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_RECLAIMABLE 1
38 #define MIGRATE_MOVABLE 2
39 #define MIGRATE_HIGHATOMIC 3
40 #define MIGRATE_RESERVE 4
41 #define MIGRATE_TYPES 5
43 #define for_each_migratetype_order(order, type) \
44 for (order = 0; order < MAX_ORDER; order++) \
45 for (type = 0; type < MIGRATE_TYPES; type++)
48 struct list_head free_list[MIGRATE_TYPES];
49 unsigned long nr_free;
55 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
56 * So add a wild amount of padding here to ensure that they fall into separate
57 * cachelines. There are very few zone structures in the machine, so space
58 * consumption is not a concern here.
60 #if defined(CONFIG_SMP)
63 } ____cacheline_internodealigned_in_smp;
64 #define ZONE_PADDING(name) struct zone_padding name;
66 #define ZONE_PADDING(name)
70 /* First 128 byte cacheline (assuming 64 bit words) */
74 NR_ANON_PAGES, /* Mapped anonymous pages */
75 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
76 only modified from process context */
80 /* Second 128 byte cacheline */
82 NR_SLAB_UNRECLAIMABLE,
83 NR_PAGETABLE, /* used for pagetables */
84 NR_UNSTABLE_NFS, /* NFS unstable pages */
88 NUMA_HIT, /* allocated in intended node */
89 NUMA_MISS, /* allocated in non intended node */
90 NUMA_FOREIGN, /* was intended here, hit elsewhere */
91 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
92 NUMA_LOCAL, /* allocation from local node */
93 NUMA_OTHER, /* allocation from other node */
95 NR_VM_ZONE_STAT_ITEMS };
97 struct per_cpu_pages {
98 int count; /* number of pages in the list */
99 int high; /* high watermark, emptying needed */
100 int batch; /* chunk size for buddy add/remove */
101 struct list_head list; /* the list of pages */
104 struct per_cpu_pageset {
105 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
111 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
113 } ____cacheline_aligned_in_smp;
116 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
118 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
122 #ifdef CONFIG_ZONE_DMA
124 * ZONE_DMA is used when there are devices that are not able
125 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
126 * carve out the portion of memory that is needed for these devices.
127 * The range is arch specific.
132 * ---------------------------
133 * parisc, ia64, sparc <4G
136 * alpha Unlimited or 0-16MB.
138 * i386, x86_64 and multiple other arches
143 #ifdef CONFIG_ZONE_DMA32
145 * x86_64 needs two ZONE_DMAs because it supports devices that are
146 * only able to do DMA to the lower 16M but also 32 bit devices that
147 * can only do DMA areas below 4G.
152 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
153 * performed on pages in ZONE_NORMAL if the DMA devices support
154 * transfers to all addressable memory.
157 #ifdef CONFIG_HIGHMEM
159 * A memory area that is only addressable by the kernel through
160 * mapping portions into its own address space. This is for example
161 * used by i386 to allow the kernel to address the memory beyond
162 * 900MB. The kernel will set up special mappings (page
163 * table entries on i386) for each page that the kernel needs to
173 * When a memory allocation must conform to specific limitations (such
174 * as being suitable for DMA) the caller will pass in hints to the
175 * allocator in the gfp_mask, in the zone modifier bits. These bits
176 * are used to select a priority ordered list of memory zones which
177 * match the requested limits. See gfp_zone() in include/linux/gfp.h
181 * Count the active zones. Note that the use of defined(X) outside
182 * #if and family is not necessarily defined so ensure we cannot use
183 * it later. Use __ZONE_COUNT to work out how many shift bits we need.
185 #define __ZONE_COUNT ( \
186 defined(CONFIG_ZONE_DMA) \
187 + defined(CONFIG_ZONE_DMA32) \
189 + defined(CONFIG_HIGHMEM) \
193 #define ZONES_SHIFT 0
194 #elif __ZONE_COUNT <= 2
195 #define ZONES_SHIFT 1
196 #elif __ZONE_COUNT <= 4
197 #define ZONES_SHIFT 2
199 #error ZONES_SHIFT -- too many zones configured adjust calculation
204 /* Fields commonly accessed by the page allocator */
205 unsigned long pages_min, pages_low, pages_high;
207 * We don't know if the memory that we're going to allocate will be freeable
208 * or/and it will be released eventually, so to avoid totally wasting several
209 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
210 * to run OOM on the lower zones despite there's tons of freeable ram
211 * on the higher zones). This array is recalculated at runtime if the
212 * sysctl_lowmem_reserve_ratio sysctl changes.
214 unsigned long lowmem_reserve[MAX_NR_ZONES];
219 * zone reclaim becomes active if more unmapped pages exist.
221 unsigned long min_unmapped_pages;
222 unsigned long min_slab_pages;
223 struct per_cpu_pageset *pageset[NR_CPUS];
225 struct per_cpu_pageset pageset[NR_CPUS];
228 * free areas of different sizes
231 #ifdef CONFIG_MEMORY_HOTPLUG
232 /* see spanned/present_pages for more description */
233 seqlock_t span_seqlock;
235 struct free_area free_area[MAX_ORDER];
237 #ifndef CONFIG_SPARSEMEM
239 * Flags for a MAX_ORDER_NR_PAGES block. See pageblock-flags.h.
240 * In SPARSEMEM, this map is stored in struct mem_section
242 unsigned long *pageblock_flags;
243 #endif /* CONFIG_SPARSEMEM */
248 /* Fields commonly accessed by the page reclaim scanner */
250 struct list_head active_list;
251 struct list_head inactive_list;
252 unsigned long nr_scan_active;
253 unsigned long nr_scan_inactive;
254 unsigned long pages_scanned; /* since last reclaim */
255 int all_unreclaimable; /* All pages pinned */
257 /* A count of how many reclaimers are scanning this zone */
258 atomic_t reclaim_in_progress;
260 /* Zone statistics */
261 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
264 * prev_priority holds the scanning priority for this zone. It is
265 * defined as the scanning priority at which we achieved our reclaim
266 * target at the previous try_to_free_pages() or balance_pgdat()
269 * We use prev_priority as a measure of how much stress page reclaim is
270 * under - it drives the swappiness decision: whether to unmap mapped
273 * Access to both this field is quite racy even on uniprocessor. But
274 * it is expected to average out OK.
280 /* Rarely used or read-mostly fields */
283 * wait_table -- the array holding the hash table
284 * wait_table_hash_nr_entries -- the size of the hash table array
285 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
287 * The purpose of all these is to keep track of the people
288 * waiting for a page to become available and make them
289 * runnable again when possible. The trouble is that this
290 * consumes a lot of space, especially when so few things
291 * wait on pages at a given time. So instead of using
292 * per-page waitqueues, we use a waitqueue hash table.
294 * The bucket discipline is to sleep on the same queue when
295 * colliding and wake all in that wait queue when removing.
296 * When something wakes, it must check to be sure its page is
297 * truly available, a la thundering herd. The cost of a
298 * collision is great, but given the expected load of the
299 * table, they should be so rare as to be outweighed by the
300 * benefits from the saved space.
302 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
303 * primary users of these fields, and in mm/page_alloc.c
304 * free_area_init_core() performs the initialization of them.
306 wait_queue_head_t * wait_table;
307 unsigned long wait_table_hash_nr_entries;
308 unsigned long wait_table_bits;
311 * Discontig memory support fields.
313 struct pglist_data *zone_pgdat;
314 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
315 unsigned long zone_start_pfn;
318 * zone_start_pfn, spanned_pages and present_pages are all
319 * protected by span_seqlock. It is a seqlock because it has
320 * to be read outside of zone->lock, and it is done in the main
321 * allocator path. But, it is written quite infrequently.
323 * The lock is declared along with zone->lock because it is
324 * frequently read in proximity to zone->lock. It's good to
325 * give them a chance of being in the same cacheline.
327 unsigned long spanned_pages; /* total size, including holes */
328 unsigned long present_pages; /* amount of memory (excluding holes) */
331 * rarely used fields:
334 } ____cacheline_internodealigned_in_smp;
337 * The "priority" of VM scanning is how much of the queues we will scan in one
338 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
339 * queues ("queue_length >> 12") during an aging round.
341 #define DEF_PRIORITY 12
343 /* Maximum number of zones on a zonelist */
344 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
349 * The NUMA zonelists are doubled becausse we need zonelists that restrict the
350 * allocations to a single node for GFP_THISNODE.
352 * [0 .. MAX_NR_ZONES -1] : Zonelists with fallback
353 * [MAZ_NR_ZONES ... MAZ_ZONELISTS -1] : No fallback (GFP_THISNODE)
355 #define MAX_ZONELISTS (2 * MAX_NR_ZONES)
359 * We cache key information from each zonelist for smaller cache
360 * footprint when scanning for free pages in get_page_from_freelist().
362 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
363 * up short of free memory since the last time (last_fullzone_zap)
364 * we zero'd fullzones.
365 * 2) The array z_to_n[] maps each zone in the zonelist to its node
366 * id, so that we can efficiently evaluate whether that node is
367 * set in the current tasks mems_allowed.
369 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
370 * indexed by a zones offset in the zonelist zones[] array.
372 * The get_page_from_freelist() routine does two scans. During the
373 * first scan, we skip zones whose corresponding bit in 'fullzones'
374 * is set or whose corresponding node in current->mems_allowed (which
375 * comes from cpusets) is not set. During the second scan, we bypass
376 * this zonelist_cache, to ensure we look methodically at each zone.
378 * Once per second, we zero out (zap) fullzones, forcing us to
379 * reconsider nodes that might have regained more free memory.
380 * The field last_full_zap is the time we last zapped fullzones.
382 * This mechanism reduces the amount of time we waste repeatedly
383 * reexaming zones for free memory when they just came up low on
384 * memory momentarilly ago.
386 * The zonelist_cache struct members logically belong in struct
387 * zonelist. However, the mempolicy zonelists constructed for
388 * MPOL_BIND are intentionally variable length (and usually much
389 * shorter). A general purpose mechanism for handling structs with
390 * multiple variable length members is more mechanism than we want
391 * here. We resort to some special case hackery instead.
393 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
394 * part because they are shorter), so we put the fixed length stuff
395 * at the front of the zonelist struct, ending in a variable length
396 * zones[], as is needed by MPOL_BIND.
398 * Then we put the optional zonelist cache on the end of the zonelist
399 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
400 * the fixed length portion at the front of the struct. This pointer
401 * both enables us to find the zonelist cache, and in the case of
402 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
403 * to know that the zonelist cache is not there.
405 * The end result is that struct zonelists come in two flavors:
406 * 1) The full, fixed length version, shown below, and
407 * 2) The custom zonelists for MPOL_BIND.
408 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
410 * Even though there may be multiple CPU cores on a node modifying
411 * fullzones or last_full_zap in the same zonelist_cache at the same
412 * time, we don't lock it. This is just hint data - if it is wrong now
413 * and then, the allocator will still function, perhaps a bit slower.
417 struct zonelist_cache {
418 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
419 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
420 unsigned long last_full_zap; /* when last zap'd (jiffies) */
423 #define MAX_ZONELISTS MAX_NR_ZONES
424 struct zonelist_cache;
428 * One allocation request operates on a zonelist. A zonelist
429 * is a list of zones, the first one is the 'goal' of the
430 * allocation, the other zones are fallback zones, in decreasing
433 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
434 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
438 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
439 struct zone *zones[MAX_ZONES_PER_ZONELIST + 1]; // NULL delimited
441 struct zonelist_cache zlcache; // optional ...
447 * Only custom zonelists like MPOL_BIND need to be filtered as part of
448 * policies. As described in the comment for struct zonelist_cache, these
449 * zonelists will not have a zlcache so zlcache_ptr will not be set. Use
450 * that to determine if the zonelists needs to be filtered or not.
452 static inline int alloc_should_filter_zonelist(struct zonelist *zonelist)
454 return !zonelist->zlcache_ptr;
457 static inline int alloc_should_filter_zonelist(struct zonelist *zonelist)
461 #endif /* CONFIG_NUMA */
463 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
464 struct node_active_region {
465 unsigned long start_pfn;
466 unsigned long end_pfn;
469 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
471 #ifndef CONFIG_DISCONTIGMEM
472 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
473 extern struct page *mem_map;
477 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
478 * (mostly NUMA machines?) to denote a higher-level memory zone than the
481 * On NUMA machines, each NUMA node would have a pg_data_t to describe
482 * it's memory layout.
484 * Memory statistics and page replacement data structures are maintained on a
488 typedef struct pglist_data {
489 struct zone node_zones[MAX_NR_ZONES];
490 struct zonelist node_zonelists[MAX_ZONELISTS];
492 #ifdef CONFIG_FLAT_NODE_MEM_MAP
493 struct page *node_mem_map;
495 struct bootmem_data *bdata;
496 #ifdef CONFIG_MEMORY_HOTPLUG
498 * Must be held any time you expect node_start_pfn, node_present_pages
499 * or node_spanned_pages stay constant. Holding this will also
500 * guarantee that any pfn_valid() stays that way.
502 * Nests above zone->lock and zone->size_seqlock.
504 spinlock_t node_size_lock;
506 unsigned long node_start_pfn;
507 unsigned long node_present_pages; /* total number of physical pages */
508 unsigned long node_spanned_pages; /* total size of physical page
509 range, including holes */
511 wait_queue_head_t kswapd_wait;
512 struct task_struct *kswapd;
513 int kswapd_max_order;
516 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
517 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
518 #ifdef CONFIG_FLAT_NODE_MEM_MAP
519 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
521 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
523 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
525 #include <linux/memory_hotplug.h>
527 void get_zone_counts(unsigned long *active, unsigned long *inactive,
528 unsigned long *free);
529 void build_all_zonelists(void);
530 void wakeup_kswapd(struct zone *zone, int order);
531 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
532 int classzone_idx, int alloc_flags);
533 enum memmap_context {
537 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
539 enum memmap_context context);
541 #ifdef CONFIG_HAVE_MEMORY_PRESENT
542 void memory_present(int nid, unsigned long start, unsigned long end);
544 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
547 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
548 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
552 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
554 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
556 static inline int populated_zone(struct zone *zone)
558 return (!!zone->present_pages);
561 extern int movable_zone;
563 static inline int zone_movable_is_highmem(void)
565 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
566 return movable_zone == ZONE_HIGHMEM;
572 static inline int is_highmem_idx(enum zone_type idx)
574 #ifdef CONFIG_HIGHMEM
575 return (idx == ZONE_HIGHMEM ||
576 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
582 static inline int is_normal_idx(enum zone_type idx)
584 return (idx == ZONE_NORMAL);
588 * is_highmem - helper function to quickly check if a struct zone is a
589 * highmem zone or not. This is an attempt to keep references
590 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
591 * @zone - pointer to struct zone variable
593 static inline int is_highmem(struct zone *zone)
595 #ifdef CONFIG_HIGHMEM
596 int zone_idx = zone - zone->zone_pgdat->node_zones;
597 return zone_idx == ZONE_HIGHMEM ||
598 (zone_idx == ZONE_MOVABLE && zone_movable_is_highmem());
604 static inline int is_normal(struct zone *zone)
606 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
609 static inline int is_dma32(struct zone *zone)
611 #ifdef CONFIG_ZONE_DMA32
612 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
618 static inline int is_dma(struct zone *zone)
620 #ifdef CONFIG_ZONE_DMA
621 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
627 /* These two functions are used to setup the per zone pages min values */
630 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
631 void __user *, size_t *, loff_t *);
632 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
633 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
634 void __user *, size_t *, loff_t *);
635 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
636 void __user *, size_t *, loff_t *);
637 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
638 struct file *, void __user *, size_t *, loff_t *);
639 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
640 struct file *, void __user *, size_t *, loff_t *);
642 extern int numa_zonelist_order_handler(struct ctl_table *, int,
643 struct file *, void __user *, size_t *, loff_t *);
644 extern char numa_zonelist_order[];
645 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
647 #include <linux/topology.h>
648 /* Returns the number of the current Node. */
650 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
653 #ifndef CONFIG_NEED_MULTIPLE_NODES
655 extern struct pglist_data contig_page_data;
656 #define NODE_DATA(nid) (&contig_page_data)
657 #define NODE_MEM_MAP(nid) mem_map
658 #define MAX_NODES_SHIFT 1
660 #else /* CONFIG_NEED_MULTIPLE_NODES */
662 #include <asm/mmzone.h>
664 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
666 extern struct pglist_data *first_online_pgdat(void);
667 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
668 extern struct zone *next_zone(struct zone *zone);
671 * for_each_pgdat - helper macro to iterate over all nodes
672 * @pgdat - pointer to a pg_data_t variable
674 #define for_each_online_pgdat(pgdat) \
675 for (pgdat = first_online_pgdat(); \
677 pgdat = next_online_pgdat(pgdat))
679 * for_each_zone - helper macro to iterate over all memory zones
680 * @zone - pointer to struct zone variable
682 * The user only needs to declare the zone variable, for_each_zone
685 #define for_each_zone(zone) \
686 for (zone = (first_online_pgdat())->node_zones; \
688 zone = next_zone(zone))
690 #ifdef CONFIG_SPARSEMEM
691 #include <asm/sparsemem.h>
694 #if BITS_PER_LONG == 32
696 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
697 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
699 #define FLAGS_RESERVED 9
701 #elif BITS_PER_LONG == 64
703 * with 64 bit flags field, there's plenty of room.
705 #define FLAGS_RESERVED 32
709 #error BITS_PER_LONG not defined
713 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
714 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
715 #define early_pfn_to_nid(nid) (0UL)
718 #ifdef CONFIG_FLATMEM
719 #define pfn_to_nid(pfn) (0)
722 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
723 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
725 #ifdef CONFIG_SPARSEMEM
728 * SECTION_SHIFT #bits space required to store a section #
730 * PA_SECTION_SHIFT physical address to/from section number
731 * PFN_SECTION_SHIFT pfn to/from section number
733 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
735 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
736 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
738 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
740 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
741 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
743 #define SECTION_BLOCKFLAGS_BITS \
744 ((1 << (PFN_SECTION_SHIFT - (MAX_ORDER-1))) * NR_PAGEBLOCK_BITS)
746 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
747 #error Allocator MAX_ORDER exceeds SECTION_SIZE
753 * This is, logically, a pointer to an array of struct
754 * pages. However, it is stored with some other magic.
755 * (see sparse.c::sparse_init_one_section())
757 * Additionally during early boot we encode node id of
758 * the location of the section here to guide allocation.
759 * (see sparse.c::memory_present())
761 * Making it a UL at least makes someone do a cast
762 * before using it wrong.
764 unsigned long section_mem_map;
766 /* See declaration of similar field in struct zone */
767 unsigned long *pageblock_flags;
770 #ifdef CONFIG_SPARSEMEM_EXTREME
771 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
773 #define SECTIONS_PER_ROOT 1
776 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
777 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
778 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
780 #ifdef CONFIG_SPARSEMEM_EXTREME
781 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
783 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
786 static inline struct mem_section *__nr_to_section(unsigned long nr)
788 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
790 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
792 extern int __section_nr(struct mem_section* ms);
795 * We use the lower bits of the mem_map pointer to store
796 * a little bit of information. There should be at least
797 * 3 bits here due to 32-bit alignment.
799 #define SECTION_MARKED_PRESENT (1UL<<0)
800 #define SECTION_HAS_MEM_MAP (1UL<<1)
801 #define SECTION_MAP_LAST_BIT (1UL<<2)
802 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
803 #define SECTION_NID_SHIFT 2
805 static inline struct page *__section_mem_map_addr(struct mem_section *section)
807 unsigned long map = section->section_mem_map;
808 map &= SECTION_MAP_MASK;
809 return (struct page *)map;
812 static inline int present_section(struct mem_section *section)
814 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
817 static inline int present_section_nr(unsigned long nr)
819 return present_section(__nr_to_section(nr));
822 static inline int valid_section(struct mem_section *section)
824 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
827 static inline int valid_section_nr(unsigned long nr)
829 return valid_section(__nr_to_section(nr));
832 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
834 return __nr_to_section(pfn_to_section_nr(pfn));
837 static inline int pfn_valid(unsigned long pfn)
839 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
841 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
844 static inline int pfn_present(unsigned long pfn)
846 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
848 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
852 * These are _only_ used during initialisation, therefore they
853 * can use __initdata ... They could have names to indicate
857 #define pfn_to_nid(pfn) \
859 unsigned long __pfn_to_nid_pfn = (pfn); \
860 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
863 #define pfn_to_nid(pfn) (0)
866 #define early_pfn_valid(pfn) pfn_valid(pfn)
867 void sparse_init(void);
869 #define sparse_init() do {} while (0)
870 #define sparse_index_init(_sec, _nid) do {} while (0)
871 #endif /* CONFIG_SPARSEMEM */
873 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
874 #define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid))
876 #define early_pfn_in_nid(pfn, nid) (1)
879 #ifndef early_pfn_valid
880 #define early_pfn_valid(pfn) (1)
883 void memory_present(int nid, unsigned long start, unsigned long end);
884 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
887 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
888 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
889 * pfn_valid_within() should be used in this case; we optimise this away
890 * when we have no holes within a MAX_ORDER_NR_PAGES block.
892 #ifdef CONFIG_HOLES_IN_ZONE
893 #define pfn_valid_within(pfn) pfn_valid(pfn)
895 #define pfn_valid_within(pfn) (1)
898 #endif /* !__ASSEMBLY__ */
899 #endif /* __KERNEL__ */
900 #endif /* _LINUX_MMZONE_H */