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)
71 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
73 # define is_migrate_cma(migratetype) false
74 # define is_migrate_cma_page(_page) false
77 static inline bool is_migrate_movable(int mt
)
79 return is_migrate_cma(mt
) || mt
== MIGRATE_MOVABLE
;
82 #define for_each_migratetype_order(order, type) \
83 for (order = 0; order < MAX_ORDER; order++) \
84 for (type = 0; type < MIGRATE_TYPES; type++)
86 extern int page_group_by_mobility_disabled
;
88 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
89 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
91 #define get_pageblock_migratetype(page) \
92 get_pfnblock_flags_mask(page, page_to_pfn(page), \
93 PB_migrate_end, MIGRATETYPE_MASK)
96 struct list_head free_list
[MIGRATE_TYPES
];
97 unsigned long nr_free
;
103 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
104 * So add a wild amount of padding here to ensure that they fall into separate
105 * cachelines. There are very few zone structures in the machine, so space
106 * consumption is not a concern here.
108 #if defined(CONFIG_SMP)
109 struct zone_padding
{
111 } ____cacheline_internodealigned_in_smp
;
112 #define ZONE_PADDING(name) struct zone_padding name;
114 #define ZONE_PADDING(name)
117 enum zone_stat_item
{
118 /* First 128 byte cacheline (assuming 64 bit words) */
120 NR_ZONE_LRU_BASE
, /* Used only for compaction and reclaim retry */
121 NR_ZONE_INACTIVE_ANON
= NR_ZONE_LRU_BASE
,
123 NR_ZONE_INACTIVE_FILE
,
126 NR_ZONE_WRITE_PENDING
, /* Count of dirty, writeback and unstable pages */
127 NR_MLOCK
, /* mlock()ed pages found and moved off LRU */
128 NR_PAGETABLE
, /* used for pagetables */
129 NR_KERNEL_STACK_KB
, /* measured in KiB */
130 /* Second 128 byte cacheline */
132 #if IS_ENABLED(CONFIG_ZSMALLOC)
133 NR_ZSPAGES
, /* allocated in zsmalloc */
136 NUMA_HIT
, /* allocated in intended node */
137 NUMA_MISS
, /* allocated in non intended node */
138 NUMA_FOREIGN
, /* was intended here, hit elsewhere */
139 NUMA_INTERLEAVE_HIT
, /* interleaver preferred this zone */
140 NUMA_LOCAL
, /* allocation from local node */
141 NUMA_OTHER
, /* allocation from other node */
144 NR_VM_ZONE_STAT_ITEMS
};
146 enum node_stat_item
{
148 NR_INACTIVE_ANON
= NR_LRU_BASE
, /* must match order of LRU_[IN]ACTIVE */
149 NR_ACTIVE_ANON
, /* " " " " " */
150 NR_INACTIVE_FILE
, /* " " " " " */
151 NR_ACTIVE_FILE
, /* " " " " " */
152 NR_UNEVICTABLE
, /* " " " " " */
154 NR_SLAB_UNRECLAIMABLE
,
155 NR_ISOLATED_ANON
, /* Temporary isolated pages from anon lru */
156 NR_ISOLATED_FILE
, /* Temporary isolated pages from file lru */
159 WORKINGSET_NODERECLAIM
,
160 NR_ANON_MAPPED
, /* Mapped anonymous pages */
161 NR_FILE_MAPPED
, /* pagecache pages mapped into pagetables.
162 only modified from process context */
166 NR_WRITEBACK_TEMP
, /* Writeback using temporary buffers */
167 NR_SHMEM
, /* shmem pages (included tmpfs/GEM pages) */
171 NR_UNSTABLE_NFS
, /* NFS unstable pages */
173 NR_VMSCAN_IMMEDIATE
, /* Prioritise for reclaim when writeback ends */
174 NR_DIRTIED
, /* page dirtyings since bootup */
175 NR_WRITTEN
, /* page writings since bootup */
176 NR_VM_NODE_STAT_ITEMS
180 * We do arithmetic on the LRU lists in various places in the code,
181 * so it is important to keep the active lists LRU_ACTIVE higher in
182 * the array than the corresponding inactive lists, and to keep
183 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
185 * This has to be kept in sync with the statistics in zone_stat_item
186 * above and the descriptions in vmstat_text in mm/vmstat.c
193 LRU_INACTIVE_ANON
= LRU_BASE
,
194 LRU_ACTIVE_ANON
= LRU_BASE
+ LRU_ACTIVE
,
195 LRU_INACTIVE_FILE
= LRU_BASE
+ LRU_FILE
,
196 LRU_ACTIVE_FILE
= LRU_BASE
+ LRU_FILE
+ LRU_ACTIVE
,
201 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
203 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
205 static inline int is_file_lru(enum lru_list lru
)
207 return (lru
== LRU_INACTIVE_FILE
|| lru
== LRU_ACTIVE_FILE
);
210 static inline int is_active_lru(enum lru_list lru
)
212 return (lru
== LRU_ACTIVE_ANON
|| lru
== LRU_ACTIVE_FILE
);
215 struct zone_reclaim_stat
{
217 * The pageout code in vmscan.c keeps track of how many of the
218 * mem/swap backed and file backed pages are referenced.
219 * The higher the rotated/scanned ratio, the more valuable
222 * The anon LRU stats live in [0], file LRU stats in [1]
224 unsigned long recent_rotated
[2];
225 unsigned long recent_scanned
[2];
229 struct list_head lists
[NR_LRU_LISTS
];
230 struct zone_reclaim_stat reclaim_stat
;
231 /* Evictions & activations on the inactive file list */
232 atomic_long_t inactive_age
;
233 /* Refaults at the time of last reclaim cycle */
234 unsigned long refaults
;
236 struct pglist_data
*pgdat
;
240 /* Mask used at gathering information at once (see memcontrol.c) */
241 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
242 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
243 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
245 /* Isolate unmapped file */
246 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
247 /* Isolate for asynchronous migration */
248 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
249 /* Isolate unevictable pages */
250 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
252 /* LRU Isolation modes. */
253 typedef unsigned __bitwise isolate_mode_t
;
255 enum zone_watermarks
{
262 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
263 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
264 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
266 struct per_cpu_pages
{
267 int count
; /* number of pages in the list */
268 int high
; /* high watermark, emptying needed */
269 int batch
; /* chunk size for buddy add/remove */
271 /* Lists of pages, one per migrate type stored on the pcp-lists */
272 struct list_head lists
[MIGRATE_PCPTYPES
];
275 struct per_cpu_pageset
{
276 struct per_cpu_pages pcp
;
282 s8 vm_stat_diff
[NR_VM_ZONE_STAT_ITEMS
];
286 struct per_cpu_nodestat
{
288 s8 vm_node_stat_diff
[NR_VM_NODE_STAT_ITEMS
];
291 #endif /* !__GENERATING_BOUNDS.H */
294 #ifdef CONFIG_ZONE_DMA
296 * ZONE_DMA is used when there are devices that are not able
297 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
298 * carve out the portion of memory that is needed for these devices.
299 * The range is arch specific.
304 * ---------------------------
305 * parisc, ia64, sparc <4G
308 * alpha Unlimited or 0-16MB.
310 * i386, x86_64 and multiple other arches
315 #ifdef CONFIG_ZONE_DMA32
317 * x86_64 needs two ZONE_DMAs because it supports devices that are
318 * only able to do DMA to the lower 16M but also 32 bit devices that
319 * can only do DMA areas below 4G.
324 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
325 * performed on pages in ZONE_NORMAL if the DMA devices support
326 * transfers to all addressable memory.
329 #ifdef CONFIG_HIGHMEM
331 * A memory area that is only addressable by the kernel through
332 * mapping portions into its own address space. This is for example
333 * used by i386 to allow the kernel to address the memory beyond
334 * 900MB. The kernel will set up special mappings (page
335 * table entries on i386) for each page that the kernel needs to
341 #ifdef CONFIG_ZONE_DEVICE
348 #ifndef __GENERATING_BOUNDS_H
351 /* Read-mostly fields */
353 /* zone watermarks, access with *_wmark_pages(zone) macros */
354 unsigned long watermark
[NR_WMARK
];
356 unsigned long nr_reserved_highatomic
;
359 * We don't know if the memory that we're going to allocate will be
360 * freeable or/and it will be released eventually, so to avoid totally
361 * wasting several GB of ram we must reserve some of the lower zone
362 * memory (otherwise we risk to run OOM on the lower zones despite
363 * there being tons of freeable ram on the higher zones). This array is
364 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
367 long lowmem_reserve
[MAX_NR_ZONES
];
372 struct pglist_data
*zone_pgdat
;
373 struct per_cpu_pageset __percpu
*pageset
;
375 #ifndef CONFIG_SPARSEMEM
377 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
378 * In SPARSEMEM, this map is stored in struct mem_section
380 unsigned long *pageblock_flags
;
381 #endif /* CONFIG_SPARSEMEM */
383 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
384 unsigned long zone_start_pfn
;
387 * spanned_pages is the total pages spanned by the zone, including
388 * holes, which is calculated as:
389 * spanned_pages = zone_end_pfn - zone_start_pfn;
391 * present_pages is physical pages existing within the zone, which
393 * present_pages = spanned_pages - absent_pages(pages in holes);
395 * managed_pages is present pages managed by the buddy system, which
396 * is calculated as (reserved_pages includes pages allocated by the
397 * bootmem allocator):
398 * managed_pages = present_pages - reserved_pages;
400 * So present_pages may be used by memory hotplug or memory power
401 * management logic to figure out unmanaged pages by checking
402 * (present_pages - managed_pages). And managed_pages should be used
403 * by page allocator and vm scanner to calculate all kinds of watermarks
408 * zone_start_pfn and spanned_pages are protected by span_seqlock.
409 * It is a seqlock because it has to be read outside of zone->lock,
410 * and it is done in the main allocator path. But, it is written
411 * quite infrequently.
413 * The span_seq lock is declared along with zone->lock because it is
414 * frequently read in proximity to zone->lock. It's good to
415 * give them a chance of being in the same cacheline.
417 * Write access to present_pages at runtime should be protected by
418 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
419 * present_pages should get_online_mems() to get a stable value.
421 * Read access to managed_pages should be safe because it's unsigned
422 * long. Write access to zone->managed_pages and totalram_pages are
423 * protected by managed_page_count_lock at runtime. Idealy only
424 * adjust_managed_page_count() should be used instead of directly
425 * touching zone->managed_pages and totalram_pages.
427 unsigned long managed_pages
;
428 unsigned long spanned_pages
;
429 unsigned long present_pages
;
433 #ifdef CONFIG_MEMORY_ISOLATION
435 * Number of isolated pageblock. It is used to solve incorrect
436 * freepage counting problem due to racy retrieving migratetype
437 * of pageblock. Protected by zone->lock.
439 unsigned long nr_isolate_pageblock
;
442 #ifdef CONFIG_MEMORY_HOTPLUG
443 /* see spanned/present_pages for more description */
444 seqlock_t span_seqlock
;
449 /* Write-intensive fields used from the page allocator */
452 /* free areas of different sizes */
453 struct free_area free_area
[MAX_ORDER
];
455 /* zone flags, see below */
458 /* Primarily protects free_area */
461 /* Write-intensive fields used by compaction and vmstats. */
465 * When free pages are below this point, additional steps are taken
466 * when reading the number of free pages to avoid per-cpu counter
467 * drift allowing watermarks to be breached
469 unsigned long percpu_drift_mark
;
471 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
472 /* pfn where compaction free scanner should start */
473 unsigned long compact_cached_free_pfn
;
474 /* pfn where async and sync compaction migration scanner should start */
475 unsigned long compact_cached_migrate_pfn
[2];
478 #ifdef CONFIG_COMPACTION
480 * On compaction failure, 1<<compact_defer_shift compactions
481 * are skipped before trying again. The number attempted since
482 * last failure is tracked with compact_considered.
484 unsigned int compact_considered
;
485 unsigned int compact_defer_shift
;
486 int compact_order_failed
;
489 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
490 /* Set to true when the PG_migrate_skip bits should be cleared */
491 bool compact_blockskip_flush
;
497 /* Zone statistics */
498 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
499 } ____cacheline_internodealigned_in_smp
;
502 PGDAT_CONGESTED
, /* pgdat has many dirty pages backed by
505 PGDAT_DIRTY
, /* reclaim scanning has recently found
506 * many dirty file pages at the tail
509 PGDAT_WRITEBACK
, /* reclaim scanning has recently found
510 * many pages under writeback
512 PGDAT_RECLAIM_LOCKED
, /* prevents concurrent reclaim */
515 static inline unsigned long zone_end_pfn(const struct zone
*zone
)
517 return zone
->zone_start_pfn
+ zone
->spanned_pages
;
520 static inline bool zone_spans_pfn(const struct zone
*zone
, unsigned long pfn
)
522 return zone
->zone_start_pfn
<= pfn
&& pfn
< zone_end_pfn(zone
);
525 static inline bool zone_is_initialized(struct zone
*zone
)
527 return zone
->initialized
;
530 static inline bool zone_is_empty(struct zone
*zone
)
532 return zone
->spanned_pages
== 0;
536 * The "priority" of VM scanning is how much of the queues we will scan in one
537 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
538 * queues ("queue_length >> 12") during an aging round.
540 #define DEF_PRIORITY 12
542 /* Maximum number of zones on a zonelist */
543 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
546 ZONELIST_FALLBACK
, /* zonelist with fallback */
549 * The NUMA zonelists are doubled because we need zonelists that
550 * restrict the allocations to a single node for __GFP_THISNODE.
552 ZONELIST_NOFALLBACK
, /* zonelist without fallback (__GFP_THISNODE) */
558 * This struct contains information about a zone in a zonelist. It is stored
559 * here to avoid dereferences into large structures and lookups of tables
562 struct zone
*zone
; /* Pointer to actual zone */
563 int zone_idx
; /* zone_idx(zoneref->zone) */
567 * One allocation request operates on a zonelist. A zonelist
568 * is a list of zones, the first one is the 'goal' of the
569 * allocation, the other zones are fallback zones, in decreasing
572 * To speed the reading of the zonelist, the zonerefs contain the zone index
573 * of the entry being read. Helper functions to access information given
574 * a struct zoneref are
576 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
577 * zonelist_zone_idx() - Return the index of the zone for an entry
578 * zonelist_node_idx() - Return the index of the node for an entry
581 struct zoneref _zonerefs
[MAX_ZONES_PER_ZONELIST
+ 1];
584 #ifndef CONFIG_DISCONTIGMEM
585 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
586 extern struct page
*mem_map
;
590 * On NUMA machines, each NUMA node would have a pg_data_t to describe
591 * it's memory layout. On UMA machines there is a single pglist_data which
592 * describes the whole memory.
594 * Memory statistics and page replacement data structures are maintained on a
598 typedef struct pglist_data
{
599 struct zone node_zones
[MAX_NR_ZONES
];
600 struct zonelist node_zonelists
[MAX_ZONELISTS
];
602 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
603 struct page
*node_mem_map
;
604 #ifdef CONFIG_PAGE_EXTENSION
605 struct page_ext
*node_page_ext
;
608 #ifndef CONFIG_NO_BOOTMEM
609 struct bootmem_data
*bdata
;
611 #ifdef CONFIG_MEMORY_HOTPLUG
613 * Must be held any time you expect node_start_pfn, node_present_pages
614 * or node_spanned_pages stay constant. Holding this will also
615 * guarantee that any pfn_valid() stays that way.
617 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
618 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
620 * Nests above zone->lock and zone->span_seqlock
622 spinlock_t node_size_lock
;
624 unsigned long node_start_pfn
;
625 unsigned long node_present_pages
; /* total number of physical pages */
626 unsigned long node_spanned_pages
; /* total size of physical page
627 range, including holes */
629 wait_queue_head_t kswapd_wait
;
630 wait_queue_head_t pfmemalloc_wait
;
631 struct task_struct
*kswapd
; /* Protected by
632 mem_hotplug_begin/end() */
634 enum zone_type kswapd_classzone_idx
;
636 int kswapd_failures
; /* Number of 'reclaimed == 0' runs */
638 #ifdef CONFIG_COMPACTION
639 int kcompactd_max_order
;
640 enum zone_type kcompactd_classzone_idx
;
641 wait_queue_head_t kcompactd_wait
;
642 struct task_struct
*kcompactd
;
644 #ifdef CONFIG_NUMA_BALANCING
645 /* Lock serializing the migrate rate limiting window */
646 spinlock_t numabalancing_migrate_lock
;
648 /* Rate limiting time interval */
649 unsigned long numabalancing_migrate_next_window
;
651 /* Number of pages migrated during the rate limiting time interval */
652 unsigned long numabalancing_migrate_nr_pages
;
655 * This is a per-node reserve of pages that are not available
656 * to userspace allocations.
658 unsigned long totalreserve_pages
;
662 * zone reclaim becomes active if more unmapped pages exist.
664 unsigned long min_unmapped_pages
;
665 unsigned long min_slab_pages
;
666 #endif /* CONFIG_NUMA */
668 /* Write-intensive fields used by page reclaim */
672 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
674 * If memory initialisation on large machines is deferred then this
675 * is the first PFN that needs to be initialised.
677 unsigned long first_deferred_pfn
;
678 /* Number of non-deferred pages */
679 unsigned long static_init_pgcnt
;
680 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
682 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
683 spinlock_t split_queue_lock
;
684 struct list_head split_queue
;
685 unsigned long split_queue_len
;
688 /* Fields commonly accessed by the page reclaim scanner */
689 struct lruvec lruvec
;
692 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
693 * this node's LRU. Maintained by the pageout code.
695 unsigned int inactive_ratio
;
701 /* Per-node vmstats */
702 struct per_cpu_nodestat __percpu
*per_cpu_nodestats
;
703 atomic_long_t vm_stat
[NR_VM_NODE_STAT_ITEMS
];
706 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
707 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
708 #ifdef CONFIG_FLAT_NODE_MEM_MAP
709 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
711 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
713 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
715 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
716 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
717 static inline spinlock_t
*zone_lru_lock(struct zone
*zone
)
719 return &zone
->zone_pgdat
->lru_lock
;
722 static inline struct lruvec
*node_lruvec(struct pglist_data
*pgdat
)
724 return &pgdat
->lruvec
;
727 static inline unsigned long pgdat_end_pfn(pg_data_t
*pgdat
)
729 return pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
732 static inline bool pgdat_is_empty(pg_data_t
*pgdat
)
734 return !pgdat
->node_start_pfn
&& !pgdat
->node_spanned_pages
;
737 static inline int zone_id(const struct zone
*zone
)
739 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
741 return zone
- pgdat
->node_zones
;
744 #ifdef CONFIG_ZONE_DEVICE
745 static inline bool is_dev_zone(const struct zone
*zone
)
747 return zone_id(zone
) == ZONE_DEVICE
;
750 static inline bool is_dev_zone(const struct zone
*zone
)
756 #include <linux/memory_hotplug.h>
758 extern struct mutex zonelists_mutex
;
759 void build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
);
760 void wakeup_kswapd(struct zone
*zone
, int order
, enum zone_type classzone_idx
);
761 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
762 int classzone_idx
, unsigned int alloc_flags
,
764 bool zone_watermark_ok(struct zone
*z
, unsigned int order
,
765 unsigned long mark
, int classzone_idx
,
766 unsigned int alloc_flags
);
767 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
768 unsigned long mark
, int classzone_idx
);
769 enum memmap_context
{
773 extern void init_currently_empty_zone(struct zone
*zone
, unsigned long start_pfn
,
776 extern void lruvec_init(struct lruvec
*lruvec
);
778 static inline struct pglist_data
*lruvec_pgdat(struct lruvec
*lruvec
)
781 return lruvec
->pgdat
;
783 return container_of(lruvec
, struct pglist_data
, lruvec
);
787 extern unsigned long lruvec_lru_size(struct lruvec
*lruvec
, enum lru_list lru
, int zone_idx
);
789 #ifdef CONFIG_HAVE_MEMORY_PRESENT
790 void memory_present(int nid
, unsigned long start
, unsigned long end
);
792 static inline void memory_present(int nid
, unsigned long start
, unsigned long end
) {}
795 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
796 int local_memory_node(int node_id
);
798 static inline int local_memory_node(int node_id
) { return node_id
; };
801 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
802 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
806 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
808 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
811 * Returns true if a zone has pages managed by the buddy allocator.
812 * All the reclaim decisions have to use this function rather than
813 * populated_zone(). If the whole zone is reserved then we can easily
814 * end up with populated_zone() && !managed_zone().
816 static inline bool managed_zone(struct zone
*zone
)
818 return zone
->managed_pages
;
821 /* Returns true if a zone has memory */
822 static inline bool populated_zone(struct zone
*zone
)
824 return zone
->present_pages
;
827 extern int movable_zone
;
829 #ifdef CONFIG_HIGHMEM
830 static inline int zone_movable_is_highmem(void)
832 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
833 return movable_zone
== ZONE_HIGHMEM
;
835 return (ZONE_MOVABLE
- 1) == ZONE_HIGHMEM
;
840 static inline int is_highmem_idx(enum zone_type idx
)
842 #ifdef CONFIG_HIGHMEM
843 return (idx
== ZONE_HIGHMEM
||
844 (idx
== ZONE_MOVABLE
&& zone_movable_is_highmem()));
851 * is_highmem - helper function to quickly check if a struct zone is a
852 * highmem zone or not. This is an attempt to keep references
853 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
854 * @zone - pointer to struct zone variable
856 static inline int is_highmem(struct zone
*zone
)
858 #ifdef CONFIG_HIGHMEM
859 return is_highmem_idx(zone_idx(zone
));
865 /* These two functions are used to setup the per zone pages min values */
867 int min_free_kbytes_sysctl_handler(struct ctl_table
*, int,
868 void __user
*, size_t *, loff_t
*);
869 int watermark_scale_factor_sysctl_handler(struct ctl_table
*, int,
870 void __user
*, size_t *, loff_t
*);
871 extern int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1];
872 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*, int,
873 void __user
*, size_t *, loff_t
*);
874 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*, int,
875 void __user
*, size_t *, loff_t
*);
876 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*, int,
877 void __user
*, size_t *, loff_t
*);
878 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*, int,
879 void __user
*, size_t *, loff_t
*);
881 extern int numa_zonelist_order_handler(struct ctl_table
*, int,
882 void __user
*, size_t *, loff_t
*);
883 extern char numa_zonelist_order
[];
884 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
886 #ifndef CONFIG_NEED_MULTIPLE_NODES
888 extern struct pglist_data contig_page_data
;
889 #define NODE_DATA(nid) (&contig_page_data)
890 #define NODE_MEM_MAP(nid) mem_map
892 #else /* CONFIG_NEED_MULTIPLE_NODES */
894 #include <asm/mmzone.h>
896 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
898 extern struct pglist_data
*first_online_pgdat(void);
899 extern struct pglist_data
*next_online_pgdat(struct pglist_data
*pgdat
);
900 extern struct zone
*next_zone(struct zone
*zone
);
903 * for_each_online_pgdat - helper macro to iterate over all online nodes
904 * @pgdat - pointer to a pg_data_t variable
906 #define for_each_online_pgdat(pgdat) \
907 for (pgdat = first_online_pgdat(); \
909 pgdat = next_online_pgdat(pgdat))
911 * for_each_zone - helper macro to iterate over all memory zones
912 * @zone - pointer to struct zone variable
914 * The user only needs to declare the zone variable, for_each_zone
917 #define for_each_zone(zone) \
918 for (zone = (first_online_pgdat())->node_zones; \
920 zone = next_zone(zone))
922 #define for_each_populated_zone(zone) \
923 for (zone = (first_online_pgdat())->node_zones; \
925 zone = next_zone(zone)) \
926 if (!populated_zone(zone)) \
930 static inline struct zone
*zonelist_zone(struct zoneref
*zoneref
)
932 return zoneref
->zone
;
935 static inline int zonelist_zone_idx(struct zoneref
*zoneref
)
937 return zoneref
->zone_idx
;
940 static inline int zonelist_node_idx(struct zoneref
*zoneref
)
943 /* zone_to_nid not available in this context */
944 return zoneref
->zone
->node
;
947 #endif /* CONFIG_NUMA */
950 struct zoneref
*__next_zones_zonelist(struct zoneref
*z
,
951 enum zone_type highest_zoneidx
,
955 * 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
956 * @z - The cursor used as a starting point for the search
957 * @highest_zoneidx - The zone index of the highest zone to return
958 * @nodes - An optional nodemask to filter the zonelist with
960 * This function returns the next zone at or below a given zone index that is
961 * within the allowed nodemask using a cursor as the starting point for the
962 * search. The zoneref returned is a cursor that represents the current zone
963 * being examined. It should be advanced by one before calling
964 * next_zones_zonelist again.
966 static __always_inline
struct zoneref
*next_zones_zonelist(struct zoneref
*z
,
967 enum zone_type highest_zoneidx
,
970 if (likely(!nodes
&& zonelist_zone_idx(z
) <= highest_zoneidx
))
972 return __next_zones_zonelist(z
, highest_zoneidx
, nodes
);
976 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
977 * @zonelist - The zonelist to search for a suitable zone
978 * @highest_zoneidx - The zone index of the highest zone to return
979 * @nodes - An optional nodemask to filter the zonelist with
980 * @return - Zoneref pointer for the first suitable zone found (see below)
982 * This function returns the first zone at or below a given zone index that is
983 * within the allowed nodemask. The zoneref returned is a cursor that can be
984 * used to iterate the zonelist with next_zones_zonelist by advancing it by
985 * one before calling.
987 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
988 * never NULL). This may happen either genuinely, or due to concurrent nodemask
989 * update due to cpuset modification.
991 static inline struct zoneref
*first_zones_zonelist(struct zonelist
*zonelist
,
992 enum zone_type highest_zoneidx
,
995 return next_zones_zonelist(zonelist
->_zonerefs
,
996 highest_zoneidx
, nodes
);
1000 * for_each_zone_zonelist_nodemask - helper macro to iterate over valid zones in a zonelist at or below a given zone index and within a nodemask
1001 * @zone - The current zone in the iterator
1002 * @z - The current pointer within zonelist->zones being iterated
1003 * @zlist - The zonelist being iterated
1004 * @highidx - The zone index of the highest zone to return
1005 * @nodemask - Nodemask allowed by the allocator
1007 * This iterator iterates though all zones at or below a given zone index and
1008 * within a given nodemask
1010 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1011 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1013 z = next_zones_zonelist(++z, highidx, nodemask), \
1014 zone = zonelist_zone(z))
1016 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1017 for (zone = z->zone; \
1019 z = next_zones_zonelist(++z, highidx, nodemask), \
1020 zone = zonelist_zone(z))
1024 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1025 * @zone - The current zone in the iterator
1026 * @z - The current pointer within zonelist->zones being iterated
1027 * @zlist - The zonelist being iterated
1028 * @highidx - The zone index of the highest zone to return
1030 * This iterator iterates though all zones at or below a given zone index.
1032 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1033 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1035 #ifdef CONFIG_SPARSEMEM
1036 #include <asm/sparsemem.h>
1039 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1040 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1041 static inline unsigned long early_pfn_to_nid(unsigned long pfn
)
1043 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA
));
1048 #ifdef CONFIG_FLATMEM
1049 #define pfn_to_nid(pfn) (0)
1052 #ifdef CONFIG_SPARSEMEM
1055 * SECTION_SHIFT #bits space required to store a section #
1057 * PA_SECTION_SHIFT physical address to/from section number
1058 * PFN_SECTION_SHIFT pfn to/from section number
1060 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1061 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1063 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1065 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1066 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1068 #define SECTION_BLOCKFLAGS_BITS \
1069 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1071 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1072 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1075 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1076 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1078 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1079 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1083 struct mem_section
{
1085 * This is, logically, a pointer to an array of struct
1086 * pages. However, it is stored with some other magic.
1087 * (see sparse.c::sparse_init_one_section())
1089 * Additionally during early boot we encode node id of
1090 * the location of the section here to guide allocation.
1091 * (see sparse.c::memory_present())
1093 * Making it a UL at least makes someone do a cast
1094 * before using it wrong.
1096 unsigned long section_mem_map
;
1098 /* See declaration of similar field in struct zone */
1099 unsigned long *pageblock_flags
;
1100 #ifdef CONFIG_PAGE_EXTENSION
1102 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1103 * section. (see page_ext.h about this.)
1105 struct page_ext
*page_ext
;
1109 * WARNING: mem_section must be a power-of-2 in size for the
1110 * calculation and use of SECTION_ROOT_MASK to make sense.
1114 #ifdef CONFIG_SPARSEMEM_EXTREME
1115 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1117 #define SECTIONS_PER_ROOT 1
1120 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1121 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1122 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1124 #ifdef CONFIG_SPARSEMEM_EXTREME
1125 extern struct mem_section
**mem_section
;
1127 extern struct mem_section mem_section
[NR_SECTION_ROOTS
][SECTIONS_PER_ROOT
];
1130 static inline struct mem_section
*__nr_to_section(unsigned long nr
)
1132 #ifdef CONFIG_SPARSEMEM_EXTREME
1136 if (!mem_section
[SECTION_NR_TO_ROOT(nr
)])
1138 return &mem_section
[SECTION_NR_TO_ROOT(nr
)][nr
& SECTION_ROOT_MASK
];
1140 extern int __section_nr(struct mem_section
* ms
);
1141 extern unsigned long usemap_size(void);
1144 * We use the lower bits of the mem_map pointer to store
1145 * a little bit of information. There should be at least
1146 * 3 bits here due to 32-bit alignment.
1148 #define SECTION_MARKED_PRESENT (1UL<<0)
1149 #define SECTION_HAS_MEM_MAP (1UL<<1)
1150 #define SECTION_IS_ONLINE (1UL<<2)
1151 #define SECTION_MAP_LAST_BIT (1UL<<3)
1152 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1153 #define SECTION_NID_SHIFT 3
1155 static inline struct page
*__section_mem_map_addr(struct mem_section
*section
)
1157 unsigned long map
= section
->section_mem_map
;
1158 map
&= SECTION_MAP_MASK
;
1159 return (struct page
*)map
;
1162 static inline int present_section(struct mem_section
*section
)
1164 return (section
&& (section
->section_mem_map
& SECTION_MARKED_PRESENT
));
1167 static inline int present_section_nr(unsigned long nr
)
1169 return present_section(__nr_to_section(nr
));
1172 static inline int valid_section(struct mem_section
*section
)
1174 return (section
&& (section
->section_mem_map
& SECTION_HAS_MEM_MAP
));
1177 static inline int valid_section_nr(unsigned long nr
)
1179 return valid_section(__nr_to_section(nr
));
1182 static inline int online_section(struct mem_section
*section
)
1184 return (section
&& (section
->section_mem_map
& SECTION_IS_ONLINE
));
1187 static inline int online_section_nr(unsigned long nr
)
1189 return online_section(__nr_to_section(nr
));
1192 #ifdef CONFIG_MEMORY_HOTPLUG
1193 void online_mem_sections(unsigned long start_pfn
, unsigned long end_pfn
);
1194 #ifdef CONFIG_MEMORY_HOTREMOVE
1195 void offline_mem_sections(unsigned long start_pfn
, unsigned long end_pfn
);
1199 static inline struct mem_section
*__pfn_to_section(unsigned long pfn
)
1201 return __nr_to_section(pfn_to_section_nr(pfn
));
1204 extern int __highest_present_section_nr
;
1206 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1207 static inline int pfn_valid(unsigned long pfn
)
1209 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1211 return valid_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1215 static inline int pfn_present(unsigned long pfn
)
1217 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1219 return present_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1223 * These are _only_ used during initialisation, therefore they
1224 * can use __initdata ... They could have names to indicate
1228 #define pfn_to_nid(pfn) \
1230 unsigned long __pfn_to_nid_pfn = (pfn); \
1231 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1234 #define pfn_to_nid(pfn) (0)
1237 #define early_pfn_valid(pfn) pfn_valid(pfn)
1238 void sparse_init(void);
1240 #define sparse_init() do {} while (0)
1241 #define sparse_index_init(_sec, _nid) do {} while (0)
1242 #endif /* CONFIG_SPARSEMEM */
1245 * During memory init memblocks map pfns to nids. The search is expensive and
1246 * this caches recent lookups. The implementation of __early_pfn_to_nid
1247 * may treat start/end as pfns or sections.
1249 struct mminit_pfnnid_cache
{
1250 unsigned long last_start
;
1251 unsigned long last_end
;
1255 #ifndef early_pfn_valid
1256 #define early_pfn_valid(pfn) (1)
1259 void memory_present(int nid
, unsigned long start
, unsigned long end
);
1260 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
1263 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1264 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1265 * pfn_valid_within() should be used in this case; we optimise this away
1266 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1268 #ifdef CONFIG_HOLES_IN_ZONE
1269 #define pfn_valid_within(pfn) pfn_valid(pfn)
1271 #define pfn_valid_within(pfn) (1)
1274 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1276 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1277 * associated with it or not. This means that a struct page exists for this
1278 * pfn. The caller cannot assume the page is fully initialized in general.
1279 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1280 * will ensure the struct page is fully online and initialized. Special pages
1281 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1283 * In FLATMEM, it is expected that holes always have valid memmap as long as
1284 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1285 * that a valid section has a memmap for the entire section.
1287 * However, an ARM, and maybe other embedded architectures in the future
1288 * free memmap backing holes to save memory on the assumption the memmap is
1289 * never used. The page_zone linkages are then broken even though pfn_valid()
1290 * returns true. A walker of the full memmap must then do this additional
1291 * check to ensure the memmap they are looking at is sane by making sure
1292 * the zone and PFN linkages are still valid. This is expensive, but walkers
1293 * of the full memmap are extremely rare.
1295 bool memmap_valid_within(unsigned long pfn
,
1296 struct page
*page
, struct zone
*zone
);
1298 static inline bool memmap_valid_within(unsigned long pfn
,
1299 struct page
*page
, struct zone
*zone
)
1303 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1305 #endif /* !__GENERATING_BOUNDS.H */
1306 #endif /* !__ASSEMBLY__ */
1307 #endif /* _LINUX_MMZONE_H */