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 <generated/bounds.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_RESERVE = 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)
69 # define is_migrate_cma(migratetype) false
72 #define for_each_migratetype_order(order, type) \
73 for (order = 0; order < MAX_ORDER; order++) \
74 for (type = 0; type < MIGRATE_TYPES; type++)
76 extern int page_group_by_mobility_disabled;
78 static inline int get_pageblock_migratetype(struct page *page)
80 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
84 struct list_head free_list[MIGRATE_TYPES];
85 unsigned long nr_free;
91 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
92 * So add a wild amount of padding here to ensure that they fall into separate
93 * cachelines. There are very few zone structures in the machine, so space
94 * consumption is not a concern here.
96 #if defined(CONFIG_SMP)
99 } ____cacheline_internodealigned_in_smp;
100 #define ZONE_PADDING(name) struct zone_padding name;
102 #define ZONE_PADDING(name)
105 enum zone_stat_item {
106 /* First 128 byte cacheline (assuming 64 bit words) */
109 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
110 NR_ACTIVE_ANON, /* " " " " " */
111 NR_INACTIVE_FILE, /* " " " " " */
112 NR_ACTIVE_FILE, /* " " " " " */
113 NR_UNEVICTABLE, /* " " " " " */
114 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
115 NR_ANON_PAGES, /* Mapped anonymous pages */
116 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
117 only modified from process context */
122 NR_SLAB_UNRECLAIMABLE,
123 NR_PAGETABLE, /* used for pagetables */
125 /* Second 128 byte cacheline */
126 NR_UNSTABLE_NFS, /* NFS unstable pages */
129 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
130 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
131 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
132 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
133 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
134 NR_DIRTIED, /* page dirtyings since bootup */
135 NR_WRITTEN, /* page writings since bootup */
137 NUMA_HIT, /* allocated in intended node */
138 NUMA_MISS, /* allocated in non intended node */
139 NUMA_FOREIGN, /* was intended here, hit elsewhere */
140 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
141 NUMA_LOCAL, /* allocation from local node */
142 NUMA_OTHER, /* allocation from other node */
144 NR_ANON_TRANSPARENT_HUGEPAGES,
146 NR_VM_ZONE_STAT_ITEMS };
149 * We do arithmetic on the LRU lists in various places in the code,
150 * so it is important to keep the active lists LRU_ACTIVE higher in
151 * the array than the corresponding inactive lists, and to keep
152 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
154 * This has to be kept in sync with the statistics in zone_stat_item
155 * above and the descriptions in vmstat_text in mm/vmstat.c
162 LRU_INACTIVE_ANON = LRU_BASE,
163 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
164 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
165 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
170 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
172 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
174 static inline int is_file_lru(enum lru_list lru)
176 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
179 static inline int is_active_lru(enum lru_list lru)
181 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
184 static inline int is_unevictable_lru(enum lru_list lru)
186 return (lru == LRU_UNEVICTABLE);
189 struct zone_reclaim_stat {
191 * The pageout code in vmscan.c keeps track of how many of the
192 * mem/swap backed and file backed pages are referenced.
193 * The higher the rotated/scanned ratio, the more valuable
196 * The anon LRU stats live in [0], file LRU stats in [1]
198 unsigned long recent_rotated[2];
199 unsigned long recent_scanned[2];
203 struct list_head lists[NR_LRU_LISTS];
204 struct zone_reclaim_stat reclaim_stat;
210 /* Mask used at gathering information at once (see memcontrol.c) */
211 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
212 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
213 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
215 /* Isolate clean file */
216 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
217 /* Isolate unmapped file */
218 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
219 /* Isolate for asynchronous migration */
220 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
221 /* Isolate unevictable pages */
222 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
224 /* LRU Isolation modes. */
225 typedef unsigned __bitwise__ isolate_mode_t;
227 enum zone_watermarks {
234 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
235 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
236 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
238 struct per_cpu_pages {
239 int count; /* number of pages in the list */
240 int high; /* high watermark, emptying needed */
241 int batch; /* chunk size for buddy add/remove */
243 /* Lists of pages, one per migrate type stored on the pcp-lists */
244 struct list_head lists[MIGRATE_PCPTYPES];
247 struct per_cpu_pageset {
248 struct per_cpu_pages pcp;
254 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
258 #endif /* !__GENERATING_BOUNDS.H */
261 #ifdef CONFIG_ZONE_DMA
263 * ZONE_DMA is used when there are devices that are not able
264 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
265 * carve out the portion of memory that is needed for these devices.
266 * The range is arch specific.
271 * ---------------------------
272 * parisc, ia64, sparc <4G
275 * alpha Unlimited or 0-16MB.
277 * i386, x86_64 and multiple other arches
282 #ifdef CONFIG_ZONE_DMA32
284 * x86_64 needs two ZONE_DMAs because it supports devices that are
285 * only able to do DMA to the lower 16M but also 32 bit devices that
286 * can only do DMA areas below 4G.
291 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
292 * performed on pages in ZONE_NORMAL if the DMA devices support
293 * transfers to all addressable memory.
296 #ifdef CONFIG_HIGHMEM
298 * A memory area that is only addressable by the kernel through
299 * mapping portions into its own address space. This is for example
300 * used by i386 to allow the kernel to address the memory beyond
301 * 900MB. The kernel will set up special mappings (page
302 * table entries on i386) for each page that the kernel needs to
311 #ifndef __GENERATING_BOUNDS_H
314 * When a memory allocation must conform to specific limitations (such
315 * as being suitable for DMA) the caller will pass in hints to the
316 * allocator in the gfp_mask, in the zone modifier bits. These bits
317 * are used to select a priority ordered list of memory zones which
318 * match the requested limits. See gfp_zone() in include/linux/gfp.h
322 #define ZONES_SHIFT 0
323 #elif MAX_NR_ZONES <= 2
324 #define ZONES_SHIFT 1
325 #elif MAX_NR_ZONES <= 4
326 #define ZONES_SHIFT 2
328 #error ZONES_SHIFT -- too many zones configured adjust calculation
332 /* Fields commonly accessed by the page allocator */
334 /* zone watermarks, access with *_wmark_pages(zone) macros */
335 unsigned long watermark[NR_WMARK];
338 * When free pages are below this point, additional steps are taken
339 * when reading the number of free pages to avoid per-cpu counter
340 * drift allowing watermarks to be breached
342 unsigned long percpu_drift_mark;
345 * We don't know if the memory that we're going to allocate will be freeable
346 * or/and it will be released eventually, so to avoid totally wasting several
347 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
348 * to run OOM on the lower zones despite there's tons of freeable ram
349 * on the higher zones). This array is recalculated at runtime if the
350 * sysctl_lowmem_reserve_ratio sysctl changes.
352 unsigned long lowmem_reserve[MAX_NR_ZONES];
355 * This is a per-zone reserve of pages that should not be
356 * considered dirtyable memory.
358 unsigned long dirty_balance_reserve;
363 * zone reclaim becomes active if more unmapped pages exist.
365 unsigned long min_unmapped_pages;
366 unsigned long min_slab_pages;
368 struct per_cpu_pageset __percpu *pageset;
370 * free areas of different sizes
373 int all_unreclaimable; /* All pages pinned */
374 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
375 /* Set to true when the PG_migrate_skip bits should be cleared */
376 bool compact_blockskip_flush;
378 /* pfns where compaction scanners should start */
379 unsigned long compact_cached_free_pfn;
380 unsigned long compact_cached_migrate_pfn;
382 #ifdef CONFIG_MEMORY_HOTPLUG
383 /* see spanned/present_pages for more description */
384 seqlock_t span_seqlock;
386 struct free_area free_area[MAX_ORDER];
388 #ifndef CONFIG_SPARSEMEM
390 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
391 * In SPARSEMEM, this map is stored in struct mem_section
393 unsigned long *pageblock_flags;
394 #endif /* CONFIG_SPARSEMEM */
396 #ifdef CONFIG_COMPACTION
398 * On compaction failure, 1<<compact_defer_shift compactions
399 * are skipped before trying again. The number attempted since
400 * last failure is tracked with compact_considered.
402 unsigned int compact_considered;
403 unsigned int compact_defer_shift;
404 int compact_order_failed;
409 /* Fields commonly accessed by the page reclaim scanner */
411 struct lruvec lruvec;
413 unsigned long pages_scanned; /* since last reclaim */
414 unsigned long flags; /* zone flags, see below */
416 /* Zone statistics */
417 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
420 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
421 * this zone's LRU. Maintained by the pageout code.
423 unsigned int inactive_ratio;
427 /* Rarely used or read-mostly fields */
430 * wait_table -- the array holding the hash table
431 * wait_table_hash_nr_entries -- the size of the hash table array
432 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
434 * The purpose of all these is to keep track of the people
435 * waiting for a page to become available and make them
436 * runnable again when possible. The trouble is that this
437 * consumes a lot of space, especially when so few things
438 * wait on pages at a given time. So instead of using
439 * per-page waitqueues, we use a waitqueue hash table.
441 * The bucket discipline is to sleep on the same queue when
442 * colliding and wake all in that wait queue when removing.
443 * When something wakes, it must check to be sure its page is
444 * truly available, a la thundering herd. The cost of a
445 * collision is great, but given the expected load of the
446 * table, they should be so rare as to be outweighed by the
447 * benefits from the saved space.
449 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
450 * primary users of these fields, and in mm/page_alloc.c
451 * free_area_init_core() performs the initialization of them.
453 wait_queue_head_t * wait_table;
454 unsigned long wait_table_hash_nr_entries;
455 unsigned long wait_table_bits;
458 * Discontig memory support fields.
460 struct pglist_data *zone_pgdat;
461 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
462 unsigned long zone_start_pfn;
465 * spanned_pages is the total pages spanned by the zone, including
466 * holes, which is calculated as:
467 * spanned_pages = zone_end_pfn - zone_start_pfn;
469 * present_pages is physical pages existing within the zone, which
471 * present_pages = spanned_pages - absent_pages(pags in holes);
473 * managed_pages is present pages managed by the buddy system, which
474 * is calculated as (reserved_pages includes pages allocated by the
475 * bootmem allocator):
476 * managed_pages = present_pages - reserved_pages;
478 * So present_pages may be used by memory hotplug or memory power
479 * management logic to figure out unmanaged pages by checking
480 * (present_pages - managed_pages). And managed_pages should be used
481 * by page allocator and vm scanner to calculate all kinds of watermarks
486 * zone_start_pfn and spanned_pages are protected by span_seqlock.
487 * It is a seqlock because it has to be read outside of zone->lock,
488 * and it is done in the main allocator path. But, it is written
489 * quite infrequently.
491 * The span_seq lock is declared along with zone->lock because it is
492 * frequently read in proximity to zone->lock. It's good to
493 * give them a chance of being in the same cacheline.
495 * Write access to present_pages and managed_pages at runtime should
496 * be protected by lock_memory_hotplug()/unlock_memory_hotplug().
497 * Any reader who can't tolerant drift of present_pages and
498 * managed_pages should hold memory hotplug lock to get a stable value.
500 unsigned long spanned_pages;
501 unsigned long present_pages;
502 unsigned long managed_pages;
505 * rarely used fields:
508 } ____cacheline_internodealigned_in_smp;
511 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
512 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
513 ZONE_CONGESTED, /* zone has many dirty pages backed by
518 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
520 set_bit(flag, &zone->flags);
523 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
525 return test_and_set_bit(flag, &zone->flags);
528 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
530 clear_bit(flag, &zone->flags);
533 static inline int zone_is_reclaim_congested(const struct zone *zone)
535 return test_bit(ZONE_CONGESTED, &zone->flags);
538 static inline int zone_is_reclaim_locked(const struct zone *zone)
540 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
543 static inline int zone_is_oom_locked(const struct zone *zone)
545 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
549 * The "priority" of VM scanning is how much of the queues we will scan in one
550 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
551 * queues ("queue_length >> 12") during an aging round.
553 #define DEF_PRIORITY 12
555 /* Maximum number of zones on a zonelist */
556 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
561 * The NUMA zonelists are doubled because we need zonelists that restrict the
562 * allocations to a single node for GFP_THISNODE.
564 * [0] : Zonelist with fallback
565 * [1] : No fallback (GFP_THISNODE)
567 #define MAX_ZONELISTS 2
571 * We cache key information from each zonelist for smaller cache
572 * footprint when scanning for free pages in get_page_from_freelist().
574 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
575 * up short of free memory since the last time (last_fullzone_zap)
576 * we zero'd fullzones.
577 * 2) The array z_to_n[] maps each zone in the zonelist to its node
578 * id, so that we can efficiently evaluate whether that node is
579 * set in the current tasks mems_allowed.
581 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
582 * indexed by a zones offset in the zonelist zones[] array.
584 * The get_page_from_freelist() routine does two scans. During the
585 * first scan, we skip zones whose corresponding bit in 'fullzones'
586 * is set or whose corresponding node in current->mems_allowed (which
587 * comes from cpusets) is not set. During the second scan, we bypass
588 * this zonelist_cache, to ensure we look methodically at each zone.
590 * Once per second, we zero out (zap) fullzones, forcing us to
591 * reconsider nodes that might have regained more free memory.
592 * The field last_full_zap is the time we last zapped fullzones.
594 * This mechanism reduces the amount of time we waste repeatedly
595 * reexaming zones for free memory when they just came up low on
596 * memory momentarilly ago.
598 * The zonelist_cache struct members logically belong in struct
599 * zonelist. However, the mempolicy zonelists constructed for
600 * MPOL_BIND are intentionally variable length (and usually much
601 * shorter). A general purpose mechanism for handling structs with
602 * multiple variable length members is more mechanism than we want
603 * here. We resort to some special case hackery instead.
605 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
606 * part because they are shorter), so we put the fixed length stuff
607 * at the front of the zonelist struct, ending in a variable length
608 * zones[], as is needed by MPOL_BIND.
610 * Then we put the optional zonelist cache on the end of the zonelist
611 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
612 * the fixed length portion at the front of the struct. This pointer
613 * both enables us to find the zonelist cache, and in the case of
614 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
615 * to know that the zonelist cache is not there.
617 * The end result is that struct zonelists come in two flavors:
618 * 1) The full, fixed length version, shown below, and
619 * 2) The custom zonelists for MPOL_BIND.
620 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
622 * Even though there may be multiple CPU cores on a node modifying
623 * fullzones or last_full_zap in the same zonelist_cache at the same
624 * time, we don't lock it. This is just hint data - if it is wrong now
625 * and then, the allocator will still function, perhaps a bit slower.
629 struct zonelist_cache {
630 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
631 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
632 unsigned long last_full_zap; /* when last zap'd (jiffies) */
635 #define MAX_ZONELISTS 1
636 struct zonelist_cache;
640 * This struct contains information about a zone in a zonelist. It is stored
641 * here to avoid dereferences into large structures and lookups of tables
644 struct zone *zone; /* Pointer to actual zone */
645 int zone_idx; /* zone_idx(zoneref->zone) */
649 * One allocation request operates on a zonelist. A zonelist
650 * is a list of zones, the first one is the 'goal' of the
651 * allocation, the other zones are fallback zones, in decreasing
654 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
655 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
657 * To speed the reading of the zonelist, the zonerefs contain the zone index
658 * of the entry being read. Helper functions to access information given
659 * a struct zoneref are
661 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
662 * zonelist_zone_idx() - Return the index of the zone for an entry
663 * zonelist_node_idx() - Return the index of the node for an entry
666 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
667 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
669 struct zonelist_cache zlcache; // optional ...
673 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
674 struct node_active_region {
675 unsigned long start_pfn;
676 unsigned long end_pfn;
679 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
681 #ifndef CONFIG_DISCONTIGMEM
682 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
683 extern struct page *mem_map;
687 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
688 * (mostly NUMA machines?) to denote a higher-level memory zone than the
691 * On NUMA machines, each NUMA node would have a pg_data_t to describe
692 * it's memory layout.
694 * Memory statistics and page replacement data structures are maintained on a
698 typedef struct pglist_data {
699 struct zone node_zones[MAX_NR_ZONES];
700 struct zonelist node_zonelists[MAX_ZONELISTS];
702 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
703 struct page *node_mem_map;
705 struct page_cgroup *node_page_cgroup;
708 #ifndef CONFIG_NO_BOOTMEM
709 struct bootmem_data *bdata;
711 #ifdef CONFIG_MEMORY_HOTPLUG
713 * Must be held any time you expect node_start_pfn, node_present_pages
714 * or node_spanned_pages stay constant. Holding this will also
715 * guarantee that any pfn_valid() stays that way.
717 * Nests above zone->lock and zone->size_seqlock.
719 spinlock_t node_size_lock;
721 unsigned long node_start_pfn;
722 unsigned long node_present_pages; /* total number of physical pages */
723 unsigned long node_spanned_pages; /* total size of physical page
724 range, including holes */
726 nodemask_t reclaim_nodes; /* Nodes allowed to reclaim from */
727 wait_queue_head_t kswapd_wait;
728 wait_queue_head_t pfmemalloc_wait;
729 struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */
730 int kswapd_max_order;
731 enum zone_type classzone_idx;
732 #ifdef CONFIG_NUMA_BALANCING
734 * Lock serializing the per destination node AutoNUMA memory
735 * migration rate limiting data.
737 spinlock_t numabalancing_migrate_lock;
739 /* Rate limiting time interval */
740 unsigned long numabalancing_migrate_next_window;
742 /* Number of pages migrated during the rate limiting time interval */
743 unsigned long numabalancing_migrate_nr_pages;
747 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
748 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
749 #ifdef CONFIG_FLAT_NODE_MEM_MAP
750 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
752 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
754 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
756 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
758 #define node_end_pfn(nid) ({\
759 pg_data_t *__pgdat = NODE_DATA(nid);\
760 __pgdat->node_start_pfn + __pgdat->node_spanned_pages;\
763 #include <linux/memory_hotplug.h>
765 extern struct mutex zonelists_mutex;
766 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
767 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
768 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
769 int classzone_idx, int alloc_flags);
770 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
771 int classzone_idx, int alloc_flags);
772 enum memmap_context {
776 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
778 enum memmap_context context);
780 extern void lruvec_init(struct lruvec *lruvec);
782 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
787 return container_of(lruvec, struct zone, lruvec);
791 #ifdef CONFIG_HAVE_MEMORY_PRESENT
792 void memory_present(int nid, unsigned long start, unsigned long end);
794 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
797 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
798 int local_memory_node(int node_id);
800 static inline int local_memory_node(int node_id) { return node_id; };
803 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
804 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
808 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
810 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
812 static inline int populated_zone(struct zone *zone)
814 return (!!zone->present_pages);
817 extern int movable_zone;
819 static inline int zone_movable_is_highmem(void)
821 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
822 return movable_zone == ZONE_HIGHMEM;
828 static inline int is_highmem_idx(enum zone_type idx)
830 #ifdef CONFIG_HIGHMEM
831 return (idx == ZONE_HIGHMEM ||
832 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
838 static inline int is_normal_idx(enum zone_type idx)
840 return (idx == ZONE_NORMAL);
844 * is_highmem - helper function to quickly check if a struct zone is a
845 * highmem zone or not. This is an attempt to keep references
846 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
847 * @zone - pointer to struct zone variable
849 static inline int is_highmem(struct zone *zone)
851 #ifdef CONFIG_HIGHMEM
852 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
853 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
854 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
855 zone_movable_is_highmem());
861 static inline int is_normal(struct zone *zone)
863 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
866 static inline int is_dma32(struct zone *zone)
868 #ifdef CONFIG_ZONE_DMA32
869 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
875 static inline int is_dma(struct zone *zone)
877 #ifdef CONFIG_ZONE_DMA
878 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
884 /* These two functions are used to setup the per zone pages min values */
886 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
887 void __user *, size_t *, loff_t *);
888 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
889 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
890 void __user *, size_t *, loff_t *);
891 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
892 void __user *, size_t *, loff_t *);
893 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
894 void __user *, size_t *, loff_t *);
895 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
896 void __user *, size_t *, loff_t *);
898 extern int numa_zonelist_order_handler(struct ctl_table *, int,
899 void __user *, size_t *, loff_t *);
900 extern char numa_zonelist_order[];
901 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
903 #ifndef CONFIG_NEED_MULTIPLE_NODES
905 extern struct pglist_data contig_page_data;
906 #define NODE_DATA(nid) (&contig_page_data)
907 #define NODE_MEM_MAP(nid) mem_map
909 #else /* CONFIG_NEED_MULTIPLE_NODES */
911 #include <asm/mmzone.h>
913 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
915 extern struct pglist_data *first_online_pgdat(void);
916 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
917 extern struct zone *next_zone(struct zone *zone);
920 * for_each_online_pgdat - helper macro to iterate over all online nodes
921 * @pgdat - pointer to a pg_data_t variable
923 #define for_each_online_pgdat(pgdat) \
924 for (pgdat = first_online_pgdat(); \
926 pgdat = next_online_pgdat(pgdat))
928 * for_each_zone - helper macro to iterate over all memory zones
929 * @zone - pointer to struct zone variable
931 * The user only needs to declare the zone variable, for_each_zone
934 #define for_each_zone(zone) \
935 for (zone = (first_online_pgdat())->node_zones; \
937 zone = next_zone(zone))
939 #define for_each_populated_zone(zone) \
940 for (zone = (first_online_pgdat())->node_zones; \
942 zone = next_zone(zone)) \
943 if (!populated_zone(zone)) \
947 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
949 return zoneref->zone;
952 static inline int zonelist_zone_idx(struct zoneref *zoneref)
954 return zoneref->zone_idx;
957 static inline int zonelist_node_idx(struct zoneref *zoneref)
960 /* zone_to_nid not available in this context */
961 return zoneref->zone->node;
964 #endif /* CONFIG_NUMA */
968 * 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
969 * @z - The cursor used as a starting point for the search
970 * @highest_zoneidx - The zone index of the highest zone to return
971 * @nodes - An optional nodemask to filter the zonelist with
972 * @zone - The first suitable zone found is returned via this parameter
974 * This function returns the next zone at or below a given zone index that is
975 * within the allowed nodemask using a cursor as the starting point for the
976 * search. The zoneref returned is a cursor that represents the current zone
977 * being examined. It should be advanced by one before calling
978 * next_zones_zonelist again.
980 struct zoneref *next_zones_zonelist(struct zoneref *z,
981 enum zone_type highest_zoneidx,
986 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
987 * @zonelist - The zonelist to search for a suitable zone
988 * @highest_zoneidx - The zone index of the highest zone to return
989 * @nodes - An optional nodemask to filter the zonelist with
990 * @zone - The first suitable zone found is returned via this parameter
992 * This function returns the first zone at or below a given zone index that is
993 * within the allowed nodemask. The zoneref returned is a cursor that can be
994 * used to iterate the zonelist with next_zones_zonelist by advancing it by
995 * one before calling.
997 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
998 enum zone_type highest_zoneidx,
1002 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1007 * 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
1008 * @zone - The current zone in the iterator
1009 * @z - The current pointer within zonelist->zones being iterated
1010 * @zlist - The zonelist being iterated
1011 * @highidx - The zone index of the highest zone to return
1012 * @nodemask - Nodemask allowed by the allocator
1014 * This iterator iterates though all zones at or below a given zone index and
1015 * within a given nodemask
1017 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1018 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1020 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1023 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1024 * @zone - The current zone in the iterator
1025 * @z - The current pointer within zonelist->zones being iterated
1026 * @zlist - The zonelist being iterated
1027 * @highidx - The zone index of the highest zone to return
1029 * This iterator iterates though all zones at or below a given zone index.
1031 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1032 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1034 #ifdef CONFIG_SPARSEMEM
1035 #include <asm/sparsemem.h>
1038 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1039 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1040 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1046 #ifdef CONFIG_FLATMEM
1047 #define pfn_to_nid(pfn) (0)
1050 #ifdef CONFIG_SPARSEMEM
1053 * SECTION_SHIFT #bits space required to store a section #
1055 * PA_SECTION_SHIFT physical address to/from section number
1056 * PFN_SECTION_SHIFT pfn to/from section number
1058 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
1060 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1061 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1063 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1065 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1066 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1068 #define SECTION_BLOCKFLAGS_BITS \
1069 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1071 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1072 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1075 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1076 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1078 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1079 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1083 struct mem_section {
1085 * This is, logically, a pointer to an array of struct
1086 * pages. However, it is stored with some other magic.
1087 * (see sparse.c::sparse_init_one_section())
1089 * Additionally during early boot we encode node id of
1090 * the location of the section here to guide allocation.
1091 * (see sparse.c::memory_present())
1093 * Making it a UL at least makes someone do a cast
1094 * before using it wrong.
1096 unsigned long section_mem_map;
1098 /* See declaration of similar field in struct zone */
1099 unsigned long *pageblock_flags;
1102 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1103 * section. (see memcontrol.h/page_cgroup.h about this.)
1105 struct page_cgroup *page_cgroup;
1110 #ifdef CONFIG_SPARSEMEM_EXTREME
1111 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1113 #define SECTIONS_PER_ROOT 1
1116 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1117 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1118 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1120 #ifdef CONFIG_SPARSEMEM_EXTREME
1121 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1123 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1126 static inline struct mem_section *__nr_to_section(unsigned long nr)
1128 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1130 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1132 extern int __section_nr(struct mem_section* ms);
1133 extern unsigned long usemap_size(void);
1136 * We use the lower bits of the mem_map pointer to store
1137 * a little bit of information. There should be at least
1138 * 3 bits here due to 32-bit alignment.
1140 #define SECTION_MARKED_PRESENT (1UL<<0)
1141 #define SECTION_HAS_MEM_MAP (1UL<<1)
1142 #define SECTION_MAP_LAST_BIT (1UL<<2)
1143 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1144 #define SECTION_NID_SHIFT 2
1146 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1148 unsigned long map = section->section_mem_map;
1149 map &= SECTION_MAP_MASK;
1150 return (struct page *)map;
1153 static inline int present_section(struct mem_section *section)
1155 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1158 static inline int present_section_nr(unsigned long nr)
1160 return present_section(__nr_to_section(nr));
1163 static inline int valid_section(struct mem_section *section)
1165 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1168 static inline int valid_section_nr(unsigned long nr)
1170 return valid_section(__nr_to_section(nr));
1173 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1175 return __nr_to_section(pfn_to_section_nr(pfn));
1178 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1179 static inline int pfn_valid(unsigned long pfn)
1181 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1183 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1187 static inline int pfn_present(unsigned long pfn)
1189 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1191 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1195 * These are _only_ used during initialisation, therefore they
1196 * can use __initdata ... They could have names to indicate
1200 #define pfn_to_nid(pfn) \
1202 unsigned long __pfn_to_nid_pfn = (pfn); \
1203 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1206 #define pfn_to_nid(pfn) (0)
1209 #define early_pfn_valid(pfn) pfn_valid(pfn)
1210 void sparse_init(void);
1212 #define sparse_init() do {} while (0)
1213 #define sparse_index_init(_sec, _nid) do {} while (0)
1214 #endif /* CONFIG_SPARSEMEM */
1216 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1217 bool early_pfn_in_nid(unsigned long pfn, int nid);
1219 #define early_pfn_in_nid(pfn, nid) (1)
1222 #ifndef early_pfn_valid
1223 #define early_pfn_valid(pfn) (1)
1226 void memory_present(int nid, unsigned long start, unsigned long end);
1227 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1230 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1231 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1232 * pfn_valid_within() should be used in this case; we optimise this away
1233 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1235 #ifdef CONFIG_HOLES_IN_ZONE
1236 #define pfn_valid_within(pfn) pfn_valid(pfn)
1238 #define pfn_valid_within(pfn) (1)
1241 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1243 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1244 * associated with it or not. In FLATMEM, it is expected that holes always
1245 * have valid memmap as long as there is valid PFNs either side of the hole.
1246 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1249 * However, an ARM, and maybe other embedded architectures in the future
1250 * free memmap backing holes to save memory on the assumption the memmap is
1251 * never used. The page_zone linkages are then broken even though pfn_valid()
1252 * returns true. A walker of the full memmap must then do this additional
1253 * check to ensure the memmap they are looking at is sane by making sure
1254 * the zone and PFN linkages are still valid. This is expensive, but walkers
1255 * of the full memmap are extremely rare.
1257 int memmap_valid_within(unsigned long pfn,
1258 struct page *page, struct zone *zone);
1260 static inline int memmap_valid_within(unsigned long pfn,
1261 struct page *page, struct zone *zone)
1265 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1267 #endif /* !__GENERATING_BOUNDS.H */
1268 #endif /* !__ASSEMBLY__ */
1269 #endif /* _LINUX_MMZONE_H */