SMP)
60 } ____cacheline_internodealigned_in_smp;
61 #define ZONE_PADDING(name) struct zone_padding name;
63 #define ZONE_PADDING(name)
67 /* First 128 byte cacheline (assuming 64 bit words) */
71 NR_ANON_PAGES, /* Mapped anonymous pages */
72 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
73 only modified from process context */
77 /* Second 128 byte cacheline */
79 NR_SLAB_UNRECLAIMABLE,
80 NR_PAGETABLE, /* used for pagetables */
81 NR_UNSTABLE_NFS, /* NFS unstable pages */
85 NUMA_HIT, /* allocated in intended node */
86 NUMA_MISS, /* allocated in non intended node */
87 NUMA_FOREIGN, /* was intended here, hit elsewhere */
88 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
89 NUMA_LOCAL, /* allocation from local node */
90 NUMA_OTHER, /* allocation from other node */
92 NR_VM_ZONE_STAT_ITEMS };
94 struct per_cpu_pages {
95 int count; /* number of pages in the list */
96 int high; /* high watermark, emptying needed */
97 int batch; /* chunk size for buddy add/remove */
98 struct list_head list; /* the list of pages */
101 struct per_cpu_pageset {
102 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
108 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
110 } ____cacheline_aligned_in_smp;
113 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
115 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
119 #ifdef CONFIG_ZONE_DMA
121 * ZONE_DMA is used when there are devices that are not able
122 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
123 * carve out the portion of memory that is needed for these devices.
124 * The range is arch specific.
129 * ---------------------------
130 * parisc, ia64, sparc <4G
133 * alpha Unlimited or 0-16MB.
135 * i386, x86_64 and multiple other arches
140 #ifdef CONFIG_ZONE_DMA32
142 * x86_64 needs two ZONE_DMAs because it supports devices that are
143 * only able to do DMA to the lower 16M but also 32 bit devices that
144 * can only do DMA areas below 4G.
149 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
150 * performed on pages in ZONE_NORMAL if the DMA devices support
151 * transfers to all addressable memory.
154 #ifdef CONFIG_HIGHMEM
156 * A memory area that is only addressable by the kernel through
157 * mapping portions into its own address space. This is for example
158 * used by i386 to allow the kernel to address the memory beyond
159 * 900MB. The kernel will set up special mappings (page
160 * table entries on i386) for each page that the kernel needs to
170 * When a memory allocation must conform to specific limitations (such
171 * as being suitable for DMA) the caller will pass in hints to the
172 * allocator in the gfp_mask, in the zone modifier bits. These bits
173 * are used to select a priority ordered list of memory zones which
174 * match the requested limits. See gfp_zone() in include/linux/gfp.h
178 * Count the active zones. Note that the use of defined(X) outside
179 * #if and family is not necessarily defined so ensure we cannot use
180 * it later. Use __ZONE_COUNT to work out how many shift bits we need.
182 #define __ZONE_COUNT ( \
183 defined(CONFIG_ZONE_DMA) \
184 + defined(CONFIG_ZONE_DMA32) \
186 + defined(CONFIG_HIGHMEM) \
190 #define ZONES_SHIFT 0
191 #elif __ZONE_COUNT <= 2
192 #define ZONES_SHIFT 1
193 #elif __ZONE_COUNT <= 4
194 #define ZONES_SHIFT 2
196 #error ZONES_SHIFT -- too many zones configured adjust calculation
201 /* Fields commonly accessed by the page allocator */
202 unsigned long pages_min, pages_low, pages_high;
204 * We don't know if the memory that we're going to allocate will be freeable
205 * or/and it will be released eventually, so to avoid totally wasting several
206 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
207 * to run OOM on the lower zones despite there's tons of freeable ram
208 * on the higher zones). This array is recalculated at runtime if the
209 * sysctl_lowmem_reserve_ratio sysctl changes.
211 unsigned long lowmem_reserve[MAX_NR_ZONES];
216 * zone reclaim becomes active if more unmapped pages exist.
218 unsigned long min_unmapped_pages;
219 unsigned long min_slab_pages;
220 struct per_cpu_pageset *pageset[NR_CPUS];
222 struct per_cpu_pageset pageset[NR_CPUS];
225 * free areas of different sizes
228 #ifdef CONFIG_MEMORY_HOTPLUG
229 /* see spanned/present_pages for more description */
230 seqlock_t span_seqlock;
232 struct free_area free_area[MAX_ORDER];
234 #ifndef CONFIG_SPARSEMEM
236 * Flags for a MAX_ORDER_NR_PAGES block. See pageblock-flags.h.
237 * In SPARSEMEM, this map is stored in struct mem_section
239 unsigned long *pageblock_flags;
240 #endif /* CONFIG_SPARSEMEM */
245 /* Fields commonly accessed by the page reclaim scanner */
247 struct list_head active_list;
248 struct list_head inactive_list;
249 unsigned long nr_scan_active;
250 unsigned long nr_scan_inactive;
251 unsigned long pages_scanned; /* since last reclaim */
252 int all_unreclaimable; /* All pages pinned */
254 /* A count of how many reclaimers are scanning this zone */
255 atomic_t reclaim_in_progress;
257 /* Zone statistics */
258 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
261 * prev_priority holds the scanning priority for this zone. It is
262 * defined as the scanning priority at which we achieved our reclaim
263 * target at the previous try_to_free_pages() or balance_pgdat()
266 * We use prev_priority as a measure of how much stress page reclaim is
267 * under - it drives the swappiness decision: whether to unmap mapped
270 * Access to both this field is quite racy even on uniprocessor. But
271 * it is expected to average out OK.
277 /* Rarely used or read-mostly fields */
280 * wait_table -- the array holding the hash table
281 * wait_table_hash_nr_entries -- the size of the hash table array
282 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
284 * The purpose of all these is to keep track of the people
285 * waiting for a page to become available and make them
286 * runnable again when possible. The trouble is that this
287 * consumes a lot of space, especially when so few things
288 * wait on pages at a given time. So instead of using
289 * per-page waitqueues, we use a waitqueue hash table.
291 * The bucket discipline is to sleep on the same queue when
292 * colliding and wake all in that wait queue when removing.
293 * When something wakes, it must check to be sure its page is
294 * truly available, a la thundering herd. The cost of a
295 * collision is great, but given the expected load of the
296 * table, they should be so rare as to be outweighed by the
297 * benefits from the saved space.
299 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
300 * primary users of these fields, and in mm/page_alloc.c
301 * free_area_init_core() performs the initialization of them.
303 wait_queue_head_t * wait_table;
304 unsigned long wait_table_hash_nr_entries;
305 unsigned long wait_table_bits;
308 * Discontig memory support fields.
310 struct pglist_data *zone_pgdat;
311 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
312 unsigned long zone_start_pfn;
315 * zone_start_pfn, spanned_pages and present_pages are all
316 * protected by span_seqlock. It is a seqlock because it has
317 * to be read outside of zone->lock, and it is done in the main
318 * allocator path. But, it is written quite infrequently.
320 * The lock is declared along with zone->lock because it is
321 * frequently read in proximity to zone->lock. It's good to
322 * give them a chance of being in the same cacheline.
324 unsigned long spanned_pages; /* total size, including holes */
325 unsigned long present_pages; /* amount of memory (excluding holes) */
328 * rarely used fields:
331 } ____cacheline_internodealigned_in_smp;
334 * The "priority" of VM scanning is how much of the queues we will scan in one
335 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
336 * queues ("queue_length >> 12") during an aging round.
338 #define DEF_PRIORITY 12
340 /* Maximum number of zones on a zonelist */
341 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
346 * The NUMA zonelists are doubled becausse we need zonelists that restrict the
347 * allocations to a single node for GFP_THISNODE.
349 * [0 .. MAX_NR_ZONES -1] : Zonelists with fallback
350 * [MAZ_NR_ZONES ... MAZ_ZONELISTS -1] : No fallback (GFP_THISNODE)
352 #define MAX_ZONELISTS (2 * MAX_NR_ZONES)
356 * We cache key information from each zonelist for smaller cache
357 * footprint when scanning for free pages in get_page_from_freelist().
359 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
360 * up short of free memory since the last time (last_fullzone_zap)
361 * we zero'd fullzones.
362 * 2) The array z_to_n[] maps each zone in the zonelist to its node
363 * id, so that we can efficiently evaluate whether that node is
364 * set in the current tasks mems_allowed.
366 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
367 * indexed by a zones offset in the zonelist zones[] array.
369 * The get_page_from_freelist() routine does two scans. During the
370 * first scan, we skip zones whose corresponding bit in 'fullzones'
371 * is set or whose corresponding node in current->mems_allowed (which
372 * comes from cpusets) is not set. During the second scan, we bypass
373 * this zonelist_cache, to ensure we look methodically at each zone.
375 * Once per second, we zero out (zap) fullzones, forcing us to
376 * reconsider nodes that might have regained more free memory.
377 * The field last_full_zap is the time we last zapped fullzones.
379 * This mechanism reduces the amount of time we waste repeatedly
380 * reexaming zones for free memory when they just came up low on
381 * memory momentarilly ago.
383 * The zonelist_cache struct members logically belong in struct
384 * zonelist. However, the mempolicy zonelists constructed for
385 * MPOL_BIND are intentionally variable length (and usually much
386 * shorter). A general purpose mechanism for handling structs with
387 * multiple variable length members is more mechanism than we want
388 * here. We resort to some special case hackery instead.
390 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
391 * part because they are shorter), so we put the fixed length stuff
392 * at the front of the zonelist struct, ending in a variable length
393 * zones[], as is needed by MPOL_BIND.
395 * Then we put the optional zonelist cache on the end of the zonelist
396 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
397 * the fixed length portion at the front of the struct. This pointer
398 * both enables us to find the zonelist cache, and in the case of
399 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
400 * to know that the zonelist cache is not there.
402 * The end result is that struct zonelists come in two flavors:
403 * 1) The full, fixed length version, shown below, and
404 * 2) The custom zonelists for MPOL_BIND.
405 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
407 * Even though there may be multiple CPU cores on a node modifying
408 * fullzones or last_full_zap in the same zonelist_cache at the same
409 * time, we don't lock it. This is just hint data - if it is wrong now
410 * and then, the allocator will still function, perhaps a bit slower.
414 struct zonelist_cache {
415 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
416 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
417 unsigned long last_full_zap; /* when last zap'd (jiffies) */
420 #define MAX_ZONELISTS MAX_NR_ZONES
421 struct zonelist_cache;
425 * One allocation request operates on a zonelist. A zonelist
426 * is a list of zones, the first one is the 'goal' of the
427 * allocation, the other zones are fallback zones, in decreasing
430 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
431 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
435 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
436 struct zone *zones[MAX_ZONES_PER_ZONELIST + 1]; // NULL delimited
438 struct zonelist_cache zlcache; // optional ...
444 * Only custom zonelists like MPOL_BIND need to be filtered as part of
445 * policies. As described in the comment for struct zonelist_cache, these
446 * zonelists will not have a zlcache so zlcache_ptr will not be set. Use
447 * that to determine if the zonelists needs to be filtered or not.
449 static inline int alloc_should_filter_zonelist(struct zonelist *zonelist)
451 return !zonelist->zlcache_ptr;
454 static inline int alloc_should_filter_zonelist(struct zonelist *zonelist)
458 #endif /* CONFIG_NUMA */
460 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
461 struct node_active_region {
462 unsigned long start_pfn;
463 unsigned long end_pfn;
466 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
468 #ifndef CONFIG_DISCONTIGMEM
469 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
470 extern struct page *mem_map;
474 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
475 * (mostly NUMA machines?) to denote a higher-level memory zone than the
478 * On NUMA machines, each NUMA node would have a pg_data_t to describe
479 * it's memory layout.
481 * Memory statistics and page replacement data structures are maintained on a
485 typedef struct pglist_data {
486 struct zone node_zones[MAX_NR_ZONES];
487 struct zonelist node_zonelists[MAX_ZONELISTS];
489 #ifdef CONFIG_FLAT_NODE_MEM_MAP
490 struct page *node_mem_map;
492 struct bootmem_data *bdata;
493 #ifdef CONFIG_MEMORY_HOTPLUG
495 * Must be held any time you expect node_start_pfn, node_present_pages
496 * or node_spanned_pages stay constant. Holding this will also
497 * guarantee that any pfn_valid() stays that way.
499 * Nests above zone->lock and zone->size_seqlock.
501 spinlock_t node_size_lock;
503 unsigned long node_start_pfn;
504 unsigned long node_present_pages; /* total number of physical pages */
505 unsigned long node_spanned_pages; /* total size of physical page
506 range, including holes */
508 wait_queue_head_t kswapd_wait;
509 struct task_struct *kswapd;
510 int kswapd_max_order;
513 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
514 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
515 #ifdef CONFIG_FLAT_NODE_MEM_MAP
516 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
518 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
520 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
522 #include <linux/memory_hotplug.h>
524 void get_zone_counts(unsigned long *active, unsigned long *inactive,
525 unsigned long *free);
526 void build_all_zonelists(void);
527 void wakeup_kswapd(struct zone *zone, int order);
528 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
529 int classzone_idx, int alloc_flags);
530 enum memmap_context {
534 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
536 enum memmap_context context);
538 #ifdef CONFIG_HAVE_MEMORY_PRESENT
539 void memory_present(int nid, unsigned long start, unsigned long end);
541 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
544 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
545 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
549 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
551 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
553 static inline int populated_zone(struct zone *zone)
555 return (!!zone->present_pages);
558 extern int movable_zone;
560 static inline int zone_movable_is_highmem(void)
562 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
563 return movable_zone == ZONE_HIGHMEM;
569 static inline int is_highmem_idx(enum zone_type idx)
571 #ifdef CONFIG_HIGHMEM
572 return (idx == ZONE_HIGHMEM ||
573 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
579 static inline int is_normal_idx(enum zone_type idx)
581 return (idx == ZONE_NORMAL);
585 * is_highmem - helper function to quickly check if a struct zone is a
586 * highmem zone or not. This is an attempt to keep references
587 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
588 * @zone - pointer to struct zone variable
590 static inline int is_highmem(struct zone *zone)
592 #ifdef CONFIG_HIGHMEM
593 int zone_idx = zone - zone->zone_pgdat->node_zones;
594 return zone_idx == ZONE_HIGHMEM ||
595 (zone_idx == ZONE_MOVABLE && zone_movable_is_highmem());
601 static inline int is_normal(struct zone *zone)
603 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
606 static inline int is_dma32(struct zone *zone)
608 #ifdef CONFIG_ZONE_DMA32
609 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
615 static inline int is_dma(struct zone *zone)
617 #ifdef CONFIG_ZONE_DMA
618 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
624 /* These two functions are used to setup the per zone pages min values */
627 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
628 void __user *, size_t *, loff_t *);
629 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
630 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
631 void __user *, size_t *, loff_t *);
632 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
633 void __user *, size_t *, loff_t *);
634 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
635 struct file *, void __user *, size_t *, loff_t *);
636 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
637 struct file *, void __user *, size_t *, loff_t *);
639 extern int numa_zonelist_order_handler(struct ctl_table *, int,
640 struct file *, void __user *, size_t *, loff_t *);
641 extern char numa_zonelist_order[];
642 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
644 #include <linux/topology.h>
645 /* Returns the number of the current Node. */
647 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
650 #ifndef CONFIG_NEED_MULTIPLE_NODES
652 extern struct pglist_data contig_page_data;
653 #define NODE_DATA(nid) (&contig_page_data)
654 #define NODE_MEM_MAP(nid) mem_map
655 #define MAX_NODES_SHIFT 1
657 #else /* CONFIG_NEED_MULTIPLE_NODES */
659 #include <asm/mmzone.h>
661 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
663 extern struct pglist_data *first_online_pgdat(void);
664 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
665 extern struct zone *next_zone(struct zone *zone);
668 * for_each_pgdat - helper macro to iterate over all nodes
669 * @pgdat - pointer to a pg_data_t variable
671 #define for_each_online_pgdat(pgdat) \
672 for (pgdat = first_online_pgdat(); \
674 pgdat = next_online_pgdat(pgdat))
676 * for_each_zone - helper macro to iterate over all memory zones
677 * @zone - pointer to struct zone variable
679 * The user only needs to declare the zone variable, for_each_zone
682 #define for_each_zone(zone) \
683 for (zone = (first_online_pgdat())->node_zones; \
685 zone = next_zone(zone))
687 #ifdef CONFIG_SPARSEMEM
688 #include <asm/sparsemem.h>
691 #if BITS_PER_LONG == 32
693 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
694 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
696 #define FLAGS_RESERVED 9
698 #elif BITS_PER_LONG == 64
700 * with 64 bit flags field, there's plenty of room.
702 #define FLAGS_RESERVED 32
706 #error BITS_PER_LONG not defined
710 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
711 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
712 #define early_pfn_to_nid(nid) (0UL)
715 #ifdef CONFIG_FLATMEM
716 #define pfn_to_nid(pfn) (0)
719 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
720 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
722 #ifdef CONFIG_SPARSEMEM
725 * SECTION_SHIFT #bits space required to store a section #
727 * PA_SECTION_SHIFT physical address to/from section number
728 * PFN_SECTION_SHIFT pfn to/from section number
730 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
732 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
733 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
735 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
737 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
738 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
740 #define SECTION_BLOCKFLAGS_BITS \
741 ((SECTION_SIZE_BITS - (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS)
743 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
744 #error Allocator MAX_ORDER exceeds SECTION_SIZE
750 * This is, logically, a pointer to an array of struct
751 * pages. However, it is stored with some other magic.
752 * (see sparse.c::sparse_init_one_section())
754 * Additionally during early boot we encode node id of
755 * the location of the section here to guide allocation.
756 * (see sparse.c::memory_present())
758 * Making it a UL at least makes someone do a cast
759 * before using it wrong.
761 unsigned long section_mem_map;
762 DECLARE_BITMAP(pageblock_flags, SECTION_BLOCKFLAGS_BITS);
765 #ifdef CONFIG_SPARSEMEM_EXTREME
766 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
768 #define SECTIONS_PER_ROOT 1
771 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
772 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
773 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
775 #ifdef CONFIG_SPARSEMEM_EXTREME
776 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
778 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
781 static inline struct mem_section *__nr_to_section(unsigned long nr)
783 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
785 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
787 extern int __section_nr(struct mem_section* ms);
790 * We use the lower bits of the mem_map pointer to store
791 * a little bit of information. There should be at least
792 * 3 bits here due to 32-bit alignment.
794 #define SECTION_MARKED_PRESENT (1UL<<0)
795 #define SECTION_HAS_MEM_MAP (1UL<<1)
796 #define SECTION_MAP_LAST_BIT (1UL<<2)
797 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
798 #define SECTION_NID_SHIFT 2
800 static inline struct page *__section_mem_map_addr(struct mem_section *section)
802 unsigned long map = section->section_mem_map;
803 map &= SECTION_MAP_MASK;
804 return (struct page *)map;
807 static inline int present_section(struct mem_section *section)
809 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
812 static inline int present_section_nr(unsigned long nr)
814 return present_section(__nr_to_section(nr));
817 static inline int valid_section(struct mem_section *section)
819 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
822 static inline int valid_section_nr(unsigned long nr)
824 return valid_section(__nr_to_section(nr));
827 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
829 return __nr_to_section(pfn_to_section_nr(pfn));
832 static inline int pfn_valid(unsigned long pfn)
834 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
836 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
839 static inline int pfn_present(unsigned long pfn)
841 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
843 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
847 * These are _only_ used during initialisation, therefore they
848 * can use __initdata ... They could have names to indicate
852 #define pfn_to_nid(pfn) \
854 unsigned long __pfn_to_nid_pfn = (pfn); \
855 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
858 #define pfn_to_nid(pfn) (0)
861 #define early_pfn_valid(pfn) pfn_valid(pfn)
862 void sparse_init(void);
864 #define sparse_init() do {} while (0)
865 #define sparse_index_init(_sec, _nid) do {} while (0)
866 #endif /* CONFIG_SPARSEMEM */
868 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
869 #define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid))
871 #define early_pfn_in_nid(pfn, nid) (1)
874 #ifndef early_pfn_valid
875 #define early_pfn_valid(pfn) (1)
878 void memory_present(int nid, unsigned long start, unsigned long end);
879 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
882 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
883 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
884 * pfn_valid_within() should be used in this case; we optimise this away
885 * when we have no holes within a MAX_ORDER_NR_PAGES block.
887 #ifdef CONFIG_HOLES_IN_ZONE
888 #define pfn_valid_within(pfn) pfn_valid(pfn)
890 #define pfn_valid_within(pfn) (1)
893 #endif /* !__ASSEMBLY__ */
894 #endif /* __KERNEL__ */
895 #endif /* _LINUX_MMZONE_H */