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 * coelesce naturally under reasonable reclaim pressure and those which
36 #define PAGE_ALLOC_COSTLY_ORDER 3
38 #define MIGRATE_UNMOVABLE 0
39 #define MIGRATE_RECLAIMABLE 1
40 #define MIGRATE_MOVABLE 2
41 #define MIGRATE_PCPTYPES 3 /* the number of types on the pcp lists */
42 #define MIGRATE_RESERVE 3
43 #define MIGRATE_ISOLATE 4 /* can't allocate from here */
44 #define MIGRATE_TYPES 5
46 #define for_each_migratetype_order(order, type) \
47 for (order = 0; order < MAX_ORDER; order++) \
48 for (type = 0; type < MIGRATE_TYPES; type++)
50 extern int page_group_by_mobility_disabled;
52 static inline int get_pageblock_migratetype(struct page *page)
54 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
58 struct list_head free_list[MIGRATE_TYPES];
59 unsigned long nr_free;
65 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
66 * So add a wild amount of padding here to ensure that they fall into separate
67 * cachelines. There are very few zone structures in the machine, so space
68 * consumption is not a concern here.
70 #if defined(CONFIG_SMP)
73 } ____cacheline_internodealigned_in_smp;
74 #define ZONE_PADDING(name) struct zone_padding name;
76 #define ZONE_PADDING(name)
80 /* First 128 byte cacheline (assuming 64 bit words) */
83 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
84 NR_ACTIVE_ANON, /* " " " " " */
85 NR_INACTIVE_FILE, /* " " " " " */
86 NR_ACTIVE_FILE, /* " " " " " */
87 NR_UNEVICTABLE, /* " " " " " */
88 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
89 NR_ANON_PAGES, /* Mapped anonymous pages */
90 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
91 only modified from process context */
96 NR_SLAB_UNRECLAIMABLE,
97 NR_PAGETABLE, /* used for pagetables */
99 /* Second 128 byte cacheline */
100 NR_UNSTABLE_NFS, /* NFS unstable pages */
103 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
104 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
105 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
106 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
107 NR_DIRTIED, /* page dirtyings since bootup */
108 NR_WRITTEN, /* page writings since bootup */
110 NUMA_HIT, /* allocated in intended node */
111 NUMA_MISS, /* allocated in non intended node */
112 NUMA_FOREIGN, /* was intended here, hit elsewhere */
113 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
114 NUMA_LOCAL, /* allocation from local node */
115 NUMA_OTHER, /* allocation from other node */
117 NR_ANON_TRANSPARENT_HUGEPAGES,
118 NR_VM_ZONE_STAT_ITEMS };
121 * We do arithmetic on the LRU lists in various places in the code,
122 * so it is important to keep the active lists LRU_ACTIVE higher in
123 * the array than the corresponding inactive lists, and to keep
124 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
126 * This has to be kept in sync with the statistics in zone_stat_item
127 * above and the descriptions in vmstat_text in mm/vmstat.c
134 LRU_INACTIVE_ANON = LRU_BASE,
135 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
136 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
137 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
142 #define for_each_lru(l) for (l = 0; l < NR_LRU_LISTS; l++)
144 #define for_each_evictable_lru(l) for (l = 0; l <= LRU_ACTIVE_FILE; l++)
146 static inline int is_file_lru(enum lru_list l)
148 return (l == LRU_INACTIVE_FILE || l == LRU_ACTIVE_FILE);
151 static inline int is_active_lru(enum lru_list l)
153 return (l == LRU_ACTIVE_ANON || l == LRU_ACTIVE_FILE);
156 static inline int is_unevictable_lru(enum lru_list l)
158 return (l == LRU_UNEVICTABLE);
161 /* Mask used at gathering information at once (see memcontrol.c) */
162 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
163 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
164 #define LRU_ALL_EVICTABLE (LRU_ALL_FILE | LRU_ALL_ANON)
165 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
167 /* Isolate inactive pages */
168 #define ISOLATE_INACTIVE ((__force isolate_mode_t)0x1)
169 /* Isolate active pages */
170 #define ISOLATE_ACTIVE ((__force isolate_mode_t)0x2)
171 /* Isolate clean file */
172 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x4)
173 /* Isolate unmapped file */
174 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x8)
176 /* LRU Isolation modes. */
177 typedef unsigned __bitwise__ isolate_mode_t;
179 enum zone_watermarks {
186 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
187 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
188 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
190 struct per_cpu_pages {
191 int count; /* number of pages in the list */
192 int high; /* high watermark, emptying needed */
193 int batch; /* chunk size for buddy add/remove */
195 /* Lists of pages, one per migrate type stored on the pcp-lists */
196 struct list_head lists[MIGRATE_PCPTYPES];
199 struct per_cpu_pageset {
200 struct per_cpu_pages pcp;
206 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
210 #endif /* !__GENERATING_BOUNDS.H */
213 #ifdef CONFIG_ZONE_DMA
215 * ZONE_DMA is used when there are devices that are not able
216 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
217 * carve out the portion of memory that is needed for these devices.
218 * The range is arch specific.
223 * ---------------------------
224 * parisc, ia64, sparc <4G
227 * alpha Unlimited or 0-16MB.
229 * i386, x86_64 and multiple other arches
234 #ifdef CONFIG_ZONE_DMA32
236 * x86_64 needs two ZONE_DMAs because it supports devices that are
237 * only able to do DMA to the lower 16M but also 32 bit devices that
238 * can only do DMA areas below 4G.
243 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
244 * performed on pages in ZONE_NORMAL if the DMA devices support
245 * transfers to all addressable memory.
248 #ifdef CONFIG_HIGHMEM
250 * A memory area that is only addressable by the kernel through
251 * mapping portions into its own address space. This is for example
252 * used by i386 to allow the kernel to address the memory beyond
253 * 900MB. The kernel will set up special mappings (page
254 * table entries on i386) for each page that the kernel needs to
263 #ifndef __GENERATING_BOUNDS_H
266 * When a memory allocation must conform to specific limitations (such
267 * as being suitable for DMA) the caller will pass in hints to the
268 * allocator in the gfp_mask, in the zone modifier bits. These bits
269 * are used to select a priority ordered list of memory zones which
270 * match the requested limits. See gfp_zone() in include/linux/gfp.h
274 #define ZONES_SHIFT 0
275 #elif MAX_NR_ZONES <= 2
276 #define ZONES_SHIFT 1
277 #elif MAX_NR_ZONES <= 4
278 #define ZONES_SHIFT 2
280 #error ZONES_SHIFT -- too many zones configured adjust calculation
283 struct zone_reclaim_stat {
285 * The pageout code in vmscan.c keeps track of how many of the
286 * mem/swap backed and file backed pages are refeferenced.
287 * The higher the rotated/scanned ratio, the more valuable
290 * The anon LRU stats live in [0], file LRU stats in [1]
292 unsigned long recent_rotated[2];
293 unsigned long recent_scanned[2];
297 /* Fields commonly accessed by the page allocator */
299 /* zone watermarks, access with *_wmark_pages(zone) macros */
300 unsigned long watermark[NR_WMARK];
303 * When free pages are below this point, additional steps are taken
304 * when reading the number of free pages to avoid per-cpu counter
305 * drift allowing watermarks to be breached
307 unsigned long percpu_drift_mark;
310 * We don't know if the memory that we're going to allocate will be freeable
311 * or/and it will be released eventually, so to avoid totally wasting several
312 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
313 * to run OOM on the lower zones despite there's tons of freeable ram
314 * on the higher zones). This array is recalculated at runtime if the
315 * sysctl_lowmem_reserve_ratio sysctl changes.
317 unsigned long lowmem_reserve[MAX_NR_ZONES];
322 * zone reclaim becomes active if more unmapped pages exist.
324 unsigned long min_unmapped_pages;
325 unsigned long min_slab_pages;
327 struct per_cpu_pageset __percpu *pageset;
329 * free areas of different sizes
332 int all_unreclaimable; /* All pages pinned */
333 #ifdef CONFIG_MEMORY_HOTPLUG
334 /* see spanned/present_pages for more description */
335 seqlock_t span_seqlock;
337 struct free_area free_area[MAX_ORDER];
339 #ifndef CONFIG_SPARSEMEM
341 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
342 * In SPARSEMEM, this map is stored in struct mem_section
344 unsigned long *pageblock_flags;
345 #endif /* CONFIG_SPARSEMEM */
347 #ifdef CONFIG_COMPACTION
349 * On compaction failure, 1<<compact_defer_shift compactions
350 * are skipped before trying again. The number attempted since
351 * last failure is tracked with compact_considered.
353 unsigned int compact_considered;
354 unsigned int compact_defer_shift;
359 /* Fields commonly accessed by the page reclaim scanner */
362 struct list_head list;
365 struct zone_reclaim_stat reclaim_stat;
367 unsigned long pages_scanned; /* since last reclaim */
368 unsigned long flags; /* zone flags, see below */
370 /* Zone statistics */
371 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
374 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
375 * this zone's LRU. Maintained by the pageout code.
377 unsigned int inactive_ratio;
381 /* Rarely used or read-mostly fields */
384 * wait_table -- the array holding the hash table
385 * wait_table_hash_nr_entries -- the size of the hash table array
386 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
388 * The purpose of all these is to keep track of the people
389 * waiting for a page to become available and make them
390 * runnable again when possible. The trouble is that this
391 * consumes a lot of space, especially when so few things
392 * wait on pages at a given time. So instead of using
393 * per-page waitqueues, we use a waitqueue hash table.
395 * The bucket discipline is to sleep on the same queue when
396 * colliding and wake all in that wait queue when removing.
397 * When something wakes, it must check to be sure its page is
398 * truly available, a la thundering herd. The cost of a
399 * collision is great, but given the expected load of the
400 * table, they should be so rare as to be outweighed by the
401 * benefits from the saved space.
403 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
404 * primary users of these fields, and in mm/page_alloc.c
405 * free_area_init_core() performs the initialization of them.
407 wait_queue_head_t * wait_table;
408 unsigned long wait_table_hash_nr_entries;
409 unsigned long wait_table_bits;
412 * Discontig memory support fields.
414 struct pglist_data *zone_pgdat;
415 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
416 unsigned long zone_start_pfn;
419 * zone_start_pfn, spanned_pages and present_pages are all
420 * protected by span_seqlock. It is a seqlock because it has
421 * to be read outside of zone->lock, and it is done in the main
422 * allocator path. But, it is written quite infrequently.
424 * The lock is declared along with zone->lock because it is
425 * frequently read in proximity to zone->lock. It's good to
426 * give them a chance of being in the same cacheline.
428 unsigned long spanned_pages; /* total size, including holes */
429 unsigned long present_pages; /* amount of memory (excluding holes) */
432 * rarely used fields:
435 } ____cacheline_internodealigned_in_smp;
438 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
439 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
440 ZONE_CONGESTED, /* zone has many dirty pages backed by
445 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
447 set_bit(flag, &zone->flags);
450 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
452 return test_and_set_bit(flag, &zone->flags);
455 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
457 clear_bit(flag, &zone->flags);
460 static inline int zone_is_reclaim_congested(const struct zone *zone)
462 return test_bit(ZONE_CONGESTED, &zone->flags);
465 static inline int zone_is_reclaim_locked(const struct zone *zone)
467 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
470 static inline int zone_is_oom_locked(const struct zone *zone)
472 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
476 * The "priority" of VM scanning is how much of the queues we will scan in one
477 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
478 * queues ("queue_length >> 12") during an aging round.
480 #define DEF_PRIORITY 12
482 /* Maximum number of zones on a zonelist */
483 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
488 * The NUMA zonelists are doubled because we need zonelists that restrict the
489 * allocations to a single node for GFP_THISNODE.
491 * [0] : Zonelist with fallback
492 * [1] : No fallback (GFP_THISNODE)
494 #define MAX_ZONELISTS 2
498 * We cache key information from each zonelist for smaller cache
499 * footprint when scanning for free pages in get_page_from_freelist().
501 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
502 * up short of free memory since the last time (last_fullzone_zap)
503 * we zero'd fullzones.
504 * 2) The array z_to_n[] maps each zone in the zonelist to its node
505 * id, so that we can efficiently evaluate whether that node is
506 * set in the current tasks mems_allowed.
508 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
509 * indexed by a zones offset in the zonelist zones[] array.
511 * The get_page_from_freelist() routine does two scans. During the
512 * first scan, we skip zones whose corresponding bit in 'fullzones'
513 * is set or whose corresponding node in current->mems_allowed (which
514 * comes from cpusets) is not set. During the second scan, we bypass
515 * this zonelist_cache, to ensure we look methodically at each zone.
517 * Once per second, we zero out (zap) fullzones, forcing us to
518 * reconsider nodes that might have regained more free memory.
519 * The field last_full_zap is the time we last zapped fullzones.
521 * This mechanism reduces the amount of time we waste repeatedly
522 * reexaming zones for free memory when they just came up low on
523 * memory momentarilly ago.
525 * The zonelist_cache struct members logically belong in struct
526 * zonelist. However, the mempolicy zonelists constructed for
527 * MPOL_BIND are intentionally variable length (and usually much
528 * shorter). A general purpose mechanism for handling structs with
529 * multiple variable length members is more mechanism than we want
530 * here. We resort to some special case hackery instead.
532 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
533 * part because they are shorter), so we put the fixed length stuff
534 * at the front of the zonelist struct, ending in a variable length
535 * zones[], as is needed by MPOL_BIND.
537 * Then we put the optional zonelist cache on the end of the zonelist
538 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
539 * the fixed length portion at the front of the struct. This pointer
540 * both enables us to find the zonelist cache, and in the case of
541 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
542 * to know that the zonelist cache is not there.
544 * The end result is that struct zonelists come in two flavors:
545 * 1) The full, fixed length version, shown below, and
546 * 2) The custom zonelists for MPOL_BIND.
547 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
549 * Even though there may be multiple CPU cores on a node modifying
550 * fullzones or last_full_zap in the same zonelist_cache at the same
551 * time, we don't lock it. This is just hint data - if it is wrong now
552 * and then, the allocator will still function, perhaps a bit slower.
556 struct zonelist_cache {
557 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
558 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
559 unsigned long last_full_zap; /* when last zap'd (jiffies) */
562 #define MAX_ZONELISTS 1
563 struct zonelist_cache;
567 * This struct contains information about a zone in a zonelist. It is stored
568 * here to avoid dereferences into large structures and lookups of tables
571 struct zone *zone; /* Pointer to actual zone */
572 int zone_idx; /* zone_idx(zoneref->zone) */
576 * One allocation request operates on a zonelist. A zonelist
577 * is a list of zones, the first one is the 'goal' of the
578 * allocation, the other zones are fallback zones, in decreasing
581 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
582 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
584 * To speed the reading of the zonelist, the zonerefs contain the zone index
585 * of the entry being read. Helper functions to access information given
586 * a struct zoneref are
588 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
589 * zonelist_zone_idx() - Return the index of the zone for an entry
590 * zonelist_node_idx() - Return the index of the node for an entry
593 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
594 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
596 struct zonelist_cache zlcache; // optional ...
600 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
601 struct node_active_region {
602 unsigned long start_pfn;
603 unsigned long end_pfn;
606 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
608 #ifndef CONFIG_DISCONTIGMEM
609 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
610 extern struct page *mem_map;
614 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
615 * (mostly NUMA machines?) to denote a higher-level memory zone than the
618 * On NUMA machines, each NUMA node would have a pg_data_t to describe
619 * it's memory layout.
621 * Memory statistics and page replacement data structures are maintained on a
625 typedef struct pglist_data {
626 struct zone node_zones[MAX_NR_ZONES];
627 struct zonelist node_zonelists[MAX_ZONELISTS];
629 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
630 struct page *node_mem_map;
631 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
632 struct page_cgroup *node_page_cgroup;
635 #ifndef CONFIG_NO_BOOTMEM
636 struct bootmem_data *bdata;
638 #ifdef CONFIG_MEMORY_HOTPLUG
640 * Must be held any time you expect node_start_pfn, node_present_pages
641 * or node_spanned_pages stay constant. Holding this will also
642 * guarantee that any pfn_valid() stays that way.
644 * Nests above zone->lock and zone->size_seqlock.
646 spinlock_t node_size_lock;
648 unsigned long node_start_pfn;
649 unsigned long node_present_pages; /* total number of physical pages */
650 unsigned long node_spanned_pages; /* total size of physical page
651 range, including holes */
653 wait_queue_head_t kswapd_wait;
654 struct task_struct *kswapd;
655 int kswapd_max_order;
656 enum zone_type classzone_idx;
659 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
660 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
661 #ifdef CONFIG_FLAT_NODE_MEM_MAP
662 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
664 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
666 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
668 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
670 #define node_end_pfn(nid) ({\
671 pg_data_t *__pgdat = NODE_DATA(nid);\
672 __pgdat->node_start_pfn + __pgdat->node_spanned_pages;\
675 #include <linux/memory_hotplug.h>
677 extern struct mutex zonelists_mutex;
678 void build_all_zonelists(void *data);
679 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
680 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
681 int classzone_idx, int alloc_flags);
682 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
683 int classzone_idx, int alloc_flags);
684 enum memmap_context {
688 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
690 enum memmap_context context);
692 #ifdef CONFIG_HAVE_MEMORY_PRESENT
693 void memory_present(int nid, unsigned long start, unsigned long end);
695 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
698 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
699 int local_memory_node(int node_id);
701 static inline int local_memory_node(int node_id) { return node_id; };
704 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
705 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
709 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
711 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
713 static inline int populated_zone(struct zone *zone)
715 return (!!zone->present_pages);
718 extern int movable_zone;
720 static inline int zone_movable_is_highmem(void)
722 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
723 return movable_zone == ZONE_HIGHMEM;
729 static inline int is_highmem_idx(enum zone_type idx)
731 #ifdef CONFIG_HIGHMEM
732 return (idx == ZONE_HIGHMEM ||
733 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
739 static inline int is_normal_idx(enum zone_type idx)
741 return (idx == ZONE_NORMAL);
745 * is_highmem - helper function to quickly check if a struct zone is a
746 * highmem zone or not. This is an attempt to keep references
747 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
748 * @zone - pointer to struct zone variable
750 static inline int is_highmem(struct zone *zone)
752 #ifdef CONFIG_HIGHMEM
753 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
754 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
755 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
756 zone_movable_is_highmem());
762 static inline int is_normal(struct zone *zone)
764 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
767 static inline int is_dma32(struct zone *zone)
769 #ifdef CONFIG_ZONE_DMA32
770 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
776 static inline int is_dma(struct zone *zone)
778 #ifdef CONFIG_ZONE_DMA
779 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
785 /* These two functions are used to setup the per zone pages min values */
787 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
788 void __user *, size_t *, loff_t *);
789 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
790 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
791 void __user *, size_t *, loff_t *);
792 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
793 void __user *, size_t *, loff_t *);
794 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
795 void __user *, size_t *, loff_t *);
796 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
797 void __user *, size_t *, loff_t *);
799 extern int numa_zonelist_order_handler(struct ctl_table *, int,
800 void __user *, size_t *, loff_t *);
801 extern char numa_zonelist_order[];
802 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
804 #ifndef CONFIG_NEED_MULTIPLE_NODES
806 extern struct pglist_data contig_page_data;
807 #define NODE_DATA(nid) (&contig_page_data)
808 #define NODE_MEM_MAP(nid) mem_map
810 #else /* CONFIG_NEED_MULTIPLE_NODES */
812 #include <asm/mmzone.h>
814 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
816 extern struct pglist_data *first_online_pgdat(void);
817 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
818 extern struct zone *next_zone(struct zone *zone);
821 * for_each_online_pgdat - helper macro to iterate over all online nodes
822 * @pgdat - pointer to a pg_data_t variable
824 #define for_each_online_pgdat(pgdat) \
825 for (pgdat = first_online_pgdat(); \
827 pgdat = next_online_pgdat(pgdat))
829 * for_each_zone - helper macro to iterate over all memory zones
830 * @zone - pointer to struct zone variable
832 * The user only needs to declare the zone variable, for_each_zone
835 #define for_each_zone(zone) \
836 for (zone = (first_online_pgdat())->node_zones; \
838 zone = next_zone(zone))
840 #define for_each_populated_zone(zone) \
841 for (zone = (first_online_pgdat())->node_zones; \
843 zone = next_zone(zone)) \
844 if (!populated_zone(zone)) \
848 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
850 return zoneref->zone;
853 static inline int zonelist_zone_idx(struct zoneref *zoneref)
855 return zoneref->zone_idx;
858 static inline int zonelist_node_idx(struct zoneref *zoneref)
861 /* zone_to_nid not available in this context */
862 return zoneref->zone->node;
865 #endif /* CONFIG_NUMA */
869 * 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
870 * @z - The cursor used as a starting point for the search
871 * @highest_zoneidx - The zone index of the highest zone to return
872 * @nodes - An optional nodemask to filter the zonelist with
873 * @zone - The first suitable zone found is returned via this parameter
875 * This function returns the next zone at or below a given zone index that is
876 * within the allowed nodemask using a cursor as the starting point for the
877 * search. The zoneref returned is a cursor that represents the current zone
878 * being examined. It should be advanced by one before calling
879 * next_zones_zonelist again.
881 struct zoneref *next_zones_zonelist(struct zoneref *z,
882 enum zone_type highest_zoneidx,
887 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
888 * @zonelist - The zonelist to search for a suitable zone
889 * @highest_zoneidx - The zone index of the highest zone to return
890 * @nodes - An optional nodemask to filter the zonelist with
891 * @zone - The first suitable zone found is returned via this parameter
893 * This function returns the first zone at or below a given zone index that is
894 * within the allowed nodemask. The zoneref returned is a cursor that can be
895 * used to iterate the zonelist with next_zones_zonelist by advancing it by
896 * one before calling.
898 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
899 enum zone_type highest_zoneidx,
903 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
908 * 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
909 * @zone - The current zone in the iterator
910 * @z - The current pointer within zonelist->zones being iterated
911 * @zlist - The zonelist being iterated
912 * @highidx - The zone index of the highest zone to return
913 * @nodemask - Nodemask allowed by the allocator
915 * This iterator iterates though all zones at or below a given zone index and
916 * within a given nodemask
918 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
919 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
921 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
924 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
925 * @zone - The current zone in the iterator
926 * @z - The current pointer within zonelist->zones being iterated
927 * @zlist - The zonelist being iterated
928 * @highidx - The zone index of the highest zone to return
930 * This iterator iterates though all zones at or below a given zone index.
932 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
933 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
935 #ifdef CONFIG_SPARSEMEM
936 #include <asm/sparsemem.h>
939 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
940 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
941 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
947 #ifdef CONFIG_FLATMEM
948 #define pfn_to_nid(pfn) (0)
951 #ifdef CONFIG_SPARSEMEM
954 * SECTION_SHIFT #bits space required to store a section #
956 * PA_SECTION_SHIFT physical address to/from section number
957 * PFN_SECTION_SHIFT pfn to/from section number
959 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
961 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
962 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
964 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
966 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
967 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
969 #define SECTION_BLOCKFLAGS_BITS \
970 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
972 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
973 #error Allocator MAX_ORDER exceeds SECTION_SIZE
976 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
977 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
979 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
980 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
986 * This is, logically, a pointer to an array of struct
987 * pages. However, it is stored with some other magic.
988 * (see sparse.c::sparse_init_one_section())
990 * Additionally during early boot we encode node id of
991 * the location of the section here to guide allocation.
992 * (see sparse.c::memory_present())
994 * Making it a UL at least makes someone do a cast
995 * before using it wrong.
997 unsigned long section_mem_map;
999 /* See declaration of similar field in struct zone */
1000 unsigned long *pageblock_flags;
1001 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1003 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1004 * section. (see memcontrol.h/page_cgroup.h about this.)
1006 struct page_cgroup *page_cgroup;
1011 #ifdef CONFIG_SPARSEMEM_EXTREME
1012 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1014 #define SECTIONS_PER_ROOT 1
1017 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1018 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1019 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1021 #ifdef CONFIG_SPARSEMEM_EXTREME
1022 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1024 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1027 static inline struct mem_section *__nr_to_section(unsigned long nr)
1029 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1031 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1033 extern int __section_nr(struct mem_section* ms);
1034 extern unsigned long usemap_size(void);
1037 * We use the lower bits of the mem_map pointer to store
1038 * a little bit of information. There should be at least
1039 * 3 bits here due to 32-bit alignment.
1041 #define SECTION_MARKED_PRESENT (1UL<<0)
1042 #define SECTION_HAS_MEM_MAP (1UL<<1)
1043 #define SECTION_MAP_LAST_BIT (1UL<<2)
1044 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1045 #define SECTION_NID_SHIFT 2
1047 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1049 unsigned long map = section->section_mem_map;
1050 map &= SECTION_MAP_MASK;
1051 return (struct page *)map;
1054 static inline int present_section(struct mem_section *section)
1056 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1059 static inline int present_section_nr(unsigned long nr)
1061 return present_section(__nr_to_section(nr));
1064 static inline int valid_section(struct mem_section *section)
1066 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1069 static inline int valid_section_nr(unsigned long nr)
1071 return valid_section(__nr_to_section(nr));
1074 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1076 return __nr_to_section(pfn_to_section_nr(pfn));
1079 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1080 static inline int pfn_valid(unsigned long pfn)
1082 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1084 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1088 static inline int pfn_present(unsigned long pfn)
1090 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1092 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1096 * These are _only_ used during initialisation, therefore they
1097 * can use __initdata ... They could have names to indicate
1101 #define pfn_to_nid(pfn) \
1103 unsigned long __pfn_to_nid_pfn = (pfn); \
1104 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1107 #define pfn_to_nid(pfn) (0)
1110 #define early_pfn_valid(pfn) pfn_valid(pfn)
1111 void sparse_init(void);
1113 #define sparse_init() do {} while (0)
1114 #define sparse_index_init(_sec, _nid) do {} while (0)
1115 #endif /* CONFIG_SPARSEMEM */
1117 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1118 bool early_pfn_in_nid(unsigned long pfn, int nid);
1120 #define early_pfn_in_nid(pfn, nid) (1)
1123 #ifndef early_pfn_valid
1124 #define early_pfn_valid(pfn) (1)
1127 void memory_present(int nid, unsigned long start, unsigned long end);
1128 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1131 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1132 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1133 * pfn_valid_within() should be used in this case; we optimise this away
1134 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1136 #ifdef CONFIG_HOLES_IN_ZONE
1137 #define pfn_valid_within(pfn) pfn_valid(pfn)
1139 #define pfn_valid_within(pfn) (1)
1142 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1144 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1145 * associated with it or not. In FLATMEM, it is expected that holes always
1146 * have valid memmap as long as there is valid PFNs either side of the hole.
1147 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1150 * However, an ARM, and maybe other embedded architectures in the future
1151 * free memmap backing holes to save memory on the assumption the memmap is
1152 * never used. The page_zone linkages are then broken even though pfn_valid()
1153 * returns true. A walker of the full memmap must then do this additional
1154 * check to ensure the memmap they are looking at is sane by making sure
1155 * the zone and PFN linkages are still valid. This is expensive, but walkers
1156 * of the full memmap are extremely rare.
1158 int memmap_valid_within(unsigned long pfn,
1159 struct page *page, struct zone *zone);
1161 static inline int memmap_valid_within(unsigned long pfn,
1162 struct page *page, struct zone *zone)
1166 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1168 #endif /* !__GENERATING_BOUNDS.H */
1169 #endif /* !__ASSEMBLY__ */
1170 #endif /* _LINUX_MMZONE_H */