xref: /linux-6.15/include/linux/mmzone.h (revision cfd6ed45)
1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
3 
4 #ifndef __ASSEMBLY__
5 #ifndef __GENERATING_BOUNDS_H
6 
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>
20 #include <asm/page.h>
21 
22 /* Free memory management - zoned buddy allocator.  */
23 #ifndef CONFIG_FORCE_MAX_ZONEORDER
24 #define MAX_ORDER 11
25 #else
26 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
27 #endif
28 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
29 
30 /*
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
34  * will not.
35  */
36 #define PAGE_ALLOC_COSTLY_ORDER 3
37 
38 enum {
39 	MIGRATE_UNMOVABLE,
40 	MIGRATE_MOVABLE,
41 	MIGRATE_RECLAIMABLE,
42 	MIGRATE_PCPTYPES,	/* the number of types on the pcp lists */
43 	MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
44 #ifdef CONFIG_CMA
45 	/*
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.
50 	 *
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
56 	 * a single pageblock.
57 	 */
58 	MIGRATE_CMA,
59 #endif
60 #ifdef CONFIG_MEMORY_ISOLATION
61 	MIGRATE_ISOLATE,	/* can't allocate from here */
62 #endif
63 	MIGRATE_TYPES
64 };
65 
66 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
67 extern char * const migratetype_names[MIGRATE_TYPES];
68 
69 #ifdef CONFIG_CMA
70 #  define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
71 #  define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
72 #else
73 #  define is_migrate_cma(migratetype) false
74 #  define is_migrate_cma_page(_page) false
75 #endif
76 
77 #define for_each_migratetype_order(order, type) \
78 	for (order = 0; order < MAX_ORDER; order++) \
79 		for (type = 0; type < MIGRATE_TYPES; type++)
80 
81 extern int page_group_by_mobility_disabled;
82 
83 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
84 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
85 
86 #define get_pageblock_migratetype(page)					\
87 	get_pfnblock_flags_mask(page, page_to_pfn(page),		\
88 			PB_migrate_end, MIGRATETYPE_MASK)
89 
90 struct free_area {
91 	struct list_head	free_list[MIGRATE_TYPES];
92 	unsigned long		nr_free;
93 };
94 
95 struct pglist_data;
96 
97 /*
98  * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
99  * So add a wild amount of padding here to ensure that they fall into separate
100  * cachelines.  There are very few zone structures in the machine, so space
101  * consumption is not a concern here.
102  */
103 #if defined(CONFIG_SMP)
104 struct zone_padding {
105 	char x[0];
106 } ____cacheline_internodealigned_in_smp;
107 #define ZONE_PADDING(name)	struct zone_padding name;
108 #else
109 #define ZONE_PADDING(name)
110 #endif
111 
112 enum zone_stat_item {
113 	/* First 128 byte cacheline (assuming 64 bit words) */
114 	NR_FREE_PAGES,
115 	NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
116 	NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
117 	NR_ZONE_ACTIVE_ANON,
118 	NR_ZONE_INACTIVE_FILE,
119 	NR_ZONE_ACTIVE_FILE,
120 	NR_ZONE_UNEVICTABLE,
121 	NR_ZONE_WRITE_PENDING,	/* Count of dirty, writeback and unstable pages */
122 	NR_MLOCK,		/* mlock()ed pages found and moved off LRU */
123 	NR_SLAB_RECLAIMABLE,
124 	NR_SLAB_UNRECLAIMABLE,
125 	NR_PAGETABLE,		/* used for pagetables */
126 	NR_KERNEL_STACK_KB,	/* measured in KiB */
127 	/* Second 128 byte cacheline */
128 	NR_BOUNCE,
129 #if IS_ENABLED(CONFIG_ZSMALLOC)
130 	NR_ZSPAGES,		/* allocated in zsmalloc */
131 #endif
132 #ifdef CONFIG_NUMA
133 	NUMA_HIT,		/* allocated in intended node */
134 	NUMA_MISS,		/* allocated in non intended node */
135 	NUMA_FOREIGN,		/* was intended here, hit elsewhere */
136 	NUMA_INTERLEAVE_HIT,	/* interleaver preferred this zone */
137 	NUMA_LOCAL,		/* allocation from local node */
138 	NUMA_OTHER,		/* allocation from other node */
139 #endif
140 	NR_FREE_CMA_PAGES,
141 	NR_VM_ZONE_STAT_ITEMS };
142 
143 enum node_stat_item {
144 	NR_LRU_BASE,
145 	NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
146 	NR_ACTIVE_ANON,		/*  "     "     "   "       "         */
147 	NR_INACTIVE_FILE,	/*  "     "     "   "       "         */
148 	NR_ACTIVE_FILE,		/*  "     "     "   "       "         */
149 	NR_UNEVICTABLE,		/*  "     "     "   "       "         */
150 	NR_ISOLATED_ANON,	/* Temporary isolated pages from anon lru */
151 	NR_ISOLATED_FILE,	/* Temporary isolated pages from file lru */
152 	NR_PAGES_SCANNED,	/* pages scanned since last reclaim */
153 	WORKINGSET_REFAULT,
154 	WORKINGSET_ACTIVATE,
155 	WORKINGSET_NODERECLAIM,
156 	NR_ANON_MAPPED,	/* Mapped anonymous pages */
157 	NR_FILE_MAPPED,	/* pagecache pages mapped into pagetables.
158 			   only modified from process context */
159 	NR_FILE_PAGES,
160 	NR_FILE_DIRTY,
161 	NR_WRITEBACK,
162 	NR_WRITEBACK_TEMP,	/* Writeback using temporary buffers */
163 	NR_SHMEM,		/* shmem pages (included tmpfs/GEM pages) */
164 	NR_SHMEM_THPS,
165 	NR_SHMEM_PMDMAPPED,
166 	NR_ANON_THPS,
167 	NR_UNSTABLE_NFS,	/* NFS unstable pages */
168 	NR_VMSCAN_WRITE,
169 	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
170 	NR_DIRTIED,		/* page dirtyings since bootup */
171 	NR_WRITTEN,		/* page writings since bootup */
172 	NR_VM_NODE_STAT_ITEMS
173 };
174 
175 /*
176  * We do arithmetic on the LRU lists in various places in the code,
177  * so it is important to keep the active lists LRU_ACTIVE higher in
178  * the array than the corresponding inactive lists, and to keep
179  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
180  *
181  * This has to be kept in sync with the statistics in zone_stat_item
182  * above and the descriptions in vmstat_text in mm/vmstat.c
183  */
184 #define LRU_BASE 0
185 #define LRU_ACTIVE 1
186 #define LRU_FILE 2
187 
188 enum lru_list {
189 	LRU_INACTIVE_ANON = LRU_BASE,
190 	LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
191 	LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
192 	LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
193 	LRU_UNEVICTABLE,
194 	NR_LRU_LISTS
195 };
196 
197 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
198 
199 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
200 
201 static inline int is_file_lru(enum lru_list lru)
202 {
203 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
204 }
205 
206 static inline int is_active_lru(enum lru_list lru)
207 {
208 	return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
209 }
210 
211 struct zone_reclaim_stat {
212 	/*
213 	 * The pageout code in vmscan.c keeps track of how many of the
214 	 * mem/swap backed and file backed pages are referenced.
215 	 * The higher the rotated/scanned ratio, the more valuable
216 	 * that cache is.
217 	 *
218 	 * The anon LRU stats live in [0], file LRU stats in [1]
219 	 */
220 	unsigned long		recent_rotated[2];
221 	unsigned long		recent_scanned[2];
222 };
223 
224 struct lruvec {
225 	struct list_head		lists[NR_LRU_LISTS];
226 	struct zone_reclaim_stat	reclaim_stat;
227 	/* Evictions & activations on the inactive file list */
228 	atomic_long_t			inactive_age;
229 #ifdef CONFIG_MEMCG
230 	struct pglist_data *pgdat;
231 #endif
232 };
233 
234 /* Mask used at gathering information at once (see memcontrol.c) */
235 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
236 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
237 #define LRU_ALL	     ((1 << NR_LRU_LISTS) - 1)
238 
239 /* Isolate unmapped file */
240 #define ISOLATE_UNMAPPED	((__force isolate_mode_t)0x2)
241 /* Isolate for asynchronous migration */
242 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
243 /* Isolate unevictable pages */
244 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
245 
246 /* LRU Isolation modes. */
247 typedef unsigned __bitwise isolate_mode_t;
248 
249 enum zone_watermarks {
250 	WMARK_MIN,
251 	WMARK_LOW,
252 	WMARK_HIGH,
253 	NR_WMARK
254 };
255 
256 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
257 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
258 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
259 
260 struct per_cpu_pages {
261 	int count;		/* number of pages in the list */
262 	int high;		/* high watermark, emptying needed */
263 	int batch;		/* chunk size for buddy add/remove */
264 
265 	/* Lists of pages, one per migrate type stored on the pcp-lists */
266 	struct list_head lists[MIGRATE_PCPTYPES];
267 };
268 
269 struct per_cpu_pageset {
270 	struct per_cpu_pages pcp;
271 #ifdef CONFIG_NUMA
272 	s8 expire;
273 #endif
274 #ifdef CONFIG_SMP
275 	s8 stat_threshold;
276 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
277 #endif
278 };
279 
280 struct per_cpu_nodestat {
281 	s8 stat_threshold;
282 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
283 };
284 
285 #endif /* !__GENERATING_BOUNDS.H */
286 
287 enum zone_type {
288 #ifdef CONFIG_ZONE_DMA
289 	/*
290 	 * ZONE_DMA is used when there are devices that are not able
291 	 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
292 	 * carve out the portion of memory that is needed for these devices.
293 	 * The range is arch specific.
294 	 *
295 	 * Some examples
296 	 *
297 	 * Architecture		Limit
298 	 * ---------------------------
299 	 * parisc, ia64, sparc	<4G
300 	 * s390			<2G
301 	 * arm			Various
302 	 * alpha		Unlimited or 0-16MB.
303 	 *
304 	 * i386, x86_64 and multiple other arches
305 	 * 			<16M.
306 	 */
307 	ZONE_DMA,
308 #endif
309 #ifdef CONFIG_ZONE_DMA32
310 	/*
311 	 * x86_64 needs two ZONE_DMAs because it supports devices that are
312 	 * only able to do DMA to the lower 16M but also 32 bit devices that
313 	 * can only do DMA areas below 4G.
314 	 */
315 	ZONE_DMA32,
316 #endif
317 	/*
318 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
319 	 * performed on pages in ZONE_NORMAL if the DMA devices support
320 	 * transfers to all addressable memory.
321 	 */
322 	ZONE_NORMAL,
323 #ifdef CONFIG_HIGHMEM
324 	/*
325 	 * A memory area that is only addressable by the kernel through
326 	 * mapping portions into its own address space. This is for example
327 	 * used by i386 to allow the kernel to address the memory beyond
328 	 * 900MB. The kernel will set up special mappings (page
329 	 * table entries on i386) for each page that the kernel needs to
330 	 * access.
331 	 */
332 	ZONE_HIGHMEM,
333 #endif
334 	ZONE_MOVABLE,
335 #ifdef CONFIG_ZONE_DEVICE
336 	ZONE_DEVICE,
337 #endif
338 	__MAX_NR_ZONES
339 
340 };
341 
342 #ifndef __GENERATING_BOUNDS_H
343 
344 struct zone {
345 	/* Read-mostly fields */
346 
347 	/* zone watermarks, access with *_wmark_pages(zone) macros */
348 	unsigned long watermark[NR_WMARK];
349 
350 	unsigned long nr_reserved_highatomic;
351 
352 	/*
353 	 * We don't know if the memory that we're going to allocate will be
354 	 * freeable or/and it will be released eventually, so to avoid totally
355 	 * wasting several GB of ram we must reserve some of the lower zone
356 	 * memory (otherwise we risk to run OOM on the lower zones despite
357 	 * there being tons of freeable ram on the higher zones).  This array is
358 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
359 	 * changes.
360 	 */
361 	long lowmem_reserve[MAX_NR_ZONES];
362 
363 #ifdef CONFIG_NUMA
364 	int node;
365 #endif
366 	struct pglist_data	*zone_pgdat;
367 	struct per_cpu_pageset __percpu *pageset;
368 
369 #ifndef CONFIG_SPARSEMEM
370 	/*
371 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
372 	 * In SPARSEMEM, this map is stored in struct mem_section
373 	 */
374 	unsigned long		*pageblock_flags;
375 #endif /* CONFIG_SPARSEMEM */
376 
377 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
378 	unsigned long		zone_start_pfn;
379 
380 	/*
381 	 * spanned_pages is the total pages spanned by the zone, including
382 	 * holes, which is calculated as:
383 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
384 	 *
385 	 * present_pages is physical pages existing within the zone, which
386 	 * is calculated as:
387 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
388 	 *
389 	 * managed_pages is present pages managed by the buddy system, which
390 	 * is calculated as (reserved_pages includes pages allocated by the
391 	 * bootmem allocator):
392 	 *	managed_pages = present_pages - reserved_pages;
393 	 *
394 	 * So present_pages may be used by memory hotplug or memory power
395 	 * management logic to figure out unmanaged pages by checking
396 	 * (present_pages - managed_pages). And managed_pages should be used
397 	 * by page allocator and vm scanner to calculate all kinds of watermarks
398 	 * and thresholds.
399 	 *
400 	 * Locking rules:
401 	 *
402 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
403 	 * It is a seqlock because it has to be read outside of zone->lock,
404 	 * and it is done in the main allocator path.  But, it is written
405 	 * quite infrequently.
406 	 *
407 	 * The span_seq lock is declared along with zone->lock because it is
408 	 * frequently read in proximity to zone->lock.  It's good to
409 	 * give them a chance of being in the same cacheline.
410 	 *
411 	 * Write access to present_pages at runtime should be protected by
412 	 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
413 	 * present_pages should get_online_mems() to get a stable value.
414 	 *
415 	 * Read access to managed_pages should be safe because it's unsigned
416 	 * long. Write access to zone->managed_pages and totalram_pages are
417 	 * protected by managed_page_count_lock at runtime. Idealy only
418 	 * adjust_managed_page_count() should be used instead of directly
419 	 * touching zone->managed_pages and totalram_pages.
420 	 */
421 	unsigned long		managed_pages;
422 	unsigned long		spanned_pages;
423 	unsigned long		present_pages;
424 
425 	const char		*name;
426 
427 #ifdef CONFIG_MEMORY_ISOLATION
428 	/*
429 	 * Number of isolated pageblock. It is used to solve incorrect
430 	 * freepage counting problem due to racy retrieving migratetype
431 	 * of pageblock. Protected by zone->lock.
432 	 */
433 	unsigned long		nr_isolate_pageblock;
434 #endif
435 
436 #ifdef CONFIG_MEMORY_HOTPLUG
437 	/* see spanned/present_pages for more description */
438 	seqlock_t		span_seqlock;
439 #endif
440 
441 	int initialized;
442 
443 	/* Write-intensive fields used from the page allocator */
444 	ZONE_PADDING(_pad1_)
445 
446 	/* free areas of different sizes */
447 	struct free_area	free_area[MAX_ORDER];
448 
449 	/* zone flags, see below */
450 	unsigned long		flags;
451 
452 	/* Primarily protects free_area */
453 	spinlock_t		lock;
454 
455 	/* Write-intensive fields used by compaction and vmstats. */
456 	ZONE_PADDING(_pad2_)
457 
458 	/*
459 	 * When free pages are below this point, additional steps are taken
460 	 * when reading the number of free pages to avoid per-cpu counter
461 	 * drift allowing watermarks to be breached
462 	 */
463 	unsigned long percpu_drift_mark;
464 
465 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
466 	/* pfn where compaction free scanner should start */
467 	unsigned long		compact_cached_free_pfn;
468 	/* pfn where async and sync compaction migration scanner should start */
469 	unsigned long		compact_cached_migrate_pfn[2];
470 #endif
471 
472 #ifdef CONFIG_COMPACTION
473 	/*
474 	 * On compaction failure, 1<<compact_defer_shift compactions
475 	 * are skipped before trying again. The number attempted since
476 	 * last failure is tracked with compact_considered.
477 	 */
478 	unsigned int		compact_considered;
479 	unsigned int		compact_defer_shift;
480 	int			compact_order_failed;
481 #endif
482 
483 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
484 	/* Set to true when the PG_migrate_skip bits should be cleared */
485 	bool			compact_blockskip_flush;
486 #endif
487 
488 	bool			contiguous;
489 
490 	ZONE_PADDING(_pad3_)
491 	/* Zone statistics */
492 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
493 } ____cacheline_internodealigned_in_smp;
494 
495 enum pgdat_flags {
496 	PGDAT_CONGESTED,		/* pgdat has many dirty pages backed by
497 					 * a congested BDI
498 					 */
499 	PGDAT_DIRTY,			/* reclaim scanning has recently found
500 					 * many dirty file pages at the tail
501 					 * of the LRU.
502 					 */
503 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
504 					 * many pages under writeback
505 					 */
506 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
507 };
508 
509 static inline unsigned long zone_end_pfn(const struct zone *zone)
510 {
511 	return zone->zone_start_pfn + zone->spanned_pages;
512 }
513 
514 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
515 {
516 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
517 }
518 
519 static inline bool zone_is_initialized(struct zone *zone)
520 {
521 	return zone->initialized;
522 }
523 
524 static inline bool zone_is_empty(struct zone *zone)
525 {
526 	return zone->spanned_pages == 0;
527 }
528 
529 /*
530  * The "priority" of VM scanning is how much of the queues we will scan in one
531  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
532  * queues ("queue_length >> 12") during an aging round.
533  */
534 #define DEF_PRIORITY 12
535 
536 /* Maximum number of zones on a zonelist */
537 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
538 
539 enum {
540 	ZONELIST_FALLBACK,	/* zonelist with fallback */
541 #ifdef CONFIG_NUMA
542 	/*
543 	 * The NUMA zonelists are doubled because we need zonelists that
544 	 * restrict the allocations to a single node for __GFP_THISNODE.
545 	 */
546 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
547 #endif
548 	MAX_ZONELISTS
549 };
550 
551 /*
552  * This struct contains information about a zone in a zonelist. It is stored
553  * here to avoid dereferences into large structures and lookups of tables
554  */
555 struct zoneref {
556 	struct zone *zone;	/* Pointer to actual zone */
557 	int zone_idx;		/* zone_idx(zoneref->zone) */
558 };
559 
560 /*
561  * One allocation request operates on a zonelist. A zonelist
562  * is a list of zones, the first one is the 'goal' of the
563  * allocation, the other zones are fallback zones, in decreasing
564  * priority.
565  *
566  * To speed the reading of the zonelist, the zonerefs contain the zone index
567  * of the entry being read. Helper functions to access information given
568  * a struct zoneref are
569  *
570  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
571  * zonelist_zone_idx()	- Return the index of the zone for an entry
572  * zonelist_node_idx()	- Return the index of the node for an entry
573  */
574 struct zonelist {
575 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
576 };
577 
578 #ifndef CONFIG_DISCONTIGMEM
579 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
580 extern struct page *mem_map;
581 #endif
582 
583 /*
584  * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
585  * (mostly NUMA machines?) to denote a higher-level memory zone than the
586  * zone denotes.
587  *
588  * On NUMA machines, each NUMA node would have a pg_data_t to describe
589  * it's memory layout.
590  *
591  * Memory statistics and page replacement data structures are maintained on a
592  * per-zone basis.
593  */
594 struct bootmem_data;
595 typedef struct pglist_data {
596 	struct zone node_zones[MAX_NR_ZONES];
597 	struct zonelist node_zonelists[MAX_ZONELISTS];
598 	int nr_zones;
599 #ifdef CONFIG_FLAT_NODE_MEM_MAP	/* means !SPARSEMEM */
600 	struct page *node_mem_map;
601 #ifdef CONFIG_PAGE_EXTENSION
602 	struct page_ext *node_page_ext;
603 #endif
604 #endif
605 #ifndef CONFIG_NO_BOOTMEM
606 	struct bootmem_data *bdata;
607 #endif
608 #ifdef CONFIG_MEMORY_HOTPLUG
609 	/*
610 	 * Must be held any time you expect node_start_pfn, node_present_pages
611 	 * or node_spanned_pages stay constant.  Holding this will also
612 	 * guarantee that any pfn_valid() stays that way.
613 	 *
614 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
615 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
616 	 *
617 	 * Nests above zone->lock and zone->span_seqlock
618 	 */
619 	spinlock_t node_size_lock;
620 #endif
621 	unsigned long node_start_pfn;
622 	unsigned long node_present_pages; /* total number of physical pages */
623 	unsigned long node_spanned_pages; /* total size of physical page
624 					     range, including holes */
625 	int node_id;
626 	wait_queue_head_t kswapd_wait;
627 	wait_queue_head_t pfmemalloc_wait;
628 	struct task_struct *kswapd;	/* Protected by
629 					   mem_hotplug_begin/end() */
630 	int kswapd_order;
631 	enum zone_type kswapd_classzone_idx;
632 
633 #ifdef CONFIG_COMPACTION
634 	int kcompactd_max_order;
635 	enum zone_type kcompactd_classzone_idx;
636 	wait_queue_head_t kcompactd_wait;
637 	struct task_struct *kcompactd;
638 #endif
639 #ifdef CONFIG_NUMA_BALANCING
640 	/* Lock serializing the migrate rate limiting window */
641 	spinlock_t numabalancing_migrate_lock;
642 
643 	/* Rate limiting time interval */
644 	unsigned long numabalancing_migrate_next_window;
645 
646 	/* Number of pages migrated during the rate limiting time interval */
647 	unsigned long numabalancing_migrate_nr_pages;
648 #endif
649 	/*
650 	 * This is a per-node reserve of pages that are not available
651 	 * to userspace allocations.
652 	 */
653 	unsigned long		totalreserve_pages;
654 
655 #ifdef CONFIG_NUMA
656 	/*
657 	 * zone reclaim becomes active if more unmapped pages exist.
658 	 */
659 	unsigned long		min_unmapped_pages;
660 	unsigned long		min_slab_pages;
661 #endif /* CONFIG_NUMA */
662 
663 	/* Write-intensive fields used by page reclaim */
664 	ZONE_PADDING(_pad1_)
665 	spinlock_t		lru_lock;
666 
667 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
668 	/*
669 	 * If memory initialisation on large machines is deferred then this
670 	 * is the first PFN that needs to be initialised.
671 	 */
672 	unsigned long first_deferred_pfn;
673 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
674 
675 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
676 	spinlock_t split_queue_lock;
677 	struct list_head split_queue;
678 	unsigned long split_queue_len;
679 #endif
680 
681 	/* Fields commonly accessed by the page reclaim scanner */
682 	struct lruvec		lruvec;
683 
684 	/*
685 	 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
686 	 * this node's LRU.  Maintained by the pageout code.
687 	 */
688 	unsigned int inactive_ratio;
689 
690 	unsigned long		flags;
691 
692 	ZONE_PADDING(_pad2_)
693 
694 	/* Per-node vmstats */
695 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
696 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
697 } pg_data_t;
698 
699 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
700 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
701 #ifdef CONFIG_FLAT_NODE_MEM_MAP
702 #define pgdat_page_nr(pgdat, pagenr)	((pgdat)->node_mem_map + (pagenr))
703 #else
704 #define pgdat_page_nr(pgdat, pagenr)	pfn_to_page((pgdat)->node_start_pfn + (pagenr))
705 #endif
706 #define nid_page_nr(nid, pagenr) 	pgdat_page_nr(NODE_DATA(nid),(pagenr))
707 
708 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
709 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
710 static inline spinlock_t *zone_lru_lock(struct zone *zone)
711 {
712 	return &zone->zone_pgdat->lru_lock;
713 }
714 
715 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
716 {
717 	return &pgdat->lruvec;
718 }
719 
720 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
721 {
722 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
723 }
724 
725 static inline bool pgdat_is_empty(pg_data_t *pgdat)
726 {
727 	return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
728 }
729 
730 static inline int zone_id(const struct zone *zone)
731 {
732 	struct pglist_data *pgdat = zone->zone_pgdat;
733 
734 	return zone - pgdat->node_zones;
735 }
736 
737 #ifdef CONFIG_ZONE_DEVICE
738 static inline bool is_dev_zone(const struct zone *zone)
739 {
740 	return zone_id(zone) == ZONE_DEVICE;
741 }
742 #else
743 static inline bool is_dev_zone(const struct zone *zone)
744 {
745 	return false;
746 }
747 #endif
748 
749 #include <linux/memory_hotplug.h>
750 
751 extern struct mutex zonelists_mutex;
752 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
753 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
754 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
755 			 int classzone_idx, unsigned int alloc_flags,
756 			 long free_pages);
757 bool zone_watermark_ok(struct zone *z, unsigned int order,
758 		unsigned long mark, int classzone_idx,
759 		unsigned int alloc_flags);
760 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
761 		unsigned long mark, int classzone_idx);
762 enum memmap_context {
763 	MEMMAP_EARLY,
764 	MEMMAP_HOTPLUG,
765 };
766 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
767 				     unsigned long size);
768 
769 extern void lruvec_init(struct lruvec *lruvec);
770 
771 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
772 {
773 #ifdef CONFIG_MEMCG
774 	return lruvec->pgdat;
775 #else
776 	return container_of(lruvec, struct pglist_data, lruvec);
777 #endif
778 }
779 
780 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
781 
782 #ifdef CONFIG_HAVE_MEMORY_PRESENT
783 void memory_present(int nid, unsigned long start, unsigned long end);
784 #else
785 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
786 #endif
787 
788 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
789 int local_memory_node(int node_id);
790 #else
791 static inline int local_memory_node(int node_id) { return node_id; };
792 #endif
793 
794 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
795 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
796 #endif
797 
798 /*
799  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
800  */
801 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
802 
803 /*
804  * Returns true if a zone has pages managed by the buddy allocator.
805  * All the reclaim decisions have to use this function rather than
806  * populated_zone(). If the whole zone is reserved then we can easily
807  * end up with populated_zone() && !managed_zone().
808  */
809 static inline bool managed_zone(struct zone *zone)
810 {
811 	return zone->managed_pages;
812 }
813 
814 /* Returns true if a zone has memory */
815 static inline bool populated_zone(struct zone *zone)
816 {
817 	return zone->present_pages;
818 }
819 
820 extern int movable_zone;
821 
822 #ifdef CONFIG_HIGHMEM
823 static inline int zone_movable_is_highmem(void)
824 {
825 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
826 	return movable_zone == ZONE_HIGHMEM;
827 #else
828 	return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
829 #endif
830 }
831 #endif
832 
833 static inline int is_highmem_idx(enum zone_type idx)
834 {
835 #ifdef CONFIG_HIGHMEM
836 	return (idx == ZONE_HIGHMEM ||
837 		(idx == ZONE_MOVABLE && zone_movable_is_highmem()));
838 #else
839 	return 0;
840 #endif
841 }
842 
843 /**
844  * is_highmem - helper function to quickly check if a struct zone is a
845  *              highmem zone or not.  This is an attempt to keep references
846  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
847  * @zone - pointer to struct zone variable
848  */
849 static inline int is_highmem(struct zone *zone)
850 {
851 #ifdef CONFIG_HIGHMEM
852 	return is_highmem_idx(zone_idx(zone));
853 #else
854 	return 0;
855 #endif
856 }
857 
858 /* These two functions are used to setup the per zone pages min values */
859 struct ctl_table;
860 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
861 					void __user *, size_t *, loff_t *);
862 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
863 					void __user *, size_t *, loff_t *);
864 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
865 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
866 					void __user *, size_t *, loff_t *);
867 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
868 					void __user *, size_t *, loff_t *);
869 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
870 			void __user *, size_t *, loff_t *);
871 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
872 			void __user *, size_t *, loff_t *);
873 
874 extern int numa_zonelist_order_handler(struct ctl_table *, int,
875 			void __user *, size_t *, loff_t *);
876 extern char numa_zonelist_order[];
877 #define NUMA_ZONELIST_ORDER_LEN 16	/* string buffer size */
878 
879 #ifndef CONFIG_NEED_MULTIPLE_NODES
880 
881 extern struct pglist_data contig_page_data;
882 #define NODE_DATA(nid)		(&contig_page_data)
883 #define NODE_MEM_MAP(nid)	mem_map
884 
885 #else /* CONFIG_NEED_MULTIPLE_NODES */
886 
887 #include <asm/mmzone.h>
888 
889 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
890 
891 extern struct pglist_data *first_online_pgdat(void);
892 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
893 extern struct zone *next_zone(struct zone *zone);
894 
895 /**
896  * for_each_online_pgdat - helper macro to iterate over all online nodes
897  * @pgdat - pointer to a pg_data_t variable
898  */
899 #define for_each_online_pgdat(pgdat)			\
900 	for (pgdat = first_online_pgdat();		\
901 	     pgdat;					\
902 	     pgdat = next_online_pgdat(pgdat))
903 /**
904  * for_each_zone - helper macro to iterate over all memory zones
905  * @zone - pointer to struct zone variable
906  *
907  * The user only needs to declare the zone variable, for_each_zone
908  * fills it in.
909  */
910 #define for_each_zone(zone)			        \
911 	for (zone = (first_online_pgdat())->node_zones; \
912 	     zone;					\
913 	     zone = next_zone(zone))
914 
915 #define for_each_populated_zone(zone)		        \
916 	for (zone = (first_online_pgdat())->node_zones; \
917 	     zone;					\
918 	     zone = next_zone(zone))			\
919 		if (!populated_zone(zone))		\
920 			; /* do nothing */		\
921 		else
922 
923 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
924 {
925 	return zoneref->zone;
926 }
927 
928 static inline int zonelist_zone_idx(struct zoneref *zoneref)
929 {
930 	return zoneref->zone_idx;
931 }
932 
933 static inline int zonelist_node_idx(struct zoneref *zoneref)
934 {
935 #ifdef CONFIG_NUMA
936 	/* zone_to_nid not available in this context */
937 	return zoneref->zone->node;
938 #else
939 	return 0;
940 #endif /* CONFIG_NUMA */
941 }
942 
943 struct zoneref *__next_zones_zonelist(struct zoneref *z,
944 					enum zone_type highest_zoneidx,
945 					nodemask_t *nodes);
946 
947 /**
948  * 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
949  * @z - The cursor used as a starting point for the search
950  * @highest_zoneidx - The zone index of the highest zone to return
951  * @nodes - An optional nodemask to filter the zonelist with
952  *
953  * This function returns the next zone at or below a given zone index that is
954  * within the allowed nodemask using a cursor as the starting point for the
955  * search. The zoneref returned is a cursor that represents the current zone
956  * being examined. It should be advanced by one before calling
957  * next_zones_zonelist again.
958  */
959 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
960 					enum zone_type highest_zoneidx,
961 					nodemask_t *nodes)
962 {
963 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
964 		return z;
965 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
966 }
967 
968 /**
969  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
970  * @zonelist - The zonelist to search for a suitable zone
971  * @highest_zoneidx - The zone index of the highest zone to return
972  * @nodes - An optional nodemask to filter the zonelist with
973  * @return - Zoneref pointer for the first suitable zone found (see below)
974  *
975  * This function returns the first zone at or below a given zone index that is
976  * within the allowed nodemask. The zoneref returned is a cursor that can be
977  * used to iterate the zonelist with next_zones_zonelist by advancing it by
978  * one before calling.
979  *
980  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
981  * never NULL). This may happen either genuinely, or due to concurrent nodemask
982  * update due to cpuset modification.
983  */
984 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
985 					enum zone_type highest_zoneidx,
986 					nodemask_t *nodes)
987 {
988 	return next_zones_zonelist(zonelist->_zonerefs,
989 							highest_zoneidx, nodes);
990 }
991 
992 /**
993  * 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
994  * @zone - The current zone in the iterator
995  * @z - The current pointer within zonelist->zones being iterated
996  * @zlist - The zonelist being iterated
997  * @highidx - The zone index of the highest zone to return
998  * @nodemask - Nodemask allowed by the allocator
999  *
1000  * This iterator iterates though all zones at or below a given zone index and
1001  * within a given nodemask
1002  */
1003 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1004 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1005 		zone;							\
1006 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1007 			zone = zonelist_zone(z))
1008 
1009 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1010 	for (zone = z->zone;	\
1011 		zone;							\
1012 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1013 			zone = zonelist_zone(z))
1014 
1015 
1016 /**
1017  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1018  * @zone - The current zone in the iterator
1019  * @z - The current pointer within zonelist->zones being iterated
1020  * @zlist - The zonelist being iterated
1021  * @highidx - The zone index of the highest zone to return
1022  *
1023  * This iterator iterates though all zones at or below a given zone index.
1024  */
1025 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1026 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1027 
1028 #ifdef CONFIG_SPARSEMEM
1029 #include <asm/sparsemem.h>
1030 #endif
1031 
1032 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1033 	!defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1034 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1035 {
1036 	return 0;
1037 }
1038 #endif
1039 
1040 #ifdef CONFIG_FLATMEM
1041 #define pfn_to_nid(pfn)		(0)
1042 #endif
1043 
1044 #ifdef CONFIG_SPARSEMEM
1045 
1046 /*
1047  * SECTION_SHIFT    		#bits space required to store a section #
1048  *
1049  * PA_SECTION_SHIFT		physical address to/from section number
1050  * PFN_SECTION_SHIFT		pfn to/from section number
1051  */
1052 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1053 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1054 
1055 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1056 
1057 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1058 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1059 
1060 #define SECTION_BLOCKFLAGS_BITS \
1061 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1062 
1063 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1064 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1065 #endif
1066 
1067 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1068 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1069 
1070 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1071 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1072 
1073 struct page;
1074 struct page_ext;
1075 struct mem_section {
1076 	/*
1077 	 * This is, logically, a pointer to an array of struct
1078 	 * pages.  However, it is stored with some other magic.
1079 	 * (see sparse.c::sparse_init_one_section())
1080 	 *
1081 	 * Additionally during early boot we encode node id of
1082 	 * the location of the section here to guide allocation.
1083 	 * (see sparse.c::memory_present())
1084 	 *
1085 	 * Making it a UL at least makes someone do a cast
1086 	 * before using it wrong.
1087 	 */
1088 	unsigned long section_mem_map;
1089 
1090 	/* See declaration of similar field in struct zone */
1091 	unsigned long *pageblock_flags;
1092 #ifdef CONFIG_PAGE_EXTENSION
1093 	/*
1094 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1095 	 * section. (see page_ext.h about this.)
1096 	 */
1097 	struct page_ext *page_ext;
1098 	unsigned long pad;
1099 #endif
1100 	/*
1101 	 * WARNING: mem_section must be a power-of-2 in size for the
1102 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1103 	 */
1104 };
1105 
1106 #ifdef CONFIG_SPARSEMEM_EXTREME
1107 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1108 #else
1109 #define SECTIONS_PER_ROOT	1
1110 #endif
1111 
1112 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1113 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1114 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1115 
1116 #ifdef CONFIG_SPARSEMEM_EXTREME
1117 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1118 #else
1119 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1120 #endif
1121 
1122 static inline struct mem_section *__nr_to_section(unsigned long nr)
1123 {
1124 	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1125 		return NULL;
1126 	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1127 }
1128 extern int __section_nr(struct mem_section* ms);
1129 extern unsigned long usemap_size(void);
1130 
1131 /*
1132  * We use the lower bits of the mem_map pointer to store
1133  * a little bit of information.  There should be at least
1134  * 3 bits here due to 32-bit alignment.
1135  */
1136 #define	SECTION_MARKED_PRESENT	(1UL<<0)
1137 #define SECTION_HAS_MEM_MAP	(1UL<<1)
1138 #define SECTION_MAP_LAST_BIT	(1UL<<2)
1139 #define SECTION_MAP_MASK	(~(SECTION_MAP_LAST_BIT-1))
1140 #define SECTION_NID_SHIFT	2
1141 
1142 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1143 {
1144 	unsigned long map = section->section_mem_map;
1145 	map &= SECTION_MAP_MASK;
1146 	return (struct page *)map;
1147 }
1148 
1149 static inline int present_section(struct mem_section *section)
1150 {
1151 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1152 }
1153 
1154 static inline int present_section_nr(unsigned long nr)
1155 {
1156 	return present_section(__nr_to_section(nr));
1157 }
1158 
1159 static inline int valid_section(struct mem_section *section)
1160 {
1161 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1162 }
1163 
1164 static inline int valid_section_nr(unsigned long nr)
1165 {
1166 	return valid_section(__nr_to_section(nr));
1167 }
1168 
1169 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1170 {
1171 	return __nr_to_section(pfn_to_section_nr(pfn));
1172 }
1173 
1174 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1175 static inline int pfn_valid(unsigned long pfn)
1176 {
1177 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1178 		return 0;
1179 	return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1180 }
1181 #endif
1182 
1183 static inline int pfn_present(unsigned long pfn)
1184 {
1185 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1186 		return 0;
1187 	return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1188 }
1189 
1190 /*
1191  * These are _only_ used during initialisation, therefore they
1192  * can use __initdata ...  They could have names to indicate
1193  * this restriction.
1194  */
1195 #ifdef CONFIG_NUMA
1196 #define pfn_to_nid(pfn)							\
1197 ({									\
1198 	unsigned long __pfn_to_nid_pfn = (pfn);				\
1199 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
1200 })
1201 #else
1202 #define pfn_to_nid(pfn)		(0)
1203 #endif
1204 
1205 #define early_pfn_valid(pfn)	pfn_valid(pfn)
1206 void sparse_init(void);
1207 #else
1208 #define sparse_init()	do {} while (0)
1209 #define sparse_index_init(_sec, _nid)  do {} while (0)
1210 #endif /* CONFIG_SPARSEMEM */
1211 
1212 /*
1213  * During memory init memblocks map pfns to nids. The search is expensive and
1214  * this caches recent lookups. The implementation of __early_pfn_to_nid
1215  * may treat start/end as pfns or sections.
1216  */
1217 struct mminit_pfnnid_cache {
1218 	unsigned long last_start;
1219 	unsigned long last_end;
1220 	int last_nid;
1221 };
1222 
1223 #ifndef early_pfn_valid
1224 #define early_pfn_valid(pfn)	(1)
1225 #endif
1226 
1227 void memory_present(int nid, unsigned long start, unsigned long end);
1228 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1229 
1230 /*
1231  * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1232  * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1233  * pfn_valid_within() should be used in this case; we optimise this away
1234  * when we have no holes within a MAX_ORDER_NR_PAGES block.
1235  */
1236 #ifdef CONFIG_HOLES_IN_ZONE
1237 #define pfn_valid_within(pfn) pfn_valid(pfn)
1238 #else
1239 #define pfn_valid_within(pfn) (1)
1240 #endif
1241 
1242 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1243 /*
1244  * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1245  * associated with it or not. In FLATMEM, it is expected that holes always
1246  * have valid memmap as long as there is valid PFNs either side of the hole.
1247  * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1248  * entire section.
1249  *
1250  * However, an ARM, and maybe other embedded architectures in the future
1251  * free memmap backing holes to save memory on the assumption the memmap is
1252  * never used. The page_zone linkages are then broken even though pfn_valid()
1253  * returns true. A walker of the full memmap must then do this additional
1254  * check to ensure the memmap they are looking at is sane by making sure
1255  * the zone and PFN linkages are still valid. This is expensive, but walkers
1256  * of the full memmap are extremely rare.
1257  */
1258 bool memmap_valid_within(unsigned long pfn,
1259 					struct page *page, struct zone *zone);
1260 #else
1261 static inline bool memmap_valid_within(unsigned long pfn,
1262 					struct page *page, struct zone *zone)
1263 {
1264 	return true;
1265 }
1266 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1267 
1268 #endif /* !__GENERATING_BOUNDS.H */
1269 #endif /* !__ASSEMBLY__ */
1270 #endif /* _LINUX_MMZONE_H */
1271