1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MMZONE_H
3 #define _LINUX_MMZONE_H
6 #ifndef __GENERATING_BOUNDS_H
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <linux/cache.h>
13 #include <linux/threads.h>
14 #include <linux/numa.h>
15 #include <linux/init.h>
16 #include <linux/seqlock.h>
17 #include <linux/nodemask.h>
18 #include <linux/pageblock-flags.h>
19 #include <linux/page-flags-layout.h>
20 #include <linux/atomic.h>
21 #include <linux/mm_types.h>
22 #include <linux/page-flags.h>
25 /* Free memory management - zoned buddy allocator. */
26 #ifndef CONFIG_FORCE_MAX_ZONEORDER
29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
36 * coalesce naturally under reasonable reclaim pressure and those which
39 #define PAGE_ALLOC_COSTLY_ORDER 3
45 MIGRATE_PCPTYPES
, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC
= MIGRATE_PCPTYPES
,
49 * MIGRATE_CMA migration type is designed to mimic the way
50 * ZONE_MOVABLE works. Only movable pages can be allocated
51 * from MIGRATE_CMA pageblocks and page allocator never
52 * implicitly change migration type of MIGRATE_CMA pageblock.
54 * The way to use it is to change migratetype of a range of
55 * pageblocks to MIGRATE_CMA which can be done by
56 * __free_pageblock_cma() function. What is important though
57 * is that a range of pageblocks must be aligned to
58 * MAX_ORDER_NR_PAGES should biggest page be bigger then
63 #ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE
, /* can't allocate from here */
69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70 extern const char * const migratetype_names
[MIGRATE_TYPES
];
73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76 # define is_migrate_cma(migratetype) false
77 # define is_migrate_cma_page(_page) false
80 static inline bool is_migrate_movable(int mt
)
82 return is_migrate_cma(mt
) || mt
== MIGRATE_MOVABLE
;
85 #define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
89 extern int page_group_by_mobility_disabled
;
91 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
92 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
94 #define get_pageblock_migratetype(page) \
95 get_pfnblock_flags_mask(page, page_to_pfn(page), \
96 PB_migrate_end, MIGRATETYPE_MASK)
99 struct list_head free_list
[MIGRATE_TYPES
];
100 unsigned long nr_free
;
103 /* Used for pages not on another list */
104 static inline void add_to_free_area(struct page
*page
, struct free_area
*area
,
107 list_add(&page
->lru
, &area
->free_list
[migratetype
]);
111 /* Used for pages not on another list */
112 static inline void add_to_free_area_tail(struct page
*page
, struct free_area
*area
,
115 list_add_tail(&page
->lru
, &area
->free_list
[migratetype
]);
119 #ifdef CONFIG_SHUFFLE_PAGE_ALLOCATOR
120 /* Used to preserve page allocation order entropy */
121 void add_to_free_area_random(struct page
*page
, struct free_area
*area
,
124 static inline void add_to_free_area_random(struct page
*page
,
125 struct free_area
*area
, int migratetype
)
127 add_to_free_area(page
, area
, migratetype
);
131 /* Used for pages which are on another list */
132 static inline void move_to_free_area(struct page
*page
, struct free_area
*area
,
135 list_move(&page
->lru
, &area
->free_list
[migratetype
]);
138 static inline struct page
*get_page_from_free_area(struct free_area
*area
,
141 return list_first_entry_or_null(&area
->free_list
[migratetype
],
145 static inline void del_page_from_free_area(struct page
*page
,
146 struct free_area
*area
)
148 list_del(&page
->lru
);
149 __ClearPageBuddy(page
);
150 set_page_private(page
, 0);
154 static inline bool free_area_empty(struct free_area
*area
, int migratetype
)
156 return list_empty(&area
->free_list
[migratetype
]);
162 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
163 * So add a wild amount of padding here to ensure that they fall into separate
164 * cachelines. There are very few zone structures in the machine, so space
165 * consumption is not a concern here.
167 #if defined(CONFIG_SMP)
168 struct zone_padding
{
170 } ____cacheline_internodealigned_in_smp
;
171 #define ZONE_PADDING(name) struct zone_padding name;
173 #define ZONE_PADDING(name)
177 enum numa_stat_item
{
178 NUMA_HIT
, /* allocated in intended node */
179 NUMA_MISS
, /* allocated in non intended node */
180 NUMA_FOREIGN
, /* was intended here, hit elsewhere */
181 NUMA_INTERLEAVE_HIT
, /* interleaver preferred this zone */
182 NUMA_LOCAL
, /* allocation from local node */
183 NUMA_OTHER
, /* allocation from other node */
184 NR_VM_NUMA_STAT_ITEMS
187 #define NR_VM_NUMA_STAT_ITEMS 0
190 enum zone_stat_item
{
191 /* First 128 byte cacheline (assuming 64 bit words) */
193 NR_ZONE_LRU_BASE
, /* Used only for compaction and reclaim retry */
194 NR_ZONE_INACTIVE_ANON
= NR_ZONE_LRU_BASE
,
196 NR_ZONE_INACTIVE_FILE
,
199 NR_ZONE_WRITE_PENDING
, /* Count of dirty, writeback and unstable pages */
200 NR_MLOCK
, /* mlock()ed pages found and moved off LRU */
201 NR_PAGETABLE
, /* used for pagetables */
202 NR_KERNEL_STACK_KB
, /* measured in KiB */
203 /* Second 128 byte cacheline */
205 #if IS_ENABLED(CONFIG_ZSMALLOC)
206 NR_ZSPAGES
, /* allocated in zsmalloc */
209 NR_VM_ZONE_STAT_ITEMS
};
211 enum node_stat_item
{
213 NR_INACTIVE_ANON
= NR_LRU_BASE
, /* must match order of LRU_[IN]ACTIVE */
214 NR_ACTIVE_ANON
, /* " " " " " */
215 NR_INACTIVE_FILE
, /* " " " " " */
216 NR_ACTIVE_FILE
, /* " " " " " */
217 NR_UNEVICTABLE
, /* " " " " " */
219 NR_SLAB_UNRECLAIMABLE
,
220 NR_ISOLATED_ANON
, /* Temporary isolated pages from anon lru */
221 NR_ISOLATED_FILE
, /* Temporary isolated pages from file lru */
226 WORKINGSET_NODERECLAIM
,
227 NR_ANON_MAPPED
, /* Mapped anonymous pages */
228 NR_FILE_MAPPED
, /* pagecache pages mapped into pagetables.
229 only modified from process context */
233 NR_WRITEBACK_TEMP
, /* Writeback using temporary buffers */
234 NR_SHMEM
, /* shmem pages (included tmpfs/GEM pages) */
238 NR_UNSTABLE_NFS
, /* NFS unstable pages */
240 NR_VMSCAN_IMMEDIATE
, /* Prioritise for reclaim when writeback ends */
241 NR_DIRTIED
, /* page dirtyings since bootup */
242 NR_WRITTEN
, /* page writings since bootup */
243 NR_KERNEL_MISC_RECLAIMABLE
, /* reclaimable non-slab kernel pages */
244 NR_VM_NODE_STAT_ITEMS
248 * We do arithmetic on the LRU lists in various places in the code,
249 * so it is important to keep the active lists LRU_ACTIVE higher in
250 * the array than the corresponding inactive lists, and to keep
251 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
253 * This has to be kept in sync with the statistics in zone_stat_item
254 * above and the descriptions in vmstat_text in mm/vmstat.c
261 LRU_INACTIVE_ANON
= LRU_BASE
,
262 LRU_ACTIVE_ANON
= LRU_BASE
+ LRU_ACTIVE
,
263 LRU_INACTIVE_FILE
= LRU_BASE
+ LRU_FILE
,
264 LRU_ACTIVE_FILE
= LRU_BASE
+ LRU_FILE
+ LRU_ACTIVE
,
269 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
271 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
273 static inline int is_file_lru(enum lru_list lru
)
275 return (lru
== LRU_INACTIVE_FILE
|| lru
== LRU_ACTIVE_FILE
);
278 static inline int is_active_lru(enum lru_list lru
)
280 return (lru
== LRU_ACTIVE_ANON
|| lru
== LRU_ACTIVE_FILE
);
283 struct zone_reclaim_stat
{
285 * The pageout code in vmscan.c keeps track of how many of the
286 * mem/swap backed and file backed pages are referenced.
287 * The higher the rotated/scanned ratio, the more valuable
290 * The anon LRU stats live in [0], file LRU stats in [1]
292 unsigned long recent_rotated
[2];
293 unsigned long recent_scanned
[2];
297 struct list_head lists
[NR_LRU_LISTS
];
298 struct zone_reclaim_stat reclaim_stat
;
299 /* Evictions & activations on the inactive file list */
300 atomic_long_t inactive_age
;
301 /* Refaults at the time of last reclaim cycle */
302 unsigned long refaults
;
304 struct pglist_data
*pgdat
;
308 /* Isolate unmapped file */
309 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
310 /* Isolate for asynchronous migration */
311 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
312 /* Isolate unevictable pages */
313 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
315 /* LRU Isolation modes. */
316 typedef unsigned __bitwise isolate_mode_t
;
318 enum zone_watermarks
{
325 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
326 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
327 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
328 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
330 struct per_cpu_pages
{
331 int count
; /* number of pages in the list */
332 int high
; /* high watermark, emptying needed */
333 int batch
; /* chunk size for buddy add/remove */
335 /* Lists of pages, one per migrate type stored on the pcp-lists */
336 struct list_head lists
[MIGRATE_PCPTYPES
];
339 struct per_cpu_pageset
{
340 struct per_cpu_pages pcp
;
343 u16 vm_numa_stat_diff
[NR_VM_NUMA_STAT_ITEMS
];
347 s8 vm_stat_diff
[NR_VM_ZONE_STAT_ITEMS
];
351 struct per_cpu_nodestat
{
353 s8 vm_node_stat_diff
[NR_VM_NODE_STAT_ITEMS
];
356 #endif /* !__GENERATING_BOUNDS.H */
359 #ifdef CONFIG_ZONE_DMA
361 * ZONE_DMA is used when there are devices that are not able
362 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
363 * carve out the portion of memory that is needed for these devices.
364 * The range is arch specific.
369 * ---------------------------
370 * parisc, ia64, sparc <4G
373 * alpha Unlimited or 0-16MB.
375 * i386, x86_64 and multiple other arches
380 #ifdef CONFIG_ZONE_DMA32
382 * x86_64 needs two ZONE_DMAs because it supports devices that are
383 * only able to do DMA to the lower 16M but also 32 bit devices that
384 * can only do DMA areas below 4G.
389 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
390 * performed on pages in ZONE_NORMAL if the DMA devices support
391 * transfers to all addressable memory.
394 #ifdef CONFIG_HIGHMEM
396 * A memory area that is only addressable by the kernel through
397 * mapping portions into its own address space. This is for example
398 * used by i386 to allow the kernel to address the memory beyond
399 * 900MB. The kernel will set up special mappings (page
400 * table entries on i386) for each page that the kernel needs to
406 #ifdef CONFIG_ZONE_DEVICE
413 #ifndef __GENERATING_BOUNDS_H
416 /* Read-mostly fields */
418 /* zone watermarks, access with *_wmark_pages(zone) macros */
419 unsigned long _watermark
[NR_WMARK
];
420 unsigned long watermark_boost
;
422 unsigned long nr_reserved_highatomic
;
425 * We don't know if the memory that we're going to allocate will be
426 * freeable or/and it will be released eventually, so to avoid totally
427 * wasting several GB of ram we must reserve some of the lower zone
428 * memory (otherwise we risk to run OOM on the lower zones despite
429 * there being tons of freeable ram on the higher zones). This array is
430 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
433 long lowmem_reserve
[MAX_NR_ZONES
];
438 struct pglist_data
*zone_pgdat
;
439 struct per_cpu_pageset __percpu
*pageset
;
441 #ifndef CONFIG_SPARSEMEM
443 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
444 * In SPARSEMEM, this map is stored in struct mem_section
446 unsigned long *pageblock_flags
;
447 #endif /* CONFIG_SPARSEMEM */
449 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
450 unsigned long zone_start_pfn
;
453 * spanned_pages is the total pages spanned by the zone, including
454 * holes, which is calculated as:
455 * spanned_pages = zone_end_pfn - zone_start_pfn;
457 * present_pages is physical pages existing within the zone, which
459 * present_pages = spanned_pages - absent_pages(pages in holes);
461 * managed_pages is present pages managed by the buddy system, which
462 * is calculated as (reserved_pages includes pages allocated by the
463 * bootmem allocator):
464 * managed_pages = present_pages - reserved_pages;
466 * So present_pages may be used by memory hotplug or memory power
467 * management logic to figure out unmanaged pages by checking
468 * (present_pages - managed_pages). And managed_pages should be used
469 * by page allocator and vm scanner to calculate all kinds of watermarks
474 * zone_start_pfn and spanned_pages are protected by span_seqlock.
475 * It is a seqlock because it has to be read outside of zone->lock,
476 * and it is done in the main allocator path. But, it is written
477 * quite infrequently.
479 * The span_seq lock is declared along with zone->lock because it is
480 * frequently read in proximity to zone->lock. It's good to
481 * give them a chance of being in the same cacheline.
483 * Write access to present_pages at runtime should be protected by
484 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
485 * present_pages should get_online_mems() to get a stable value.
487 atomic_long_t managed_pages
;
488 unsigned long spanned_pages
;
489 unsigned long present_pages
;
493 #ifdef CONFIG_MEMORY_ISOLATION
495 * Number of isolated pageblock. It is used to solve incorrect
496 * freepage counting problem due to racy retrieving migratetype
497 * of pageblock. Protected by zone->lock.
499 unsigned long nr_isolate_pageblock
;
502 #ifdef CONFIG_MEMORY_HOTPLUG
503 /* see spanned/present_pages for more description */
504 seqlock_t span_seqlock
;
509 /* Write-intensive fields used from the page allocator */
512 /* free areas of different sizes */
513 struct free_area free_area
[MAX_ORDER
];
515 /* zone flags, see below */
518 /* Primarily protects free_area */
521 /* Write-intensive fields used by compaction and vmstats. */
525 * When free pages are below this point, additional steps are taken
526 * when reading the number of free pages to avoid per-cpu counter
527 * drift allowing watermarks to be breached
529 unsigned long percpu_drift_mark
;
531 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
532 /* pfn where compaction free scanner should start */
533 unsigned long compact_cached_free_pfn
;
534 /* pfn where async and sync compaction migration scanner should start */
535 unsigned long compact_cached_migrate_pfn
[2];
536 unsigned long compact_init_migrate_pfn
;
537 unsigned long compact_init_free_pfn
;
540 #ifdef CONFIG_COMPACTION
542 * On compaction failure, 1<<compact_defer_shift compactions
543 * are skipped before trying again. The number attempted since
544 * last failure is tracked with compact_considered.
546 unsigned int compact_considered
;
547 unsigned int compact_defer_shift
;
548 int compact_order_failed
;
551 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
552 /* Set to true when the PG_migrate_skip bits should be cleared */
553 bool compact_blockskip_flush
;
559 /* Zone statistics */
560 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
561 atomic_long_t vm_numa_stat
[NR_VM_NUMA_STAT_ITEMS
];
562 } ____cacheline_internodealigned_in_smp
;
565 PGDAT_CONGESTED
, /* pgdat has many dirty pages backed by
568 PGDAT_DIRTY
, /* reclaim scanning has recently found
569 * many dirty file pages at the tail
572 PGDAT_WRITEBACK
, /* reclaim scanning has recently found
573 * many pages under writeback
575 PGDAT_RECLAIM_LOCKED
, /* prevents concurrent reclaim */
579 ZONE_BOOSTED_WATERMARK
, /* zone recently boosted watermarks.
580 * Cleared when kswapd is woken.
584 static inline unsigned long zone_managed_pages(struct zone
*zone
)
586 return (unsigned long)atomic_long_read(&zone
->managed_pages
);
589 static inline unsigned long zone_end_pfn(const struct zone
*zone
)
591 return zone
->zone_start_pfn
+ zone
->spanned_pages
;
594 static inline bool zone_spans_pfn(const struct zone
*zone
, unsigned long pfn
)
596 return zone
->zone_start_pfn
<= pfn
&& pfn
< zone_end_pfn(zone
);
599 static inline bool zone_is_initialized(struct zone
*zone
)
601 return zone
->initialized
;
604 static inline bool zone_is_empty(struct zone
*zone
)
606 return zone
->spanned_pages
== 0;
610 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
611 * intersection with the given zone
613 static inline bool zone_intersects(struct zone
*zone
,
614 unsigned long start_pfn
, unsigned long nr_pages
)
616 if (zone_is_empty(zone
))
618 if (start_pfn
>= zone_end_pfn(zone
) ||
619 start_pfn
+ nr_pages
<= zone
->zone_start_pfn
)
626 * The "priority" of VM scanning is how much of the queues we will scan in one
627 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
628 * queues ("queue_length >> 12") during an aging round.
630 #define DEF_PRIORITY 12
632 /* Maximum number of zones on a zonelist */
633 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
636 ZONELIST_FALLBACK
, /* zonelist with fallback */
639 * The NUMA zonelists are doubled because we need zonelists that
640 * restrict the allocations to a single node for __GFP_THISNODE.
642 ZONELIST_NOFALLBACK
, /* zonelist without fallback (__GFP_THISNODE) */
648 * This struct contains information about a zone in a zonelist. It is stored
649 * here to avoid dereferences into large structures and lookups of tables
652 struct zone
*zone
; /* Pointer to actual zone */
653 int zone_idx
; /* zone_idx(zoneref->zone) */
657 * One allocation request operates on a zonelist. A zonelist
658 * is a list of zones, the first one is the 'goal' of the
659 * allocation, the other zones are fallback zones, in decreasing
662 * To speed the reading of the zonelist, the zonerefs contain the zone index
663 * of the entry being read. Helper functions to access information given
664 * a struct zoneref are
666 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
667 * zonelist_zone_idx() - Return the index of the zone for an entry
668 * zonelist_node_idx() - Return the index of the node for an entry
671 struct zoneref _zonerefs
[MAX_ZONES_PER_ZONELIST
+ 1];
674 #ifndef CONFIG_DISCONTIGMEM
675 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
676 extern struct page
*mem_map
;
680 * On NUMA machines, each NUMA node would have a pg_data_t to describe
681 * it's memory layout. On UMA machines there is a single pglist_data which
682 * describes the whole memory.
684 * Memory statistics and page replacement data structures are maintained on a
688 typedef struct pglist_data
{
689 struct zone node_zones
[MAX_NR_ZONES
];
690 struct zonelist node_zonelists
[MAX_ZONELISTS
];
692 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
693 struct page
*node_mem_map
;
694 #ifdef CONFIG_PAGE_EXTENSION
695 struct page_ext
*node_page_ext
;
698 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
700 * Must be held any time you expect node_start_pfn,
701 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
703 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
704 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
705 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
707 * Nests above zone->lock and zone->span_seqlock
709 spinlock_t node_size_lock
;
711 unsigned long node_start_pfn
;
712 unsigned long node_present_pages
; /* total number of physical pages */
713 unsigned long node_spanned_pages
; /* total size of physical page
714 range, including holes */
716 wait_queue_head_t kswapd_wait
;
717 wait_queue_head_t pfmemalloc_wait
;
718 struct task_struct
*kswapd
; /* Protected by
719 mem_hotplug_begin/end() */
721 enum zone_type kswapd_classzone_idx
;
723 int kswapd_failures
; /* Number of 'reclaimed == 0' runs */
725 #ifdef CONFIG_COMPACTION
726 int kcompactd_max_order
;
727 enum zone_type kcompactd_classzone_idx
;
728 wait_queue_head_t kcompactd_wait
;
729 struct task_struct
*kcompactd
;
732 * This is a per-node reserve of pages that are not available
733 * to userspace allocations.
735 unsigned long totalreserve_pages
;
739 * zone reclaim becomes active if more unmapped pages exist.
741 unsigned long min_unmapped_pages
;
742 unsigned long min_slab_pages
;
743 #endif /* CONFIG_NUMA */
745 /* Write-intensive fields used by page reclaim */
749 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
751 * If memory initialisation on large machines is deferred then this
752 * is the first PFN that needs to be initialised.
754 unsigned long first_deferred_pfn
;
755 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
757 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
758 spinlock_t split_queue_lock
;
759 struct list_head split_queue
;
760 unsigned long split_queue_len
;
763 /* Fields commonly accessed by the page reclaim scanner */
764 struct lruvec lruvec
;
770 /* Per-node vmstats */
771 struct per_cpu_nodestat __percpu
*per_cpu_nodestats
;
772 atomic_long_t vm_stat
[NR_VM_NODE_STAT_ITEMS
];
775 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
776 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
777 #ifdef CONFIG_FLAT_NODE_MEM_MAP
778 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
780 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
782 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
784 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
785 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
787 static inline struct lruvec
*node_lruvec(struct pglist_data
*pgdat
)
789 return &pgdat
->lruvec
;
792 static inline unsigned long pgdat_end_pfn(pg_data_t
*pgdat
)
794 return pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
797 static inline bool pgdat_is_empty(pg_data_t
*pgdat
)
799 return !pgdat
->node_start_pfn
&& !pgdat
->node_spanned_pages
;
802 #include <linux/memory_hotplug.h>
804 void build_all_zonelists(pg_data_t
*pgdat
);
805 void wakeup_kswapd(struct zone
*zone
, gfp_t gfp_mask
, int order
,
806 enum zone_type classzone_idx
);
807 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
808 int classzone_idx
, unsigned int alloc_flags
,
810 bool zone_watermark_ok(struct zone
*z
, unsigned int order
,
811 unsigned long mark
, int classzone_idx
,
812 unsigned int alloc_flags
);
813 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
814 unsigned long mark
, int classzone_idx
);
815 enum memmap_context
{
819 extern void init_currently_empty_zone(struct zone
*zone
, unsigned long start_pfn
,
822 extern void lruvec_init(struct lruvec
*lruvec
);
824 static inline struct pglist_data
*lruvec_pgdat(struct lruvec
*lruvec
)
827 return lruvec
->pgdat
;
829 return container_of(lruvec
, struct pglist_data
, lruvec
);
833 extern unsigned long lruvec_lru_size(struct lruvec
*lruvec
, enum lru_list lru
, int zone_idx
);
835 #ifdef CONFIG_HAVE_MEMORY_PRESENT
836 void memory_present(int nid
, unsigned long start
, unsigned long end
);
838 static inline void memory_present(int nid
, unsigned long start
, unsigned long end
) {}
841 #if defined(CONFIG_SPARSEMEM)
842 void memblocks_present(void);
844 static inline void memblocks_present(void) {}
847 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
848 int local_memory_node(int node_id
);
850 static inline int local_memory_node(int node_id
) { return node_id
; };
854 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
856 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
858 #ifdef CONFIG_ZONE_DEVICE
859 static inline bool is_dev_zone(const struct zone
*zone
)
861 return zone_idx(zone
) == ZONE_DEVICE
;
864 static inline bool is_dev_zone(const struct zone
*zone
)
871 * Returns true if a zone has pages managed by the buddy allocator.
872 * All the reclaim decisions have to use this function rather than
873 * populated_zone(). If the whole zone is reserved then we can easily
874 * end up with populated_zone() && !managed_zone().
876 static inline bool managed_zone(struct zone
*zone
)
878 return zone_managed_pages(zone
);
881 /* Returns true if a zone has memory */
882 static inline bool populated_zone(struct zone
*zone
)
884 return zone
->present_pages
;
888 static inline int zone_to_nid(struct zone
*zone
)
893 static inline void zone_set_nid(struct zone
*zone
, int nid
)
898 static inline int zone_to_nid(struct zone
*zone
)
903 static inline void zone_set_nid(struct zone
*zone
, int nid
) {}
906 extern int movable_zone
;
908 #ifdef CONFIG_HIGHMEM
909 static inline int zone_movable_is_highmem(void)
911 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
912 return movable_zone
== ZONE_HIGHMEM
;
914 return (ZONE_MOVABLE
- 1) == ZONE_HIGHMEM
;
919 static inline int is_highmem_idx(enum zone_type idx
)
921 #ifdef CONFIG_HIGHMEM
922 return (idx
== ZONE_HIGHMEM
||
923 (idx
== ZONE_MOVABLE
&& zone_movable_is_highmem()));
930 * is_highmem - helper function to quickly check if a struct zone is a
931 * highmem zone or not. This is an attempt to keep references
932 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
933 * @zone - pointer to struct zone variable
935 static inline int is_highmem(struct zone
*zone
)
937 #ifdef CONFIG_HIGHMEM
938 return is_highmem_idx(zone_idx(zone
));
944 /* These two functions are used to setup the per zone pages min values */
946 int min_free_kbytes_sysctl_handler(struct ctl_table
*, int,
947 void __user
*, size_t *, loff_t
*);
948 int watermark_boost_factor_sysctl_handler(struct ctl_table
*, int,
949 void __user
*, size_t *, loff_t
*);
950 int watermark_scale_factor_sysctl_handler(struct ctl_table
*, int,
951 void __user
*, size_t *, loff_t
*);
952 extern int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
];
953 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*, int,
954 void __user
*, size_t *, loff_t
*);
955 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*, int,
956 void __user
*, size_t *, loff_t
*);
957 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*, int,
958 void __user
*, size_t *, loff_t
*);
959 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*, int,
960 void __user
*, size_t *, loff_t
*);
962 extern int numa_zonelist_order_handler(struct ctl_table
*, int,
963 void __user
*, size_t *, loff_t
*);
964 extern char numa_zonelist_order
[];
965 #define NUMA_ZONELIST_ORDER_LEN 16
967 #ifndef CONFIG_NEED_MULTIPLE_NODES
969 extern struct pglist_data contig_page_data
;
970 #define NODE_DATA(nid) (&contig_page_data)
971 #define NODE_MEM_MAP(nid) mem_map
973 #else /* CONFIG_NEED_MULTIPLE_NODES */
975 #include <asm/mmzone.h>
977 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
979 extern struct pglist_data
*first_online_pgdat(void);
980 extern struct pglist_data
*next_online_pgdat(struct pglist_data
*pgdat
);
981 extern struct zone
*next_zone(struct zone
*zone
);
984 * for_each_online_pgdat - helper macro to iterate over all online nodes
985 * @pgdat - pointer to a pg_data_t variable
987 #define for_each_online_pgdat(pgdat) \
988 for (pgdat = first_online_pgdat(); \
990 pgdat = next_online_pgdat(pgdat))
992 * for_each_zone - helper macro to iterate over all memory zones
993 * @zone - pointer to struct zone variable
995 * The user only needs to declare the zone variable, for_each_zone
998 #define for_each_zone(zone) \
999 for (zone = (first_online_pgdat())->node_zones; \
1001 zone = next_zone(zone))
1003 #define for_each_populated_zone(zone) \
1004 for (zone = (first_online_pgdat())->node_zones; \
1006 zone = next_zone(zone)) \
1007 if (!populated_zone(zone)) \
1008 ; /* do nothing */ \
1011 static inline struct zone
*zonelist_zone(struct zoneref
*zoneref
)
1013 return zoneref
->zone
;
1016 static inline int zonelist_zone_idx(struct zoneref
*zoneref
)
1018 return zoneref
->zone_idx
;
1021 static inline int zonelist_node_idx(struct zoneref
*zoneref
)
1023 return zone_to_nid(zoneref
->zone
);
1026 struct zoneref
*__next_zones_zonelist(struct zoneref
*z
,
1027 enum zone_type highest_zoneidx
,
1031 * 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
1032 * @z - The cursor used as a starting point for the search
1033 * @highest_zoneidx - The zone index of the highest zone to return
1034 * @nodes - An optional nodemask to filter the zonelist with
1036 * This function returns the next zone at or below a given zone index that is
1037 * within the allowed nodemask using a cursor as the starting point for the
1038 * search. The zoneref returned is a cursor that represents the current zone
1039 * being examined. It should be advanced by one before calling
1040 * next_zones_zonelist again.
1042 static __always_inline
struct zoneref
*next_zones_zonelist(struct zoneref
*z
,
1043 enum zone_type highest_zoneidx
,
1046 if (likely(!nodes
&& zonelist_zone_idx(z
) <= highest_zoneidx
))
1048 return __next_zones_zonelist(z
, highest_zoneidx
, nodes
);
1052 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1053 * @zonelist - The zonelist to search for a suitable zone
1054 * @highest_zoneidx - The zone index of the highest zone to return
1055 * @nodes - An optional nodemask to filter the zonelist with
1056 * @return - Zoneref pointer for the first suitable zone found (see below)
1058 * This function returns the first zone at or below a given zone index that is
1059 * within the allowed nodemask. The zoneref returned is a cursor that can be
1060 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1061 * one before calling.
1063 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1064 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1065 * update due to cpuset modification.
1067 static inline struct zoneref
*first_zones_zonelist(struct zonelist
*zonelist
,
1068 enum zone_type highest_zoneidx
,
1071 return next_zones_zonelist(zonelist
->_zonerefs
,
1072 highest_zoneidx
, nodes
);
1076 * 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
1077 * @zone - The current zone in the iterator
1078 * @z - The current pointer within zonelist->zones being iterated
1079 * @zlist - The zonelist being iterated
1080 * @highidx - The zone index of the highest zone to return
1081 * @nodemask - Nodemask allowed by the allocator
1083 * This iterator iterates though all zones at or below a given zone index and
1084 * within a given nodemask
1086 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1087 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1089 z = next_zones_zonelist(++z, highidx, nodemask), \
1090 zone = zonelist_zone(z))
1092 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1093 for (zone = z->zone; \
1095 z = next_zones_zonelist(++z, highidx, nodemask), \
1096 zone = zonelist_zone(z))
1100 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1101 * @zone - The current zone in the iterator
1102 * @z - The current pointer within zonelist->zones being iterated
1103 * @zlist - The zonelist being iterated
1104 * @highidx - The zone index of the highest zone to return
1106 * This iterator iterates though all zones at or below a given zone index.
1108 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1109 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1111 #ifdef CONFIG_SPARSEMEM
1112 #include <asm/sparsemem.h>
1115 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1116 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1117 static inline unsigned long early_pfn_to_nid(unsigned long pfn
)
1119 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA
));
1124 #ifdef CONFIG_FLATMEM
1125 #define pfn_to_nid(pfn) (0)
1128 #ifdef CONFIG_SPARSEMEM
1131 * SECTION_SHIFT #bits space required to store a section #
1133 * PA_SECTION_SHIFT physical address to/from section number
1134 * PFN_SECTION_SHIFT pfn to/from section number
1136 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1137 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1139 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1141 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1142 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1144 #define SECTION_BLOCKFLAGS_BITS \
1145 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1147 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1148 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1151 static inline unsigned long pfn_to_section_nr(unsigned long pfn
)
1153 return pfn
>> PFN_SECTION_SHIFT
;
1155 static inline unsigned long section_nr_to_pfn(unsigned long sec
)
1157 return sec
<< PFN_SECTION_SHIFT
;
1160 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1161 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1165 struct mem_section
{
1167 * This is, logically, a pointer to an array of struct
1168 * pages. However, it is stored with some other magic.
1169 * (see sparse.c::sparse_init_one_section())
1171 * Additionally during early boot we encode node id of
1172 * the location of the section here to guide allocation.
1173 * (see sparse.c::memory_present())
1175 * Making it a UL at least makes someone do a cast
1176 * before using it wrong.
1178 unsigned long section_mem_map
;
1180 /* See declaration of similar field in struct zone */
1181 unsigned long *pageblock_flags
;
1182 #ifdef CONFIG_PAGE_EXTENSION
1184 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1185 * section. (see page_ext.h about this.)
1187 struct page_ext
*page_ext
;
1191 * WARNING: mem_section must be a power-of-2 in size for the
1192 * calculation and use of SECTION_ROOT_MASK to make sense.
1196 #ifdef CONFIG_SPARSEMEM_EXTREME
1197 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1199 #define SECTIONS_PER_ROOT 1
1202 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1203 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1204 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1206 #ifdef CONFIG_SPARSEMEM_EXTREME
1207 extern struct mem_section
**mem_section
;
1209 extern struct mem_section mem_section
[NR_SECTION_ROOTS
][SECTIONS_PER_ROOT
];
1212 static inline struct mem_section
*__nr_to_section(unsigned long nr
)
1214 #ifdef CONFIG_SPARSEMEM_EXTREME
1218 if (!mem_section
[SECTION_NR_TO_ROOT(nr
)])
1220 return &mem_section
[SECTION_NR_TO_ROOT(nr
)][nr
& SECTION_ROOT_MASK
];
1222 extern int __section_nr(struct mem_section
* ms
);
1223 extern unsigned long usemap_size(void);
1226 * We use the lower bits of the mem_map pointer to store
1227 * a little bit of information. The pointer is calculated
1228 * as mem_map - section_nr_to_pfn(pnum). The result is
1229 * aligned to the minimum alignment of the two values:
1230 * 1. All mem_map arrays are page-aligned.
1231 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1232 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1233 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1234 * worst combination is powerpc with 256k pages,
1235 * which results in PFN_SECTION_SHIFT equal 6.
1236 * To sum it up, at least 6 bits are available.
1238 #define SECTION_MARKED_PRESENT (1UL<<0)
1239 #define SECTION_HAS_MEM_MAP (1UL<<1)
1240 #define SECTION_IS_ONLINE (1UL<<2)
1241 #define SECTION_MAP_LAST_BIT (1UL<<3)
1242 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1243 #define SECTION_NID_SHIFT 3
1245 static inline struct page
*__section_mem_map_addr(struct mem_section
*section
)
1247 unsigned long map
= section
->section_mem_map
;
1248 map
&= SECTION_MAP_MASK
;
1249 return (struct page
*)map
;
1252 static inline int present_section(struct mem_section
*section
)
1254 return (section
&& (section
->section_mem_map
& SECTION_MARKED_PRESENT
));
1257 static inline int present_section_nr(unsigned long nr
)
1259 return present_section(__nr_to_section(nr
));
1262 static inline int valid_section(struct mem_section
*section
)
1264 return (section
&& (section
->section_mem_map
& SECTION_HAS_MEM_MAP
));
1267 static inline int valid_section_nr(unsigned long nr
)
1269 return valid_section(__nr_to_section(nr
));
1272 static inline int online_section(struct mem_section
*section
)
1274 return (section
&& (section
->section_mem_map
& SECTION_IS_ONLINE
));
1277 static inline int online_section_nr(unsigned long nr
)
1279 return online_section(__nr_to_section(nr
));
1282 #ifdef CONFIG_MEMORY_HOTPLUG
1283 void online_mem_sections(unsigned long start_pfn
, unsigned long end_pfn
);
1284 #ifdef CONFIG_MEMORY_HOTREMOVE
1285 void offline_mem_sections(unsigned long start_pfn
, unsigned long end_pfn
);
1289 static inline struct mem_section
*__pfn_to_section(unsigned long pfn
)
1291 return __nr_to_section(pfn_to_section_nr(pfn
));
1294 extern int __highest_present_section_nr
;
1296 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1297 static inline int pfn_valid(unsigned long pfn
)
1299 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1301 return valid_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1305 static inline int pfn_present(unsigned long pfn
)
1307 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1309 return present_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1313 * These are _only_ used during initialisation, therefore they
1314 * can use __initdata ... They could have names to indicate
1318 #define pfn_to_nid(pfn) \
1320 unsigned long __pfn_to_nid_pfn = (pfn); \
1321 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1324 #define pfn_to_nid(pfn) (0)
1327 #define early_pfn_valid(pfn) pfn_valid(pfn)
1328 void sparse_init(void);
1330 #define sparse_init() do {} while (0)
1331 #define sparse_index_init(_sec, _nid) do {} while (0)
1332 #define pfn_present pfn_valid
1333 #endif /* CONFIG_SPARSEMEM */
1336 * During memory init memblocks map pfns to nids. The search is expensive and
1337 * this caches recent lookups. The implementation of __early_pfn_to_nid
1338 * may treat start/end as pfns or sections.
1340 struct mminit_pfnnid_cache
{
1341 unsigned long last_start
;
1342 unsigned long last_end
;
1346 #ifndef early_pfn_valid
1347 #define early_pfn_valid(pfn) (1)
1350 void memory_present(int nid
, unsigned long start
, unsigned long end
);
1353 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1354 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1355 * pfn_valid_within() should be used in this case; we optimise this away
1356 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1358 #ifdef CONFIG_HOLES_IN_ZONE
1359 #define pfn_valid_within(pfn) pfn_valid(pfn)
1361 #define pfn_valid_within(pfn) (1)
1364 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1366 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1367 * associated with it or not. This means that a struct page exists for this
1368 * pfn. The caller cannot assume the page is fully initialized in general.
1369 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1370 * will ensure the struct page is fully online and initialized. Special pages
1371 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1373 * In FLATMEM, it is expected that holes always have valid memmap as long as
1374 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1375 * that a valid section has a memmap for the entire section.
1377 * However, an ARM, and maybe other embedded architectures in the future
1378 * free memmap backing holes to save memory on the assumption the memmap is
1379 * never used. The page_zone linkages are then broken even though pfn_valid()
1380 * returns true. A walker of the full memmap must then do this additional
1381 * check to ensure the memmap they are looking at is sane by making sure
1382 * the zone and PFN linkages are still valid. This is expensive, but walkers
1383 * of the full memmap are extremely rare.
1385 bool memmap_valid_within(unsigned long pfn
,
1386 struct page
*page
, struct zone
*zone
);
1388 static inline bool memmap_valid_within(unsigned long pfn
,
1389 struct page
*page
, struct zone
*zone
)
1393 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1395 #endif /* !__GENERATING_BOUNDS.H */
1396 #endif /* !__ASSEMBLY__ */
1397 #endif /* _LINUX_MMZONE_H */