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 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
37 #define MIGRATE_UNMOVABLE 0
38 #define MIGRATE_RECLAIMABLE 1
39 #define MIGRATE_MOVABLE 2
40 #define MIGRATE_TYPES 3
42 #define MIGRATE_UNMOVABLE 0
43 #define MIGRATE_UNRECLAIMABLE 0
44 #define MIGRATE_MOVABLE 0
45 #define MIGRATE_TYPES 1
48 #define for_each_migratetype_order(order, type) \
49 for (order = 0; order < MAX_ORDER; order++) \
50 for (type = 0; type < MIGRATE_TYPES; type++)
53 struct list_head free_list[MIGRATE_TYPES];
54 unsigned long nr_free;
60 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
61 * So add a wild amount of padding here to ensure that they fall into separate
62 * cachelines. There are very few zone structures in the machine, so space
63 * consumption is not a concern here.
65 #if defined(CONFIG_SMP)
68 } ____cacheline_internodealigned_in_smp;
69 #define ZONE_PADDING(name) struct zone_padding name;
71 #define ZONE_PADDING(name)
75 /* First 128 byte cacheline (assuming 64 bit words) */
79 NR_ANON_PAGES, /* Mapped anonymous pages */
80 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
81 only modified from process context */
85 /* Second 128 byte cacheline */
87 NR_SLAB_UNRECLAIMABLE,
88 NR_PAGETABLE, /* used for pagetables */
89 NR_UNSTABLE_NFS, /* NFS unstable pages */
93 NUMA_HIT, /* allocated in intended node */
94 NUMA_MISS, /* allocated in non intended node */
95 NUMA_FOREIGN, /* was intended here, hit elsewhere */
96 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
97 NUMA_LOCAL, /* allocation from local node */
98 NUMA_OTHER, /* allocation from other node */
100 NR_VM_ZONE_STAT_ITEMS };
102 struct per_cpu_pages {
103 int count; /* number of pages in the list */
104 int high; /* high watermark, emptying needed */
105 int batch; /* chunk size for buddy add/remove */
106 struct list_head list; /* the list of pages */
109 struct per_cpu_pageset {
110 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
116 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
118 } ____cacheline_aligned_in_smp;
121 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
123 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
127 #ifdef CONFIG_ZONE_DMA
129 * ZONE_DMA is used when there are devices that are not able
130 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
131 * carve out the portion of memory that is needed for these devices.
132 * The range is arch specific.
137 * ---------------------------
138 * parisc, ia64, sparc <4G
141 * alpha Unlimited or 0-16MB.
143 * i386, x86_64 and multiple other arches
148 #ifdef CONFIG_ZONE_DMA32
150 * x86_64 needs two ZONE_DMAs because it supports devices that are
151 * only able to do DMA to the lower 16M but also 32 bit devices that
152 * can only do DMA areas below 4G.
157 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
158 * performed on pages in ZONE_NORMAL if the DMA devices support
159 * transfers to all addressable memory.
162 #ifdef CONFIG_HIGHMEM
164 * A memory area that is only addressable by the kernel through
165 * mapping portions into its own address space. This is for example
166 * used by i386 to allow the kernel to address the memory beyond
167 * 900MB. The kernel will set up special mappings (page
168 * table entries on i386) for each page that the kernel needs to
178 * When a memory allocation must conform to specific limitations (such
179 * as being suitable for DMA) the caller will pass in hints to the
180 * allocator in the gfp_mask, in the zone modifier bits. These bits
181 * are used to select a priority ordered list of memory zones which
182 * match the requested limits. See gfp_zone() in include/linux/gfp.h
186 * Count the active zones. Note that the use of defined(X) outside
187 * #if and family is not necessarily defined so ensure we cannot use
188 * it later. Use __ZONE_COUNT to work out how many shift bits we need.
190 #define __ZONE_COUNT ( \
191 defined(CONFIG_ZONE_DMA) \
192 + defined(CONFIG_ZONE_DMA32) \
194 + defined(CONFIG_HIGHMEM) \
198 #define ZONES_SHIFT 0
199 #elif __ZONE_COUNT <= 2
200 #define ZONES_SHIFT 1
201 #elif __ZONE_COUNT <= 4
202 #define ZONES_SHIFT 2
204 #error ZONES_SHIFT -- too many zones configured adjust calculation
209 /* Fields commonly accessed by the page allocator */
210 unsigned long pages_min, pages_low, pages_high;
212 * We don't know if the memory that we're going to allocate will be freeable
213 * or/and it will be released eventually, so to avoid totally wasting several
214 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
215 * to run OOM on the lower zones despite there's tons of freeable ram
216 * on the higher zones). This array is recalculated at runtime if the
217 * sysctl_lowmem_reserve_ratio sysctl changes.
219 unsigned long lowmem_reserve[MAX_NR_ZONES];
224 * zone reclaim becomes active if more unmapped pages exist.
226 unsigned long min_unmapped_pages;
227 unsigned long min_slab_pages;
228 struct per_cpu_pageset *pageset[NR_CPUS];
230 struct per_cpu_pageset pageset[NR_CPUS];
233 * free areas of different sizes
236 #ifdef CONFIG_MEMORY_HOTPLUG
237 /* see spanned/present_pages for more description */
238 seqlock_t span_seqlock;
240 struct free_area free_area[MAX_ORDER];
242 #ifndef CONFIG_SPARSEMEM
244 * Flags for a MAX_ORDER_NR_PAGES block. See pageblock-flags.h.
245 * In SPARSEMEM, this map is stored in struct mem_section
247 unsigned long *pageblock_flags;
248 #endif /* CONFIG_SPARSEMEM */
253 /* Fields commonly accessed by the page reclaim scanner */
255 struct list_head active_list;
256 struct list_head inactive_list;
257 unsigned long nr_scan_active;
258 unsigned long nr_scan_inactive;
259 unsigned long pages_scanned; /* since last reclaim */
260 int all_unreclaimable; /* All pages pinned */
262 /* A count of how many reclaimers are scanning this zone */
263 atomic_t reclaim_in_progress;
265 /* Zone statistics */
266 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
269 * prev_priority holds the scanning priority for this zone. It is
270 * defined as the scanning priority at which we achieved our reclaim
271 * target at the previous try_to_free_pages() or balance_pgdat()
274 * We use prev_priority as a measure of how much stress page reclaim is
275 * under - it drives the swappiness decision: whether to unmap mapped
278 * Access to both this field is quite racy even on uniprocessor. But
279 * it is expected to average out OK.
285 /* Rarely used or read-mostly fields */
288 * wait_table -- the array holding the hash table
289 * wait_table_hash_nr_entries -- the size of the hash table array
290 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
292 * The purpose of all these is to keep track of the people
293 * waiting for a page to become available and make them
294 * runnable again when possible. The trouble is that this
295 * consumes a lot of space, especially when so few things
296 * wait on pages at a given time. So instead of using
297 * per-page waitqueues, we use a waitqueue hash table.
299 * The bucket discipline is to sleep on the same queue when
300 * colliding and wake all in that wait queue when removing.
301 * When something wakes, it must check to be sure its page is
302 * truly available, a la thundering herd. The cost of a
303 * collision is great, but given the expected load of the
304 * table, they should be so rare as to be outweighed by the
305 * benefits from the saved space.
307 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
308 * primary users of these fields, and in mm/page_alloc.c
309 * free_area_init_core() performs the initialization of them.
311 wait_queue_head_t * wait_table;
312 unsigned long wait_table_hash_nr_entries;
313 unsigned long wait_table_bits;
316 * Discontig memory support fields.
318 struct pglist_data *zone_pgdat;
319 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
320 unsigned long zone_start_pfn;
323 * zone_start_pfn, spanned_pages and present_pages are all
324 * protected by span_seqlock. It is a seqlock because it has
325 * to be read outside of zone->lock, and it is done in the main
326 * allocator path. But, it is written quite infrequently.
328 * The lock is declared along with zone->lock because it is
329 * frequently read in proximity to zone->lock. It's good to
330 * give them a chance of being in the same cacheline.
332 unsigned long spanned_pages; /* total size, including holes */
333 unsigned long present_pages; /* amount of memory (excluding holes) */
336 * rarely used fields:
339 } ____cacheline_internodealigned_in_smp;
342 * The "priority" of VM scanning is how much of the queues we will scan in one
343 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
344 * queues ("queue_length >> 12") during an aging round.
346 #define DEF_PRIORITY 12
348 /* Maximum number of zones on a zonelist */
349 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
354 * The NUMA zonelists are doubled becausse we need zonelists that restrict the
355 * allocations to a single node for GFP_THISNODE.
357 * [0 .. MAX_NR_ZONES -1] : Zonelists with fallback
358 * [MAZ_NR_ZONES ... MAZ_ZONELISTS -1] : No fallback (GFP_THISNODE)
360 #define MAX_ZONELISTS (2 * MAX_NR_ZONES)
364 * We cache key information from each zonelist for smaller cache
365 * footprint when scanning for free pages in get_page_from_freelist().
367 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
368 * up short of free memory since the last time (last_fullzone_zap)
369 * we zero'd fullzones.
370 * 2) The array z_to_n[] maps each zone in the zonelist to its node
371 * id, so that we can efficiently evaluate whether that node is
372 * set in the current tasks mems_allowed.
374 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
375 * indexed by a zones offset in the zonelist zones[] array.
377 * The get_page_from_freelist() routine does two scans. During the
378 * first scan, we skip zones whose corresponding bit in 'fullzones'
379 * is set or whose corresponding node in current->mems_allowed (which
380 * comes from cpusets) is not set. During the second scan, we bypass
381 * this zonelist_cache, to ensure we look methodically at each zone.
383 * Once per second, we zero out (zap) fullzones, forcing us to
384 * reconsider nodes that might have regained more free memory.
385 * The field last_full_zap is the time we last zapped fullzones.
387 * This mechanism reduces the amount of time we waste repeatedly
388 * reexaming zones for free memory when they just came up low on
389 * memory momentarilly ago.
391 * The zonelist_cache struct members logically belong in struct
392 * zonelist. However, the mempolicy zonelists constructed for
393 * MPOL_BIND are intentionally variable length (and usually much
394 * shorter). A general purpose mechanism for handling structs with
395 * multiple variable length members is more mechanism than we want
396 * here. We resort to some special case hackery instead.
398 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
399 * part because they are shorter), so we put the fixed length stuff
400 * at the front of the zonelist struct, ending in a variable length
401 * zones[], as is needed by MPOL_BIND.
403 * Then we put the optional zonelist cache on the end of the zonelist
404 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
405 * the fixed length portion at the front of the struct. This pointer
406 * both enables us to find the zonelist cache, and in the case of
407 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
408 * to know that the zonelist cache is not there.
410 * The end result is that struct zonelists come in two flavors:
411 * 1) The full, fixed length version, shown below, and
412 * 2) The custom zonelists for MPOL_BIND.
413 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
415 * Even though there may be multiple CPU cores on a node modifying
416 * fullzones or last_full_zap in the same zonelist_cache at the same
417 * time, we don't lock it. This is just hint data - if it is wrong now
418 * and then, the allocator will still function, perhaps a bit slower.
422 struct zonelist_cache {
423 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
424 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
425 unsigned long last_full_zap; /* when last zap'd (jiffies) */
428 #define MAX_ZONELISTS MAX_NR_ZONES
429 struct zonelist_cache;
433 * One allocation request operates on a zonelist. A zonelist
434 * is a list of zones, the first one is the 'goal' of the
435 * allocation, the other zones are fallback zones, in decreasing
438 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
439 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
443 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
444 struct zone *zones[MAX_ZONES_PER_ZONELIST + 1]; // NULL delimited
446 struct zonelist_cache zlcache; // optional ...
452 * Only custom zonelists like MPOL_BIND need to be filtered as part of
453 * policies. As described in the comment for struct zonelist_cache, these
454 * zonelists will not have a zlcache so zlcache_ptr will not be set. Use
455 * that to determine if the zonelists needs to be filtered or not.
457 static inline int alloc_should_filter_zonelist(struct zonelist *zonelist)
459 return !zonelist->zlcache_ptr;
462 static inline int alloc_should_filter_zonelist(struct zonelist *zonelist)
466 #endif /* CONFIG_NUMA */
468 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
469 struct node_active_region {
470 unsigned long start_pfn;
471 unsigned long end_pfn;
474 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
476 #ifndef CONFIG_DISCONTIGMEM
477 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
478 extern struct page *mem_map;
482 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
483 * (mostly NUMA machines?) to denote a higher-level memory zone than the
486 * On NUMA machines, each NUMA node would have a pg_data_t to describe
487 * it's memory layout.
489 * Memory statistics and page replacement data structures are maintained on a
493 typedef struct pglist_data {
494 struct zone node_zones[MAX_NR_ZONES];
495 struct zonelist node_zonelists[MAX_ZONELISTS];
497 #ifdef CONFIG_FLAT_NODE_MEM_MAP
498 struct page *node_mem_map;
500 struct bootmem_data *bdata;
501 #ifdef CONFIG_MEMORY_HOTPLUG
503 * Must be held any time you expect node_start_pfn, node_present_pages
504 * or node_spanned_pages stay constant. Holding this will also
505 * guarantee that any pfn_valid() stays that way.
507 * Nests above zone->lock and zone->size_seqlock.
509 spinlock_t node_size_lock;
511 unsigned long node_start_pfn;
512 unsigned long node_present_pages; /* total number of physical pages */
513 unsigned long node_spanned_pages; /* total size of physical page
514 range, including holes */
516 wait_queue_head_t kswapd_wait;
517 struct task_struct *kswapd;
518 int kswapd_max_order;
521 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
522 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
523 #ifdef CONFIG_FLAT_NODE_MEM_MAP
524 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
526 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
528 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
530 #include <linux/memory_hotplug.h>
532 void get_zone_counts(unsigned long *active, unsigned long *inactive,
533 unsigned long *free);
534 void build_all_zonelists(void);
535 void wakeup_kswapd(struct zone *zone, int order);
536 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
537 int classzone_idx, int alloc_flags);
538 enum memmap_context {
542 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
544 enum memmap_context context);
546 #ifdef CONFIG_HAVE_MEMORY_PRESENT
547 void memory_present(int nid, unsigned long start, unsigned long end);
549 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
552 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
553 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
557 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
559 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
561 static inline int populated_zone(struct zone *zone)
563 return (!!zone->present_pages);
566 extern int movable_zone;
568 static inline int zone_movable_is_highmem(void)
570 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
571 return movable_zone == ZONE_HIGHMEM;
577 static inline int is_highmem_idx(enum zone_type idx)
579 #ifdef CONFIG_HIGHMEM
580 return (idx == ZONE_HIGHMEM ||
581 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
587 static inline int is_normal_idx(enum zone_type idx)
589 return (idx == ZONE_NORMAL);
593 * is_highmem - helper function to quickly check if a struct zone is a
594 * highmem zone or not. This is an attempt to keep references
595 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
596 * @zone - pointer to struct zone variable
598 static inline int is_highmem(struct zone *zone)
600 #ifdef CONFIG_HIGHMEM
601 int zone_idx = zone - zone->zone_pgdat->node_zones;
602 return zone_idx == ZONE_HIGHMEM ||
603 (zone_idx == ZONE_MOVABLE && zone_movable_is_highmem());
609 static inline int is_normal(struct zone *zone)
611 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
614 static inline int is_dma32(struct zone *zone)
616 #ifdef CONFIG_ZONE_DMA32
617 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
623 static inline int is_dma(struct zone *zone)
625 #ifdef CONFIG_ZONE_DMA
626 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
632 /* These two functions are used to setup the per zone pages min values */
635 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
636 void __user *, size_t *, loff_t *);
637 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
638 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
639 void __user *, size_t *, loff_t *);
640 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
641 void __user *, size_t *, loff_t *);
642 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
643 struct file *, void __user *, size_t *, loff_t *);
644 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
645 struct file *, void __user *, size_t *, loff_t *);
647 extern int numa_zonelist_order_handler(struct ctl_table *, int,
648 struct file *, void __user *, size_t *, loff_t *);
649 extern char numa_zonelist_order[];
650 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
652 #include <linux/topology.h>
653 /* Returns the number of the current Node. */
655 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
658 #ifndef CONFIG_NEED_MULTIPLE_NODES
660 extern struct pglist_data contig_page_data;
661 #define NODE_DATA(nid) (&contig_page_data)
662 #define NODE_MEM_MAP(nid) mem_map
663 #define MAX_NODES_SHIFT 1
665 #else /* CONFIG_NEED_MULTIPLE_NODES */
667 #include <asm/mmzone.h>
669 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
671 extern struct pglist_data *first_online_pgdat(void);
672 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
673 extern struct zone *next_zone(struct zone *zone);
676 * for_each_pgdat - helper macro to iterate over all nodes
677 * @pgdat - pointer to a pg_data_t variable
679 #define for_each_online_pgdat(pgdat) \
680 for (pgdat = first_online_pgdat(); \
682 pgdat = next_online_pgdat(pgdat))
684 * for_each_zone - helper macro to iterate over all memory zones
685 * @zone - pointer to struct zone variable
687 * The user only needs to declare the zone variable, for_each_zone
690 #define for_each_zone(zone) \
691 for (zone = (first_online_pgdat())->node_zones; \
693 zone = next_zone(zone))
695 #ifdef CONFIG_SPARSEMEM
696 #include <asm/sparsemem.h>
699 #if BITS_PER_LONG == 32
701 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
702 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
704 #define FLAGS_RESERVED 9
706 #elif BITS_PER_LONG == 64
708 * with 64 bit flags field, there's plenty of room.
710 #define FLAGS_RESERVED 32
714 #error BITS_PER_LONG not defined
718 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
719 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
720 #define early_pfn_to_nid(nid) (0UL)
723 #ifdef CONFIG_FLATMEM
724 #define pfn_to_nid(pfn) (0)
727 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
728 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
730 #ifdef CONFIG_SPARSEMEM
733 * SECTION_SHIFT #bits space required to store a section #
735 * PA_SECTION_SHIFT physical address to/from section number
736 * PFN_SECTION_SHIFT pfn to/from section number
738 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
740 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
741 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
743 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
745 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
746 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
748 #define SECTION_BLOCKFLAGS_BITS \
749 ((SECTION_SIZE_BITS - (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS)
751 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
752 #error Allocator MAX_ORDER exceeds SECTION_SIZE
758 * This is, logically, a pointer to an array of struct
759 * pages. However, it is stored with some other magic.
760 * (see sparse.c::sparse_init_one_section())
762 * Additionally during early boot we encode node id of
763 * the location of the section here to guide allocation.
764 * (see sparse.c::memory_present())
766 * Making it a UL at least makes someone do a cast
767 * before using it wrong.
769 unsigned long section_mem_map;
770 DECLARE_BITMAP(pageblock_flags, SECTION_BLOCKFLAGS_BITS);
773 #ifdef CONFIG_SPARSEMEM_EXTREME
774 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
776 #define SECTIONS_PER_ROOT 1
779 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
780 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
781 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
783 #ifdef CONFIG_SPARSEMEM_EXTREME
784 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
786 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
789 static inline struct mem_section *__nr_to_section(unsigned long nr)
791 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
793 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
795 extern int __section_nr(struct mem_section* ms);
798 * We use the lower bits of the mem_map pointer to store
799 * a little bit of information. There should be at least
800 * 3 bits here due to 32-bit alignment.
802 #define SECTION_MARKED_PRESENT (1UL<<0)
803 #define SECTION_HAS_MEM_MAP (1UL<<1)
804 #define SECTION_MAP_LAST_BIT (1UL<<2)
805 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
806 #define SECTION_NID_SHIFT 2
808 static inline struct page *__section_mem_map_addr(struct mem_section *section)
810 unsigned long map = section->section_mem_map;
811 map &= SECTION_MAP_MASK;
812 return (struct page *)map;
815 static inline int present_section(struct mem_section *section)
817 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
820 static inline int present_section_nr(unsigned long nr)
822 return present_section(__nr_to_section(nr));
825 static inline int valid_section(struct mem_section *section)
827 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
830 static inline int valid_section_nr(unsigned long nr)
832 return valid_section(__nr_to_section(nr));
835 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
837 return __nr_to_section(pfn_to_section_nr(pfn));
840 static inline int pfn_valid(unsigned long pfn)
842 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
844 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
847 static inline int pfn_present(unsigned long pfn)
849 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
851 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
855 * These are _only_ used during initialisation, therefore they
856 * can use __initdata ... They could have names to indicate
860 #define pfn_to_nid(pfn) \
862 unsigned long __pfn_to_nid_pfn = (pfn); \
863 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
866 #define pfn_to_nid(pfn) (0)
869 #define early_pfn_valid(pfn) pfn_valid(pfn)
870 void sparse_init(void);
872 #define sparse_init() do {} while (0)
873 #define sparse_index_init(_sec, _nid) do {} while (0)
874 #endif /* CONFIG_SPARSEMEM */
876 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
877 #define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid))
879 #define early_pfn_in_nid(pfn, nid) (1)
882 #ifndef early_pfn_valid
883 #define early_pfn_valid(pfn) (1)
886 void memory_present(int nid, unsigned long start, unsigned long end);
887 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
890 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
891 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
892 * pfn_valid_within() should be used in this case; we optimise this away
893 * when we have no holes within a MAX_ORDER_NR_PAGES block.
895 #ifdef CONFIG_HOLES_IN_ZONE
896 #define pfn_valid_within(pfn) pfn_valid(pfn)
898 #define pfn_valid_within(pfn) (1)
901 #endif /* !__ASSEMBLY__ */
902 #endif /* __KERNEL__ */
903 #endif /* _LINUX_MMZONE_H */