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_MOVABLE 1
39 #define MIGRATE_TYPES 2
41 #define MIGRATE_UNMOVABLE 0
42 #define MIGRATE_MOVABLE 0
43 #define MIGRATE_TYPES 1
46 #define for_each_migratetype_order(order, type) \
47 for (order = 0; order < MAX_ORDER; order++) \
48 for (type = 0; type < MIGRATE_TYPES; type++)
51 struct list_head free_list[MIGRATE_TYPES];
52 unsigned long nr_free;
58 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
59 * So add a wild amount of padding here to ensure that they fall into separate
60 * cachelines. There are very few zone structures in the machine, so space
61 * consumption is not a concern here.
63 #if defined(CONFIG_SMP)
66 } ____cacheline_internodealigned_in_smp;
67 #define ZONE_PADDING(name) struct zone_padding name;
69 #define ZONE_PADDING(name)
73 /* First 128 byte cacheline (assuming 64 bit words) */
77 NR_ANON_PAGES, /* Mapped anonymous pages */
78 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
79 only modified from process context */
83 /* Second 128 byte cacheline */
85 NR_SLAB_UNRECLAIMABLE,
86 NR_PAGETABLE, /* used for pagetables */
87 NR_UNSTABLE_NFS, /* NFS unstable pages */
91 NUMA_HIT, /* allocated in intended node */
92 NUMA_MISS, /* allocated in non intended node */
93 NUMA_FOREIGN, /* was intended here, hit elsewhere */
94 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
95 NUMA_LOCAL, /* allocation from local node */
96 NUMA_OTHER, /* allocation from other node */
98 NR_VM_ZONE_STAT_ITEMS };
100 struct per_cpu_pages {
101 int count; /* number of pages in the list */
102 int high; /* high watermark, emptying needed */
103 int batch; /* chunk size for buddy add/remove */
104 struct list_head list; /* the list of pages */
107 struct per_cpu_pageset {
108 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
114 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
116 } ____cacheline_aligned_in_smp;
119 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
121 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
125 #ifdef CONFIG_ZONE_DMA
127 * ZONE_DMA is used when there are devices that are not able
128 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
129 * carve out the portion of memory that is needed for these devices.
130 * The range is arch specific.
135 * ---------------------------
136 * parisc, ia64, sparc <4G
139 * alpha Unlimited or 0-16MB.
141 * i386, x86_64 and multiple other arches
146 #ifdef CONFIG_ZONE_DMA32
148 * x86_64 needs two ZONE_DMAs because it supports devices that are
149 * only able to do DMA to the lower 16M but also 32 bit devices that
150 * can only do DMA areas below 4G.
155 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
156 * performed on pages in ZONE_NORMAL if the DMA devices support
157 * transfers to all addressable memory.
160 #ifdef CONFIG_HIGHMEM
162 * A memory area that is only addressable by the kernel through
163 * mapping portions into its own address space. This is for example
164 * used by i386 to allow the kernel to address the memory beyond
165 * 900MB. The kernel will set up special mappings (page
166 * table entries on i386) for each page that the kernel needs to
176 * When a memory allocation must conform to specific limitations (such
177 * as being suitable for DMA) the caller will pass in hints to the
178 * allocator in the gfp_mask, in the zone modifier bits. These bits
179 * are used to select a priority ordered list of memory zones which
180 * match the requested limits. See gfp_zone() in include/linux/gfp.h
184 * Count the active zones. Note that the use of defined(X) outside
185 * #if and family is not necessarily defined so ensure we cannot use
186 * it later. Use __ZONE_COUNT to work out how many shift bits we need.
188 #define __ZONE_COUNT ( \
189 defined(CONFIG_ZONE_DMA) \
190 + defined(CONFIG_ZONE_DMA32) \
192 + defined(CONFIG_HIGHMEM) \
196 #define ZONES_SHIFT 0
197 #elif __ZONE_COUNT <= 2
198 #define ZONES_SHIFT 1
199 #elif __ZONE_COUNT <= 4
200 #define ZONES_SHIFT 2
202 #error ZONES_SHIFT -- too many zones configured adjust calculation
207 /* Fields commonly accessed by the page allocator */
208 unsigned long pages_min, pages_low, pages_high;
210 * We don't know if the memory that we're going to allocate will be freeable
211 * or/and it will be released eventually, so to avoid totally wasting several
212 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
213 * to run OOM on the lower zones despite there's tons of freeable ram
214 * on the higher zones). This array is recalculated at runtime if the
215 * sysctl_lowmem_reserve_ratio sysctl changes.
217 unsigned long lowmem_reserve[MAX_NR_ZONES];
222 * zone reclaim becomes active if more unmapped pages exist.
224 unsigned long min_unmapped_pages;
225 unsigned long min_slab_pages;
226 struct per_cpu_pageset *pageset[NR_CPUS];
228 struct per_cpu_pageset pageset[NR_CPUS];
231 * free areas of different sizes
234 #ifdef CONFIG_MEMORY_HOTPLUG
235 /* see spanned/present_pages for more description */
236 seqlock_t span_seqlock;
238 struct free_area free_area[MAX_ORDER];
240 #ifndef CONFIG_SPARSEMEM
242 * Flags for a MAX_ORDER_NR_PAGES block. See pageblock-flags.h.
243 * In SPARSEMEM, this map is stored in struct mem_section
245 unsigned long *pageblock_flags;
246 #endif /* CONFIG_SPARSEMEM */
251 /* Fields commonly accessed by the page reclaim scanner */
253 struct list_head active_list;
254 struct list_head inactive_list;
255 unsigned long nr_scan_active;
256 unsigned long nr_scan_inactive;
257 unsigned long pages_scanned; /* since last reclaim */
258 int all_unreclaimable; /* All pages pinned */
260 /* A count of how many reclaimers are scanning this zone */
261 atomic_t reclaim_in_progress;
263 /* Zone statistics */
264 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
267 * prev_priority holds the scanning priority for this zone. It is
268 * defined as the scanning priority at which we achieved our reclaim
269 * target at the previous try_to_free_pages() or balance_pgdat()
272 * We use prev_priority as a measure of how much stress page reclaim is
273 * under - it drives the swappiness decision: whether to unmap mapped
276 * Access to both this field is quite racy even on uniprocessor. But
277 * it is expected to average out OK.
283 /* Rarely used or read-mostly fields */
286 * wait_table -- the array holding the hash table
287 * wait_table_hash_nr_entries -- the size of the hash table array
288 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
290 * The purpose of all these is to keep track of the people
291 * waiting for a page to become available and make them
292 * runnable again when possible. The trouble is that this
293 * consumes a lot of space, especially when so few things
294 * wait on pages at a given time. So instead of using
295 * per-page waitqueues, we use a waitqueue hash table.
297 * The bucket discipline is to sleep on the same queue when
298 * colliding and wake all in that wait queue when removing.
299 * When something wakes, it must check to be sure its page is
300 * truly available, a la thundering herd. The cost of a
301 * collision is great, but given the expected load of the
302 * table, they should be so rare as to be outweighed by the
303 * benefits from the saved space.
305 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
306 * primary users of these fields, and in mm/page_alloc.c
307 * free_area_init_core() performs the initialization of them.
309 wait_queue_head_t * wait_table;
310 unsigned long wait_table_hash_nr_entries;
311 unsigned long wait_table_bits;
314 * Discontig memory support fields.
316 struct pglist_data *zone_pgdat;
317 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
318 unsigned long zone_start_pfn;
321 * zone_start_pfn, spanned_pages and present_pages are all
322 * protected by span_seqlock. It is a seqlock because it has
323 * to be read outside of zone->lock, and it is done in the main
324 * allocator path. But, it is written quite infrequently.
326 * The lock is declared along with zone->lock because it is
327 * frequently read in proximity to zone->lock. It's good to
328 * give them a chance of being in the same cacheline.
330 unsigned long spanned_pages; /* total size, including holes */
331 unsigned long present_pages; /* amount of memory (excluding holes) */
334 * rarely used fields:
337 } ____cacheline_internodealigned_in_smp;
340 * The "priority" of VM scanning is how much of the queues we will scan in one
341 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
342 * queues ("queue_length >> 12") during an aging round.
344 #define DEF_PRIORITY 12
346 /* Maximum number of zones on a zonelist */
347 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
352 * The NUMA zonelists are doubled becausse we need zonelists that restrict the
353 * allocations to a single node for GFP_THISNODE.
355 * [0 .. MAX_NR_ZONES -1] : Zonelists with fallback
356 * [MAZ_NR_ZONES ... MAZ_ZONELISTS -1] : No fallback (GFP_THISNODE)
358 #define MAX_ZONELISTS (2 * MAX_NR_ZONES)
362 * We cache key information from each zonelist for smaller cache
363 * footprint when scanning for free pages in get_page_from_freelist().
365 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
366 * up short of free memory since the last time (last_fullzone_zap)
367 * we zero'd fullzones.
368 * 2) The array z_to_n[] maps each zone in the zonelist to its node
369 * id, so that we can efficiently evaluate whether that node is
370 * set in the current tasks mems_allowed.
372 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
373 * indexed by a zones offset in the zonelist zones[] array.
375 * The get_page_from_freelist() routine does two scans. During the
376 * first scan, we skip zones whose corresponding bit in 'fullzones'
377 * is set or whose corresponding node in current->mems_allowed (which
378 * comes from cpusets) is not set. During the second scan, we bypass
379 * this zonelist_cache, to ensure we look methodically at each zone.
381 * Once per second, we zero out (zap) fullzones, forcing us to
382 * reconsider nodes that might have regained more free memory.
383 * The field last_full_zap is the time we last zapped fullzones.
385 * This mechanism reduces the amount of time we waste repeatedly
386 * reexaming zones for free memory when they just came up low on
387 * memory momentarilly ago.
389 * The zonelist_cache struct members logically belong in struct
390 * zonelist. However, the mempolicy zonelists constructed for
391 * MPOL_BIND are intentionally variable length (and usually much
392 * shorter). A general purpose mechanism for handling structs with
393 * multiple variable length members is more mechanism than we want
394 * here. We resort to some special case hackery instead.
396 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
397 * part because they are shorter), so we put the fixed length stuff
398 * at the front of the zonelist struct, ending in a variable length
399 * zones[], as is needed by MPOL_BIND.
401 * Then we put the optional zonelist cache on the end of the zonelist
402 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
403 * the fixed length portion at the front of the struct. This pointer
404 * both enables us to find the zonelist cache, and in the case of
405 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
406 * to know that the zonelist cache is not there.
408 * The end result is that struct zonelists come in two flavors:
409 * 1) The full, fixed length version, shown below, and
410 * 2) The custom zonelists for MPOL_BIND.
411 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
413 * Even though there may be multiple CPU cores on a node modifying
414 * fullzones or last_full_zap in the same zonelist_cache at the same
415 * time, we don't lock it. This is just hint data - if it is wrong now
416 * and then, the allocator will still function, perhaps a bit slower.
420 struct zonelist_cache {
421 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
422 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
423 unsigned long last_full_zap; /* when last zap'd (jiffies) */
426 #define MAX_ZONELISTS MAX_NR_ZONES
427 struct zonelist_cache;
431 * One allocation request operates on a zonelist. A zonelist
432 * is a list of zones, the first one is the 'goal' of the
433 * allocation, the other zones are fallback zones, in decreasing
436 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
437 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
441 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
442 struct zone *zones[MAX_ZONES_PER_ZONELIST + 1]; // NULL delimited
444 struct zonelist_cache zlcache; // optional ...
450 * Only custom zonelists like MPOL_BIND need to be filtered as part of
451 * policies. As described in the comment for struct zonelist_cache, these
452 * zonelists will not have a zlcache so zlcache_ptr will not be set. Use
453 * that to determine if the zonelists needs to be filtered or not.
455 static inline int alloc_should_filter_zonelist(struct zonelist *zonelist)
457 return !zonelist->zlcache_ptr;
460 static inline int alloc_should_filter_zonelist(struct zonelist *zonelist)
464 #endif /* CONFIG_NUMA */
466 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
467 struct node_active_region {
468 unsigned long start_pfn;
469 unsigned long end_pfn;
472 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
474 #ifndef CONFIG_DISCONTIGMEM
475 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
476 extern struct page *mem_map;
480 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
481 * (mostly NUMA machines?) to denote a higher-level memory zone than the
484 * On NUMA machines, each NUMA node would have a pg_data_t to describe
485 * it's memory layout.
487 * Memory statistics and page replacement data structures are maintained on a
491 typedef struct pglist_data {
492 struct zone node_zones[MAX_NR_ZONES];
493 struct zonelist node_zonelists[MAX_ZONELISTS];
495 #ifdef CONFIG_FLAT_NODE_MEM_MAP
496 struct page *node_mem_map;
498 struct bootmem_data *bdata;
499 #ifdef CONFIG_MEMORY_HOTPLUG
501 * Must be held any time you expect node_start_pfn, node_present_pages
502 * or node_spanned_pages stay constant. Holding this will also
503 * guarantee that any pfn_valid() stays that way.
505 * Nests above zone->lock and zone->size_seqlock.
507 spinlock_t node_size_lock;
509 unsigned long node_start_pfn;
510 unsigned long node_present_pages; /* total number of physical pages */
511 unsigned long node_spanned_pages; /* total size of physical page
512 range, including holes */
514 wait_queue_head_t kswapd_wait;
515 struct task_struct *kswapd;
516 int kswapd_max_order;
519 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
520 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
521 #ifdef CONFIG_FLAT_NODE_MEM_MAP
522 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
524 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
526 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
528 #include <linux/memory_hotplug.h>
530 void get_zone_counts(unsigned long *active, unsigned long *inactive,
531 unsigned long *free);
532 void build_all_zonelists(void);
533 void wakeup_kswapd(struct zone *zone, int order);
534 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
535 int classzone_idx, int alloc_flags);
536 enum memmap_context {
540 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
542 enum memmap_context context);
544 #ifdef CONFIG_HAVE_MEMORY_PRESENT
545 void memory_present(int nid, unsigned long start, unsigned long end);
547 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
550 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
551 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
555 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
557 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
559 static inline int populated_zone(struct zone *zone)
561 return (!!zone->present_pages);
564 extern int movable_zone;
566 static inline int zone_movable_is_highmem(void)
568 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
569 return movable_zone == ZONE_HIGHMEM;
575 static inline int is_highmem_idx(enum zone_type idx)
577 #ifdef CONFIG_HIGHMEM
578 return (idx == ZONE_HIGHMEM ||
579 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
585 static inline int is_normal_idx(enum zone_type idx)
587 return (idx == ZONE_NORMAL);
591 * is_highmem - helper function to quickly check if a struct zone is a
592 * highmem zone or not. This is an attempt to keep references
593 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
594 * @zone - pointer to struct zone variable
596 static inline int is_highmem(struct zone *zone)
598 #ifdef CONFIG_HIGHMEM
599 int zone_idx = zone - zone->zone_pgdat->node_zones;
600 return zone_idx == ZONE_HIGHMEM ||
601 (zone_idx == ZONE_MOVABLE && zone_movable_is_highmem());
607 static inline int is_normal(struct zone *zone)
609 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
612 static inline int is_dma32(struct zone *zone)
614 #ifdef CONFIG_ZONE_DMA32
615 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
621 static inline int is_dma(struct zone *zone)
623 #ifdef CONFIG_ZONE_DMA
624 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
630 /* These two functions are used to setup the per zone pages min values */
633 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
634 void __user *, size_t *, loff_t *);
635 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
636 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
637 void __user *, size_t *, loff_t *);
638 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
639 void __user *, size_t *, loff_t *);
640 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
641 struct file *, void __user *, size_t *, loff_t *);
642 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
643 struct file *, void __user *, size_t *, loff_t *);
645 extern int numa_zonelist_order_handler(struct ctl_table *, int,
646 struct file *, void __user *, size_t *, loff_t *);
647 extern char numa_zonelist_order[];
648 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
650 #include <linux/topology.h>
651 /* Returns the number of the current Node. */
653 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
656 #ifndef CONFIG_NEED_MULTIPLE_NODES
658 extern struct pglist_data contig_page_data;
659 #define NODE_DATA(nid) (&contig_page_data)
660 #define NODE_MEM_MAP(nid) mem_map
661 #define MAX_NODES_SHIFT 1
663 #else /* CONFIG_NEED_MULTIPLE_NODES */
665 #include <asm/mmzone.h>
667 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
669 extern struct pglist_data *first_online_pgdat(void);
670 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
671 extern struct zone *next_zone(struct zone *zone);
674 * for_each_pgdat - helper macro to iterate over all nodes
675 * @pgdat - pointer to a pg_data_t variable
677 #define for_each_online_pgdat(pgdat) \
678 for (pgdat = first_online_pgdat(); \
680 pgdat = next_online_pgdat(pgdat))
682 * for_each_zone - helper macro to iterate over all memory zones
683 * @zone - pointer to struct zone variable
685 * The user only needs to declare the zone variable, for_each_zone
688 #define for_each_zone(zone) \
689 for (zone = (first_online_pgdat())->node_zones; \
691 zone = next_zone(zone))
693 #ifdef CONFIG_SPARSEMEM
694 #include <asm/sparsemem.h>
697 #if BITS_PER_LONG == 32
699 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
700 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
702 #define FLAGS_RESERVED 9
704 #elif BITS_PER_LONG == 64
706 * with 64 bit flags field, there's plenty of room.
708 #define FLAGS_RESERVED 32
712 #error BITS_PER_LONG not defined
716 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
717 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
718 #define early_pfn_to_nid(nid) (0UL)
721 #ifdef CONFIG_FLATMEM
722 #define pfn_to_nid(pfn) (0)
725 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
726 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
728 #ifdef CONFIG_SPARSEMEM
731 * SECTION_SHIFT #bits space required to store a section #
733 * PA_SECTION_SHIFT physical address to/from section number
734 * PFN_SECTION_SHIFT pfn to/from section number
736 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
738 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
739 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
741 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
743 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
744 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
746 #define SECTION_BLOCKFLAGS_BITS \
747 ((SECTION_SIZE_BITS - (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS)
749 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
750 #error Allocator MAX_ORDER exceeds SECTION_SIZE
756 * This is, logically, a pointer to an array of struct
757 * pages. However, it is stored with some other magic.
758 * (see sparse.c::sparse_init_one_section())
760 * Additionally during early boot we encode node id of
761 * the location of the section here to guide allocation.
762 * (see sparse.c::memory_present())
764 * Making it a UL at least makes someone do a cast
765 * before using it wrong.
767 unsigned long section_mem_map;
768 DECLARE_BITMAP(pageblock_flags, SECTION_BLOCKFLAGS_BITS);
771 #ifdef CONFIG_SPARSEMEM_EXTREME
772 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
774 #define SECTIONS_PER_ROOT 1
777 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
778 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
779 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
781 #ifdef CONFIG_SPARSEMEM_EXTREME
782 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
784 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
787 static inline struct mem_section *__nr_to_section(unsigned long nr)
789 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
791 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
793 extern int __section_nr(struct mem_section* ms);
796 * We use the lower bits of the mem_map pointer to store
797 * a little bit of information. There should be at least
798 * 3 bits here due to 32-bit alignment.
800 #define SECTION_MARKED_PRESENT (1UL<<0)
801 #define SECTION_HAS_MEM_MAP (1UL<<1)
802 #define SECTION_MAP_LAST_BIT (1UL<<2)
803 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
804 #define SECTION_NID_SHIFT 2
806 static inline struct page *__section_mem_map_addr(struct mem_section *section)
808 unsigned long map = section->section_mem_map;
809 map &= SECTION_MAP_MASK;
810 return (struct page *)map;
813 static inline int present_section(struct mem_section *section)
815 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
818 static inline int present_section_nr(unsigned long nr)
820 return present_section(__nr_to_section(nr));
823 static inline int valid_section(struct mem_section *section)
825 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
828 static inline int valid_section_nr(unsigned long nr)
830 return valid_section(__nr_to_section(nr));
833 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
835 return __nr_to_section(pfn_to_section_nr(pfn));
838 static inline int pfn_valid(unsigned long pfn)
840 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
842 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
845 static inline int pfn_present(unsigned long pfn)
847 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
849 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
853 * These are _only_ used during initialisation, therefore they
854 * can use __initdata ... They could have names to indicate
858 #define pfn_to_nid(pfn) \
860 unsigned long __pfn_to_nid_pfn = (pfn); \
861 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
864 #define pfn_to_nid(pfn) (0)
867 #define early_pfn_valid(pfn) pfn_valid(pfn)
868 void sparse_init(void);
870 #define sparse_init() do {} while (0)
871 #define sparse_index_init(_sec, _nid) do {} while (0)
872 #endif /* CONFIG_SPARSEMEM */
874 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
875 #define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid))
877 #define early_pfn_in_nid(pfn, nid) (1)
880 #ifndef early_pfn_valid
881 #define early_pfn_valid(pfn) (1)
884 void memory_present(int nid, unsigned long start, unsigned long end);
885 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
888 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
889 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
890 * pfn_valid_within() should be used in this case; we optimise this away
891 * when we have no holes within a MAX_ORDER_NR_PAGES block.
893 #ifdef CONFIG_HOLES_IN_ZONE
894 #define pfn_valid_within(pfn) pfn_valid(pfn)
896 #define pfn_valid_within(pfn) (1)
899 #endif /* !__ASSEMBLY__ */
900 #endif /* __KERNEL__ */
901 #endif /* _LINUX_MMZONE_H */