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_HIGHATOMIC
= 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 */
66 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
67 extern char * const migratetype_names
[MIGRATE_TYPES
];
70 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
72 # define is_migrate_cma(migratetype) false
75 #define for_each_migratetype_order(order, type) \
76 for (order = 0; order < MAX_ORDER; order++) \
77 for (type = 0; type < MIGRATE_TYPES; type++)
79 extern int page_group_by_mobility_disabled
;
81 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
82 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
84 #define get_pageblock_migratetype(page) \
85 get_pfnblock_flags_mask(page, page_to_pfn(page), \
86 PB_migrate_end, MIGRATETYPE_MASK)
89 struct list_head free_list
[MIGRATE_TYPES
];
90 unsigned long nr_free
;
96 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
97 * So add a wild amount of padding here to ensure that they fall into separate
98 * cachelines. There are very few zone structures in the machine, so space
99 * consumption is not a concern here.
101 #if defined(CONFIG_SMP)
102 struct zone_padding
{
104 } ____cacheline_internodealigned_in_smp
;
105 #define ZONE_PADDING(name) struct zone_padding name;
107 #define ZONE_PADDING(name)
110 enum zone_stat_item
{
111 /* First 128 byte cacheline (assuming 64 bit words) */
114 NR_ZONE_LRU_BASE
, /* Used only for compaction and reclaim retry */
115 NR_ZONE_LRU_ANON
= NR_ZONE_LRU_BASE
,
117 NR_MLOCK
, /* mlock()ed pages found and moved off LRU */
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_SHMEM
, /* shmem pages (included tmpfs/GEM pages) */
132 NR_DIRTIED
, /* page dirtyings since bootup */
133 NR_WRITTEN
, /* page writings since bootup */
134 #if IS_ENABLED(CONFIG_ZSMALLOC)
135 NR_ZSPAGES
, /* allocated in zsmalloc */
138 NUMA_HIT
, /* allocated in intended node */
139 NUMA_MISS
, /* allocated in non intended node */
140 NUMA_FOREIGN
, /* was intended here, hit elsewhere */
141 NUMA_INTERLEAVE_HIT
, /* interleaver preferred this zone */
142 NUMA_LOCAL
, /* allocation from local node */
143 NUMA_OTHER
, /* allocation from other node */
149 NR_VM_ZONE_STAT_ITEMS
};
151 enum node_stat_item
{
153 NR_INACTIVE_ANON
= NR_LRU_BASE
, /* must match order of LRU_[IN]ACTIVE */
154 NR_ACTIVE_ANON
, /* " " " " " */
155 NR_INACTIVE_FILE
, /* " " " " " */
156 NR_ACTIVE_FILE
, /* " " " " " */
157 NR_UNEVICTABLE
, /* " " " " " */
158 NR_ISOLATED_ANON
, /* Temporary isolated pages from anon lru */
159 NR_ISOLATED_FILE
, /* Temporary isolated pages from file lru */
160 NR_PAGES_SCANNED
, /* pages scanned since last reclaim */
163 WORKINGSET_NODERECLAIM
,
164 NR_ANON_PAGES
, /* Mapped anonymous pages */
165 NR_FILE_MAPPED
, /* pagecache pages mapped into pagetables.
166 only modified from process context */
167 NR_VM_NODE_STAT_ITEMS
171 * We do arithmetic on the LRU lists in various places in the code,
172 * so it is important to keep the active lists LRU_ACTIVE higher in
173 * the array than the corresponding inactive lists, and to keep
174 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
176 * This has to be kept in sync with the statistics in zone_stat_item
177 * above and the descriptions in vmstat_text in mm/vmstat.c
184 LRU_INACTIVE_ANON
= LRU_BASE
,
185 LRU_ACTIVE_ANON
= LRU_BASE
+ LRU_ACTIVE
,
186 LRU_INACTIVE_FILE
= LRU_BASE
+ LRU_FILE
,
187 LRU_ACTIVE_FILE
= LRU_BASE
+ LRU_FILE
+ LRU_ACTIVE
,
192 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
194 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
196 static inline int is_file_lru(enum lru_list lru
)
198 return (lru
== LRU_INACTIVE_FILE
|| lru
== LRU_ACTIVE_FILE
);
201 static inline int is_active_lru(enum lru_list lru
)
203 return (lru
== LRU_ACTIVE_ANON
|| lru
== LRU_ACTIVE_FILE
);
206 struct zone_reclaim_stat
{
208 * The pageout code in vmscan.c keeps track of how many of the
209 * mem/swap backed and file backed pages are referenced.
210 * The higher the rotated/scanned ratio, the more valuable
213 * The anon LRU stats live in [0], file LRU stats in [1]
215 unsigned long recent_rotated
[2];
216 unsigned long recent_scanned
[2];
220 struct list_head lists
[NR_LRU_LISTS
];
221 struct zone_reclaim_stat reclaim_stat
;
222 /* Evictions & activations on the inactive file list */
223 atomic_long_t inactive_age
;
225 struct pglist_data
*pgdat
;
229 /* Mask used at gathering information at once (see memcontrol.c) */
230 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
231 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
232 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
234 /* Isolate clean file */
235 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
236 /* Isolate unmapped file */
237 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
238 /* Isolate for asynchronous migration */
239 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
240 /* Isolate unevictable pages */
241 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
243 /* LRU Isolation modes. */
244 typedef unsigned __bitwise__ isolate_mode_t
;
246 enum zone_watermarks
{
253 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
254 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
255 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
257 struct per_cpu_pages
{
258 int count
; /* number of pages in the list */
259 int high
; /* high watermark, emptying needed */
260 int batch
; /* chunk size for buddy add/remove */
262 /* Lists of pages, one per migrate type stored on the pcp-lists */
263 struct list_head lists
[MIGRATE_PCPTYPES
];
266 struct per_cpu_pageset
{
267 struct per_cpu_pages pcp
;
273 s8 vm_stat_diff
[NR_VM_ZONE_STAT_ITEMS
];
277 struct per_cpu_nodestat
{
279 s8 vm_node_stat_diff
[NR_VM_NODE_STAT_ITEMS
];
282 #endif /* !__GENERATING_BOUNDS.H */
285 #ifdef CONFIG_ZONE_DMA
287 * ZONE_DMA is used when there are devices that are not able
288 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
289 * carve out the portion of memory that is needed for these devices.
290 * The range is arch specific.
295 * ---------------------------
296 * parisc, ia64, sparc <4G
299 * alpha Unlimited or 0-16MB.
301 * i386, x86_64 and multiple other arches
306 #ifdef CONFIG_ZONE_DMA32
308 * x86_64 needs two ZONE_DMAs because it supports devices that are
309 * only able to do DMA to the lower 16M but also 32 bit devices that
310 * can only do DMA areas below 4G.
315 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
316 * performed on pages in ZONE_NORMAL if the DMA devices support
317 * transfers to all addressable memory.
320 #ifdef CONFIG_HIGHMEM
322 * A memory area that is only addressable by the kernel through
323 * mapping portions into its own address space. This is for example
324 * used by i386 to allow the kernel to address the memory beyond
325 * 900MB. The kernel will set up special mappings (page
326 * table entries on i386) for each page that the kernel needs to
332 #ifdef CONFIG_ZONE_DEVICE
339 #ifndef __GENERATING_BOUNDS_H
342 /* Read-mostly fields */
344 /* zone watermarks, access with *_wmark_pages(zone) macros */
345 unsigned long watermark
[NR_WMARK
];
347 unsigned long nr_reserved_highatomic
;
350 * We don't know if the memory that we're going to allocate will be
351 * freeable or/and it will be released eventually, so to avoid totally
352 * wasting several GB of ram we must reserve some of the lower zone
353 * memory (otherwise we risk to run OOM on the lower zones despite
354 * there being tons of freeable ram on the higher zones). This array is
355 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
358 long lowmem_reserve
[MAX_NR_ZONES
];
363 struct pglist_data
*zone_pgdat
;
364 struct per_cpu_pageset __percpu
*pageset
;
366 #ifndef CONFIG_SPARSEMEM
368 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
369 * In SPARSEMEM, this map is stored in struct mem_section
371 unsigned long *pageblock_flags
;
372 #endif /* CONFIG_SPARSEMEM */
376 * zone reclaim becomes active if more unmapped pages exist.
378 unsigned long min_unmapped_pages
;
379 unsigned long min_slab_pages
;
380 #endif /* CONFIG_NUMA */
382 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
383 unsigned long zone_start_pfn
;
386 * spanned_pages is the total pages spanned by the zone, including
387 * holes, which is calculated as:
388 * spanned_pages = zone_end_pfn - zone_start_pfn;
390 * present_pages is physical pages existing within the zone, which
392 * present_pages = spanned_pages - absent_pages(pages in holes);
394 * managed_pages is present pages managed by the buddy system, which
395 * is calculated as (reserved_pages includes pages allocated by the
396 * bootmem allocator):
397 * managed_pages = present_pages - reserved_pages;
399 * So present_pages may be used by memory hotplug or memory power
400 * management logic to figure out unmanaged pages by checking
401 * (present_pages - managed_pages). And managed_pages should be used
402 * by page allocator and vm scanner to calculate all kinds of watermarks
407 * zone_start_pfn and spanned_pages are protected by span_seqlock.
408 * It is a seqlock because it has to be read outside of zone->lock,
409 * and it is done in the main allocator path. But, it is written
410 * quite infrequently.
412 * The span_seq lock is declared along with zone->lock because it is
413 * frequently read in proximity to zone->lock. It's good to
414 * give them a chance of being in the same cacheline.
416 * Write access to present_pages at runtime should be protected by
417 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
418 * present_pages should get_online_mems() to get a stable value.
420 * Read access to managed_pages should be safe because it's unsigned
421 * long. Write access to zone->managed_pages and totalram_pages are
422 * protected by managed_page_count_lock at runtime. Idealy only
423 * adjust_managed_page_count() should be used instead of directly
424 * touching zone->managed_pages and totalram_pages.
426 unsigned long managed_pages
;
427 unsigned long spanned_pages
;
428 unsigned long present_pages
;
432 #ifdef CONFIG_MEMORY_ISOLATION
434 * Number of isolated pageblock. It is used to solve incorrect
435 * freepage counting problem due to racy retrieving migratetype
436 * of pageblock. Protected by zone->lock.
438 unsigned long nr_isolate_pageblock
;
441 #ifdef CONFIG_MEMORY_HOTPLUG
442 /* see spanned/present_pages for more description */
443 seqlock_t span_seqlock
;
447 * wait_table -- the array holding the hash table
448 * wait_table_hash_nr_entries -- the size of the hash table array
449 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
451 * The purpose of all these is to keep track of the people
452 * waiting for a page to become available and make them
453 * runnable again when possible. The trouble is that this
454 * consumes a lot of space, especially when so few things
455 * wait on pages at a given time. So instead of using
456 * per-page waitqueues, we use a waitqueue hash table.
458 * The bucket discipline is to sleep on the same queue when
459 * colliding and wake all in that wait queue when removing.
460 * When something wakes, it must check to be sure its page is
461 * truly available, a la thundering herd. The cost of a
462 * collision is great, but given the expected load of the
463 * table, they should be so rare as to be outweighed by the
464 * benefits from the saved space.
466 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
467 * primary users of these fields, and in mm/page_alloc.c
468 * free_area_init_core() performs the initialization of them.
470 wait_queue_head_t
*wait_table
;
471 unsigned long wait_table_hash_nr_entries
;
472 unsigned long wait_table_bits
;
474 /* Write-intensive fields used from the page allocator */
477 /* free areas of different sizes */
478 struct free_area free_area
[MAX_ORDER
];
480 /* zone flags, see below */
483 /* Primarily protects free_area */
486 /* Write-intensive fields used by compaction and vmstats. */
490 * When free pages are below this point, additional steps are taken
491 * when reading the number of free pages to avoid per-cpu counter
492 * drift allowing watermarks to be breached
494 unsigned long percpu_drift_mark
;
496 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
497 /* pfn where compaction free scanner should start */
498 unsigned long compact_cached_free_pfn
;
499 /* pfn where async and sync compaction migration scanner should start */
500 unsigned long compact_cached_migrate_pfn
[2];
503 #ifdef CONFIG_COMPACTION
505 * On compaction failure, 1<<compact_defer_shift compactions
506 * are skipped before trying again. The number attempted since
507 * last failure is tracked with compact_considered.
509 unsigned int compact_considered
;
510 unsigned int compact_defer_shift
;
511 int compact_order_failed
;
514 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
515 /* Set to true when the PG_migrate_skip bits should be cleared */
516 bool compact_blockskip_flush
;
522 /* Zone statistics */
523 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
524 } ____cacheline_internodealigned_in_smp
;
527 ZONE_RECLAIM_LOCKED
, /* prevents concurrent reclaim */
528 ZONE_FAIR_DEPLETED
, /* fair zone policy batch depleted */
532 PGDAT_CONGESTED
, /* pgdat has many dirty pages backed by
535 PGDAT_DIRTY
, /* reclaim scanning has recently found
536 * many dirty file pages at the tail
539 PGDAT_WRITEBACK
, /* reclaim scanning has recently found
540 * many pages under writeback
544 static inline unsigned long zone_end_pfn(const struct zone
*zone
)
546 return zone
->zone_start_pfn
+ zone
->spanned_pages
;
549 static inline bool zone_spans_pfn(const struct zone
*zone
, unsigned long pfn
)
551 return zone
->zone_start_pfn
<= pfn
&& pfn
< zone_end_pfn(zone
);
554 static inline bool zone_is_initialized(struct zone
*zone
)
556 return !!zone
->wait_table
;
559 static inline bool zone_is_empty(struct zone
*zone
)
561 return zone
->spanned_pages
== 0;
565 * The "priority" of VM scanning is how much of the queues we will scan in one
566 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
567 * queues ("queue_length >> 12") during an aging round.
569 #define DEF_PRIORITY 12
571 /* Maximum number of zones on a zonelist */
572 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
575 ZONELIST_FALLBACK
, /* zonelist with fallback */
578 * The NUMA zonelists are doubled because we need zonelists that
579 * restrict the allocations to a single node for __GFP_THISNODE.
581 ZONELIST_NOFALLBACK
, /* zonelist without fallback (__GFP_THISNODE) */
587 * This struct contains information about a zone in a zonelist. It is stored
588 * here to avoid dereferences into large structures and lookups of tables
591 struct zone
*zone
; /* Pointer to actual zone */
592 int zone_idx
; /* zone_idx(zoneref->zone) */
596 * One allocation request operates on a zonelist. A zonelist
597 * is a list of zones, the first one is the 'goal' of the
598 * allocation, the other zones are fallback zones, in decreasing
601 * To speed the reading of the zonelist, the zonerefs contain the zone index
602 * of the entry being read. Helper functions to access information given
603 * a struct zoneref are
605 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
606 * zonelist_zone_idx() - Return the index of the zone for an entry
607 * zonelist_node_idx() - Return the index of the node for an entry
610 struct zoneref _zonerefs
[MAX_ZONES_PER_ZONELIST
+ 1];
613 #ifndef CONFIG_DISCONTIGMEM
614 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
615 extern struct page
*mem_map
;
619 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
620 * (mostly NUMA machines?) to denote a higher-level memory zone than the
623 * On NUMA machines, each NUMA node would have a pg_data_t to describe
624 * it's memory layout.
626 * Memory statistics and page replacement data structures are maintained on a
630 typedef struct pglist_data
{
631 struct zone node_zones
[MAX_NR_ZONES
];
632 struct zonelist node_zonelists
[MAX_ZONELISTS
];
634 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
635 struct page
*node_mem_map
;
636 #ifdef CONFIG_PAGE_EXTENSION
637 struct page_ext
*node_page_ext
;
640 #ifndef CONFIG_NO_BOOTMEM
641 struct bootmem_data
*bdata
;
643 #ifdef CONFIG_MEMORY_HOTPLUG
645 * Must be held any time you expect node_start_pfn, node_present_pages
646 * or node_spanned_pages stay constant. Holding this will also
647 * guarantee that any pfn_valid() stays that way.
649 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
650 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
652 * Nests above zone->lock and zone->span_seqlock
654 spinlock_t node_size_lock
;
656 unsigned long node_start_pfn
;
657 unsigned long node_present_pages
; /* total number of physical pages */
658 unsigned long node_spanned_pages
; /* total size of physical page
659 range, including holes */
661 wait_queue_head_t kswapd_wait
;
662 wait_queue_head_t pfmemalloc_wait
;
663 struct task_struct
*kswapd
; /* Protected by
664 mem_hotplug_begin/end() */
666 enum zone_type kswapd_classzone_idx
;
668 #ifdef CONFIG_COMPACTION
669 int kcompactd_max_order
;
670 enum zone_type kcompactd_classzone_idx
;
671 wait_queue_head_t kcompactd_wait
;
672 struct task_struct
*kcompactd
;
674 #ifdef CONFIG_NUMA_BALANCING
675 /* Lock serializing the migrate rate limiting window */
676 spinlock_t numabalancing_migrate_lock
;
678 /* Rate limiting time interval */
679 unsigned long numabalancing_migrate_next_window
;
681 /* Number of pages migrated during the rate limiting time interval */
682 unsigned long numabalancing_migrate_nr_pages
;
685 * This is a per-node reserve of pages that are not available
686 * to userspace allocations.
688 unsigned long totalreserve_pages
;
690 /* Write-intensive fields used by page reclaim */
694 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
696 * If memory initialisation on large machines is deferred then this
697 * is the first PFN that needs to be initialised.
699 unsigned long first_deferred_pfn
;
700 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
702 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
703 spinlock_t split_queue_lock
;
704 struct list_head split_queue
;
705 unsigned long split_queue_len
;
708 /* Fields commonly accessed by the page reclaim scanner */
709 struct lruvec lruvec
;
712 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
713 * this node's LRU. Maintained by the pageout code.
715 unsigned int inactive_ratio
;
721 /* Per-node vmstats */
722 struct per_cpu_nodestat __percpu
*per_cpu_nodestats
;
723 atomic_long_t vm_stat
[NR_VM_NODE_STAT_ITEMS
];
726 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
727 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
728 #ifdef CONFIG_FLAT_NODE_MEM_MAP
729 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
731 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
733 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
735 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
736 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
737 static inline spinlock_t
*zone_lru_lock(struct zone
*zone
)
739 return &zone
->zone_pgdat
->lru_lock
;
742 static inline struct lruvec
*node_lruvec(struct pglist_data
*pgdat
)
744 return &pgdat
->lruvec
;
747 static inline unsigned long pgdat_end_pfn(pg_data_t
*pgdat
)
749 return pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
752 static inline bool pgdat_is_empty(pg_data_t
*pgdat
)
754 return !pgdat
->node_start_pfn
&& !pgdat
->node_spanned_pages
;
757 static inline int zone_id(const struct zone
*zone
)
759 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
761 return zone
- pgdat
->node_zones
;
764 #ifdef CONFIG_ZONE_DEVICE
765 static inline bool is_dev_zone(const struct zone
*zone
)
767 return zone_id(zone
) == ZONE_DEVICE
;
770 static inline bool is_dev_zone(const struct zone
*zone
)
776 #include <linux/memory_hotplug.h>
778 extern struct mutex zonelists_mutex
;
779 void build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
);
780 void wakeup_kswapd(struct zone
*zone
, int order
, enum zone_type classzone_idx
);
781 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
782 int classzone_idx
, unsigned int alloc_flags
,
784 bool zone_watermark_ok(struct zone
*z
, unsigned int order
,
785 unsigned long mark
, int classzone_idx
,
786 unsigned int alloc_flags
);
787 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
788 unsigned long mark
, int classzone_idx
);
789 enum memmap_context
{
793 extern int init_currently_empty_zone(struct zone
*zone
, unsigned long start_pfn
,
796 extern void lruvec_init(struct lruvec
*lruvec
);
798 static inline struct pglist_data
*lruvec_pgdat(struct lruvec
*lruvec
)
801 return lruvec
->pgdat
;
803 return container_of(lruvec
, struct pglist_data
, lruvec
);
807 extern unsigned long lruvec_lru_size(struct lruvec
*lruvec
, enum lru_list lru
);
809 #ifdef CONFIG_HAVE_MEMORY_PRESENT
810 void memory_present(int nid
, unsigned long start
, unsigned long end
);
812 static inline void memory_present(int nid
, unsigned long start
, unsigned long end
) {}
815 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
816 int local_memory_node(int node_id
);
818 static inline int local_memory_node(int node_id
) { return node_id
; };
821 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
822 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
826 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
828 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
830 static inline int populated_zone(struct zone
*zone
)
832 return (!!zone
->present_pages
);
835 extern int movable_zone
;
837 #ifdef CONFIG_HIGHMEM
838 static inline int zone_movable_is_highmem(void)
840 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
841 return movable_zone
== ZONE_HIGHMEM
;
843 return (ZONE_MOVABLE
- 1) == ZONE_HIGHMEM
;
848 static inline int is_highmem_idx(enum zone_type idx
)
850 #ifdef CONFIG_HIGHMEM
851 return (idx
== ZONE_HIGHMEM
||
852 (idx
== ZONE_MOVABLE
&& zone_movable_is_highmem()));
859 * is_highmem - helper function to quickly check if a struct zone is a
860 * highmem zone or not. This is an attempt to keep references
861 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
862 * @zone - pointer to struct zone variable
864 static inline int is_highmem(struct zone
*zone
)
866 #ifdef CONFIG_HIGHMEM
867 return is_highmem_idx(zone_idx(zone
));
873 /* These two functions are used to setup the per zone pages min values */
875 int min_free_kbytes_sysctl_handler(struct ctl_table
*, int,
876 void __user
*, size_t *, loff_t
*);
877 int watermark_scale_factor_sysctl_handler(struct ctl_table
*, int,
878 void __user
*, size_t *, loff_t
*);
879 extern int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1];
880 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*, int,
881 void __user
*, size_t *, loff_t
*);
882 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*, int,
883 void __user
*, size_t *, loff_t
*);
884 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*, int,
885 void __user
*, size_t *, loff_t
*);
886 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*, int,
887 void __user
*, size_t *, loff_t
*);
889 extern int numa_zonelist_order_handler(struct ctl_table
*, int,
890 void __user
*, size_t *, loff_t
*);
891 extern char numa_zonelist_order
[];
892 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
894 #ifndef CONFIG_NEED_MULTIPLE_NODES
896 extern struct pglist_data contig_page_data
;
897 #define NODE_DATA(nid) (&contig_page_data)
898 #define NODE_MEM_MAP(nid) mem_map
900 #else /* CONFIG_NEED_MULTIPLE_NODES */
902 #include <asm/mmzone.h>
904 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
906 extern struct pglist_data
*first_online_pgdat(void);
907 extern struct pglist_data
*next_online_pgdat(struct pglist_data
*pgdat
);
908 extern struct zone
*next_zone(struct zone
*zone
);
911 * for_each_online_pgdat - helper macro to iterate over all online nodes
912 * @pgdat - pointer to a pg_data_t variable
914 #define for_each_online_pgdat(pgdat) \
915 for (pgdat = first_online_pgdat(); \
917 pgdat = next_online_pgdat(pgdat))
919 * for_each_zone - helper macro to iterate over all memory zones
920 * @zone - pointer to struct zone variable
922 * The user only needs to declare the zone variable, for_each_zone
925 #define for_each_zone(zone) \
926 for (zone = (first_online_pgdat())->node_zones; \
928 zone = next_zone(zone))
930 #define for_each_populated_zone(zone) \
931 for (zone = (first_online_pgdat())->node_zones; \
933 zone = next_zone(zone)) \
934 if (!populated_zone(zone)) \
938 static inline struct zone
*zonelist_zone(struct zoneref
*zoneref
)
940 return zoneref
->zone
;
943 static inline int zonelist_zone_idx(struct zoneref
*zoneref
)
945 return zoneref
->zone_idx
;
948 static inline int zonelist_node_idx(struct zoneref
*zoneref
)
951 /* zone_to_nid not available in this context */
952 return zoneref
->zone
->node
;
955 #endif /* CONFIG_NUMA */
958 struct zoneref
*__next_zones_zonelist(struct zoneref
*z
,
959 enum zone_type highest_zoneidx
,
963 * 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
964 * @z - The cursor used as a starting point for the search
965 * @highest_zoneidx - The zone index of the highest zone to return
966 * @nodes - An optional nodemask to filter the zonelist with
968 * This function returns the next zone at or below a given zone index that is
969 * within the allowed nodemask using a cursor as the starting point for the
970 * search. The zoneref returned is a cursor that represents the current zone
971 * being examined. It should be advanced by one before calling
972 * next_zones_zonelist again.
974 static __always_inline
struct zoneref
*next_zones_zonelist(struct zoneref
*z
,
975 enum zone_type highest_zoneidx
,
978 if (likely(!nodes
&& zonelist_zone_idx(z
) <= highest_zoneidx
))
980 return __next_zones_zonelist(z
, highest_zoneidx
, nodes
);
984 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
985 * @zonelist - The zonelist to search for a suitable zone
986 * @highest_zoneidx - The zone index of the highest zone to return
987 * @nodes - An optional nodemask to filter the zonelist with
988 * @zone - The first suitable zone found is returned via this parameter
990 * This function returns the first zone at or below a given zone index that is
991 * within the allowed nodemask. The zoneref returned is a cursor that can be
992 * used to iterate the zonelist with next_zones_zonelist by advancing it by
993 * one before calling.
995 static inline struct zoneref
*first_zones_zonelist(struct zonelist
*zonelist
,
996 enum zone_type highest_zoneidx
,
999 return next_zones_zonelist(zonelist
->_zonerefs
,
1000 highest_zoneidx
, nodes
);
1004 * 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
1005 * @zone - The current zone in the iterator
1006 * @z - The current pointer within zonelist->zones being iterated
1007 * @zlist - The zonelist being iterated
1008 * @highidx - The zone index of the highest zone to return
1009 * @nodemask - Nodemask allowed by the allocator
1011 * This iterator iterates though all zones at or below a given zone index and
1012 * within a given nodemask
1014 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1015 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1017 z = next_zones_zonelist(++z, highidx, nodemask), \
1018 zone = zonelist_zone(z))
1020 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1021 for (zone = z->zone; \
1023 z = next_zones_zonelist(++z, highidx, nodemask), \
1024 zone = zonelist_zone(z))
1028 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1029 * @zone - The current zone in the iterator
1030 * @z - The current pointer within zonelist->zones being iterated
1031 * @zlist - The zonelist being iterated
1032 * @highidx - The zone index of the highest zone to return
1034 * This iterator iterates though all zones at or below a given zone index.
1036 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1037 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1039 #ifdef CONFIG_SPARSEMEM
1040 #include <asm/sparsemem.h>
1043 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1044 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1045 static inline unsigned long early_pfn_to_nid(unsigned long pfn
)
1051 #ifdef CONFIG_FLATMEM
1052 #define pfn_to_nid(pfn) (0)
1055 #ifdef CONFIG_SPARSEMEM
1058 * SECTION_SHIFT #bits space required to store a section #
1060 * PA_SECTION_SHIFT physical address to/from section number
1061 * PFN_SECTION_SHIFT pfn to/from section number
1063 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1064 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1066 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1068 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1069 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1071 #define SECTION_BLOCKFLAGS_BITS \
1072 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1074 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1075 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1078 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1079 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1081 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1082 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1086 struct mem_section
{
1088 * This is, logically, a pointer to an array of struct
1089 * pages. However, it is stored with some other magic.
1090 * (see sparse.c::sparse_init_one_section())
1092 * Additionally during early boot we encode node id of
1093 * the location of the section here to guide allocation.
1094 * (see sparse.c::memory_present())
1096 * Making it a UL at least makes someone do a cast
1097 * before using it wrong.
1099 unsigned long section_mem_map
;
1101 /* See declaration of similar field in struct zone */
1102 unsigned long *pageblock_flags
;
1103 #ifdef CONFIG_PAGE_EXTENSION
1105 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1106 * section. (see page_ext.h about this.)
1108 struct page_ext
*page_ext
;
1112 * WARNING: mem_section must be a power-of-2 in size for the
1113 * calculation and use of SECTION_ROOT_MASK to make sense.
1117 #ifdef CONFIG_SPARSEMEM_EXTREME
1118 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1120 #define SECTIONS_PER_ROOT 1
1123 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1124 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1125 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1127 #ifdef CONFIG_SPARSEMEM_EXTREME
1128 extern struct mem_section
*mem_section
[NR_SECTION_ROOTS
];
1130 extern struct mem_section mem_section
[NR_SECTION_ROOTS
][SECTIONS_PER_ROOT
];
1133 static inline struct mem_section
*__nr_to_section(unsigned long nr
)
1135 if (!mem_section
[SECTION_NR_TO_ROOT(nr
)])
1137 return &mem_section
[SECTION_NR_TO_ROOT(nr
)][nr
& SECTION_ROOT_MASK
];
1139 extern int __section_nr(struct mem_section
* ms
);
1140 extern unsigned long usemap_size(void);
1143 * We use the lower bits of the mem_map pointer to store
1144 * a little bit of information. There should be at least
1145 * 3 bits here due to 32-bit alignment.
1147 #define SECTION_MARKED_PRESENT (1UL<<0)
1148 #define SECTION_HAS_MEM_MAP (1UL<<1)
1149 #define SECTION_MAP_LAST_BIT (1UL<<2)
1150 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1151 #define SECTION_NID_SHIFT 2
1153 static inline struct page
*__section_mem_map_addr(struct mem_section
*section
)
1155 unsigned long map
= section
->section_mem_map
;
1156 map
&= SECTION_MAP_MASK
;
1157 return (struct page
*)map
;
1160 static inline int present_section(struct mem_section
*section
)
1162 return (section
&& (section
->section_mem_map
& SECTION_MARKED_PRESENT
));
1165 static inline int present_section_nr(unsigned long nr
)
1167 return present_section(__nr_to_section(nr
));
1170 static inline int valid_section(struct mem_section
*section
)
1172 return (section
&& (section
->section_mem_map
& SECTION_HAS_MEM_MAP
));
1175 static inline int valid_section_nr(unsigned long nr
)
1177 return valid_section(__nr_to_section(nr
));
1180 static inline struct mem_section
*__pfn_to_section(unsigned long pfn
)
1182 return __nr_to_section(pfn_to_section_nr(pfn
));
1185 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1186 static inline int pfn_valid(unsigned long pfn
)
1188 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1190 return valid_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1194 static inline int pfn_present(unsigned long pfn
)
1196 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1198 return present_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1202 * These are _only_ used during initialisation, therefore they
1203 * can use __initdata ... They could have names to indicate
1207 #define pfn_to_nid(pfn) \
1209 unsigned long __pfn_to_nid_pfn = (pfn); \
1210 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1213 #define pfn_to_nid(pfn) (0)
1216 #define early_pfn_valid(pfn) pfn_valid(pfn)
1217 void sparse_init(void);
1219 #define sparse_init() do {} while (0)
1220 #define sparse_index_init(_sec, _nid) do {} while (0)
1221 #endif /* CONFIG_SPARSEMEM */
1224 * During memory init memblocks map pfns to nids. The search is expensive and
1225 * this caches recent lookups. The implementation of __early_pfn_to_nid
1226 * may treat start/end as pfns or sections.
1228 struct mminit_pfnnid_cache
{
1229 unsigned long last_start
;
1230 unsigned long last_end
;
1234 #ifndef early_pfn_valid
1235 #define early_pfn_valid(pfn) (1)
1238 void memory_present(int nid
, unsigned long start
, unsigned long end
);
1239 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
1242 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1243 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1244 * pfn_valid_within() should be used in this case; we optimise this away
1245 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1247 #ifdef CONFIG_HOLES_IN_ZONE
1248 #define pfn_valid_within(pfn) pfn_valid(pfn)
1250 #define pfn_valid_within(pfn) (1)
1253 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1255 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1256 * associated with it or not. In FLATMEM, it is expected that holes always
1257 * have valid memmap as long as there is valid PFNs either side of the hole.
1258 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1261 * However, an ARM, and maybe other embedded architectures in the future
1262 * free memmap backing holes to save memory on the assumption the memmap is
1263 * never used. The page_zone linkages are then broken even though pfn_valid()
1264 * returns true. A walker of the full memmap must then do this additional
1265 * check to ensure the memmap they are looking at is sane by making sure
1266 * the zone and PFN linkages are still valid. This is expensive, but walkers
1267 * of the full memmap are extremely rare.
1269 bool memmap_valid_within(unsigned long pfn
,
1270 struct page
*page
, struct zone
*zone
);
1272 static inline bool memmap_valid_within(unsigned long pfn
,
1273 struct page
*page
, struct zone
*zone
)
1277 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1279 #endif /* !__GENERATING_BOUNDS.H */
1280 #endif /* !__ASSEMBLY__ */
1281 #endif /* _LINUX_MMZONE_H */