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>
23 /* Free memory management - zoned buddy allocator. */
24 #ifndef CONFIG_FORCE_MAX_ZONEORDER
27 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
29 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
32 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
33 * costly to service. That is between allocation orders which should
34 * coalesce naturally under reasonable reclaim pressure and those which
37 #define PAGE_ALLOC_COSTLY_ORDER 3
43 MIGRATE_PCPTYPES
, /* the number of types on the pcp lists */
44 MIGRATE_HIGHATOMIC
= MIGRATE_PCPTYPES
,
47 * MIGRATE_CMA migration type is designed to mimic the way
48 * ZONE_MOVABLE works. Only movable pages can be allocated
49 * from MIGRATE_CMA pageblocks and page allocator never
50 * implicitly change migration type of MIGRATE_CMA pageblock.
52 * The way to use it is to change migratetype of a range of
53 * pageblocks to MIGRATE_CMA which can be done by
54 * __free_pageblock_cma() function. What is important though
55 * is that a range of pageblocks must be aligned to
56 * MAX_ORDER_NR_PAGES should biggest page be bigger then
61 #ifdef CONFIG_MEMORY_ISOLATION
62 MIGRATE_ISOLATE
, /* can't allocate from here */
67 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
68 extern char * const migratetype_names
[MIGRATE_TYPES
];
71 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
72 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
74 # define is_migrate_cma(migratetype) false
75 # define is_migrate_cma_page(_page) false
78 static inline bool is_migrate_movable(int mt
)
80 return is_migrate_cma(mt
) || mt
== MIGRATE_MOVABLE
;
83 #define for_each_migratetype_order(order, type) \
84 for (order = 0; order < MAX_ORDER; order++) \
85 for (type = 0; type < MIGRATE_TYPES; type++)
87 extern int page_group_by_mobility_disabled
;
89 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
90 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
92 #define get_pageblock_migratetype(page) \
93 get_pfnblock_flags_mask(page, page_to_pfn(page), \
94 PB_migrate_end, MIGRATETYPE_MASK)
97 struct list_head free_list
[MIGRATE_TYPES
];
98 unsigned long nr_free
;
104 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
105 * So add a wild amount of padding here to ensure that they fall into separate
106 * cachelines. There are very few zone structures in the machine, so space
107 * consumption is not a concern here.
109 #if defined(CONFIG_SMP)
110 struct zone_padding
{
112 } ____cacheline_internodealigned_in_smp
;
113 #define ZONE_PADDING(name) struct zone_padding name;
115 #define ZONE_PADDING(name)
119 enum numa_stat_item
{
120 NUMA_HIT
, /* allocated in intended node */
121 NUMA_MISS
, /* allocated in non intended node */
122 NUMA_FOREIGN
, /* was intended here, hit elsewhere */
123 NUMA_INTERLEAVE_HIT
, /* interleaver preferred this zone */
124 NUMA_LOCAL
, /* allocation from local node */
125 NUMA_OTHER
, /* allocation from other node */
126 NR_VM_NUMA_STAT_ITEMS
129 #define NR_VM_NUMA_STAT_ITEMS 0
132 enum zone_stat_item
{
133 /* First 128 byte cacheline (assuming 64 bit words) */
135 NR_ZONE_LRU_BASE
, /* Used only for compaction and reclaim retry */
136 NR_ZONE_INACTIVE_ANON
= NR_ZONE_LRU_BASE
,
138 NR_ZONE_INACTIVE_FILE
,
141 NR_ZONE_WRITE_PENDING
, /* Count of dirty, writeback and unstable pages */
142 NR_MLOCK
, /* mlock()ed pages found and moved off LRU */
143 NR_PAGETABLE
, /* used for pagetables */
144 NR_KERNEL_STACK_KB
, /* measured in KiB */
145 /* Second 128 byte cacheline */
147 #if IS_ENABLED(CONFIG_ZSMALLOC)
148 NR_ZSPAGES
, /* allocated in zsmalloc */
151 NR_VM_ZONE_STAT_ITEMS
};
153 enum node_stat_item
{
155 NR_INACTIVE_ANON
= NR_LRU_BASE
, /* must match order of LRU_[IN]ACTIVE */
156 NR_ACTIVE_ANON
, /* " " " " " */
157 NR_INACTIVE_FILE
, /* " " " " " */
158 NR_ACTIVE_FILE
, /* " " " " " */
159 NR_UNEVICTABLE
, /* " " " " " */
161 NR_SLAB_UNRECLAIMABLE
,
162 NR_ISOLATED_ANON
, /* Temporary isolated pages from anon lru */
163 NR_ISOLATED_FILE
, /* Temporary isolated pages from file lru */
168 WORKINGSET_NODERECLAIM
,
169 NR_ANON_MAPPED
, /* Mapped anonymous pages */
170 NR_FILE_MAPPED
, /* pagecache pages mapped into pagetables.
171 only modified from process context */
175 NR_WRITEBACK_TEMP
, /* Writeback using temporary buffers */
176 NR_SHMEM
, /* shmem pages (included tmpfs/GEM pages) */
180 NR_UNSTABLE_NFS
, /* NFS unstable pages */
182 NR_VMSCAN_IMMEDIATE
, /* Prioritise for reclaim when writeback ends */
183 NR_DIRTIED
, /* page dirtyings since bootup */
184 NR_WRITTEN
, /* page writings since bootup */
185 NR_KERNEL_MISC_RECLAIMABLE
, /* reclaimable non-slab kernel pages */
186 NR_VM_NODE_STAT_ITEMS
190 * We do arithmetic on the LRU lists in various places in the code,
191 * so it is important to keep the active lists LRU_ACTIVE higher in
192 * the array than the corresponding inactive lists, and to keep
193 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
195 * This has to be kept in sync with the statistics in zone_stat_item
196 * above and the descriptions in vmstat_text in mm/vmstat.c
203 LRU_INACTIVE_ANON
= LRU_BASE
,
204 LRU_ACTIVE_ANON
= LRU_BASE
+ LRU_ACTIVE
,
205 LRU_INACTIVE_FILE
= LRU_BASE
+ LRU_FILE
,
206 LRU_ACTIVE_FILE
= LRU_BASE
+ LRU_FILE
+ LRU_ACTIVE
,
211 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
213 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
215 static inline int is_file_lru(enum lru_list lru
)
217 return (lru
== LRU_INACTIVE_FILE
|| lru
== LRU_ACTIVE_FILE
);
220 static inline int is_active_lru(enum lru_list lru
)
222 return (lru
== LRU_ACTIVE_ANON
|| lru
== LRU_ACTIVE_FILE
);
225 struct zone_reclaim_stat
{
227 * The pageout code in vmscan.c keeps track of how many of the
228 * mem/swap backed and file backed pages are referenced.
229 * The higher the rotated/scanned ratio, the more valuable
232 * The anon LRU stats live in [0], file LRU stats in [1]
234 unsigned long recent_rotated
[2];
235 unsigned long recent_scanned
[2];
239 struct list_head lists
[NR_LRU_LISTS
];
240 struct zone_reclaim_stat reclaim_stat
;
241 /* Evictions & activations on the inactive file list */
242 atomic_long_t inactive_age
;
243 /* Refaults at the time of last reclaim cycle */
244 unsigned long refaults
;
246 struct pglist_data
*pgdat
;
250 /* Mask used at gathering information at once (see memcontrol.c) */
251 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
252 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
253 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
255 /* Isolate unmapped file */
256 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
257 /* Isolate for asynchronous migration */
258 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
259 /* Isolate unevictable pages */
260 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
262 /* LRU Isolation modes. */
263 typedef unsigned __bitwise isolate_mode_t
;
265 enum zone_watermarks
{
272 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
273 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
274 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
276 struct per_cpu_pages
{
277 int count
; /* number of pages in the list */
278 int high
; /* high watermark, emptying needed */
279 int batch
; /* chunk size for buddy add/remove */
281 /* Lists of pages, one per migrate type stored on the pcp-lists */
282 struct list_head lists
[MIGRATE_PCPTYPES
];
285 struct per_cpu_pageset
{
286 struct per_cpu_pages pcp
;
289 u16 vm_numa_stat_diff
[NR_VM_NUMA_STAT_ITEMS
];
293 s8 vm_stat_diff
[NR_VM_ZONE_STAT_ITEMS
];
297 struct per_cpu_nodestat
{
299 s8 vm_node_stat_diff
[NR_VM_NODE_STAT_ITEMS
];
302 #endif /* !__GENERATING_BOUNDS.H */
305 #ifdef CONFIG_ZONE_DMA
307 * ZONE_DMA is used when there are devices that are not able
308 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
309 * carve out the portion of memory that is needed for these devices.
310 * The range is arch specific.
315 * ---------------------------
316 * parisc, ia64, sparc <4G
319 * alpha Unlimited or 0-16MB.
321 * i386, x86_64 and multiple other arches
326 #ifdef CONFIG_ZONE_DMA32
328 * x86_64 needs two ZONE_DMAs because it supports devices that are
329 * only able to do DMA to the lower 16M but also 32 bit devices that
330 * can only do DMA areas below 4G.
335 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
336 * performed on pages in ZONE_NORMAL if the DMA devices support
337 * transfers to all addressable memory.
340 #ifdef CONFIG_HIGHMEM
342 * A memory area that is only addressable by the kernel through
343 * mapping portions into its own address space. This is for example
344 * used by i386 to allow the kernel to address the memory beyond
345 * 900MB. The kernel will set up special mappings (page
346 * table entries on i386) for each page that the kernel needs to
352 #ifdef CONFIG_ZONE_DEVICE
359 #ifndef __GENERATING_BOUNDS_H
362 /* Read-mostly fields */
364 /* zone watermarks, access with *_wmark_pages(zone) macros */
365 unsigned long watermark
[NR_WMARK
];
367 unsigned long nr_reserved_highatomic
;
370 * We don't know if the memory that we're going to allocate will be
371 * freeable or/and it will be released eventually, so to avoid totally
372 * wasting several GB of ram we must reserve some of the lower zone
373 * memory (otherwise we risk to run OOM on the lower zones despite
374 * there being tons of freeable ram on the higher zones). This array is
375 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
378 long lowmem_reserve
[MAX_NR_ZONES
];
383 struct pglist_data
*zone_pgdat
;
384 struct per_cpu_pageset __percpu
*pageset
;
386 #ifndef CONFIG_SPARSEMEM
388 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
389 * In SPARSEMEM, this map is stored in struct mem_section
391 unsigned long *pageblock_flags
;
392 #endif /* CONFIG_SPARSEMEM */
394 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
395 unsigned long zone_start_pfn
;
398 * spanned_pages is the total pages spanned by the zone, including
399 * holes, which is calculated as:
400 * spanned_pages = zone_end_pfn - zone_start_pfn;
402 * present_pages is physical pages existing within the zone, which
404 * present_pages = spanned_pages - absent_pages(pages in holes);
406 * managed_pages is present pages managed by the buddy system, which
407 * is calculated as (reserved_pages includes pages allocated by the
408 * bootmem allocator):
409 * managed_pages = present_pages - reserved_pages;
411 * So present_pages may be used by memory hotplug or memory power
412 * management logic to figure out unmanaged pages by checking
413 * (present_pages - managed_pages). And managed_pages should be used
414 * by page allocator and vm scanner to calculate all kinds of watermarks
419 * zone_start_pfn and spanned_pages are protected by span_seqlock.
420 * It is a seqlock because it has to be read outside of zone->lock,
421 * and it is done in the main allocator path. But, it is written
422 * quite infrequently.
424 * The span_seq lock is declared along with zone->lock because it is
425 * frequently read in proximity to zone->lock. It's good to
426 * give them a chance of being in the same cacheline.
428 * Write access to present_pages at runtime should be protected by
429 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
430 * present_pages should get_online_mems() to get a stable value.
432 * Read access to managed_pages should be safe because it's unsigned
433 * long. Write access to zone->managed_pages and totalram_pages are
434 * protected by managed_page_count_lock at runtime. Idealy only
435 * adjust_managed_page_count() should be used instead of directly
436 * touching zone->managed_pages and totalram_pages.
438 unsigned long managed_pages
;
439 unsigned long spanned_pages
;
440 unsigned long present_pages
;
444 #ifdef CONFIG_MEMORY_ISOLATION
446 * Number of isolated pageblock. It is used to solve incorrect
447 * freepage counting problem due to racy retrieving migratetype
448 * of pageblock. Protected by zone->lock.
450 unsigned long nr_isolate_pageblock
;
453 #ifdef CONFIG_MEMORY_HOTPLUG
454 /* see spanned/present_pages for more description */
455 seqlock_t span_seqlock
;
460 /* Write-intensive fields used from the page allocator */
463 /* free areas of different sizes */
464 struct free_area free_area
[MAX_ORDER
];
466 /* zone flags, see below */
469 /* Primarily protects free_area */
472 /* Write-intensive fields used by compaction and vmstats. */
476 * When free pages are below this point, additional steps are taken
477 * when reading the number of free pages to avoid per-cpu counter
478 * drift allowing watermarks to be breached
480 unsigned long percpu_drift_mark
;
482 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
483 /* pfn where compaction free scanner should start */
484 unsigned long compact_cached_free_pfn
;
485 /* pfn where async and sync compaction migration scanner should start */
486 unsigned long compact_cached_migrate_pfn
[2];
489 #ifdef CONFIG_COMPACTION
491 * On compaction failure, 1<<compact_defer_shift compactions
492 * are skipped before trying again. The number attempted since
493 * last failure is tracked with compact_considered.
495 unsigned int compact_considered
;
496 unsigned int compact_defer_shift
;
497 int compact_order_failed
;
500 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
501 /* Set to true when the PG_migrate_skip bits should be cleared */
502 bool compact_blockskip_flush
;
508 /* Zone statistics */
509 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
510 atomic_long_t vm_numa_stat
[NR_VM_NUMA_STAT_ITEMS
];
511 } ____cacheline_internodealigned_in_smp
;
514 PGDAT_CONGESTED
, /* pgdat has many dirty pages backed by
517 PGDAT_DIRTY
, /* reclaim scanning has recently found
518 * many dirty file pages at the tail
521 PGDAT_WRITEBACK
, /* reclaim scanning has recently found
522 * many pages under writeback
524 PGDAT_RECLAIM_LOCKED
, /* prevents concurrent reclaim */
527 static inline unsigned long zone_end_pfn(const struct zone
*zone
)
529 return zone
->zone_start_pfn
+ zone
->spanned_pages
;
532 static inline bool zone_spans_pfn(const struct zone
*zone
, unsigned long pfn
)
534 return zone
->zone_start_pfn
<= pfn
&& pfn
< zone_end_pfn(zone
);
537 static inline bool zone_is_initialized(struct zone
*zone
)
539 return zone
->initialized
;
542 static inline bool zone_is_empty(struct zone
*zone
)
544 return zone
->spanned_pages
== 0;
548 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
549 * intersection with the given zone
551 static inline bool zone_intersects(struct zone
*zone
,
552 unsigned long start_pfn
, unsigned long nr_pages
)
554 if (zone_is_empty(zone
))
556 if (start_pfn
>= zone_end_pfn(zone
) ||
557 start_pfn
+ nr_pages
<= zone
->zone_start_pfn
)
564 * The "priority" of VM scanning is how much of the queues we will scan in one
565 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
566 * queues ("queue_length >> 12") during an aging round.
568 #define DEF_PRIORITY 12
570 /* Maximum number of zones on a zonelist */
571 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
574 ZONELIST_FALLBACK
, /* zonelist with fallback */
577 * The NUMA zonelists are doubled because we need zonelists that
578 * restrict the allocations to a single node for __GFP_THISNODE.
580 ZONELIST_NOFALLBACK
, /* zonelist without fallback (__GFP_THISNODE) */
586 * This struct contains information about a zone in a zonelist. It is stored
587 * here to avoid dereferences into large structures and lookups of tables
590 struct zone
*zone
; /* Pointer to actual zone */
591 int zone_idx
; /* zone_idx(zoneref->zone) */
595 * One allocation request operates on a zonelist. A zonelist
596 * is a list of zones, the first one is the 'goal' of the
597 * allocation, the other zones are fallback zones, in decreasing
600 * To speed the reading of the zonelist, the zonerefs contain the zone index
601 * of the entry being read. Helper functions to access information given
602 * a struct zoneref are
604 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
605 * zonelist_zone_idx() - Return the index of the zone for an entry
606 * zonelist_node_idx() - Return the index of the node for an entry
609 struct zoneref _zonerefs
[MAX_ZONES_PER_ZONELIST
+ 1];
612 #ifndef CONFIG_DISCONTIGMEM
613 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
614 extern struct page
*mem_map
;
618 * On NUMA machines, each NUMA node would have a pg_data_t to describe
619 * it's memory layout. On UMA machines there is a single pglist_data which
620 * describes the whole memory.
622 * Memory statistics and page replacement data structures are maintained on a
626 typedef struct pglist_data
{
627 struct zone node_zones
[MAX_NR_ZONES
];
628 struct zonelist node_zonelists
[MAX_ZONELISTS
];
630 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
631 struct page
*node_mem_map
;
632 #ifdef CONFIG_PAGE_EXTENSION
633 struct page_ext
*node_page_ext
;
636 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
638 * Must be held any time you expect node_start_pfn, node_present_pages
639 * or node_spanned_pages stay constant. Holding this will also
640 * guarantee that any pfn_valid() stays that way.
642 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
643 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
644 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
646 * Nests above zone->lock and zone->span_seqlock
648 spinlock_t node_size_lock
;
650 unsigned long node_start_pfn
;
651 unsigned long node_present_pages
; /* total number of physical pages */
652 unsigned long node_spanned_pages
; /* total size of physical page
653 range, including holes */
655 wait_queue_head_t kswapd_wait
;
656 wait_queue_head_t pfmemalloc_wait
;
657 struct task_struct
*kswapd
; /* Protected by
658 mem_hotplug_begin/end() */
660 enum zone_type kswapd_classzone_idx
;
662 int kswapd_failures
; /* Number of 'reclaimed == 0' runs */
664 #ifdef CONFIG_COMPACTION
665 int kcompactd_max_order
;
666 enum zone_type kcompactd_classzone_idx
;
667 wait_queue_head_t kcompactd_wait
;
668 struct task_struct
*kcompactd
;
671 * This is a per-node reserve of pages that are not available
672 * to userspace allocations.
674 unsigned long totalreserve_pages
;
678 * zone reclaim becomes active if more unmapped pages exist.
680 unsigned long min_unmapped_pages
;
681 unsigned long min_slab_pages
;
682 #endif /* CONFIG_NUMA */
684 /* Write-intensive fields used by page reclaim */
688 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
690 * If memory initialisation on large machines is deferred then this
691 * is the first PFN that needs to be initialised.
693 unsigned long first_deferred_pfn
;
694 /* Number of non-deferred pages */
695 unsigned long static_init_pgcnt
;
696 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
698 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
699 spinlock_t split_queue_lock
;
700 struct list_head split_queue
;
701 unsigned long split_queue_len
;
704 /* Fields commonly accessed by the page reclaim scanner */
705 struct lruvec lruvec
;
711 /* Per-node vmstats */
712 struct per_cpu_nodestat __percpu
*per_cpu_nodestats
;
713 atomic_long_t vm_stat
[NR_VM_NODE_STAT_ITEMS
];
716 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
717 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
718 #ifdef CONFIG_FLAT_NODE_MEM_MAP
719 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
721 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
723 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
725 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
726 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
727 static inline spinlock_t
*zone_lru_lock(struct zone
*zone
)
729 return &zone
->zone_pgdat
->lru_lock
;
732 static inline struct lruvec
*node_lruvec(struct pglist_data
*pgdat
)
734 return &pgdat
->lruvec
;
737 static inline unsigned long pgdat_end_pfn(pg_data_t
*pgdat
)
739 return pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
742 static inline bool pgdat_is_empty(pg_data_t
*pgdat
)
744 return !pgdat
->node_start_pfn
&& !pgdat
->node_spanned_pages
;
747 #include <linux/memory_hotplug.h>
749 void build_all_zonelists(pg_data_t
*pgdat
);
750 void wakeup_kswapd(struct zone
*zone
, gfp_t gfp_mask
, int order
,
751 enum zone_type classzone_idx
);
752 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
753 int classzone_idx
, unsigned int alloc_flags
,
755 bool zone_watermark_ok(struct zone
*z
, unsigned int order
,
756 unsigned long mark
, int classzone_idx
,
757 unsigned int alloc_flags
);
758 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
759 unsigned long mark
, int classzone_idx
);
760 enum memmap_context
{
764 extern void init_currently_empty_zone(struct zone
*zone
, unsigned long start_pfn
,
767 extern void lruvec_init(struct lruvec
*lruvec
);
769 static inline struct pglist_data
*lruvec_pgdat(struct lruvec
*lruvec
)
772 return lruvec
->pgdat
;
774 return container_of(lruvec
, struct pglist_data
, lruvec
);
778 extern unsigned long lruvec_lru_size(struct lruvec
*lruvec
, enum lru_list lru
, int zone_idx
);
780 #ifdef CONFIG_HAVE_MEMORY_PRESENT
781 void memory_present(int nid
, unsigned long start
, unsigned long end
);
783 static inline void memory_present(int nid
, unsigned long start
, unsigned long end
) {}
786 #if defined(CONFIG_SPARSEMEM)
787 void memblocks_present(void);
789 static inline void memblocks_present(void) {}
792 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
793 int local_memory_node(int node_id
);
795 static inline int local_memory_node(int node_id
) { return node_id
; };
799 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
801 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
803 #ifdef CONFIG_ZONE_DEVICE
804 static inline bool is_dev_zone(const struct zone
*zone
)
806 return zone_idx(zone
) == ZONE_DEVICE
;
809 static inline bool is_dev_zone(const struct zone
*zone
)
816 * Returns true if a zone has pages managed by the buddy allocator.
817 * All the reclaim decisions have to use this function rather than
818 * populated_zone(). If the whole zone is reserved then we can easily
819 * end up with populated_zone() && !managed_zone().
821 static inline bool managed_zone(struct zone
*zone
)
823 return zone
->managed_pages
;
826 /* Returns true if a zone has memory */
827 static inline bool populated_zone(struct zone
*zone
)
829 return zone
->present_pages
;
833 static inline int zone_to_nid(struct zone
*zone
)
838 static inline void zone_set_nid(struct zone
*zone
, int nid
)
843 static inline int zone_to_nid(struct zone
*zone
)
848 static inline void zone_set_nid(struct zone
*zone
, int nid
) {}
851 extern int movable_zone
;
853 #ifdef CONFIG_HIGHMEM
854 static inline int zone_movable_is_highmem(void)
856 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
857 return movable_zone
== ZONE_HIGHMEM
;
859 return (ZONE_MOVABLE
- 1) == ZONE_HIGHMEM
;
864 static inline int is_highmem_idx(enum zone_type idx
)
866 #ifdef CONFIG_HIGHMEM
867 return (idx
== ZONE_HIGHMEM
||
868 (idx
== ZONE_MOVABLE
&& zone_movable_is_highmem()));
875 * is_highmem - helper function to quickly check if a struct zone is a
876 * highmem zone or not. This is an attempt to keep references
877 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
878 * @zone - pointer to struct zone variable
880 static inline int is_highmem(struct zone
*zone
)
882 #ifdef CONFIG_HIGHMEM
883 return is_highmem_idx(zone_idx(zone
));
889 /* These two functions are used to setup the per zone pages min values */
891 int min_free_kbytes_sysctl_handler(struct ctl_table
*, int,
892 void __user
*, size_t *, loff_t
*);
893 int watermark_scale_factor_sysctl_handler(struct ctl_table
*, int,
894 void __user
*, size_t *, loff_t
*);
895 extern int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
];
896 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*, int,
897 void __user
*, size_t *, loff_t
*);
898 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*, int,
899 void __user
*, size_t *, loff_t
*);
900 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*, int,
901 void __user
*, size_t *, loff_t
*);
902 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*, int,
903 void __user
*, size_t *, loff_t
*);
905 extern int numa_zonelist_order_handler(struct ctl_table
*, int,
906 void __user
*, size_t *, loff_t
*);
907 extern char numa_zonelist_order
[];
908 #define NUMA_ZONELIST_ORDER_LEN 16
910 #ifndef CONFIG_NEED_MULTIPLE_NODES
912 extern struct pglist_data contig_page_data
;
913 #define NODE_DATA(nid) (&contig_page_data)
914 #define NODE_MEM_MAP(nid) mem_map
916 #else /* CONFIG_NEED_MULTIPLE_NODES */
918 #include <asm/mmzone.h>
920 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
922 extern struct pglist_data
*first_online_pgdat(void);
923 extern struct pglist_data
*next_online_pgdat(struct pglist_data
*pgdat
);
924 extern struct zone
*next_zone(struct zone
*zone
);
927 * for_each_online_pgdat - helper macro to iterate over all online nodes
928 * @pgdat - pointer to a pg_data_t variable
930 #define for_each_online_pgdat(pgdat) \
931 for (pgdat = first_online_pgdat(); \
933 pgdat = next_online_pgdat(pgdat))
935 * for_each_zone - helper macro to iterate over all memory zones
936 * @zone - pointer to struct zone variable
938 * The user only needs to declare the zone variable, for_each_zone
941 #define for_each_zone(zone) \
942 for (zone = (first_online_pgdat())->node_zones; \
944 zone = next_zone(zone))
946 #define for_each_populated_zone(zone) \
947 for (zone = (first_online_pgdat())->node_zones; \
949 zone = next_zone(zone)) \
950 if (!populated_zone(zone)) \
954 static inline struct zone
*zonelist_zone(struct zoneref
*zoneref
)
956 return zoneref
->zone
;
959 static inline int zonelist_zone_idx(struct zoneref
*zoneref
)
961 return zoneref
->zone_idx
;
964 static inline int zonelist_node_idx(struct zoneref
*zoneref
)
966 return zone_to_nid(zoneref
->zone
);
969 struct zoneref
*__next_zones_zonelist(struct zoneref
*z
,
970 enum zone_type highest_zoneidx
,
974 * 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
975 * @z - The cursor used as a starting point for the search
976 * @highest_zoneidx - The zone index of the highest zone to return
977 * @nodes - An optional nodemask to filter the zonelist with
979 * This function returns the next zone at or below a given zone index that is
980 * within the allowed nodemask using a cursor as the starting point for the
981 * search. The zoneref returned is a cursor that represents the current zone
982 * being examined. It should be advanced by one before calling
983 * next_zones_zonelist again.
985 static __always_inline
struct zoneref
*next_zones_zonelist(struct zoneref
*z
,
986 enum zone_type highest_zoneidx
,
989 if (likely(!nodes
&& zonelist_zone_idx(z
) <= highest_zoneidx
))
991 return __next_zones_zonelist(z
, highest_zoneidx
, nodes
);
995 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
996 * @zonelist - The zonelist to search for a suitable zone
997 * @highest_zoneidx - The zone index of the highest zone to return
998 * @nodes - An optional nodemask to filter the zonelist with
999 * @return - Zoneref pointer for the first suitable zone found (see below)
1001 * This function returns the first zone at or below a given zone index that is
1002 * within the allowed nodemask. The zoneref returned is a cursor that can be
1003 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1004 * one before calling.
1006 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1007 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1008 * update due to cpuset modification.
1010 static inline struct zoneref
*first_zones_zonelist(struct zonelist
*zonelist
,
1011 enum zone_type highest_zoneidx
,
1014 return next_zones_zonelist(zonelist
->_zonerefs
,
1015 highest_zoneidx
, nodes
);
1019 * 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
1020 * @zone - The current zone in the iterator
1021 * @z - The current pointer within zonelist->zones being iterated
1022 * @zlist - The zonelist being iterated
1023 * @highidx - The zone index of the highest zone to return
1024 * @nodemask - Nodemask allowed by the allocator
1026 * This iterator iterates though all zones at or below a given zone index and
1027 * within a given nodemask
1029 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1030 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1032 z = next_zones_zonelist(++z, highidx, nodemask), \
1033 zone = zonelist_zone(z))
1035 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1036 for (zone = z->zone; \
1038 z = next_zones_zonelist(++z, highidx, nodemask), \
1039 zone = zonelist_zone(z))
1043 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1044 * @zone - The current zone in the iterator
1045 * @z - The current pointer within zonelist->zones being iterated
1046 * @zlist - The zonelist being iterated
1047 * @highidx - The zone index of the highest zone to return
1049 * This iterator iterates though all zones at or below a given zone index.
1051 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1052 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1054 #ifdef CONFIG_SPARSEMEM
1055 #include <asm/sparsemem.h>
1058 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1059 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1060 static inline unsigned long early_pfn_to_nid(unsigned long pfn
)
1062 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA
));
1067 #ifdef CONFIG_FLATMEM
1068 #define pfn_to_nid(pfn) (0)
1071 #ifdef CONFIG_SPARSEMEM
1074 * SECTION_SHIFT #bits space required to store a section #
1076 * PA_SECTION_SHIFT physical address to/from section number
1077 * PFN_SECTION_SHIFT pfn to/from section number
1079 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1080 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1082 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1084 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1085 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1087 #define SECTION_BLOCKFLAGS_BITS \
1088 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1090 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1091 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1094 static inline unsigned long pfn_to_section_nr(unsigned long pfn
)
1096 return pfn
>> PFN_SECTION_SHIFT
;
1098 static inline unsigned long section_nr_to_pfn(unsigned long sec
)
1100 return sec
<< PFN_SECTION_SHIFT
;
1103 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1104 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1108 struct mem_section
{
1110 * This is, logically, a pointer to an array of struct
1111 * pages. However, it is stored with some other magic.
1112 * (see sparse.c::sparse_init_one_section())
1114 * Additionally during early boot we encode node id of
1115 * the location of the section here to guide allocation.
1116 * (see sparse.c::memory_present())
1118 * Making it a UL at least makes someone do a cast
1119 * before using it wrong.
1121 unsigned long section_mem_map
;
1123 /* See declaration of similar field in struct zone */
1124 unsigned long *pageblock_flags
;
1125 #ifdef CONFIG_PAGE_EXTENSION
1127 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1128 * section. (see page_ext.h about this.)
1130 struct page_ext
*page_ext
;
1134 * WARNING: mem_section must be a power-of-2 in size for the
1135 * calculation and use of SECTION_ROOT_MASK to make sense.
1139 #ifdef CONFIG_SPARSEMEM_EXTREME
1140 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1142 #define SECTIONS_PER_ROOT 1
1145 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1146 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1147 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1149 #ifdef CONFIG_SPARSEMEM_EXTREME
1150 extern struct mem_section
**mem_section
;
1152 extern struct mem_section mem_section
[NR_SECTION_ROOTS
][SECTIONS_PER_ROOT
];
1155 static inline struct mem_section
*__nr_to_section(unsigned long nr
)
1157 #ifdef CONFIG_SPARSEMEM_EXTREME
1161 if (!mem_section
[SECTION_NR_TO_ROOT(nr
)])
1163 return &mem_section
[SECTION_NR_TO_ROOT(nr
)][nr
& SECTION_ROOT_MASK
];
1165 extern int __section_nr(struct mem_section
* ms
);
1166 extern unsigned long usemap_size(void);
1169 * We use the lower bits of the mem_map pointer to store
1170 * a little bit of information. The pointer is calculated
1171 * as mem_map - section_nr_to_pfn(pnum). The result is
1172 * aligned to the minimum alignment of the two values:
1173 * 1. All mem_map arrays are page-aligned.
1174 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1175 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1176 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1177 * worst combination is powerpc with 256k pages,
1178 * which results in PFN_SECTION_SHIFT equal 6.
1179 * To sum it up, at least 6 bits are available.
1181 #define SECTION_MARKED_PRESENT (1UL<<0)
1182 #define SECTION_HAS_MEM_MAP (1UL<<1)
1183 #define SECTION_IS_ONLINE (1UL<<2)
1184 #define SECTION_MAP_LAST_BIT (1UL<<3)
1185 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1186 #define SECTION_NID_SHIFT 3
1188 static inline struct page
*__section_mem_map_addr(struct mem_section
*section
)
1190 unsigned long map
= section
->section_mem_map
;
1191 map
&= SECTION_MAP_MASK
;
1192 return (struct page
*)map
;
1195 static inline int present_section(struct mem_section
*section
)
1197 return (section
&& (section
->section_mem_map
& SECTION_MARKED_PRESENT
));
1200 static inline int present_section_nr(unsigned long nr
)
1202 return present_section(__nr_to_section(nr
));
1205 static inline int valid_section(struct mem_section
*section
)
1207 return (section
&& (section
->section_mem_map
& SECTION_HAS_MEM_MAP
));
1210 static inline int valid_section_nr(unsigned long nr
)
1212 return valid_section(__nr_to_section(nr
));
1215 static inline int online_section(struct mem_section
*section
)
1217 return (section
&& (section
->section_mem_map
& SECTION_IS_ONLINE
));
1220 static inline int online_section_nr(unsigned long nr
)
1222 return online_section(__nr_to_section(nr
));
1225 #ifdef CONFIG_MEMORY_HOTPLUG
1226 void online_mem_sections(unsigned long start_pfn
, unsigned long end_pfn
);
1227 #ifdef CONFIG_MEMORY_HOTREMOVE
1228 void offline_mem_sections(unsigned long start_pfn
, unsigned long end_pfn
);
1232 static inline struct mem_section
*__pfn_to_section(unsigned long pfn
)
1234 return __nr_to_section(pfn_to_section_nr(pfn
));
1237 extern int __highest_present_section_nr
;
1239 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1240 static inline int pfn_valid(unsigned long pfn
)
1242 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1244 return valid_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1248 static inline int pfn_present(unsigned long pfn
)
1250 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1252 return present_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1256 * These are _only_ used during initialisation, therefore they
1257 * can use __initdata ... They could have names to indicate
1261 #define pfn_to_nid(pfn) \
1263 unsigned long __pfn_to_nid_pfn = (pfn); \
1264 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1267 #define pfn_to_nid(pfn) (0)
1270 #define early_pfn_valid(pfn) pfn_valid(pfn)
1271 void sparse_init(void);
1273 #define sparse_init() do {} while (0)
1274 #define sparse_index_init(_sec, _nid) do {} while (0)
1275 #endif /* CONFIG_SPARSEMEM */
1278 * During memory init memblocks map pfns to nids. The search is expensive and
1279 * this caches recent lookups. The implementation of __early_pfn_to_nid
1280 * may treat start/end as pfns or sections.
1282 struct mminit_pfnnid_cache
{
1283 unsigned long last_start
;
1284 unsigned long last_end
;
1288 #ifndef early_pfn_valid
1289 #define early_pfn_valid(pfn) (1)
1292 void memory_present(int nid
, unsigned long start
, unsigned long end
);
1295 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1296 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1297 * pfn_valid_within() should be used in this case; we optimise this away
1298 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1300 #ifdef CONFIG_HOLES_IN_ZONE
1301 #define pfn_valid_within(pfn) pfn_valid(pfn)
1303 #define pfn_valid_within(pfn) (1)
1306 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1308 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1309 * associated with it or not. This means that a struct page exists for this
1310 * pfn. The caller cannot assume the page is fully initialized in general.
1311 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1312 * will ensure the struct page is fully online and initialized. Special pages
1313 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1315 * In FLATMEM, it is expected that holes always have valid memmap as long as
1316 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1317 * that a valid section has a memmap for the entire section.
1319 * However, an ARM, and maybe other embedded architectures in the future
1320 * free memmap backing holes to save memory on the assumption the memmap is
1321 * never used. The page_zone linkages are then broken even though pfn_valid()
1322 * returns true. A walker of the full memmap must then do this additional
1323 * check to ensure the memmap they are looking at is sane by making sure
1324 * the zone and PFN linkages are still valid. This is expensive, but walkers
1325 * of the full memmap are extremely rare.
1327 bool memmap_valid_within(unsigned long pfn
,
1328 struct page
*page
, struct zone
*zone
);
1330 static inline bool memmap_valid_within(unsigned long pfn
,
1331 struct page
*page
, struct zone
*zone
)
1335 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1337 #endif /* !__GENERATING_BOUNDS.H */
1338 #endif /* !__ASSEMBLY__ */
1339 #endif /* _LINUX_MMZONE_H */