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)
66 # define cma_wmark_pages(zone) zone->min_cma_pages
68 # define is_migrate_cma(migratetype) false
69 # define cma_wmark_pages(zone) 0
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,
145 NR_VM_ZONE_STAT_ITEMS };
148 * We do arithmetic on the LRU lists in various places in the code,
149 * so it is important to keep the active lists LRU_ACTIVE higher in
150 * the array than the corresponding inactive lists, and to keep
151 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
153 * This has to be kept in sync with the statistics in zone_stat_item
154 * above and the descriptions in vmstat_text in mm/vmstat.c
161 LRU_INACTIVE_ANON = LRU_BASE,
162 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
163 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
164 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
169 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
171 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
173 static inline int is_file_lru(enum lru_list lru)
175 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
178 static inline int is_active_lru(enum lru_list lru)
180 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
183 static inline int is_unevictable_lru(enum lru_list lru)
185 return (lru == LRU_UNEVICTABLE);
188 struct zone_reclaim_stat {
190 * The pageout code in vmscan.c keeps track of how many of the
191 * mem/swap backed and file backed pages are refeferenced.
192 * The higher the rotated/scanned ratio, the more valuable
195 * The anon LRU stats live in [0], file LRU stats in [1]
197 unsigned long recent_rotated[2];
198 unsigned long recent_scanned[2];
202 struct list_head lists[NR_LRU_LISTS];
203 struct zone_reclaim_stat reclaim_stat;
206 /* Mask used at gathering information at once (see memcontrol.c) */
207 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
208 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
209 #define LRU_ALL_EVICTABLE (LRU_ALL_FILE | LRU_ALL_ANON)
210 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
212 /* Isolate clean file */
213 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
214 /* Isolate unmapped file */
215 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
216 /* Isolate for asynchronous migration */
217 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
219 /* LRU Isolation modes. */
220 typedef unsigned __bitwise__ isolate_mode_t;
222 enum zone_watermarks {
229 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
230 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
231 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
233 struct per_cpu_pages {
234 int count; /* number of pages in the list */
235 int high; /* high watermark, emptying needed */
236 int batch; /* chunk size for buddy add/remove */
238 /* Lists of pages, one per migrate type stored on the pcp-lists */
239 struct list_head lists[MIGRATE_PCPTYPES];
242 struct per_cpu_pageset {
243 struct per_cpu_pages pcp;
249 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
253 #endif /* !__GENERATING_BOUNDS.H */
256 #ifdef CONFIG_ZONE_DMA
258 * ZONE_DMA is used when there are devices that are not able
259 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
260 * carve out the portion of memory that is needed for these devices.
261 * The range is arch specific.
266 * ---------------------------
267 * parisc, ia64, sparc <4G
270 * alpha Unlimited or 0-16MB.
272 * i386, x86_64 and multiple other arches
277 #ifdef CONFIG_ZONE_DMA32
279 * x86_64 needs two ZONE_DMAs because it supports devices that are
280 * only able to do DMA to the lower 16M but also 32 bit devices that
281 * can only do DMA areas below 4G.
286 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
287 * performed on pages in ZONE_NORMAL if the DMA devices support
288 * transfers to all addressable memory.
291 #ifdef CONFIG_HIGHMEM
293 * A memory area that is only addressable by the kernel through
294 * mapping portions into its own address space. This is for example
295 * used by i386 to allow the kernel to address the memory beyond
296 * 900MB. The kernel will set up special mappings (page
297 * table entries on i386) for each page that the kernel needs to
306 #ifndef __GENERATING_BOUNDS_H
309 * When a memory allocation must conform to specific limitations (such
310 * as being suitable for DMA) the caller will pass in hints to the
311 * allocator in the gfp_mask, in the zone modifier bits. These bits
312 * are used to select a priority ordered list of memory zones which
313 * match the requested limits. See gfp_zone() in include/linux/gfp.h
317 #define ZONES_SHIFT 0
318 #elif MAX_NR_ZONES <= 2
319 #define ZONES_SHIFT 1
320 #elif MAX_NR_ZONES <= 4
321 #define ZONES_SHIFT 2
323 #error ZONES_SHIFT -- too many zones configured adjust calculation
327 /* Fields commonly accessed by the page allocator */
329 /* zone watermarks, access with *_wmark_pages(zone) macros */
330 unsigned long watermark[NR_WMARK];
333 * When free pages are below this point, additional steps are taken
334 * when reading the number of free pages to avoid per-cpu counter
335 * drift allowing watermarks to be breached
337 unsigned long percpu_drift_mark;
340 * We don't know if the memory that we're going to allocate will be freeable
341 * or/and it will be released eventually, so to avoid totally wasting several
342 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
343 * to run OOM on the lower zones despite there's tons of freeable ram
344 * on the higher zones). This array is recalculated at runtime if the
345 * sysctl_lowmem_reserve_ratio sysctl changes.
347 unsigned long lowmem_reserve[MAX_NR_ZONES];
350 * This is a per-zone reserve of pages that should not be
351 * considered dirtyable memory.
353 unsigned long dirty_balance_reserve;
358 * zone reclaim becomes active if more unmapped pages exist.
360 unsigned long min_unmapped_pages;
361 unsigned long min_slab_pages;
363 struct per_cpu_pageset __percpu *pageset;
365 * free areas of different sizes
368 int all_unreclaimable; /* All pages pinned */
369 #ifdef CONFIG_MEMORY_HOTPLUG
370 /* see spanned/present_pages for more description */
371 seqlock_t span_seqlock;
375 * CMA needs to increase watermark levels during the allocation
376 * process to make sure that the system is not starved.
378 unsigned long min_cma_pages;
380 struct free_area free_area[MAX_ORDER];
382 #ifndef CONFIG_SPARSEMEM
384 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
385 * In SPARSEMEM, this map is stored in struct mem_section
387 unsigned long *pageblock_flags;
388 #endif /* CONFIG_SPARSEMEM */
390 #ifdef CONFIG_COMPACTION
392 * On compaction failure, 1<<compact_defer_shift compactions
393 * are skipped before trying again. The number attempted since
394 * last failure is tracked with compact_considered.
396 unsigned int compact_considered;
397 unsigned int compact_defer_shift;
398 int compact_order_failed;
403 /* Fields commonly accessed by the page reclaim scanner */
405 struct lruvec lruvec;
407 unsigned long pages_scanned; /* since last reclaim */
408 unsigned long flags; /* zone flags, see below */
410 /* Zone statistics */
411 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
414 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
415 * this zone's LRU. Maintained by the pageout code.
417 unsigned int inactive_ratio;
421 /* Rarely used or read-mostly fields */
424 * wait_table -- the array holding the hash table
425 * wait_table_hash_nr_entries -- the size of the hash table array
426 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
428 * The purpose of all these is to keep track of the people
429 * waiting for a page to become available and make them
430 * runnable again when possible. The trouble is that this
431 * consumes a lot of space, especially when so few things
432 * wait on pages at a given time. So instead of using
433 * per-page waitqueues, we use a waitqueue hash table.
435 * The bucket discipline is to sleep on the same queue when
436 * colliding and wake all in that wait queue when removing.
437 * When something wakes, it must check to be sure its page is
438 * truly available, a la thundering herd. The cost of a
439 * collision is great, but given the expected load of the
440 * table, they should be so rare as to be outweighed by the
441 * benefits from the saved space.
443 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
444 * primary users of these fields, and in mm/page_alloc.c
445 * free_area_init_core() performs the initialization of them.
447 wait_queue_head_t * wait_table;
448 unsigned long wait_table_hash_nr_entries;
449 unsigned long wait_table_bits;
452 * Discontig memory support fields.
454 struct pglist_data *zone_pgdat;
455 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
456 unsigned long zone_start_pfn;
459 * zone_start_pfn, spanned_pages and present_pages are all
460 * protected by span_seqlock. It is a seqlock because it has
461 * to be read outside of zone->lock, and it is done in the main
462 * allocator path. But, it is written quite infrequently.
464 * The lock is declared along with zone->lock because it is
465 * frequently read in proximity to zone->lock. It's good to
466 * give them a chance of being in the same cacheline.
468 unsigned long spanned_pages; /* total size, including holes */
469 unsigned long present_pages; /* amount of memory (excluding holes) */
472 * rarely used fields:
475 } ____cacheline_internodealigned_in_smp;
478 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
479 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
480 ZONE_CONGESTED, /* zone has many dirty pages backed by
485 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
487 set_bit(flag, &zone->flags);
490 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
492 return test_and_set_bit(flag, &zone->flags);
495 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
497 clear_bit(flag, &zone->flags);
500 static inline int zone_is_reclaim_congested(const struct zone *zone)
502 return test_bit(ZONE_CONGESTED, &zone->flags);
505 static inline int zone_is_reclaim_locked(const struct zone *zone)
507 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
510 static inline int zone_is_oom_locked(const struct zone *zone)
512 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
516 * The "priority" of VM scanning is how much of the queues we will scan in one
517 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
518 * queues ("queue_length >> 12") during an aging round.
520 #define DEF_PRIORITY 12
522 /* Maximum number of zones on a zonelist */
523 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
528 * The NUMA zonelists are doubled because we need zonelists that restrict the
529 * allocations to a single node for GFP_THISNODE.
531 * [0] : Zonelist with fallback
532 * [1] : No fallback (GFP_THISNODE)
534 #define MAX_ZONELISTS 2
538 * We cache key information from each zonelist for smaller cache
539 * footprint when scanning for free pages in get_page_from_freelist().
541 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
542 * up short of free memory since the last time (last_fullzone_zap)
543 * we zero'd fullzones.
544 * 2) The array z_to_n[] maps each zone in the zonelist to its node
545 * id, so that we can efficiently evaluate whether that node is
546 * set in the current tasks mems_allowed.
548 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
549 * indexed by a zones offset in the zonelist zones[] array.
551 * The get_page_from_freelist() routine does two scans. During the
552 * first scan, we skip zones whose corresponding bit in 'fullzones'
553 * is set or whose corresponding node in current->mems_allowed (which
554 * comes from cpusets) is not set. During the second scan, we bypass
555 * this zonelist_cache, to ensure we look methodically at each zone.
557 * Once per second, we zero out (zap) fullzones, forcing us to
558 * reconsider nodes that might have regained more free memory.
559 * The field last_full_zap is the time we last zapped fullzones.
561 * This mechanism reduces the amount of time we waste repeatedly
562 * reexaming zones for free memory when they just came up low on
563 * memory momentarilly ago.
565 * The zonelist_cache struct members logically belong in struct
566 * zonelist. However, the mempolicy zonelists constructed for
567 * MPOL_BIND are intentionally variable length (and usually much
568 * shorter). A general purpose mechanism for handling structs with
569 * multiple variable length members is more mechanism than we want
570 * here. We resort to some special case hackery instead.
572 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
573 * part because they are shorter), so we put the fixed length stuff
574 * at the front of the zonelist struct, ending in a variable length
575 * zones[], as is needed by MPOL_BIND.
577 * Then we put the optional zonelist cache on the end of the zonelist
578 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
579 * the fixed length portion at the front of the struct. This pointer
580 * both enables us to find the zonelist cache, and in the case of
581 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
582 * to know that the zonelist cache is not there.
584 * The end result is that struct zonelists come in two flavors:
585 * 1) The full, fixed length version, shown below, and
586 * 2) The custom zonelists for MPOL_BIND.
587 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
589 * Even though there may be multiple CPU cores on a node modifying
590 * fullzones or last_full_zap in the same zonelist_cache at the same
591 * time, we don't lock it. This is just hint data - if it is wrong now
592 * and then, the allocator will still function, perhaps a bit slower.
596 struct zonelist_cache {
597 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
598 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
599 unsigned long last_full_zap; /* when last zap'd (jiffies) */
602 #define MAX_ZONELISTS 1
603 struct zonelist_cache;
607 * This struct contains information about a zone in a zonelist. It is stored
608 * here to avoid dereferences into large structures and lookups of tables
611 struct zone *zone; /* Pointer to actual zone */
612 int zone_idx; /* zone_idx(zoneref->zone) */
616 * One allocation request operates on a zonelist. A zonelist
617 * is a list of zones, the first one is the 'goal' of the
618 * allocation, the other zones are fallback zones, in decreasing
621 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
622 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
624 * To speed the reading of the zonelist, the zonerefs contain the zone index
625 * of the entry being read. Helper functions to access information given
626 * a struct zoneref are
628 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
629 * zonelist_zone_idx() - Return the index of the zone for an entry
630 * zonelist_node_idx() - Return the index of the node for an entry
633 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
634 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
636 struct zonelist_cache zlcache; // optional ...
640 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
641 struct node_active_region {
642 unsigned long start_pfn;
643 unsigned long end_pfn;
646 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
648 #ifndef CONFIG_DISCONTIGMEM
649 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
650 extern struct page *mem_map;
654 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
655 * (mostly NUMA machines?) to denote a higher-level memory zone than the
658 * On NUMA machines, each NUMA node would have a pg_data_t to describe
659 * it's memory layout.
661 * Memory statistics and page replacement data structures are maintained on a
665 typedef struct pglist_data {
666 struct zone node_zones[MAX_NR_ZONES];
667 struct zonelist node_zonelists[MAX_ZONELISTS];
669 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
670 struct page *node_mem_map;
671 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
672 struct page_cgroup *node_page_cgroup;
675 #ifndef CONFIG_NO_BOOTMEM
676 struct bootmem_data *bdata;
678 #ifdef CONFIG_MEMORY_HOTPLUG
680 * Must be held any time you expect node_start_pfn, node_present_pages
681 * or node_spanned_pages stay constant. Holding this will also
682 * guarantee that any pfn_valid() stays that way.
684 * Nests above zone->lock and zone->size_seqlock.
686 spinlock_t node_size_lock;
688 unsigned long node_start_pfn;
689 unsigned long node_present_pages; /* total number of physical pages */
690 unsigned long node_spanned_pages; /* total size of physical page
691 range, including holes */
693 wait_queue_head_t kswapd_wait;
694 struct task_struct *kswapd;
695 int kswapd_max_order;
696 enum zone_type classzone_idx;
699 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
700 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
701 #ifdef CONFIG_FLAT_NODE_MEM_MAP
702 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
704 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
706 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
708 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
710 #define node_end_pfn(nid) ({\
711 pg_data_t *__pgdat = NODE_DATA(nid);\
712 __pgdat->node_start_pfn + __pgdat->node_spanned_pages;\
715 #include <linux/memory_hotplug.h>
717 extern struct mutex zonelists_mutex;
718 void build_all_zonelists(void *data);
719 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
720 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
721 int classzone_idx, int alloc_flags);
722 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
723 int classzone_idx, int alloc_flags);
724 enum memmap_context {
728 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
730 enum memmap_context context);
732 #ifdef CONFIG_HAVE_MEMORY_PRESENT
733 void memory_present(int nid, unsigned long start, unsigned long end);
735 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
738 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
739 int local_memory_node(int node_id);
741 static inline int local_memory_node(int node_id) { return node_id; };
744 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
745 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
749 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
751 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
753 static inline int populated_zone(struct zone *zone)
755 return (!!zone->present_pages);
758 extern int movable_zone;
760 static inline int zone_movable_is_highmem(void)
762 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE)
763 return movable_zone == ZONE_HIGHMEM;
769 static inline int is_highmem_idx(enum zone_type idx)
771 #ifdef CONFIG_HIGHMEM
772 return (idx == ZONE_HIGHMEM ||
773 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
779 static inline int is_normal_idx(enum zone_type idx)
781 return (idx == ZONE_NORMAL);
785 * is_highmem - helper function to quickly check if a struct zone is a
786 * highmem zone or not. This is an attempt to keep references
787 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
788 * @zone - pointer to struct zone variable
790 static inline int is_highmem(struct zone *zone)
792 #ifdef CONFIG_HIGHMEM
793 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
794 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
795 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
796 zone_movable_is_highmem());
802 static inline int is_normal(struct zone *zone)
804 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
807 static inline int is_dma32(struct zone *zone)
809 #ifdef CONFIG_ZONE_DMA32
810 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
816 static inline int is_dma(struct zone *zone)
818 #ifdef CONFIG_ZONE_DMA
819 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
825 /* These two functions are used to setup the per zone pages min values */
827 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
828 void __user *, size_t *, loff_t *);
829 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
830 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
831 void __user *, size_t *, loff_t *);
832 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
833 void __user *, size_t *, loff_t *);
834 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
835 void __user *, size_t *, loff_t *);
836 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
837 void __user *, size_t *, loff_t *);
839 extern int numa_zonelist_order_handler(struct ctl_table *, int,
840 void __user *, size_t *, loff_t *);
841 extern char numa_zonelist_order[];
842 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
844 #ifndef CONFIG_NEED_MULTIPLE_NODES
846 extern struct pglist_data contig_page_data;
847 #define NODE_DATA(nid) (&contig_page_data)
848 #define NODE_MEM_MAP(nid) mem_map
850 #else /* CONFIG_NEED_MULTIPLE_NODES */
852 #include <asm/mmzone.h>
854 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
856 extern struct pglist_data *first_online_pgdat(void);
857 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
858 extern struct zone *next_zone(struct zone *zone);
861 * for_each_online_pgdat - helper macro to iterate over all online nodes
862 * @pgdat - pointer to a pg_data_t variable
864 #define for_each_online_pgdat(pgdat) \
865 for (pgdat = first_online_pgdat(); \
867 pgdat = next_online_pgdat(pgdat))
869 * for_each_zone - helper macro to iterate over all memory zones
870 * @zone - pointer to struct zone variable
872 * The user only needs to declare the zone variable, for_each_zone
875 #define for_each_zone(zone) \
876 for (zone = (first_online_pgdat())->node_zones; \
878 zone = next_zone(zone))
880 #define for_each_populated_zone(zone) \
881 for (zone = (first_online_pgdat())->node_zones; \
883 zone = next_zone(zone)) \
884 if (!populated_zone(zone)) \
888 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
890 return zoneref->zone;
893 static inline int zonelist_zone_idx(struct zoneref *zoneref)
895 return zoneref->zone_idx;
898 static inline int zonelist_node_idx(struct zoneref *zoneref)
901 /* zone_to_nid not available in this context */
902 return zoneref->zone->node;
905 #endif /* CONFIG_NUMA */
909 * 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
910 * @z - The cursor used as a starting point for the search
911 * @highest_zoneidx - The zone index of the highest zone to return
912 * @nodes - An optional nodemask to filter the zonelist with
913 * @zone - The first suitable zone found is returned via this parameter
915 * This function returns the next zone at or below a given zone index that is
916 * within the allowed nodemask using a cursor as the starting point for the
917 * search. The zoneref returned is a cursor that represents the current zone
918 * being examined. It should be advanced by one before calling
919 * next_zones_zonelist again.
921 struct zoneref *next_zones_zonelist(struct zoneref *z,
922 enum zone_type highest_zoneidx,
927 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
928 * @zonelist - The zonelist to search for a suitable zone
929 * @highest_zoneidx - The zone index of the highest zone to return
930 * @nodes - An optional nodemask to filter the zonelist with
931 * @zone - The first suitable zone found is returned via this parameter
933 * This function returns the first zone at or below a given zone index that is
934 * within the allowed nodemask. The zoneref returned is a cursor that can be
935 * used to iterate the zonelist with next_zones_zonelist by advancing it by
936 * one before calling.
938 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
939 enum zone_type highest_zoneidx,
943 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
948 * 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
949 * @zone - The current zone in the iterator
950 * @z - The current pointer within zonelist->zones being iterated
951 * @zlist - The zonelist being iterated
952 * @highidx - The zone index of the highest zone to return
953 * @nodemask - Nodemask allowed by the allocator
955 * This iterator iterates though all zones at or below a given zone index and
956 * within a given nodemask
958 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
959 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
961 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
964 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
965 * @zone - The current zone in the iterator
966 * @z - The current pointer within zonelist->zones being iterated
967 * @zlist - The zonelist being iterated
968 * @highidx - The zone index of the highest zone to return
970 * This iterator iterates though all zones at or below a given zone index.
972 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
973 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
975 #ifdef CONFIG_SPARSEMEM
976 #include <asm/sparsemem.h>
979 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
980 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
981 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
987 #ifdef CONFIG_FLATMEM
988 #define pfn_to_nid(pfn) (0)
991 #ifdef CONFIG_SPARSEMEM
994 * SECTION_SHIFT #bits space required to store a section #
996 * PA_SECTION_SHIFT physical address to/from section number
997 * PFN_SECTION_SHIFT pfn to/from section number
999 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
1001 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1002 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1004 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1006 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1007 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1009 #define SECTION_BLOCKFLAGS_BITS \
1010 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1012 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1013 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1016 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1017 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1019 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1020 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1024 struct mem_section {
1026 * This is, logically, a pointer to an array of struct
1027 * pages. However, it is stored with some other magic.
1028 * (see sparse.c::sparse_init_one_section())
1030 * Additionally during early boot we encode node id of
1031 * the location of the section here to guide allocation.
1032 * (see sparse.c::memory_present())
1034 * Making it a UL at least makes someone do a cast
1035 * before using it wrong.
1037 unsigned long section_mem_map;
1039 /* See declaration of similar field in struct zone */
1040 unsigned long *pageblock_flags;
1041 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1043 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1044 * section. (see memcontrol.h/page_cgroup.h about this.)
1046 struct page_cgroup *page_cgroup;
1051 #ifdef CONFIG_SPARSEMEM_EXTREME
1052 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1054 #define SECTIONS_PER_ROOT 1
1057 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1058 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1059 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1061 #ifdef CONFIG_SPARSEMEM_EXTREME
1062 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1064 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1067 static inline struct mem_section *__nr_to_section(unsigned long nr)
1069 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1071 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1073 extern int __section_nr(struct mem_section* ms);
1074 extern unsigned long usemap_size(void);
1077 * We use the lower bits of the mem_map pointer to store
1078 * a little bit of information. There should be at least
1079 * 3 bits here due to 32-bit alignment.
1081 #define SECTION_MARKED_PRESENT (1UL<<0)
1082 #define SECTION_HAS_MEM_MAP (1UL<<1)
1083 #define SECTION_MAP_LAST_BIT (1UL<<2)
1084 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1085 #define SECTION_NID_SHIFT 2
1087 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1089 unsigned long map = section->section_mem_map;
1090 map &= SECTION_MAP_MASK;
1091 return (struct page *)map;
1094 static inline int present_section(struct mem_section *section)
1096 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1099 static inline int present_section_nr(unsigned long nr)
1101 return present_section(__nr_to_section(nr));
1104 static inline int valid_section(struct mem_section *section)
1106 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1109 static inline int valid_section_nr(unsigned long nr)
1111 return valid_section(__nr_to_section(nr));
1114 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1116 return __nr_to_section(pfn_to_section_nr(pfn));
1119 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1120 static inline int pfn_valid(unsigned long pfn)
1122 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1124 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1128 static inline int pfn_present(unsigned long pfn)
1130 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1132 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1136 * These are _only_ used during initialisation, therefore they
1137 * can use __initdata ... They could have names to indicate
1141 #define pfn_to_nid(pfn) \
1143 unsigned long __pfn_to_nid_pfn = (pfn); \
1144 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1147 #define pfn_to_nid(pfn) (0)
1150 #define early_pfn_valid(pfn) pfn_valid(pfn)
1151 void sparse_init(void);
1153 #define sparse_init() do {} while (0)
1154 #define sparse_index_init(_sec, _nid) do {} while (0)
1155 #endif /* CONFIG_SPARSEMEM */
1157 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1158 bool early_pfn_in_nid(unsigned long pfn, int nid);
1160 #define early_pfn_in_nid(pfn, nid) (1)
1163 #ifndef early_pfn_valid
1164 #define early_pfn_valid(pfn) (1)
1167 void memory_present(int nid, unsigned long start, unsigned long end);
1168 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1171 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1172 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1173 * pfn_valid_within() should be used in this case; we optimise this away
1174 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1176 #ifdef CONFIG_HOLES_IN_ZONE
1177 #define pfn_valid_within(pfn) pfn_valid(pfn)
1179 #define pfn_valid_within(pfn) (1)
1182 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1184 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1185 * associated with it or not. In FLATMEM, it is expected that holes always
1186 * have valid memmap as long as there is valid PFNs either side of the hole.
1187 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1190 * However, an ARM, and maybe other embedded architectures in the future
1191 * free memmap backing holes to save memory on the assumption the memmap is
1192 * never used. The page_zone linkages are then broken even though pfn_valid()
1193 * returns true. A walker of the full memmap must then do this additional
1194 * check to ensure the memmap they are looking at is sane by making sure
1195 * the zone and PFN linkages are still valid. This is expensive, but walkers
1196 * of the full memmap are extremely rare.
1198 int memmap_valid_within(unsigned long pfn,
1199 struct page *page, struct zone *zone);
1201 static inline int memmap_valid_within(unsigned long pfn,
1202 struct page *page, struct zone *zone)
1206 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1208 #endif /* !__GENERATING_BOUNDS.H */
1209 #endif /* !__ASSEMBLY__ */
1210 #endif /* _LINUX_MMZONE_H */