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 MIGRATE_ISOLATE, /* can't allocate from here */
65 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
67 # define is_migrate_cma(migratetype) false
70 #define for_each_migratetype_order(order, type) \
71 for (order = 0; order < MAX_ORDER; order++) \
72 for (type = 0; type < MIGRATE_TYPES; type++)
74 extern int page_group_by_mobility_disabled;
76 static inline int get_pageblock_migratetype(struct page *page)
78 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
82 struct list_head free_list[MIGRATE_TYPES];
83 unsigned long nr_free;
89 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
90 * So add a wild amount of padding here to ensure that they fall into separate
91 * cachelines. There are very few zone structures in the machine, so space
92 * consumption is not a concern here.
94 #if defined(CONFIG_SMP)
97 } ____cacheline_internodealigned_in_smp;
98 #define ZONE_PADDING(name) struct zone_padding name;
100 #define ZONE_PADDING(name)
103 enum zone_stat_item {
104 /* First 128 byte cacheline (assuming 64 bit words) */
107 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
108 NR_ACTIVE_ANON, /* " " " " " */
109 NR_INACTIVE_FILE, /* " " " " " */
110 NR_ACTIVE_FILE, /* " " " " " */
111 NR_UNEVICTABLE, /* " " " " " */
112 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
113 NR_ANON_PAGES, /* Mapped anonymous pages */
114 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
115 only modified from process context */
120 NR_SLAB_UNRECLAIMABLE,
121 NR_PAGETABLE, /* used for pagetables */
123 /* Second 128 byte cacheline */
124 NR_UNSTABLE_NFS, /* NFS unstable pages */
127 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
128 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
129 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
130 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
131 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
132 NR_DIRTIED, /* page dirtyings since bootup */
133 NR_WRITTEN, /* page writings since bootup */
135 NUMA_HIT, /* allocated in intended node */
136 NUMA_MISS, /* allocated in non intended node */
137 NUMA_FOREIGN, /* was intended here, hit elsewhere */
138 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
139 NUMA_LOCAL, /* allocation from local node */
140 NUMA_OTHER, /* allocation from other node */
142 NR_ANON_TRANSPARENT_HUGEPAGES,
144 NR_VM_ZONE_STAT_ITEMS };
147 * We do arithmetic on the LRU lists in various places in the code,
148 * so it is important to keep the active lists LRU_ACTIVE higher in
149 * the array than the corresponding inactive lists, and to keep
150 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
152 * This has to be kept in sync with the statistics in zone_stat_item
153 * above and the descriptions in vmstat_text in mm/vmstat.c
160 LRU_INACTIVE_ANON = LRU_BASE,
161 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
162 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
163 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
168 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
170 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
172 static inline int is_file_lru(enum lru_list lru)
174 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
177 static inline int is_active_lru(enum lru_list lru)
179 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
182 static inline int is_unevictable_lru(enum lru_list lru)
184 return (lru == LRU_UNEVICTABLE);
187 struct zone_reclaim_stat {
189 * The pageout code in vmscan.c keeps track of how many of the
190 * mem/swap backed and file backed pages are referenced.
191 * The higher the rotated/scanned ratio, the more valuable
194 * The anon LRU stats live in [0], file LRU stats in [1]
196 unsigned long recent_rotated[2];
197 unsigned long recent_scanned[2];
201 struct list_head lists[NR_LRU_LISTS];
202 struct zone_reclaim_stat reclaim_stat;
208 /* Mask used at gathering information at once (see memcontrol.c) */
209 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
210 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
211 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
213 /* Isolate clean file */
214 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
215 /* Isolate unmapped file */
216 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
217 /* Isolate for asynchronous migration */
218 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
219 /* Isolate unevictable pages */
220 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
222 /* LRU Isolation modes. */
223 typedef unsigned __bitwise__ isolate_mode_t;
225 enum zone_watermarks {
232 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
233 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
234 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
236 struct per_cpu_pages {
237 int count; /* number of pages in the list */
238 int high; /* high watermark, emptying needed */
239 int batch; /* chunk size for buddy add/remove */
241 /* Lists of pages, one per migrate type stored on the pcp-lists */
242 struct list_head lists[MIGRATE_PCPTYPES];
245 struct per_cpu_pageset {
246 struct per_cpu_pages pcp;
252 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
256 #endif /* !__GENERATING_BOUNDS.H */
259 #ifdef CONFIG_ZONE_DMA
261 * ZONE_DMA is used when there are devices that are not able
262 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
263 * carve out the portion of memory that is needed for these devices.
264 * The range is arch specific.
269 * ---------------------------
270 * parisc, ia64, sparc <4G
273 * alpha Unlimited or 0-16MB.
275 * i386, x86_64 and multiple other arches
280 #ifdef CONFIG_ZONE_DMA32
282 * x86_64 needs two ZONE_DMAs because it supports devices that are
283 * only able to do DMA to the lower 16M but also 32 bit devices that
284 * can only do DMA areas below 4G.
289 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
290 * performed on pages in ZONE_NORMAL if the DMA devices support
291 * transfers to all addressable memory.
294 #ifdef CONFIG_HIGHMEM
296 * A memory area that is only addressable by the kernel through
297 * mapping portions into its own address space. This is for example
298 * used by i386 to allow the kernel to address the memory beyond
299 * 900MB. The kernel will set up special mappings (page
300 * table entries on i386) for each page that the kernel needs to
309 #ifndef __GENERATING_BOUNDS_H
312 * When a memory allocation must conform to specific limitations (such
313 * as being suitable for DMA) the caller will pass in hints to the
314 * allocator in the gfp_mask, in the zone modifier bits. These bits
315 * are used to select a priority ordered list of memory zones which
316 * match the requested limits. See gfp_zone() in include/linux/gfp.h
320 #define ZONES_SHIFT 0
321 #elif MAX_NR_ZONES <= 2
322 #define ZONES_SHIFT 1
323 #elif MAX_NR_ZONES <= 4
324 #define ZONES_SHIFT 2
326 #error ZONES_SHIFT -- too many zones configured adjust calculation
330 /* Fields commonly accessed by the page allocator */
332 /* zone watermarks, access with *_wmark_pages(zone) macros */
333 unsigned long watermark[NR_WMARK];
336 * When free pages are below this point, additional steps are taken
337 * when reading the number of free pages to avoid per-cpu counter
338 * drift allowing watermarks to be breached
340 unsigned long percpu_drift_mark;
343 * We don't know if the memory that we're going to allocate will be freeable
344 * or/and it will be released eventually, so to avoid totally wasting several
345 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
346 * to run OOM on the lower zones despite there's tons of freeable ram
347 * on the higher zones). This array is recalculated at runtime if the
348 * sysctl_lowmem_reserve_ratio sysctl changes.
350 unsigned long lowmem_reserve[MAX_NR_ZONES];
353 * This is a per-zone reserve of pages that should not be
354 * considered dirtyable memory.
356 unsigned long dirty_balance_reserve;
361 * zone reclaim becomes active if more unmapped pages exist.
363 unsigned long min_unmapped_pages;
364 unsigned long min_slab_pages;
366 struct per_cpu_pageset __percpu *pageset;
368 * free areas of different sizes
371 int all_unreclaimable; /* All pages pinned */
372 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
373 /* Set to true when the PG_migrate_skip bits should be cleared */
374 bool compact_blockskip_flush;
376 /* pfns where compaction scanners should start */
377 unsigned long compact_cached_free_pfn;
378 unsigned long compact_cached_migrate_pfn;
380 #ifdef CONFIG_MEMORY_HOTPLUG
381 /* see spanned/present_pages for more description */
382 seqlock_t span_seqlock;
384 struct free_area free_area[MAX_ORDER];
386 #ifndef CONFIG_SPARSEMEM
388 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
389 * In SPARSEMEM, this map is stored in struct mem_section
391 unsigned long *pageblock_flags;
392 #endif /* CONFIG_SPARSEMEM */
394 #ifdef CONFIG_COMPACTION
396 * On compaction failure, 1<<compact_defer_shift compactions
397 * are skipped before trying again. The number attempted since
398 * last failure is tracked with compact_considered.
400 unsigned int compact_considered;
401 unsigned int compact_defer_shift;
402 int compact_order_failed;
407 /* Fields commonly accessed by the page reclaim scanner */
409 struct lruvec lruvec;
411 unsigned long pages_scanned; /* since last reclaim */
412 unsigned long flags; /* zone flags, see below */
414 /* Zone statistics */
415 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
418 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
419 * this zone's LRU. Maintained by the pageout code.
421 unsigned int inactive_ratio;
425 /* Rarely used or read-mostly fields */
428 * wait_table -- the array holding the hash table
429 * wait_table_hash_nr_entries -- the size of the hash table array
430 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
432 * The purpose of all these is to keep track of the people
433 * waiting for a page to become available and make them
434 * runnable again when possible. The trouble is that this
435 * consumes a lot of space, especially when so few things
436 * wait on pages at a given time. So instead of using
437 * per-page waitqueues, we use a waitqueue hash table.
439 * The bucket discipline is to sleep on the same queue when
440 * colliding and wake all in that wait queue when removing.
441 * When something wakes, it must check to be sure its page is
442 * truly available, a la thundering herd. The cost of a
443 * collision is great, but given the expected load of the
444 * table, they should be so rare as to be outweighed by the
445 * benefits from the saved space.
447 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
448 * primary users of these fields, and in mm/page_alloc.c
449 * free_area_init_core() performs the initialization of them.
451 wait_queue_head_t * wait_table;
452 unsigned long wait_table_hash_nr_entries;
453 unsigned long wait_table_bits;
456 * Discontig memory support fields.
458 struct pglist_data *zone_pgdat;
459 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
460 unsigned long zone_start_pfn;
463 * spanned_pages is the total pages spanned by the zone, including
464 * holes, which is calculated as:
465 * spanned_pages = zone_end_pfn - zone_start_pfn;
467 * present_pages is physical pages existing within the zone, which
469 * present_pages = spanned_pages - absent_pages(pags in holes);
471 * managed_pages is present pages managed by the buddy system, which
472 * is calculated as (reserved_pages includes pages allocated by the
473 * bootmem allocator):
474 * managed_pages = present_pages - reserved_pages;
476 * So present_pages may be used by memory hotplug or memory power
477 * management logic to figure out unmanaged pages by checking
478 * (present_pages - managed_pages). And managed_pages should be used
479 * by page allocator and vm scanner to calculate all kinds of watermarks
484 * zone_start_pfn and spanned_pages are protected by span_seqlock.
485 * It is a seqlock because it has to be read outside of zone->lock,
486 * and it is done in the main allocator path. But, it is written
487 * quite infrequently.
489 * The span_seq lock is declared along with zone->lock because it is
490 * frequently read in proximity to zone->lock. It's good to
491 * give them a chance of being in the same cacheline.
493 * Write access to present_pages and managed_pages at runtime should
494 * be protected by lock_memory_hotplug()/unlock_memory_hotplug().
495 * Any reader who can't tolerant drift of present_pages and
496 * managed_pages should hold memory hotplug lock to get a stable value.
498 unsigned long spanned_pages;
499 unsigned long present_pages;
500 unsigned long managed_pages;
503 * rarely used fields:
506 #ifdef CONFIG_MEMORY_ISOLATION
508 * the number of MIGRATE_ISOLATE *pageblock*.
509 * We need this for free page counting. Look at zone_watermark_ok_safe.
510 * It's protected by zone->lock
512 int nr_pageblock_isolate;
514 } ____cacheline_internodealigned_in_smp;
517 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
518 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
519 ZONE_CONGESTED, /* zone has many dirty pages backed by
524 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
526 set_bit(flag, &zone->flags);
529 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
531 return test_and_set_bit(flag, &zone->flags);
534 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
536 clear_bit(flag, &zone->flags);
539 static inline int zone_is_reclaim_congested(const struct zone *zone)
541 return test_bit(ZONE_CONGESTED, &zone->flags);
544 static inline int zone_is_reclaim_locked(const struct zone *zone)
546 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
549 static inline int zone_is_oom_locked(const struct zone *zone)
551 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
555 * The "priority" of VM scanning is how much of the queues we will scan in one
556 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
557 * queues ("queue_length >> 12") during an aging round.
559 #define DEF_PRIORITY 12
561 /* Maximum number of zones on a zonelist */
562 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
567 * The NUMA zonelists are doubled because we need zonelists that restrict the
568 * allocations to a single node for GFP_THISNODE.
570 * [0] : Zonelist with fallback
571 * [1] : No fallback (GFP_THISNODE)
573 #define MAX_ZONELISTS 2
577 * We cache key information from each zonelist for smaller cache
578 * footprint when scanning for free pages in get_page_from_freelist().
580 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
581 * up short of free memory since the last time (last_fullzone_zap)
582 * we zero'd fullzones.
583 * 2) The array z_to_n[] maps each zone in the zonelist to its node
584 * id, so that we can efficiently evaluate whether that node is
585 * set in the current tasks mems_allowed.
587 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
588 * indexed by a zones offset in the zonelist zones[] array.
590 * The get_page_from_freelist() routine does two scans. During the
591 * first scan, we skip zones whose corresponding bit in 'fullzones'
592 * is set or whose corresponding node in current->mems_allowed (which
593 * comes from cpusets) is not set. During the second scan, we bypass
594 * this zonelist_cache, to ensure we look methodically at each zone.
596 * Once per second, we zero out (zap) fullzones, forcing us to
597 * reconsider nodes that might have regained more free memory.
598 * The field last_full_zap is the time we last zapped fullzones.
600 * This mechanism reduces the amount of time we waste repeatedly
601 * reexaming zones for free memory when they just came up low on
602 * memory momentarilly ago.
604 * The zonelist_cache struct members logically belong in struct
605 * zonelist. However, the mempolicy zonelists constructed for
606 * MPOL_BIND are intentionally variable length (and usually much
607 * shorter). A general purpose mechanism for handling structs with
608 * multiple variable length members is more mechanism than we want
609 * here. We resort to some special case hackery instead.
611 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
612 * part because they are shorter), so we put the fixed length stuff
613 * at the front of the zonelist struct, ending in a variable length
614 * zones[], as is needed by MPOL_BIND.
616 * Then we put the optional zonelist cache on the end of the zonelist
617 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
618 * the fixed length portion at the front of the struct. This pointer
619 * both enables us to find the zonelist cache, and in the case of
620 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
621 * to know that the zonelist cache is not there.
623 * The end result is that struct zonelists come in two flavors:
624 * 1) The full, fixed length version, shown below, and
625 * 2) The custom zonelists for MPOL_BIND.
626 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
628 * Even though there may be multiple CPU cores on a node modifying
629 * fullzones or last_full_zap in the same zonelist_cache at the same
630 * time, we don't lock it. This is just hint data - if it is wrong now
631 * and then, the allocator will still function, perhaps a bit slower.
635 struct zonelist_cache {
636 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
637 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
638 unsigned long last_full_zap; /* when last zap'd (jiffies) */
641 #define MAX_ZONELISTS 1
642 struct zonelist_cache;
646 * This struct contains information about a zone in a zonelist. It is stored
647 * here to avoid dereferences into large structures and lookups of tables
650 struct zone *zone; /* Pointer to actual zone */
651 int zone_idx; /* zone_idx(zoneref->zone) */
655 * One allocation request operates on a zonelist. A zonelist
656 * is a list of zones, the first one is the 'goal' of the
657 * allocation, the other zones are fallback zones, in decreasing
660 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
661 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
663 * To speed the reading of the zonelist, the zonerefs contain the zone index
664 * of the entry being read. Helper functions to access information given
665 * a struct zoneref are
667 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
668 * zonelist_zone_idx() - Return the index of the zone for an entry
669 * zonelist_node_idx() - Return the index of the node for an entry
672 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
673 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
675 struct zonelist_cache zlcache; // optional ...
679 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
680 struct node_active_region {
681 unsigned long start_pfn;
682 unsigned long end_pfn;
685 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
687 #ifndef CONFIG_DISCONTIGMEM
688 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
689 extern struct page *mem_map;
693 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
694 * (mostly NUMA machines?) to denote a higher-level memory zone than the
697 * On NUMA machines, each NUMA node would have a pg_data_t to describe
698 * it's memory layout.
700 * Memory statistics and page replacement data structures are maintained on a
704 typedef struct pglist_data {
705 struct zone node_zones[MAX_NR_ZONES];
706 struct zonelist node_zonelists[MAX_ZONELISTS];
708 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
709 struct page *node_mem_map;
711 struct page_cgroup *node_page_cgroup;
714 #ifndef CONFIG_NO_BOOTMEM
715 struct bootmem_data *bdata;
717 #ifdef CONFIG_MEMORY_HOTPLUG
719 * Must be held any time you expect node_start_pfn, node_present_pages
720 * or node_spanned_pages stay constant. Holding this will also
721 * guarantee that any pfn_valid() stays that way.
723 * Nests above zone->lock and zone->size_seqlock.
725 spinlock_t node_size_lock;
727 unsigned long node_start_pfn;
728 unsigned long node_present_pages; /* total number of physical pages */
729 unsigned long node_spanned_pages; /* total size of physical page
730 range, including holes */
732 nodemask_t reclaim_nodes; /* Nodes allowed to reclaim from */
733 wait_queue_head_t kswapd_wait;
734 wait_queue_head_t pfmemalloc_wait;
735 struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */
736 int kswapd_max_order;
737 enum zone_type classzone_idx;
740 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
741 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
742 #ifdef CONFIG_FLAT_NODE_MEM_MAP
743 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
745 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
747 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
749 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
751 #define node_end_pfn(nid) ({\
752 pg_data_t *__pgdat = NODE_DATA(nid);\
753 __pgdat->node_start_pfn + __pgdat->node_spanned_pages;\
756 #include <linux/memory_hotplug.h>
758 extern struct mutex zonelists_mutex;
759 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
760 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
761 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
762 int classzone_idx, int alloc_flags);
763 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
764 int classzone_idx, int alloc_flags);
765 enum memmap_context {
769 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
771 enum memmap_context context);
773 extern void lruvec_init(struct lruvec *lruvec);
775 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
780 return container_of(lruvec, struct zone, lruvec);
784 #ifdef CONFIG_HAVE_MEMORY_PRESENT
785 void memory_present(int nid, unsigned long start, unsigned long end);
787 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
790 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
791 int local_memory_node(int node_id);
793 static inline int local_memory_node(int node_id) { return node_id; };
796 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
797 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
801 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
803 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
805 static inline int populated_zone(struct zone *zone)
807 return (!!zone->present_pages);
810 extern int movable_zone;
812 static inline int zone_movable_is_highmem(void)
814 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
815 return movable_zone == ZONE_HIGHMEM;
821 static inline int is_highmem_idx(enum zone_type idx)
823 #ifdef CONFIG_HIGHMEM
824 return (idx == ZONE_HIGHMEM ||
825 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
831 static inline int is_normal_idx(enum zone_type idx)
833 return (idx == ZONE_NORMAL);
837 * is_highmem - helper function to quickly check if a struct zone is a
838 * highmem zone or not. This is an attempt to keep references
839 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
840 * @zone - pointer to struct zone variable
842 static inline int is_highmem(struct zone *zone)
844 #ifdef CONFIG_HIGHMEM
845 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
846 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
847 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
848 zone_movable_is_highmem());
854 static inline int is_normal(struct zone *zone)
856 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
859 static inline int is_dma32(struct zone *zone)
861 #ifdef CONFIG_ZONE_DMA32
862 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
868 static inline int is_dma(struct zone *zone)
870 #ifdef CONFIG_ZONE_DMA
871 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
877 /* These two functions are used to setup the per zone pages min values */
879 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
880 void __user *, size_t *, loff_t *);
881 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
882 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
883 void __user *, size_t *, loff_t *);
884 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
885 void __user *, size_t *, loff_t *);
886 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
887 void __user *, size_t *, loff_t *);
888 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
889 void __user *, size_t *, loff_t *);
891 extern int numa_zonelist_order_handler(struct ctl_table *, int,
892 void __user *, size_t *, loff_t *);
893 extern char numa_zonelist_order[];
894 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
896 #ifndef CONFIG_NEED_MULTIPLE_NODES
898 extern struct pglist_data contig_page_data;
899 #define NODE_DATA(nid) (&contig_page_data)
900 #define NODE_MEM_MAP(nid) mem_map
902 #else /* CONFIG_NEED_MULTIPLE_NODES */
904 #include <asm/mmzone.h>
906 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
908 extern struct pglist_data *first_online_pgdat(void);
909 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
910 extern struct zone *next_zone(struct zone *zone);
913 * for_each_online_pgdat - helper macro to iterate over all online nodes
914 * @pgdat - pointer to a pg_data_t variable
916 #define for_each_online_pgdat(pgdat) \
917 for (pgdat = first_online_pgdat(); \
919 pgdat = next_online_pgdat(pgdat))
921 * for_each_zone - helper macro to iterate over all memory zones
922 * @zone - pointer to struct zone variable
924 * The user only needs to declare the zone variable, for_each_zone
927 #define for_each_zone(zone) \
928 for (zone = (first_online_pgdat())->node_zones; \
930 zone = next_zone(zone))
932 #define for_each_populated_zone(zone) \
933 for (zone = (first_online_pgdat())->node_zones; \
935 zone = next_zone(zone)) \
936 if (!populated_zone(zone)) \
940 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
942 return zoneref->zone;
945 static inline int zonelist_zone_idx(struct zoneref *zoneref)
947 return zoneref->zone_idx;
950 static inline int zonelist_node_idx(struct zoneref *zoneref)
953 /* zone_to_nid not available in this context */
954 return zoneref->zone->node;
957 #endif /* CONFIG_NUMA */
961 * 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
962 * @z - The cursor used as a starting point for the search
963 * @highest_zoneidx - The zone index of the highest zone to return
964 * @nodes - An optional nodemask to filter the zonelist with
965 * @zone - The first suitable zone found is returned via this parameter
967 * This function returns the next zone at or below a given zone index that is
968 * within the allowed nodemask using a cursor as the starting point for the
969 * search. The zoneref returned is a cursor that represents the current zone
970 * being examined. It should be advanced by one before calling
971 * next_zones_zonelist again.
973 struct zoneref *next_zones_zonelist(struct zoneref *z,
974 enum zone_type highest_zoneidx,
979 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
980 * @zonelist - The zonelist to search for a suitable zone
981 * @highest_zoneidx - The zone index of the highest zone to return
982 * @nodes - An optional nodemask to filter the zonelist with
983 * @zone - The first suitable zone found is returned via this parameter
985 * This function returns the first zone at or below a given zone index that is
986 * within the allowed nodemask. The zoneref returned is a cursor that can be
987 * used to iterate the zonelist with next_zones_zonelist by advancing it by
988 * one before calling.
990 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
991 enum zone_type highest_zoneidx,
995 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1000 * 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
1001 * @zone - The current zone in the iterator
1002 * @z - The current pointer within zonelist->zones being iterated
1003 * @zlist - The zonelist being iterated
1004 * @highidx - The zone index of the highest zone to return
1005 * @nodemask - Nodemask allowed by the allocator
1007 * This iterator iterates though all zones at or below a given zone index and
1008 * within a given nodemask
1010 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1011 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1013 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1016 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1017 * @zone - The current zone in the iterator
1018 * @z - The current pointer within zonelist->zones being iterated
1019 * @zlist - The zonelist being iterated
1020 * @highidx - The zone index of the highest zone to return
1022 * This iterator iterates though all zones at or below a given zone index.
1024 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1025 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1027 #ifdef CONFIG_SPARSEMEM
1028 #include <asm/sparsemem.h>
1031 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1032 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1033 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1039 #ifdef CONFIG_FLATMEM
1040 #define pfn_to_nid(pfn) (0)
1043 #ifdef CONFIG_SPARSEMEM
1046 * SECTION_SHIFT #bits space required to store a section #
1048 * PA_SECTION_SHIFT physical address to/from section number
1049 * PFN_SECTION_SHIFT pfn to/from section number
1051 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
1053 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1054 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1056 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1058 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1059 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1061 #define SECTION_BLOCKFLAGS_BITS \
1062 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1064 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1065 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1068 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1069 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1071 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1072 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1076 struct mem_section {
1078 * This is, logically, a pointer to an array of struct
1079 * pages. However, it is stored with some other magic.
1080 * (see sparse.c::sparse_init_one_section())
1082 * Additionally during early boot we encode node id of
1083 * the location of the section here to guide allocation.
1084 * (see sparse.c::memory_present())
1086 * Making it a UL at least makes someone do a cast
1087 * before using it wrong.
1089 unsigned long section_mem_map;
1091 /* See declaration of similar field in struct zone */
1092 unsigned long *pageblock_flags;
1095 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1096 * section. (see memcontrol.h/page_cgroup.h about this.)
1098 struct page_cgroup *page_cgroup;
1103 #ifdef CONFIG_SPARSEMEM_EXTREME
1104 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1106 #define SECTIONS_PER_ROOT 1
1109 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1110 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1111 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1113 #ifdef CONFIG_SPARSEMEM_EXTREME
1114 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1116 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1119 static inline struct mem_section *__nr_to_section(unsigned long nr)
1121 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1123 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1125 extern int __section_nr(struct mem_section* ms);
1126 extern unsigned long usemap_size(void);
1129 * We use the lower bits of the mem_map pointer to store
1130 * a little bit of information. There should be at least
1131 * 3 bits here due to 32-bit alignment.
1133 #define SECTION_MARKED_PRESENT (1UL<<0)
1134 #define SECTION_HAS_MEM_MAP (1UL<<1)
1135 #define SECTION_MAP_LAST_BIT (1UL<<2)
1136 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1137 #define SECTION_NID_SHIFT 2
1139 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1141 unsigned long map = section->section_mem_map;
1142 map &= SECTION_MAP_MASK;
1143 return (struct page *)map;
1146 static inline int present_section(struct mem_section *section)
1148 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1151 static inline int present_section_nr(unsigned long nr)
1153 return present_section(__nr_to_section(nr));
1156 static inline int valid_section(struct mem_section *section)
1158 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1161 static inline int valid_section_nr(unsigned long nr)
1163 return valid_section(__nr_to_section(nr));
1166 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1168 return __nr_to_section(pfn_to_section_nr(pfn));
1171 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1172 static inline int pfn_valid(unsigned long pfn)
1174 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1176 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1180 static inline int pfn_present(unsigned long pfn)
1182 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1184 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1188 * These are _only_ used during initialisation, therefore they
1189 * can use __initdata ... They could have names to indicate
1193 #define pfn_to_nid(pfn) \
1195 unsigned long __pfn_to_nid_pfn = (pfn); \
1196 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1199 #define pfn_to_nid(pfn) (0)
1202 #define early_pfn_valid(pfn) pfn_valid(pfn)
1203 void sparse_init(void);
1205 #define sparse_init() do {} while (0)
1206 #define sparse_index_init(_sec, _nid) do {} while (0)
1207 #endif /* CONFIG_SPARSEMEM */
1209 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1210 bool early_pfn_in_nid(unsigned long pfn, int nid);
1212 #define early_pfn_in_nid(pfn, nid) (1)
1215 #ifndef early_pfn_valid
1216 #define early_pfn_valid(pfn) (1)
1219 void memory_present(int nid, unsigned long start, unsigned long end);
1220 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1223 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1224 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1225 * pfn_valid_within() should be used in this case; we optimise this away
1226 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1228 #ifdef CONFIG_HOLES_IN_ZONE
1229 #define pfn_valid_within(pfn) pfn_valid(pfn)
1231 #define pfn_valid_within(pfn) (1)
1234 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1236 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1237 * associated with it or not. In FLATMEM, it is expected that holes always
1238 * have valid memmap as long as there is valid PFNs either side of the hole.
1239 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1242 * However, an ARM, and maybe other embedded architectures in the future
1243 * free memmap backing holes to save memory on the assumption the memmap is
1244 * never used. The page_zone linkages are then broken even though pfn_valid()
1245 * returns true. A walker of the full memmap must then do this additional
1246 * check to ensure the memmap they are looking at is sane by making sure
1247 * the zone and PFN linkages are still valid. This is expensive, but walkers
1248 * of the full memmap are extremely rare.
1250 int memmap_valid_within(unsigned long pfn,
1251 struct page *page, struct zone *zone);
1253 static inline int memmap_valid_within(unsigned long pfn,
1254 struct page *page, struct zone *zone)
1258 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1260 #endif /* !__GENERATING_BOUNDS.H */
1261 #endif /* !__ASSEMBLY__ */
1262 #endif /* _LINUX_MMZONE_H */