1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
5 #ifndef __GENERATING_BOUNDS_H
7 #include <linux/spinlock.h>
8 #include <linux/list.h>
9 #include <linux/wait.h>
10 #include <linux/bitops.h>
11 #include <linux/cache.h>
12 #include <linux/threads.h>
13 #include <linux/numa.h>
14 #include <linux/init.h>
15 #include <linux/seqlock.h>
16 #include <linux/nodemask.h>
17 #include <linux/pageblock-flags.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/atomic.h>
22 /* Free memory management - zoned buddy allocator. */
23 #ifndef CONFIG_FORCE_MAX_ZONEORDER
26 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
28 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
31 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
32 * costly to service. That is between allocation orders which should
33 * coalesce naturally under reasonable reclaim pressure and those which
36 #define PAGE_ALLOC_COSTLY_ORDER 3
42 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
43 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
46 * MIGRATE_CMA migration type is designed to mimic the way
47 * ZONE_MOVABLE works. Only movable pages can be allocated
48 * from MIGRATE_CMA pageblocks and page allocator never
49 * implicitly change migration type of MIGRATE_CMA pageblock.
51 * The way to use it is to change migratetype of a range of
52 * pageblocks to MIGRATE_CMA which can be done by
53 * __free_pageblock_cma() function. What is important though
54 * is that a range of pageblocks must be aligned to
55 * MAX_ORDER_NR_PAGES should biggest page be bigger then
60 #ifdef CONFIG_MEMORY_ISOLATION
61 MIGRATE_ISOLATE, /* can't allocate from here */
67 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
68 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
70 # define is_migrate_cma(migratetype) false
71 # define is_migrate_cma_page(_page) false
74 #define for_each_migratetype_order(order, type) \
75 for (order = 0; order < MAX_ORDER; order++) \
76 for (type = 0; type < MIGRATE_TYPES; type++)
78 extern int page_group_by_mobility_disabled;
80 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
81 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
83 #define get_pageblock_migratetype(page) \
84 get_pfnblock_flags_mask(page, page_to_pfn(page), \
85 PB_migrate_end, MIGRATETYPE_MASK)
87 static inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
89 BUILD_BUG_ON(PB_migrate_end - PB_migrate != 2);
90 return get_pfnblock_flags_mask(page, pfn, PB_migrate_end,
95 struct list_head free_list[MIGRATE_TYPES];
96 unsigned long nr_free;
102 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
103 * So add a wild amount of padding here to ensure that they fall into separate
104 * cachelines. There are very few zone structures in the machine, so space
105 * consumption is not a concern here.
107 #if defined(CONFIG_SMP)
108 struct zone_padding {
110 } ____cacheline_internodealigned_in_smp;
111 #define ZONE_PADDING(name) struct zone_padding name;
113 #define ZONE_PADDING(name)
116 enum zone_stat_item {
117 /* First 128 byte cacheline (assuming 64 bit words) */
121 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
122 NR_ACTIVE_ANON, /* " " " " " */
123 NR_INACTIVE_FILE, /* " " " " " */
124 NR_ACTIVE_FILE, /* " " " " " */
125 NR_UNEVICTABLE, /* " " " " " */
126 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
127 NR_ANON_PAGES, /* Mapped anonymous pages */
128 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
129 only modified from process context */
134 NR_SLAB_UNRECLAIMABLE,
135 NR_PAGETABLE, /* used for pagetables */
137 /* Second 128 byte cacheline */
138 NR_UNSTABLE_NFS, /* NFS unstable pages */
141 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
142 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
143 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
144 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
145 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
146 NR_DIRTIED, /* page dirtyings since bootup */
147 NR_WRITTEN, /* page writings since bootup */
148 NR_PAGES_SCANNED, /* pages scanned since last reclaim */
150 NUMA_HIT, /* allocated in intended node */
151 NUMA_MISS, /* allocated in non intended node */
152 NUMA_FOREIGN, /* was intended here, hit elsewhere */
153 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
154 NUMA_LOCAL, /* allocation from local node */
155 NUMA_OTHER, /* allocation from other node */
159 WORKINGSET_NODERECLAIM,
160 NR_ANON_TRANSPARENT_HUGEPAGES,
162 NR_VM_ZONE_STAT_ITEMS };
165 * We do arithmetic on the LRU lists in various places in the code,
166 * so it is important to keep the active lists LRU_ACTIVE higher in
167 * the array than the corresponding inactive lists, and to keep
168 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
170 * This has to be kept in sync with the statistics in zone_stat_item
171 * above and the descriptions in vmstat_text in mm/vmstat.c
178 LRU_INACTIVE_ANON = LRU_BASE,
179 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
180 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
181 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
186 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
188 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
190 static inline int is_file_lru(enum lru_list lru)
192 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
195 static inline int is_active_lru(enum lru_list lru)
197 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
200 static inline int is_unevictable_lru(enum lru_list lru)
202 return (lru == LRU_UNEVICTABLE);
205 struct zone_reclaim_stat {
207 * The pageout code in vmscan.c keeps track of how many of the
208 * mem/swap backed and file backed pages are referenced.
209 * The higher the rotated/scanned ratio, the more valuable
212 * The anon LRU stats live in [0], file LRU stats in [1]
214 unsigned long recent_rotated[2];
215 unsigned long recent_scanned[2];
219 struct list_head lists[NR_LRU_LISTS];
220 struct zone_reclaim_stat reclaim_stat;
226 /* Mask used at gathering information at once (see memcontrol.c) */
227 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
228 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
229 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
231 /* Isolate clean file */
232 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
233 /* Isolate unmapped file */
234 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
235 /* Isolate for asynchronous migration */
236 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
237 /* Isolate unevictable pages */
238 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
240 /* LRU Isolation modes. */
241 typedef unsigned __bitwise__ isolate_mode_t;
243 enum zone_watermarks {
250 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
251 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
252 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
254 struct per_cpu_pages {
255 int count; /* number of pages in the list */
256 int high; /* high watermark, emptying needed */
257 int batch; /* chunk size for buddy add/remove */
259 /* Lists of pages, one per migrate type stored on the pcp-lists */
260 struct list_head lists[MIGRATE_PCPTYPES];
263 struct per_cpu_pageset {
264 struct per_cpu_pages pcp;
270 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
274 #endif /* !__GENERATING_BOUNDS.H */
277 #ifdef CONFIG_ZONE_DMA
279 * ZONE_DMA is used when there are devices that are not able
280 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
281 * carve out the portion of memory that is needed for these devices.
282 * The range is arch specific.
287 * ---------------------------
288 * parisc, ia64, sparc <4G
291 * alpha Unlimited or 0-16MB.
293 * i386, x86_64 and multiple other arches
298 #ifdef CONFIG_ZONE_DMA32
300 * x86_64 needs two ZONE_DMAs because it supports devices that are
301 * only able to do DMA to the lower 16M but also 32 bit devices that
302 * can only do DMA areas below 4G.
307 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
308 * performed on pages in ZONE_NORMAL if the DMA devices support
309 * transfers to all addressable memory.
312 #ifdef CONFIG_HIGHMEM
314 * A memory area that is only addressable by the kernel through
315 * mapping portions into its own address space. This is for example
316 * used by i386 to allow the kernel to address the memory beyond
317 * 900MB. The kernel will set up special mappings (page
318 * table entries on i386) for each page that the kernel needs to
324 #ifdef CONFIG_ZONE_DEVICE
331 #ifndef __GENERATING_BOUNDS_H
334 /* Read-mostly fields */
336 /* zone watermarks, access with *_wmark_pages(zone) macros */
337 unsigned long watermark[NR_WMARK];
339 unsigned long nr_reserved_highatomic;
342 * We don't know if the memory that we're going to allocate will be
343 * freeable or/and it will be released eventually, so to avoid totally
344 * wasting several GB of ram we must reserve some of the lower zone
345 * memory (otherwise we risk to run OOM on the lower zones despite
346 * there being tons of freeable ram on the higher zones). This array is
347 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
350 long lowmem_reserve[MAX_NR_ZONES];
357 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
358 * this zone's LRU. Maintained by the pageout code.
360 unsigned int inactive_ratio;
362 struct pglist_data *zone_pgdat;
363 struct per_cpu_pageset __percpu *pageset;
366 * This is a per-zone reserve of pages that should not be
367 * considered dirtyable memory.
369 unsigned long dirty_balance_reserve;
371 #ifndef CONFIG_SPARSEMEM
373 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
374 * In SPARSEMEM, this map is stored in struct mem_section
376 unsigned long *pageblock_flags;
377 #endif /* CONFIG_SPARSEMEM */
381 * zone reclaim becomes active if more unmapped pages exist.
383 unsigned long min_unmapped_pages;
384 unsigned long min_slab_pages;
385 #endif /* CONFIG_NUMA */
387 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
388 unsigned long zone_start_pfn;
391 * spanned_pages is the total pages spanned by the zone, including
392 * holes, which is calculated as:
393 * spanned_pages = zone_end_pfn - zone_start_pfn;
395 * present_pages is physical pages existing within the zone, which
397 * present_pages = spanned_pages - absent_pages(pages in holes);
399 * managed_pages is present pages managed by the buddy system, which
400 * is calculated as (reserved_pages includes pages allocated by the
401 * bootmem allocator):
402 * managed_pages = present_pages - reserved_pages;
404 * So present_pages may be used by memory hotplug or memory power
405 * management logic to figure out unmanaged pages by checking
406 * (present_pages - managed_pages). And managed_pages should be used
407 * by page allocator and vm scanner to calculate all kinds of watermarks
412 * zone_start_pfn and spanned_pages are protected by span_seqlock.
413 * It is a seqlock because it has to be read outside of zone->lock,
414 * and it is done in the main allocator path. But, it is written
415 * quite infrequently.
417 * The span_seq lock is declared along with zone->lock because it is
418 * frequently read in proximity to zone->lock. It's good to
419 * give them a chance of being in the same cacheline.
421 * Write access to present_pages at runtime should be protected by
422 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
423 * present_pages should get_online_mems() to get a stable value.
425 * Read access to managed_pages should be safe because it's unsigned
426 * long. Write access to zone->managed_pages and totalram_pages are
427 * protected by managed_page_count_lock at runtime. Idealy only
428 * adjust_managed_page_count() should be used instead of directly
429 * touching zone->managed_pages and totalram_pages.
431 unsigned long managed_pages;
432 unsigned long spanned_pages;
433 unsigned long present_pages;
437 #ifdef CONFIG_MEMORY_ISOLATION
439 * Number of isolated pageblock. It is used to solve incorrect
440 * freepage counting problem due to racy retrieving migratetype
441 * of pageblock. Protected by zone->lock.
443 unsigned long nr_isolate_pageblock;
446 #ifdef CONFIG_MEMORY_HOTPLUG
447 /* see spanned/present_pages for more description */
448 seqlock_t span_seqlock;
452 * wait_table -- the array holding the hash table
453 * wait_table_hash_nr_entries -- the size of the hash table array
454 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
456 * The purpose of all these is to keep track of the people
457 * waiting for a page to become available and make them
458 * runnable again when possible. The trouble is that this
459 * consumes a lot of space, especially when so few things
460 * wait on pages at a given time. So instead of using
461 * per-page waitqueues, we use a waitqueue hash table.
463 * The bucket discipline is to sleep on the same queue when
464 * colliding and wake all in that wait queue when removing.
465 * When something wakes, it must check to be sure its page is
466 * truly available, a la thundering herd. The cost of a
467 * collision is great, but given the expected load of the
468 * table, they should be so rare as to be outweighed by the
469 * benefits from the saved space.
471 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
472 * primary users of these fields, and in mm/page_alloc.c
473 * free_area_init_core() performs the initialization of them.
475 wait_queue_head_t *wait_table;
476 unsigned long wait_table_hash_nr_entries;
477 unsigned long wait_table_bits;
480 /* free areas of different sizes */
481 struct free_area free_area[MAX_ORDER];
483 /* zone flags, see below */
486 /* Write-intensive fields used from the page allocator */
491 /* Write-intensive fields used by page reclaim */
493 /* Fields commonly accessed by the page reclaim scanner */
495 struct lruvec lruvec;
497 /* Evictions & activations on the inactive file list */
498 atomic_long_t inactive_age;
501 * When free pages are below this point, additional steps are taken
502 * when reading the number of free pages to avoid per-cpu counter
503 * drift allowing watermarks to be breached
505 unsigned long percpu_drift_mark;
507 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
508 /* pfn where compaction free scanner should start */
509 unsigned long compact_cached_free_pfn;
510 /* pfn where async and sync compaction migration scanner should start */
511 unsigned long compact_cached_migrate_pfn[2];
514 #ifdef CONFIG_COMPACTION
516 * On compaction failure, 1<<compact_defer_shift compactions
517 * are skipped before trying again. The number attempted since
518 * last failure is tracked with compact_considered.
520 unsigned int compact_considered;
521 unsigned int compact_defer_shift;
522 int compact_order_failed;
525 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
526 /* Set to true when the PG_migrate_skip bits should be cleared */
527 bool compact_blockskip_flush;
531 /* Zone statistics */
532 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
533 } ____cacheline_internodealigned_in_smp;
536 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
537 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
538 ZONE_CONGESTED, /* zone has many dirty pages backed by
541 ZONE_DIRTY, /* reclaim scanning has recently found
542 * many dirty file pages at the tail
545 ZONE_WRITEBACK, /* reclaim scanning has recently found
546 * many pages under writeback
548 ZONE_FAIR_DEPLETED, /* fair zone policy batch depleted */
551 static inline unsigned long zone_end_pfn(const struct zone *zone)
553 return zone->zone_start_pfn + zone->spanned_pages;
556 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
558 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
561 static inline bool zone_is_initialized(struct zone *zone)
563 return !!zone->wait_table;
566 static inline bool zone_is_empty(struct zone *zone)
568 return zone->spanned_pages == 0;
572 * The "priority" of VM scanning is how much of the queues we will scan in one
573 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
574 * queues ("queue_length >> 12") during an aging round.
576 #define DEF_PRIORITY 12
578 /* Maximum number of zones on a zonelist */
579 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
584 * The NUMA zonelists are doubled because we need zonelists that restrict the
585 * allocations to a single node for __GFP_THISNODE.
587 * [0] : Zonelist with fallback
588 * [1] : No fallback (__GFP_THISNODE)
590 #define MAX_ZONELISTS 2
592 #define MAX_ZONELISTS 1
596 * This struct contains information about a zone in a zonelist. It is stored
597 * here to avoid dereferences into large structures and lookups of tables
600 struct zone *zone; /* Pointer to actual zone */
601 int zone_idx; /* zone_idx(zoneref->zone) */
605 * One allocation request operates on a zonelist. A zonelist
606 * is a list of zones, the first one is the 'goal' of the
607 * allocation, the other zones are fallback zones, in decreasing
610 * To speed the reading of the zonelist, the zonerefs contain the zone index
611 * of the entry being read. Helper functions to access information given
612 * a struct zoneref are
614 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
615 * zonelist_zone_idx() - Return the index of the zone for an entry
616 * zonelist_node_idx() - Return the index of the node for an entry
619 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
622 #ifndef CONFIG_DISCONTIGMEM
623 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
624 extern struct page *mem_map;
628 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
629 * (mostly NUMA machines?) to denote a higher-level memory zone than the
632 * On NUMA machines, each NUMA node would have a pg_data_t to describe
633 * it's memory layout.
635 * Memory statistics and page replacement data structures are maintained on a
639 typedef struct pglist_data {
640 struct zone node_zones[MAX_NR_ZONES];
641 struct zonelist node_zonelists[MAX_ZONELISTS];
643 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
644 struct page *node_mem_map;
645 #ifdef CONFIG_PAGE_EXTENSION
646 struct page_ext *node_page_ext;
649 #ifndef CONFIG_NO_BOOTMEM
650 struct bootmem_data *bdata;
652 #ifdef CONFIG_MEMORY_HOTPLUG
654 * Must be held any time you expect node_start_pfn, node_present_pages
655 * or node_spanned_pages stay constant. Holding this will also
656 * guarantee that any pfn_valid() stays that way.
658 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
659 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
661 * Nests above zone->lock and zone->span_seqlock
663 spinlock_t node_size_lock;
665 unsigned long node_start_pfn;
666 unsigned long node_present_pages; /* total number of physical pages */
667 unsigned long node_spanned_pages; /* total size of physical page
668 range, including holes */
670 wait_queue_head_t kswapd_wait;
671 wait_queue_head_t pfmemalloc_wait;
672 struct task_struct *kswapd; /* Protected by
673 mem_hotplug_begin/end() */
674 int kswapd_max_order;
675 enum zone_type classzone_idx;
676 #ifdef CONFIG_NUMA_BALANCING
677 /* Lock serializing the migrate rate limiting window */
678 spinlock_t numabalancing_migrate_lock;
680 /* Rate limiting time interval */
681 unsigned long numabalancing_migrate_next_window;
683 /* Number of pages migrated during the rate limiting time interval */
684 unsigned long numabalancing_migrate_nr_pages;
687 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
689 * If memory initialisation on large machines is deferred then this
690 * is the first PFN that needs to be initialised.
692 unsigned long first_deferred_pfn;
693 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
696 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
697 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
698 #ifdef CONFIG_FLAT_NODE_MEM_MAP
699 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
701 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
703 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
705 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
706 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
708 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
710 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
713 static inline bool pgdat_is_empty(pg_data_t *pgdat)
715 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
718 static inline int zone_id(const struct zone *zone)
720 struct pglist_data *pgdat = zone->zone_pgdat;
722 return zone - pgdat->node_zones;
725 #ifdef CONFIG_ZONE_DEVICE
726 static inline bool is_dev_zone(const struct zone *zone)
728 return zone_id(zone) == ZONE_DEVICE;
731 static inline bool is_dev_zone(const struct zone *zone)
737 #include <linux/memory_hotplug.h>
739 extern struct mutex zonelists_mutex;
740 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
741 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
742 bool zone_watermark_ok(struct zone *z, unsigned int order,
743 unsigned long mark, int classzone_idx, int alloc_flags);
744 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
745 unsigned long mark, int classzone_idx);
746 enum memmap_context {
750 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
753 extern void lruvec_init(struct lruvec *lruvec);
755 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
760 return container_of(lruvec, struct zone, lruvec);
764 #ifdef CONFIG_HAVE_MEMORY_PRESENT
765 void memory_present(int nid, unsigned long start, unsigned long end);
767 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
770 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
771 int local_memory_node(int node_id);
773 static inline int local_memory_node(int node_id) { return node_id; };
776 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
777 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
781 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
783 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
785 static inline int populated_zone(struct zone *zone)
787 return (!!zone->present_pages);
790 extern int movable_zone;
792 #ifdef CONFIG_HIGHMEM
793 static inline int zone_movable_is_highmem(void)
795 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
796 return movable_zone == ZONE_HIGHMEM;
798 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
803 static inline int is_highmem_idx(enum zone_type idx)
805 #ifdef CONFIG_HIGHMEM
806 return (idx == ZONE_HIGHMEM ||
807 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
814 * is_highmem - helper function to quickly check if a struct zone is a
815 * highmem zone or not. This is an attempt to keep references
816 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
817 * @zone - pointer to struct zone variable
819 static inline int is_highmem(struct zone *zone)
821 #ifdef CONFIG_HIGHMEM
822 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
823 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
824 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
825 zone_movable_is_highmem());
831 /* These two functions are used to setup the per zone pages min values */
833 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
834 void __user *, size_t *, loff_t *);
835 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
836 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
837 void __user *, size_t *, loff_t *);
838 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
839 void __user *, size_t *, loff_t *);
840 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
841 void __user *, size_t *, loff_t *);
842 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
843 void __user *, size_t *, loff_t *);
845 extern int numa_zonelist_order_handler(struct ctl_table *, int,
846 void __user *, size_t *, loff_t *);
847 extern char numa_zonelist_order[];
848 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
850 #ifndef CONFIG_NEED_MULTIPLE_NODES
852 extern struct pglist_data contig_page_data;
853 #define NODE_DATA(nid) (&contig_page_data)
854 #define NODE_MEM_MAP(nid) mem_map
856 #else /* CONFIG_NEED_MULTIPLE_NODES */
858 #include <asm/mmzone.h>
860 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
862 extern struct pglist_data *first_online_pgdat(void);
863 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
864 extern struct zone *next_zone(struct zone *zone);
867 * for_each_online_pgdat - helper macro to iterate over all online nodes
868 * @pgdat - pointer to a pg_data_t variable
870 #define for_each_online_pgdat(pgdat) \
871 for (pgdat = first_online_pgdat(); \
873 pgdat = next_online_pgdat(pgdat))
875 * for_each_zone - helper macro to iterate over all memory zones
876 * @zone - pointer to struct zone variable
878 * The user only needs to declare the zone variable, for_each_zone
881 #define for_each_zone(zone) \
882 for (zone = (first_online_pgdat())->node_zones; \
884 zone = next_zone(zone))
886 #define for_each_populated_zone(zone) \
887 for (zone = (first_online_pgdat())->node_zones; \
889 zone = next_zone(zone)) \
890 if (!populated_zone(zone)) \
894 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
896 return zoneref->zone;
899 static inline int zonelist_zone_idx(struct zoneref *zoneref)
901 return zoneref->zone_idx;
904 static inline int zonelist_node_idx(struct zoneref *zoneref)
907 /* zone_to_nid not available in this context */
908 return zoneref->zone->node;
911 #endif /* CONFIG_NUMA */
915 * next_zones_zonelist - Returns the next zone at or below highest_zoneidx within the allowed nodemask using a cursor within a zonelist as a starting point
916 * @z - The cursor used as a starting point for the search
917 * @highest_zoneidx - The zone index of the highest zone to return
918 * @nodes - An optional nodemask to filter the zonelist with
920 * This function returns the next zone at or below a given zone index that is
921 * within the allowed nodemask using a cursor as the starting point for the
922 * search. The zoneref returned is a cursor that represents the current zone
923 * being examined. It should be advanced by one before calling
924 * next_zones_zonelist again.
926 struct zoneref *next_zones_zonelist(struct zoneref *z,
927 enum zone_type highest_zoneidx,
931 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
932 * @zonelist - The zonelist to search for a suitable zone
933 * @highest_zoneidx - The zone index of the highest zone to return
934 * @nodes - An optional nodemask to filter the zonelist with
935 * @zone - The first suitable zone found is returned via this parameter
937 * This function returns the first zone at or below a given zone index that is
938 * within the allowed nodemask. The zoneref returned is a cursor that can be
939 * used to iterate the zonelist with next_zones_zonelist by advancing it by
940 * one before calling.
942 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
943 enum zone_type highest_zoneidx,
947 struct zoneref *z = next_zones_zonelist(zonelist->_zonerefs,
948 highest_zoneidx, nodes);
949 *zone = zonelist_zone(z);
954 * for_each_zone_zonelist_nodemask - helper macro to iterate over valid zones in a zonelist at or below a given zone index and within a nodemask
955 * @zone - The current zone in the iterator
956 * @z - The current pointer within zonelist->zones being iterated
957 * @zlist - The zonelist being iterated
958 * @highidx - The zone index of the highest zone to return
959 * @nodemask - Nodemask allowed by the allocator
961 * This iterator iterates though all zones at or below a given zone index and
962 * within a given nodemask
964 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
965 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
967 z = next_zones_zonelist(++z, highidx, nodemask), \
968 zone = zonelist_zone(z)) \
971 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
972 * @zone - The current zone in the iterator
973 * @z - The current pointer within zonelist->zones being iterated
974 * @zlist - The zonelist being iterated
975 * @highidx - The zone index of the highest zone to return
977 * This iterator iterates though all zones at or below a given zone index.
979 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
980 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
982 #ifdef CONFIG_SPARSEMEM
983 #include <asm/sparsemem.h>
986 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
987 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
988 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
994 #ifdef CONFIG_FLATMEM
995 #define pfn_to_nid(pfn) (0)
998 #ifdef CONFIG_SPARSEMEM
1001 * SECTION_SHIFT #bits space required to store a section #
1003 * PA_SECTION_SHIFT physical address to/from section number
1004 * PFN_SECTION_SHIFT pfn to/from section number
1006 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1007 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1009 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1011 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1012 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1014 #define SECTION_BLOCKFLAGS_BITS \
1015 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1017 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1018 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1021 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1022 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1024 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1025 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1029 struct mem_section {
1031 * This is, logically, a pointer to an array of struct
1032 * pages. However, it is stored with some other magic.
1033 * (see sparse.c::sparse_init_one_section())
1035 * Additionally during early boot we encode node id of
1036 * the location of the section here to guide allocation.
1037 * (see sparse.c::memory_present())
1039 * Making it a UL at least makes someone do a cast
1040 * before using it wrong.
1042 unsigned long section_mem_map;
1044 /* See declaration of similar field in struct zone */
1045 unsigned long *pageblock_flags;
1046 #ifdef CONFIG_PAGE_EXTENSION
1048 * If !SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1049 * section. (see page_ext.h about this.)
1051 struct page_ext *page_ext;
1055 * WARNING: mem_section must be a power-of-2 in size for the
1056 * calculation and use of SECTION_ROOT_MASK to make sense.
1060 #ifdef CONFIG_SPARSEMEM_EXTREME
1061 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1063 #define SECTIONS_PER_ROOT 1
1066 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1067 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1068 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1070 #ifdef CONFIG_SPARSEMEM_EXTREME
1071 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1073 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1076 static inline struct mem_section *__nr_to_section(unsigned long nr)
1078 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1080 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1082 extern int __section_nr(struct mem_section* ms);
1083 extern unsigned long usemap_size(void);
1086 * We use the lower bits of the mem_map pointer to store
1087 * a little bit of information. There should be at least
1088 * 3 bits here due to 32-bit alignment.
1090 #define SECTION_MARKED_PRESENT (1UL<<0)
1091 #define SECTION_HAS_MEM_MAP (1UL<<1)
1092 #define SECTION_MAP_LAST_BIT (1UL<<2)
1093 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1094 #define SECTION_NID_SHIFT 2
1096 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1098 unsigned long map = section->section_mem_map;
1099 map &= SECTION_MAP_MASK;
1100 return (struct page *)map;
1103 static inline int present_section(struct mem_section *section)
1105 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1108 static inline int present_section_nr(unsigned long nr)
1110 return present_section(__nr_to_section(nr));
1113 static inline int valid_section(struct mem_section *section)
1115 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1118 static inline int valid_section_nr(unsigned long nr)
1120 return valid_section(__nr_to_section(nr));
1123 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1125 return __nr_to_section(pfn_to_section_nr(pfn));
1128 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1129 static inline int pfn_valid(unsigned long pfn)
1131 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1133 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1137 static inline int pfn_present(unsigned long pfn)
1139 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1141 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1145 * These are _only_ used during initialisation, therefore they
1146 * can use __initdata ... They could have names to indicate
1150 #define pfn_to_nid(pfn) \
1152 unsigned long __pfn_to_nid_pfn = (pfn); \
1153 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1156 #define pfn_to_nid(pfn) (0)
1159 #define early_pfn_valid(pfn) pfn_valid(pfn)
1160 void sparse_init(void);
1162 #define sparse_init() do {} while (0)
1163 #define sparse_index_init(_sec, _nid) do {} while (0)
1164 #endif /* CONFIG_SPARSEMEM */
1167 * During memory init memblocks map pfns to nids. The search is expensive and
1168 * this caches recent lookups. The implementation of __early_pfn_to_nid
1169 * may treat start/end as pfns or sections.
1171 struct mminit_pfnnid_cache {
1172 unsigned long last_start;
1173 unsigned long last_end;
1177 #ifndef early_pfn_valid
1178 #define early_pfn_valid(pfn) (1)
1181 void memory_present(int nid, unsigned long start, unsigned long end);
1182 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1185 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1186 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1187 * pfn_valid_within() should be used in this case; we optimise this away
1188 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1190 #ifdef CONFIG_HOLES_IN_ZONE
1191 #define pfn_valid_within(pfn) pfn_valid(pfn)
1193 #define pfn_valid_within(pfn) (1)
1196 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1198 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1199 * associated with it or not. In FLATMEM, it is expected that holes always
1200 * have valid memmap as long as there is valid PFNs either side of the hole.
1201 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1204 * However, an ARM, and maybe other embedded architectures in the future
1205 * free memmap backing holes to save memory on the assumption the memmap is
1206 * never used. The page_zone linkages are then broken even though pfn_valid()
1207 * returns true. A walker of the full memmap must then do this additional
1208 * check to ensure the memmap they are looking at is sane by making sure
1209 * the zone and PFN linkages are still valid. This is expensive, but walkers
1210 * of the full memmap are extremely rare.
1212 int memmap_valid_within(unsigned long pfn,
1213 struct page *page, struct zone *zone);
1215 static inline int memmap_valid_within(unsigned long pfn,
1216 struct page *page, struct zone *zone)
1220 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1222 #endif /* !__GENERATING_BOUNDS.H */
1223 #endif /* !__ASSEMBLY__ */
1224 #endif /* _LINUX_MMZONE_H */