1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
5 #ifndef __GENERATING_BOUNDS_H
7 #include <linux/spinlock.h>
8 #include <linux/list.h>
9 #include <linux/wait.h>
10 #include <linux/bitops.h>
11 #include <linux/cache.h>
12 #include <linux/threads.h>
13 #include <linux/numa.h>
14 #include <linux/init.h>
15 #include <linux/seqlock.h>
16 #include <linux/nodemask.h>
17 #include <linux/pageblock-flags.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/atomic.h>
22 /* Free memory management - zoned buddy allocator. */
23 #ifndef CONFIG_FORCE_MAX_ZONEORDER
26 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
28 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
31 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
32 * costly to service. That is between allocation orders which should
33 * coalesce naturally under reasonable reclaim pressure and those which
36 #define PAGE_ALLOC_COSTLY_ORDER 3
42 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
43 MIGRATE_RESERVE = MIGRATE_PCPTYPES,
46 * MIGRATE_CMA migration type is designed to mimic the way
47 * ZONE_MOVABLE works. Only movable pages can be allocated
48 * from MIGRATE_CMA pageblocks and page allocator never
49 * implicitly change migration type of MIGRATE_CMA pageblock.
51 * The way to use it is to change migratetype of a range of
52 * pageblocks to MIGRATE_CMA which can be done by
53 * __free_pageblock_cma() function. What is important though
54 * is that a range of pageblocks must be aligned to
55 * MAX_ORDER_NR_PAGES should biggest page be bigger then
60 #ifdef CONFIG_MEMORY_ISOLATION
61 MIGRATE_ISOLATE, /* can't allocate from here */
67 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
69 # define is_migrate_cma(migratetype) false
72 #define for_each_migratetype_order(order, type) \
73 for (order = 0; order < MAX_ORDER; order++) \
74 for (type = 0; type < MIGRATE_TYPES; type++)
76 extern int page_group_by_mobility_disabled;
78 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
79 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
81 #define get_pageblock_migratetype(page) \
82 get_pfnblock_flags_mask(page, page_to_pfn(page), \
83 PB_migrate_end, MIGRATETYPE_MASK)
85 static inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
87 BUILD_BUG_ON(PB_migrate_end - PB_migrate != 2);
88 return get_pfnblock_flags_mask(page, pfn, PB_migrate_end,
93 struct list_head free_list[MIGRATE_TYPES];
94 unsigned long nr_free;
100 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
101 * So add a wild amount of padding here to ensure that they fall into separate
102 * cachelines. There are very few zone structures in the machine, so space
103 * consumption is not a concern here.
105 #if defined(CONFIG_SMP)
106 struct zone_padding {
108 } ____cacheline_internodealigned_in_smp;
109 #define ZONE_PADDING(name) struct zone_padding name;
111 #define ZONE_PADDING(name)
114 enum zone_stat_item {
115 /* First 128 byte cacheline (assuming 64 bit words) */
119 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
120 NR_ACTIVE_ANON, /* " " " " " */
121 NR_INACTIVE_FILE, /* " " " " " */
122 NR_ACTIVE_FILE, /* " " " " " */
123 NR_UNEVICTABLE, /* " " " " " */
124 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
125 NR_ANON_PAGES, /* Mapped anonymous pages */
126 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
127 only modified from process context */
132 NR_SLAB_UNRECLAIMABLE,
133 NR_PAGETABLE, /* used for pagetables */
135 /* Second 128 byte cacheline */
136 NR_UNSTABLE_NFS, /* NFS unstable pages */
139 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
140 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
141 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
142 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
143 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
144 NR_DIRTIED, /* page dirtyings since bootup */
145 NR_WRITTEN, /* page writings since bootup */
147 NUMA_HIT, /* allocated in intended node */
148 NUMA_MISS, /* allocated in non intended node */
149 NUMA_FOREIGN, /* was intended here, hit elsewhere */
150 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
151 NUMA_LOCAL, /* allocation from local node */
152 NUMA_OTHER, /* allocation from other node */
156 WORKINGSET_NODERECLAIM,
157 NR_ANON_TRANSPARENT_HUGEPAGES,
159 NR_VM_ZONE_STAT_ITEMS };
162 * We do arithmetic on the LRU lists in various places in the code,
163 * so it is important to keep the active lists LRU_ACTIVE higher in
164 * the array than the corresponding inactive lists, and to keep
165 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
167 * This has to be kept in sync with the statistics in zone_stat_item
168 * above and the descriptions in vmstat_text in mm/vmstat.c
175 LRU_INACTIVE_ANON = LRU_BASE,
176 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
177 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
178 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
183 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
185 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
187 static inline int is_file_lru(enum lru_list lru)
189 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
192 static inline int is_active_lru(enum lru_list lru)
194 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
197 static inline int is_unevictable_lru(enum lru_list lru)
199 return (lru == LRU_UNEVICTABLE);
202 struct zone_reclaim_stat {
204 * The pageout code in vmscan.c keeps track of how many of the
205 * mem/swap backed and file backed pages are referenced.
206 * The higher the rotated/scanned ratio, the more valuable
209 * The anon LRU stats live in [0], file LRU stats in [1]
211 unsigned long recent_rotated[2];
212 unsigned long recent_scanned[2];
216 struct list_head lists[NR_LRU_LISTS];
217 struct zone_reclaim_stat reclaim_stat;
223 /* Mask used at gathering information at once (see memcontrol.c) */
224 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
225 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
226 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
228 /* Isolate clean file */
229 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
230 /* Isolate unmapped file */
231 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
232 /* Isolate for asynchronous migration */
233 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
234 /* Isolate unevictable pages */
235 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
237 /* LRU Isolation modes. */
238 typedef unsigned __bitwise__ isolate_mode_t;
240 enum zone_watermarks {
247 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
248 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
249 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
251 struct per_cpu_pages {
252 int count; /* number of pages in the list */
253 int high; /* high watermark, emptying needed */
254 int batch; /* chunk size for buddy add/remove */
256 /* Lists of pages, one per migrate type stored on the pcp-lists */
257 struct list_head lists[MIGRATE_PCPTYPES];
260 struct per_cpu_pageset {
261 struct per_cpu_pages pcp;
267 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
271 #endif /* !__GENERATING_BOUNDS.H */
274 #ifdef CONFIG_ZONE_DMA
276 * ZONE_DMA is used when there are devices that are not able
277 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
278 * carve out the portion of memory that is needed for these devices.
279 * The range is arch specific.
284 * ---------------------------
285 * parisc, ia64, sparc <4G
288 * alpha Unlimited or 0-16MB.
290 * i386, x86_64 and multiple other arches
295 #ifdef CONFIG_ZONE_DMA32
297 * x86_64 needs two ZONE_DMAs because it supports devices that are
298 * only able to do DMA to the lower 16M but also 32 bit devices that
299 * can only do DMA areas below 4G.
304 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
305 * performed on pages in ZONE_NORMAL if the DMA devices support
306 * transfers to all addressable memory.
309 #ifdef CONFIG_HIGHMEM
311 * A memory area that is only addressable by the kernel through
312 * mapping portions into its own address space. This is for example
313 * used by i386 to allow the kernel to address the memory beyond
314 * 900MB. The kernel will set up special mappings (page
315 * table entries on i386) for each page that the kernel needs to
324 #ifndef __GENERATING_BOUNDS_H
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 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
369 /* Set to true when the PG_migrate_skip bits should be cleared */
370 bool compact_blockskip_flush;
372 /* pfn where compaction free scanner should start */
373 unsigned long compact_cached_free_pfn;
374 /* pfn where async and sync compaction migration scanner should start */
375 unsigned long compact_cached_migrate_pfn[2];
377 #ifdef CONFIG_MEMORY_HOTPLUG
378 /* see spanned/present_pages for more description */
379 seqlock_t span_seqlock;
381 struct free_area free_area[MAX_ORDER];
383 #ifndef CONFIG_SPARSEMEM
385 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
386 * In SPARSEMEM, this map is stored in struct mem_section
388 unsigned long *pageblock_flags;
389 #endif /* CONFIG_SPARSEMEM */
391 #ifdef CONFIG_COMPACTION
393 * On compaction failure, 1<<compact_defer_shift compactions
394 * are skipped before trying again. The number attempted since
395 * last failure is tracked with compact_considered.
397 unsigned int compact_considered;
398 unsigned int compact_defer_shift;
399 int compact_order_failed;
404 /* Fields commonly accessed by the page reclaim scanner */
406 struct lruvec lruvec;
408 /* Evictions & activations on the inactive file list */
409 atomic_long_t inactive_age;
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(pages 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 at runtime should be protected by
494 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
495 * present_pages should get_online_mems() to get a stable value.
497 * Read access to managed_pages should be safe because it's unsigned
498 * long. Write access to zone->managed_pages and totalram_pages are
499 * protected by managed_page_count_lock at runtime. Idealy only
500 * adjust_managed_page_count() should be used instead of directly
501 * touching zone->managed_pages and totalram_pages.
503 unsigned long spanned_pages;
504 unsigned long present_pages;
505 unsigned long managed_pages;
508 * Number of MIGRATE_RESEVE page block. To maintain for just
509 * optimization. Protected by zone->lock.
511 int nr_migrate_reserve_block;
514 * rarely used fields:
517 } ____cacheline_internodealigned_in_smp;
520 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
521 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
522 ZONE_CONGESTED, /* zone has many dirty pages backed by
525 ZONE_TAIL_LRU_DIRTY, /* reclaim scanning has recently found
526 * many dirty file pages at the tail
529 ZONE_WRITEBACK, /* reclaim scanning has recently found
530 * many pages under writeback
534 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
536 set_bit(flag, &zone->flags);
539 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
541 return test_and_set_bit(flag, &zone->flags);
544 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
546 clear_bit(flag, &zone->flags);
549 static inline int zone_is_reclaim_congested(const struct zone *zone)
551 return test_bit(ZONE_CONGESTED, &zone->flags);
554 static inline int zone_is_reclaim_dirty(const struct zone *zone)
556 return test_bit(ZONE_TAIL_LRU_DIRTY, &zone->flags);
559 static inline int zone_is_reclaim_writeback(const struct zone *zone)
561 return test_bit(ZONE_WRITEBACK, &zone->flags);
564 static inline int zone_is_reclaim_locked(const struct zone *zone)
566 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
569 static inline int zone_is_oom_locked(const struct zone *zone)
571 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
574 static inline unsigned long zone_end_pfn(const struct zone *zone)
576 return zone->zone_start_pfn + zone->spanned_pages;
579 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
581 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
584 static inline bool zone_is_initialized(struct zone *zone)
586 return !!zone->wait_table;
589 static inline bool zone_is_empty(struct zone *zone)
591 return zone->spanned_pages == 0;
595 * The "priority" of VM scanning is how much of the queues we will scan in one
596 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
597 * queues ("queue_length >> 12") during an aging round.
599 #define DEF_PRIORITY 12
601 /* Maximum number of zones on a zonelist */
602 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
607 * The NUMA zonelists are doubled because we need zonelists that restrict the
608 * allocations to a single node for __GFP_THISNODE.
610 * [0] : Zonelist with fallback
611 * [1] : No fallback (__GFP_THISNODE)
613 #define MAX_ZONELISTS 2
617 * We cache key information from each zonelist for smaller cache
618 * footprint when scanning for free pages in get_page_from_freelist().
620 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
621 * up short of free memory since the last time (last_fullzone_zap)
622 * we zero'd fullzones.
623 * 2) The array z_to_n[] maps each zone in the zonelist to its node
624 * id, so that we can efficiently evaluate whether that node is
625 * set in the current tasks mems_allowed.
627 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
628 * indexed by a zones offset in the zonelist zones[] array.
630 * The get_page_from_freelist() routine does two scans. During the
631 * first scan, we skip zones whose corresponding bit in 'fullzones'
632 * is set or whose corresponding node in current->mems_allowed (which
633 * comes from cpusets) is not set. During the second scan, we bypass
634 * this zonelist_cache, to ensure we look methodically at each zone.
636 * Once per second, we zero out (zap) fullzones, forcing us to
637 * reconsider nodes that might have regained more free memory.
638 * The field last_full_zap is the time we last zapped fullzones.
640 * This mechanism reduces the amount of time we waste repeatedly
641 * reexaming zones for free memory when they just came up low on
642 * memory momentarilly ago.
644 * The zonelist_cache struct members logically belong in struct
645 * zonelist. However, the mempolicy zonelists constructed for
646 * MPOL_BIND are intentionally variable length (and usually much
647 * shorter). A general purpose mechanism for handling structs with
648 * multiple variable length members is more mechanism than we want
649 * here. We resort to some special case hackery instead.
651 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
652 * part because they are shorter), so we put the fixed length stuff
653 * at the front of the zonelist struct, ending in a variable length
654 * zones[], as is needed by MPOL_BIND.
656 * Then we put the optional zonelist cache on the end of the zonelist
657 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
658 * the fixed length portion at the front of the struct. This pointer
659 * both enables us to find the zonelist cache, and in the case of
660 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
661 * to know that the zonelist cache is not there.
663 * The end result is that struct zonelists come in two flavors:
664 * 1) The full, fixed length version, shown below, and
665 * 2) The custom zonelists for MPOL_BIND.
666 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
668 * Even though there may be multiple CPU cores on a node modifying
669 * fullzones or last_full_zap in the same zonelist_cache at the same
670 * time, we don't lock it. This is just hint data - if it is wrong now
671 * and then, the allocator will still function, perhaps a bit slower.
675 struct zonelist_cache {
676 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
677 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
678 unsigned long last_full_zap; /* when last zap'd (jiffies) */
681 #define MAX_ZONELISTS 1
682 struct zonelist_cache;
686 * This struct contains information about a zone in a zonelist. It is stored
687 * here to avoid dereferences into large structures and lookups of tables
690 struct zone *zone; /* Pointer to actual zone */
691 int zone_idx; /* zone_idx(zoneref->zone) */
695 * One allocation request operates on a zonelist. A zonelist
696 * is a list of zones, the first one is the 'goal' of the
697 * allocation, the other zones are fallback zones, in decreasing
700 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
701 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
703 * To speed the reading of the zonelist, the zonerefs contain the zone index
704 * of the entry being read. Helper functions to access information given
705 * a struct zoneref are
707 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
708 * zonelist_zone_idx() - Return the index of the zone for an entry
709 * zonelist_node_idx() - Return the index of the node for an entry
712 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
713 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
715 struct zonelist_cache zlcache; // optional ...
719 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
720 struct node_active_region {
721 unsigned long start_pfn;
722 unsigned long end_pfn;
725 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
727 #ifndef CONFIG_DISCONTIGMEM
728 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
729 extern struct page *mem_map;
733 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
734 * (mostly NUMA machines?) to denote a higher-level memory zone than the
737 * On NUMA machines, each NUMA node would have a pg_data_t to describe
738 * it's memory layout.
740 * Memory statistics and page replacement data structures are maintained on a
744 typedef struct pglist_data {
745 struct zone node_zones[MAX_NR_ZONES];
746 struct zonelist node_zonelists[MAX_ZONELISTS];
748 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
749 struct page *node_mem_map;
751 struct page_cgroup *node_page_cgroup;
754 #ifndef CONFIG_NO_BOOTMEM
755 struct bootmem_data *bdata;
757 #ifdef CONFIG_MEMORY_HOTPLUG
759 * Must be held any time you expect node_start_pfn, node_present_pages
760 * or node_spanned_pages stay constant. Holding this will also
761 * guarantee that any pfn_valid() stays that way.
763 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
764 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
766 * Nests above zone->lock and zone->span_seqlock
768 spinlock_t node_size_lock;
770 unsigned long node_start_pfn;
771 unsigned long node_present_pages; /* total number of physical pages */
772 unsigned long node_spanned_pages; /* total size of physical page
773 range, including holes */
775 wait_queue_head_t kswapd_wait;
776 wait_queue_head_t pfmemalloc_wait;
777 struct task_struct *kswapd; /* Protected by
778 mem_hotplug_begin/end() */
779 int kswapd_max_order;
780 enum zone_type classzone_idx;
781 #ifdef CONFIG_NUMA_BALANCING
782 /* Lock serializing the migrate rate limiting window */
783 spinlock_t numabalancing_migrate_lock;
785 /* Rate limiting time interval */
786 unsigned long numabalancing_migrate_next_window;
788 /* Number of pages migrated during the rate limiting time interval */
789 unsigned long numabalancing_migrate_nr_pages;
793 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
794 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
795 #ifdef CONFIG_FLAT_NODE_MEM_MAP
796 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
798 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
800 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
802 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
803 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
805 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
807 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
810 static inline bool pgdat_is_empty(pg_data_t *pgdat)
812 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
815 #include <linux/memory_hotplug.h>
817 extern struct mutex zonelists_mutex;
818 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
819 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
820 bool zone_watermark_ok(struct zone *z, unsigned int order,
821 unsigned long mark, int classzone_idx, int alloc_flags);
822 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
823 unsigned long mark, int classzone_idx, int alloc_flags);
824 enum memmap_context {
828 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
830 enum memmap_context context);
832 extern void lruvec_init(struct lruvec *lruvec);
834 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
839 return container_of(lruvec, struct zone, lruvec);
843 #ifdef CONFIG_HAVE_MEMORY_PRESENT
844 void memory_present(int nid, unsigned long start, unsigned long end);
846 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
849 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
850 int local_memory_node(int node_id);
852 static inline int local_memory_node(int node_id) { return node_id; };
855 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
856 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
860 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
862 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
864 static inline int populated_zone(struct zone *zone)
866 return (!!zone->present_pages);
869 extern int movable_zone;
871 static inline int zone_movable_is_highmem(void)
873 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
874 return movable_zone == ZONE_HIGHMEM;
880 static inline int is_highmem_idx(enum zone_type idx)
882 #ifdef CONFIG_HIGHMEM
883 return (idx == ZONE_HIGHMEM ||
884 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
891 * is_highmem - helper function to quickly check if a struct zone is a
892 * highmem zone or not. This is an attempt to keep references
893 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
894 * @zone - pointer to struct zone variable
896 static inline int is_highmem(struct zone *zone)
898 #ifdef CONFIG_HIGHMEM
899 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
900 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
901 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
902 zone_movable_is_highmem());
908 /* These two functions are used to setup the per zone pages min values */
910 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
911 void __user *, size_t *, loff_t *);
912 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
913 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
914 void __user *, size_t *, loff_t *);
915 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
916 void __user *, size_t *, loff_t *);
917 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
918 void __user *, size_t *, loff_t *);
919 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
920 void __user *, size_t *, loff_t *);
922 extern int numa_zonelist_order_handler(struct ctl_table *, int,
923 void __user *, size_t *, loff_t *);
924 extern char numa_zonelist_order[];
925 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
927 #ifndef CONFIG_NEED_MULTIPLE_NODES
929 extern struct pglist_data contig_page_data;
930 #define NODE_DATA(nid) (&contig_page_data)
931 #define NODE_MEM_MAP(nid) mem_map
933 #else /* CONFIG_NEED_MULTIPLE_NODES */
935 #include <asm/mmzone.h>
937 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
939 extern struct pglist_data *first_online_pgdat(void);
940 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
941 extern struct zone *next_zone(struct zone *zone);
944 * for_each_online_pgdat - helper macro to iterate over all online nodes
945 * @pgdat - pointer to a pg_data_t variable
947 #define for_each_online_pgdat(pgdat) \
948 for (pgdat = first_online_pgdat(); \
950 pgdat = next_online_pgdat(pgdat))
952 * for_each_zone - helper macro to iterate over all memory zones
953 * @zone - pointer to struct zone variable
955 * The user only needs to declare the zone variable, for_each_zone
958 #define for_each_zone(zone) \
959 for (zone = (first_online_pgdat())->node_zones; \
961 zone = next_zone(zone))
963 #define for_each_populated_zone(zone) \
964 for (zone = (first_online_pgdat())->node_zones; \
966 zone = next_zone(zone)) \
967 if (!populated_zone(zone)) \
971 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
973 return zoneref->zone;
976 static inline int zonelist_zone_idx(struct zoneref *zoneref)
978 return zoneref->zone_idx;
981 static inline int zonelist_node_idx(struct zoneref *zoneref)
984 /* zone_to_nid not available in this context */
985 return zoneref->zone->node;
988 #endif /* CONFIG_NUMA */
992 * 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
993 * @z - The cursor used as a starting point for the search
994 * @highest_zoneidx - The zone index of the highest zone to return
995 * @nodes - An optional nodemask to filter the zonelist with
996 * @zone - The first suitable zone found is returned via this parameter
998 * This function returns the next zone at or below a given zone index that is
999 * within the allowed nodemask using a cursor as the starting point for the
1000 * search. The zoneref returned is a cursor that represents the current zone
1001 * being examined. It should be advanced by one before calling
1002 * next_zones_zonelist again.
1004 struct zoneref *next_zones_zonelist(struct zoneref *z,
1005 enum zone_type highest_zoneidx,
1007 struct zone **zone);
1010 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1011 * @zonelist - The zonelist to search for a suitable zone
1012 * @highest_zoneidx - The zone index of the highest zone to return
1013 * @nodes - An optional nodemask to filter the zonelist with
1014 * @zone - The first suitable zone found is returned via this parameter
1016 * This function returns the first zone at or below a given zone index that is
1017 * within the allowed nodemask. The zoneref returned is a cursor that can be
1018 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1019 * one before calling.
1021 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1022 enum zone_type highest_zoneidx,
1026 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1031 * 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
1032 * @zone - The current zone in the iterator
1033 * @z - The current pointer within zonelist->zones being iterated
1034 * @zlist - The zonelist being iterated
1035 * @highidx - The zone index of the highest zone to return
1036 * @nodemask - Nodemask allowed by the allocator
1038 * This iterator iterates though all zones at or below a given zone index and
1039 * within a given nodemask
1041 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1042 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1044 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1047 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1048 * @zone - The current zone in the iterator
1049 * @z - The current pointer within zonelist->zones being iterated
1050 * @zlist - The zonelist being iterated
1051 * @highidx - The zone index of the highest zone to return
1053 * This iterator iterates though all zones at or below a given zone index.
1055 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1056 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1058 #ifdef CONFIG_SPARSEMEM
1059 #include <asm/sparsemem.h>
1062 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1063 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1064 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1070 #ifdef CONFIG_FLATMEM
1071 #define pfn_to_nid(pfn) (0)
1074 #ifdef CONFIG_SPARSEMEM
1077 * SECTION_SHIFT #bits space required to store a section #
1079 * PA_SECTION_SHIFT physical address to/from section number
1080 * PFN_SECTION_SHIFT pfn to/from section number
1082 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1083 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1085 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1087 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1088 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1090 #define SECTION_BLOCKFLAGS_BITS \
1091 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1093 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1094 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1097 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1098 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1100 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1101 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1105 struct mem_section {
1107 * This is, logically, a pointer to an array of struct
1108 * pages. However, it is stored with some other magic.
1109 * (see sparse.c::sparse_init_one_section())
1111 * Additionally during early boot we encode node id of
1112 * the location of the section here to guide allocation.
1113 * (see sparse.c::memory_present())
1115 * Making it a UL at least makes someone do a cast
1116 * before using it wrong.
1118 unsigned long section_mem_map;
1120 /* See declaration of similar field in struct zone */
1121 unsigned long *pageblock_flags;
1124 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1125 * section. (see memcontrol.h/page_cgroup.h about this.)
1127 struct page_cgroup *page_cgroup;
1131 * WARNING: mem_section must be a power-of-2 in size for the
1132 * calculation and use of SECTION_ROOT_MASK to make sense.
1136 #ifdef CONFIG_SPARSEMEM_EXTREME
1137 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1139 #define SECTIONS_PER_ROOT 1
1142 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1143 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1144 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1146 #ifdef CONFIG_SPARSEMEM_EXTREME
1147 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1149 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1152 static inline struct mem_section *__nr_to_section(unsigned long nr)
1154 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1156 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1158 extern int __section_nr(struct mem_section* ms);
1159 extern unsigned long usemap_size(void);
1162 * We use the lower bits of the mem_map pointer to store
1163 * a little bit of information. There should be at least
1164 * 3 bits here due to 32-bit alignment.
1166 #define SECTION_MARKED_PRESENT (1UL<<0)
1167 #define SECTION_HAS_MEM_MAP (1UL<<1)
1168 #define SECTION_MAP_LAST_BIT (1UL<<2)
1169 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1170 #define SECTION_NID_SHIFT 2
1172 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1174 unsigned long map = section->section_mem_map;
1175 map &= SECTION_MAP_MASK;
1176 return (struct page *)map;
1179 static inline int present_section(struct mem_section *section)
1181 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1184 static inline int present_section_nr(unsigned long nr)
1186 return present_section(__nr_to_section(nr));
1189 static inline int valid_section(struct mem_section *section)
1191 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1194 static inline int valid_section_nr(unsigned long nr)
1196 return valid_section(__nr_to_section(nr));
1199 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1201 return __nr_to_section(pfn_to_section_nr(pfn));
1204 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1205 static inline int pfn_valid(unsigned long pfn)
1207 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1209 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1213 static inline int pfn_present(unsigned long pfn)
1215 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1217 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1221 * These are _only_ used during initialisation, therefore they
1222 * can use __initdata ... They could have names to indicate
1226 #define pfn_to_nid(pfn) \
1228 unsigned long __pfn_to_nid_pfn = (pfn); \
1229 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1232 #define pfn_to_nid(pfn) (0)
1235 #define early_pfn_valid(pfn) pfn_valid(pfn)
1236 void sparse_init(void);
1238 #define sparse_init() do {} while (0)
1239 #define sparse_index_init(_sec, _nid) do {} while (0)
1240 #endif /* CONFIG_SPARSEMEM */
1242 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1243 bool early_pfn_in_nid(unsigned long pfn, int nid);
1245 #define early_pfn_in_nid(pfn, nid) (1)
1248 #ifndef early_pfn_valid
1249 #define early_pfn_valid(pfn) (1)
1252 void memory_present(int nid, unsigned long start, unsigned long end);
1253 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1256 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1257 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1258 * pfn_valid_within() should be used in this case; we optimise this away
1259 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1261 #ifdef CONFIG_HOLES_IN_ZONE
1262 #define pfn_valid_within(pfn) pfn_valid(pfn)
1264 #define pfn_valid_within(pfn) (1)
1267 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1269 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1270 * associated with it or not. In FLATMEM, it is expected that holes always
1271 * have valid memmap as long as there is valid PFNs either side of the hole.
1272 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1275 * However, an ARM, and maybe other embedded architectures in the future
1276 * free memmap backing holes to save memory on the assumption the memmap is
1277 * never used. The page_zone linkages are then broken even though pfn_valid()
1278 * returns true. A walker of the full memmap must then do this additional
1279 * check to ensure the memmap they are looking at is sane by making sure
1280 * the zone and PFN linkages are still valid. This is expensive, but walkers
1281 * of the full memmap are extremely rare.
1283 int memmap_valid_within(unsigned long pfn,
1284 struct page *page, struct zone *zone);
1286 static inline int memmap_valid_within(unsigned long pfn,
1287 struct page *page, struct zone *zone)
1291 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1293 #endif /* !__GENERATING_BOUNDS.H */
1294 #endif /* !__ASSEMBLY__ */
1295 #endif /* _LINUX_MMZONE_H */