xref: /linux-6.15/include/linux/mmzone.h (revision 6852aa02)
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 migratetype {
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 	WORKINGSET_REFAULT,
153 	WORKINGSET_ACTIVATE,
154 	WORKINGSET_NODERECLAIM,
155 	NR_ANON_MAPPED,	/* Mapped anonymous pages */
156 	NR_FILE_MAPPED,	/* pagecache pages mapped into pagetables.
157 			   only modified from process context */
158 	NR_FILE_PAGES,
159 	NR_FILE_DIRTY,
160 	NR_WRITEBACK,
161 	NR_WRITEBACK_TEMP,	/* Writeback using temporary buffers */
162 	NR_SHMEM,		/* shmem pages (included tmpfs/GEM pages) */
163 	NR_SHMEM_THPS,
164 	NR_SHMEM_PMDMAPPED,
165 	NR_ANON_THPS,
166 	NR_UNSTABLE_NFS,	/* NFS unstable pages */
167 	NR_VMSCAN_WRITE,
168 	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
169 	NR_DIRTIED,		/* page dirtyings since bootup */
170 	NR_WRITTEN,		/* page writings since bootup */
171 	NR_VM_NODE_STAT_ITEMS
172 };
173 
174 /*
175  * We do arithmetic on the LRU lists in various places in the code,
176  * so it is important to keep the active lists LRU_ACTIVE higher in
177  * the array than the corresponding inactive lists, and to keep
178  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
179  *
180  * This has to be kept in sync with the statistics in zone_stat_item
181  * above and the descriptions in vmstat_text in mm/vmstat.c
182  */
183 #define LRU_BASE 0
184 #define LRU_ACTIVE 1
185 #define LRU_FILE 2
186 
187 enum lru_list {
188 	LRU_INACTIVE_ANON = LRU_BASE,
189 	LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
190 	LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
191 	LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
192 	LRU_UNEVICTABLE,
193 	NR_LRU_LISTS
194 };
195 
196 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
197 
198 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
199 
200 static inline int is_file_lru(enum lru_list lru)
201 {
202 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
203 }
204 
205 static inline int is_active_lru(enum lru_list lru)
206 {
207 	return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
208 }
209 
210 struct zone_reclaim_stat {
211 	/*
212 	 * The pageout code in vmscan.c keeps track of how many of the
213 	 * mem/swap backed and file backed pages are referenced.
214 	 * The higher the rotated/scanned ratio, the more valuable
215 	 * that cache is.
216 	 *
217 	 * The anon LRU stats live in [0], file LRU stats in [1]
218 	 */
219 	unsigned long		recent_rotated[2];
220 	unsigned long		recent_scanned[2];
221 };
222 
223 struct lruvec {
224 	struct list_head		lists[NR_LRU_LISTS];
225 	struct zone_reclaim_stat	reclaim_stat;
226 	/* Evictions & activations on the inactive file list */
227 	atomic_long_t			inactive_age;
228 	/* Refaults at the time of last reclaim cycle */
229 	unsigned long			refaults;
230 #ifdef CONFIG_MEMCG
231 	struct pglist_data *pgdat;
232 #endif
233 };
234 
235 /* Mask used at gathering information at once (see memcontrol.c) */
236 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
237 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
238 #define LRU_ALL	     ((1 << NR_LRU_LISTS) - 1)
239 
240 /* Isolate unmapped file */
241 #define ISOLATE_UNMAPPED	((__force isolate_mode_t)0x2)
242 /* Isolate for asynchronous migration */
243 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
244 /* Isolate unevictable pages */
245 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
246 
247 /* LRU Isolation modes. */
248 typedef unsigned __bitwise isolate_mode_t;
249 
250 enum zone_watermarks {
251 	WMARK_MIN,
252 	WMARK_LOW,
253 	WMARK_HIGH,
254 	NR_WMARK
255 };
256 
257 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
258 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
259 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
260 
261 struct per_cpu_pages {
262 	int count;		/* number of pages in the list */
263 	int high;		/* high watermark, emptying needed */
264 	int batch;		/* chunk size for buddy add/remove */
265 
266 	/* Lists of pages, one per migrate type stored on the pcp-lists */
267 	struct list_head lists[MIGRATE_PCPTYPES];
268 };
269 
270 struct per_cpu_pageset {
271 	struct per_cpu_pages pcp;
272 #ifdef CONFIG_NUMA
273 	s8 expire;
274 #endif
275 #ifdef CONFIG_SMP
276 	s8 stat_threshold;
277 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
278 #endif
279 };
280 
281 struct per_cpu_nodestat {
282 	s8 stat_threshold;
283 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
284 };
285 
286 #endif /* !__GENERATING_BOUNDS.H */
287 
288 enum zone_type {
289 #ifdef CONFIG_ZONE_DMA
290 	/*
291 	 * ZONE_DMA is used when there are devices that are not able
292 	 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
293 	 * carve out the portion of memory that is needed for these devices.
294 	 * The range is arch specific.
295 	 *
296 	 * Some examples
297 	 *
298 	 * Architecture		Limit
299 	 * ---------------------------
300 	 * parisc, ia64, sparc	<4G
301 	 * s390			<2G
302 	 * arm			Various
303 	 * alpha		Unlimited or 0-16MB.
304 	 *
305 	 * i386, x86_64 and multiple other arches
306 	 * 			<16M.
307 	 */
308 	ZONE_DMA,
309 #endif
310 #ifdef CONFIG_ZONE_DMA32
311 	/*
312 	 * x86_64 needs two ZONE_DMAs because it supports devices that are
313 	 * only able to do DMA to the lower 16M but also 32 bit devices that
314 	 * can only do DMA areas below 4G.
315 	 */
316 	ZONE_DMA32,
317 #endif
318 	/*
319 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
320 	 * performed on pages in ZONE_NORMAL if the DMA devices support
321 	 * transfers to all addressable memory.
322 	 */
323 	ZONE_NORMAL,
324 #ifdef CONFIG_HIGHMEM
325 	/*
326 	 * A memory area that is only addressable by the kernel through
327 	 * mapping portions into its own address space. This is for example
328 	 * used by i386 to allow the kernel to address the memory beyond
329 	 * 900MB. The kernel will set up special mappings (page
330 	 * table entries on i386) for each page that the kernel needs to
331 	 * access.
332 	 */
333 	ZONE_HIGHMEM,
334 #endif
335 	ZONE_MOVABLE,
336 #ifdef CONFIG_ZONE_DEVICE
337 	ZONE_DEVICE,
338 #endif
339 	__MAX_NR_ZONES
340 
341 };
342 
343 #ifndef __GENERATING_BOUNDS_H
344 
345 struct zone {
346 	/* Read-mostly fields */
347 
348 	/* zone watermarks, access with *_wmark_pages(zone) macros */
349 	unsigned long watermark[NR_WMARK];
350 
351 	unsigned long nr_reserved_highatomic;
352 
353 	/*
354 	 * We don't know if the memory that we're going to allocate will be
355 	 * freeable or/and it will be released eventually, so to avoid totally
356 	 * wasting several GB of ram we must reserve some of the lower zone
357 	 * memory (otherwise we risk to run OOM on the lower zones despite
358 	 * there being tons of freeable ram on the higher zones).  This array is
359 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
360 	 * changes.
361 	 */
362 	long lowmem_reserve[MAX_NR_ZONES];
363 
364 #ifdef CONFIG_NUMA
365 	int node;
366 #endif
367 	struct pglist_data	*zone_pgdat;
368 	struct per_cpu_pageset __percpu *pageset;
369 
370 #ifndef CONFIG_SPARSEMEM
371 	/*
372 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
373 	 * In SPARSEMEM, this map is stored in struct mem_section
374 	 */
375 	unsigned long		*pageblock_flags;
376 #endif /* CONFIG_SPARSEMEM */
377 
378 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
379 	unsigned long		zone_start_pfn;
380 
381 	/*
382 	 * spanned_pages is the total pages spanned by the zone, including
383 	 * holes, which is calculated as:
384 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
385 	 *
386 	 * present_pages is physical pages existing within the zone, which
387 	 * is calculated as:
388 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
389 	 *
390 	 * managed_pages is present pages managed by the buddy system, which
391 	 * is calculated as (reserved_pages includes pages allocated by the
392 	 * bootmem allocator):
393 	 *	managed_pages = present_pages - reserved_pages;
394 	 *
395 	 * So present_pages may be used by memory hotplug or memory power
396 	 * management logic to figure out unmanaged pages by checking
397 	 * (present_pages - managed_pages). And managed_pages should be used
398 	 * by page allocator and vm scanner to calculate all kinds of watermarks
399 	 * and thresholds.
400 	 *
401 	 * Locking rules:
402 	 *
403 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
404 	 * It is a seqlock because it has to be read outside of zone->lock,
405 	 * and it is done in the main allocator path.  But, it is written
406 	 * quite infrequently.
407 	 *
408 	 * The span_seq lock is declared along with zone->lock because it is
409 	 * frequently read in proximity to zone->lock.  It's good to
410 	 * give them a chance of being in the same cacheline.
411 	 *
412 	 * Write access to present_pages at runtime should be protected by
413 	 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
414 	 * present_pages should get_online_mems() to get a stable value.
415 	 *
416 	 * Read access to managed_pages should be safe because it's unsigned
417 	 * long. Write access to zone->managed_pages and totalram_pages are
418 	 * protected by managed_page_count_lock at runtime. Idealy only
419 	 * adjust_managed_page_count() should be used instead of directly
420 	 * touching zone->managed_pages and totalram_pages.
421 	 */
422 	unsigned long		managed_pages;
423 	unsigned long		spanned_pages;
424 	unsigned long		present_pages;
425 
426 	const char		*name;
427 
428 #ifdef CONFIG_MEMORY_ISOLATION
429 	/*
430 	 * Number of isolated pageblock. It is used to solve incorrect
431 	 * freepage counting problem due to racy retrieving migratetype
432 	 * of pageblock. Protected by zone->lock.
433 	 */
434 	unsigned long		nr_isolate_pageblock;
435 #endif
436 
437 #ifdef CONFIG_MEMORY_HOTPLUG
438 	/* see spanned/present_pages for more description */
439 	seqlock_t		span_seqlock;
440 #endif
441 
442 	int initialized;
443 
444 	/* Write-intensive fields used from the page allocator */
445 	ZONE_PADDING(_pad1_)
446 
447 	/* free areas of different sizes */
448 	struct free_area	free_area[MAX_ORDER];
449 
450 	/* zone flags, see below */
451 	unsigned long		flags;
452 
453 	/* Primarily protects free_area */
454 	spinlock_t		lock;
455 
456 	/* Write-intensive fields used by compaction and vmstats. */
457 	ZONE_PADDING(_pad2_)
458 
459 	/*
460 	 * When free pages are below this point, additional steps are taken
461 	 * when reading the number of free pages to avoid per-cpu counter
462 	 * drift allowing watermarks to be breached
463 	 */
464 	unsigned long percpu_drift_mark;
465 
466 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
467 	/* pfn where compaction free scanner should start */
468 	unsigned long		compact_cached_free_pfn;
469 	/* pfn where async and sync compaction migration scanner should start */
470 	unsigned long		compact_cached_migrate_pfn[2];
471 #endif
472 
473 #ifdef CONFIG_COMPACTION
474 	/*
475 	 * On compaction failure, 1<<compact_defer_shift compactions
476 	 * are skipped before trying again. The number attempted since
477 	 * last failure is tracked with compact_considered.
478 	 */
479 	unsigned int		compact_considered;
480 	unsigned int		compact_defer_shift;
481 	int			compact_order_failed;
482 #endif
483 
484 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
485 	/* Set to true when the PG_migrate_skip bits should be cleared */
486 	bool			compact_blockskip_flush;
487 #endif
488 
489 	bool			contiguous;
490 
491 	ZONE_PADDING(_pad3_)
492 	/* Zone statistics */
493 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
494 } ____cacheline_internodealigned_in_smp;
495 
496 enum pgdat_flags {
497 	PGDAT_CONGESTED,		/* pgdat has many dirty pages backed by
498 					 * a congested BDI
499 					 */
500 	PGDAT_DIRTY,			/* reclaim scanning has recently found
501 					 * many dirty file pages at the tail
502 					 * of the LRU.
503 					 */
504 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
505 					 * many pages under writeback
506 					 */
507 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
508 };
509 
510 static inline unsigned long zone_end_pfn(const struct zone *zone)
511 {
512 	return zone->zone_start_pfn + zone->spanned_pages;
513 }
514 
515 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
516 {
517 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
518 }
519 
520 static inline bool zone_is_initialized(struct zone *zone)
521 {
522 	return zone->initialized;
523 }
524 
525 static inline bool zone_is_empty(struct zone *zone)
526 {
527 	return zone->spanned_pages == 0;
528 }
529 
530 /*
531  * The "priority" of VM scanning is how much of the queues we will scan in one
532  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
533  * queues ("queue_length >> 12") during an aging round.
534  */
535 #define DEF_PRIORITY 12
536 
537 /* Maximum number of zones on a zonelist */
538 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
539 
540 enum {
541 	ZONELIST_FALLBACK,	/* zonelist with fallback */
542 #ifdef CONFIG_NUMA
543 	/*
544 	 * The NUMA zonelists are doubled because we need zonelists that
545 	 * restrict the allocations to a single node for __GFP_THISNODE.
546 	 */
547 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
548 #endif
549 	MAX_ZONELISTS
550 };
551 
552 /*
553  * This struct contains information about a zone in a zonelist. It is stored
554  * here to avoid dereferences into large structures and lookups of tables
555  */
556 struct zoneref {
557 	struct zone *zone;	/* Pointer to actual zone */
558 	int zone_idx;		/* zone_idx(zoneref->zone) */
559 };
560 
561 /*
562  * One allocation request operates on a zonelist. A zonelist
563  * is a list of zones, the first one is the 'goal' of the
564  * allocation, the other zones are fallback zones, in decreasing
565  * priority.
566  *
567  * To speed the reading of the zonelist, the zonerefs contain the zone index
568  * of the entry being read. Helper functions to access information given
569  * a struct zoneref are
570  *
571  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
572  * zonelist_zone_idx()	- Return the index of the zone for an entry
573  * zonelist_node_idx()	- Return the index of the node for an entry
574  */
575 struct zonelist {
576 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
577 };
578 
579 #ifndef CONFIG_DISCONTIGMEM
580 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
581 extern struct page *mem_map;
582 #endif
583 
584 /*
585  * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
586  * (mostly NUMA machines?) to denote a higher-level memory zone than the
587  * zone denotes.
588  *
589  * On NUMA machines, each NUMA node would have a pg_data_t to describe
590  * it's memory layout.
591  *
592  * Memory statistics and page replacement data structures are maintained on a
593  * per-zone basis.
594  */
595 struct bootmem_data;
596 typedef struct pglist_data {
597 	struct zone node_zones[MAX_NR_ZONES];
598 	struct zonelist node_zonelists[MAX_ZONELISTS];
599 	int nr_zones;
600 #ifdef CONFIG_FLAT_NODE_MEM_MAP	/* means !SPARSEMEM */
601 	struct page *node_mem_map;
602 #ifdef CONFIG_PAGE_EXTENSION
603 	struct page_ext *node_page_ext;
604 #endif
605 #endif
606 #ifndef CONFIG_NO_BOOTMEM
607 	struct bootmem_data *bdata;
608 #endif
609 #ifdef CONFIG_MEMORY_HOTPLUG
610 	/*
611 	 * Must be held any time you expect node_start_pfn, node_present_pages
612 	 * or node_spanned_pages stay constant.  Holding this will also
613 	 * guarantee that any pfn_valid() stays that way.
614 	 *
615 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
616 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
617 	 *
618 	 * Nests above zone->lock and zone->span_seqlock
619 	 */
620 	spinlock_t node_size_lock;
621 #endif
622 	unsigned long node_start_pfn;
623 	unsigned long node_present_pages; /* total number of physical pages */
624 	unsigned long node_spanned_pages; /* total size of physical page
625 					     range, including holes */
626 	int node_id;
627 	wait_queue_head_t kswapd_wait;
628 	wait_queue_head_t pfmemalloc_wait;
629 	struct task_struct *kswapd;	/* Protected by
630 					   mem_hotplug_begin/end() */
631 	int kswapd_order;
632 	enum zone_type kswapd_classzone_idx;
633 
634 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
635 
636 #ifdef CONFIG_COMPACTION
637 	int kcompactd_max_order;
638 	enum zone_type kcompactd_classzone_idx;
639 	wait_queue_head_t kcompactd_wait;
640 	struct task_struct *kcompactd;
641 #endif
642 #ifdef CONFIG_NUMA_BALANCING
643 	/* Lock serializing the migrate rate limiting window */
644 	spinlock_t numabalancing_migrate_lock;
645 
646 	/* Rate limiting time interval */
647 	unsigned long numabalancing_migrate_next_window;
648 
649 	/* Number of pages migrated during the rate limiting time interval */
650 	unsigned long numabalancing_migrate_nr_pages;
651 #endif
652 	/*
653 	 * This is a per-node reserve of pages that are not available
654 	 * to userspace allocations.
655 	 */
656 	unsigned long		totalreserve_pages;
657 
658 #ifdef CONFIG_NUMA
659 	/*
660 	 * zone reclaim becomes active if more unmapped pages exist.
661 	 */
662 	unsigned long		min_unmapped_pages;
663 	unsigned long		min_slab_pages;
664 #endif /* CONFIG_NUMA */
665 
666 	/* Write-intensive fields used by page reclaim */
667 	ZONE_PADDING(_pad1_)
668 	spinlock_t		lru_lock;
669 
670 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
671 	/*
672 	 * If memory initialisation on large machines is deferred then this
673 	 * is the first PFN that needs to be initialised.
674 	 */
675 	unsigned long first_deferred_pfn;
676 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
677 
678 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
679 	spinlock_t split_queue_lock;
680 	struct list_head split_queue;
681 	unsigned long split_queue_len;
682 #endif
683 
684 	/* Fields commonly accessed by the page reclaim scanner */
685 	struct lruvec		lruvec;
686 
687 	/*
688 	 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
689 	 * this node's LRU.  Maintained by the pageout code.
690 	 */
691 	unsigned int inactive_ratio;
692 
693 	unsigned long		flags;
694 
695 	ZONE_PADDING(_pad2_)
696 
697 	/* Per-node vmstats */
698 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
699 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
700 } pg_data_t;
701 
702 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
703 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
704 #ifdef CONFIG_FLAT_NODE_MEM_MAP
705 #define pgdat_page_nr(pgdat, pagenr)	((pgdat)->node_mem_map + (pagenr))
706 #else
707 #define pgdat_page_nr(pgdat, pagenr)	pfn_to_page((pgdat)->node_start_pfn + (pagenr))
708 #endif
709 #define nid_page_nr(nid, pagenr) 	pgdat_page_nr(NODE_DATA(nid),(pagenr))
710 
711 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
712 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
713 static inline spinlock_t *zone_lru_lock(struct zone *zone)
714 {
715 	return &zone->zone_pgdat->lru_lock;
716 }
717 
718 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
719 {
720 	return &pgdat->lruvec;
721 }
722 
723 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
724 {
725 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
726 }
727 
728 static inline bool pgdat_is_empty(pg_data_t *pgdat)
729 {
730 	return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
731 }
732 
733 static inline int zone_id(const struct zone *zone)
734 {
735 	struct pglist_data *pgdat = zone->zone_pgdat;
736 
737 	return zone - pgdat->node_zones;
738 }
739 
740 #ifdef CONFIG_ZONE_DEVICE
741 static inline bool is_dev_zone(const struct zone *zone)
742 {
743 	return zone_id(zone) == ZONE_DEVICE;
744 }
745 #else
746 static inline bool is_dev_zone(const struct zone *zone)
747 {
748 	return false;
749 }
750 #endif
751 
752 #include <linux/memory_hotplug.h>
753 
754 extern struct mutex zonelists_mutex;
755 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
756 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
757 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
758 			 int classzone_idx, unsigned int alloc_flags,
759 			 long free_pages);
760 bool zone_watermark_ok(struct zone *z, unsigned int order,
761 		unsigned long mark, int classzone_idx,
762 		unsigned int alloc_flags);
763 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
764 		unsigned long mark, int classzone_idx);
765 enum memmap_context {
766 	MEMMAP_EARLY,
767 	MEMMAP_HOTPLUG,
768 };
769 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
770 				     unsigned long size);
771 
772 extern void lruvec_init(struct lruvec *lruvec);
773 
774 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
775 {
776 #ifdef CONFIG_MEMCG
777 	return lruvec->pgdat;
778 #else
779 	return container_of(lruvec, struct pglist_data, lruvec);
780 #endif
781 }
782 
783 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
784 
785 #ifdef CONFIG_HAVE_MEMORY_PRESENT
786 void memory_present(int nid, unsigned long start, unsigned long end);
787 #else
788 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
789 #endif
790 
791 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
792 int local_memory_node(int node_id);
793 #else
794 static inline int local_memory_node(int node_id) { return node_id; };
795 #endif
796 
797 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
798 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
799 #endif
800 
801 /*
802  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
803  */
804 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
805 
806 /*
807  * Returns true if a zone has pages managed by the buddy allocator.
808  * All the reclaim decisions have to use this function rather than
809  * populated_zone(). If the whole zone is reserved then we can easily
810  * end up with populated_zone() && !managed_zone().
811  */
812 static inline bool managed_zone(struct zone *zone)
813 {
814 	return zone->managed_pages;
815 }
816 
817 /* Returns true if a zone has memory */
818 static inline bool populated_zone(struct zone *zone)
819 {
820 	return zone->present_pages;
821 }
822 
823 extern int movable_zone;
824 
825 #ifdef CONFIG_HIGHMEM
826 static inline int zone_movable_is_highmem(void)
827 {
828 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
829 	return movable_zone == ZONE_HIGHMEM;
830 #else
831 	return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
832 #endif
833 }
834 #endif
835 
836 static inline int is_highmem_idx(enum zone_type idx)
837 {
838 #ifdef CONFIG_HIGHMEM
839 	return (idx == ZONE_HIGHMEM ||
840 		(idx == ZONE_MOVABLE && zone_movable_is_highmem()));
841 #else
842 	return 0;
843 #endif
844 }
845 
846 /**
847  * is_highmem - helper function to quickly check if a struct zone is a
848  *              highmem zone or not.  This is an attempt to keep references
849  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
850  * @zone - pointer to struct zone variable
851  */
852 static inline int is_highmem(struct zone *zone)
853 {
854 #ifdef CONFIG_HIGHMEM
855 	return is_highmem_idx(zone_idx(zone));
856 #else
857 	return 0;
858 #endif
859 }
860 
861 /* These two functions are used to setup the per zone pages min values */
862 struct ctl_table;
863 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
864 					void __user *, size_t *, loff_t *);
865 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
866 					void __user *, size_t *, loff_t *);
867 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
868 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
869 					void __user *, size_t *, loff_t *);
870 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
871 					void __user *, size_t *, loff_t *);
872 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
873 			void __user *, size_t *, loff_t *);
874 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
875 			void __user *, size_t *, loff_t *);
876 
877 extern int numa_zonelist_order_handler(struct ctl_table *, int,
878 			void __user *, size_t *, loff_t *);
879 extern char numa_zonelist_order[];
880 #define NUMA_ZONELIST_ORDER_LEN 16	/* string buffer size */
881 
882 #ifndef CONFIG_NEED_MULTIPLE_NODES
883 
884 extern struct pglist_data contig_page_data;
885 #define NODE_DATA(nid)		(&contig_page_data)
886 #define NODE_MEM_MAP(nid)	mem_map
887 
888 #else /* CONFIG_NEED_MULTIPLE_NODES */
889 
890 #include <asm/mmzone.h>
891 
892 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
893 
894 extern struct pglist_data *first_online_pgdat(void);
895 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
896 extern struct zone *next_zone(struct zone *zone);
897 
898 /**
899  * for_each_online_pgdat - helper macro to iterate over all online nodes
900  * @pgdat - pointer to a pg_data_t variable
901  */
902 #define for_each_online_pgdat(pgdat)			\
903 	for (pgdat = first_online_pgdat();		\
904 	     pgdat;					\
905 	     pgdat = next_online_pgdat(pgdat))
906 /**
907  * for_each_zone - helper macro to iterate over all memory zones
908  * @zone - pointer to struct zone variable
909  *
910  * The user only needs to declare the zone variable, for_each_zone
911  * fills it in.
912  */
913 #define for_each_zone(zone)			        \
914 	for (zone = (first_online_pgdat())->node_zones; \
915 	     zone;					\
916 	     zone = next_zone(zone))
917 
918 #define for_each_populated_zone(zone)		        \
919 	for (zone = (first_online_pgdat())->node_zones; \
920 	     zone;					\
921 	     zone = next_zone(zone))			\
922 		if (!populated_zone(zone))		\
923 			; /* do nothing */		\
924 		else
925 
926 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
927 {
928 	return zoneref->zone;
929 }
930 
931 static inline int zonelist_zone_idx(struct zoneref *zoneref)
932 {
933 	return zoneref->zone_idx;
934 }
935 
936 static inline int zonelist_node_idx(struct zoneref *zoneref)
937 {
938 #ifdef CONFIG_NUMA
939 	/* zone_to_nid not available in this context */
940 	return zoneref->zone->node;
941 #else
942 	return 0;
943 #endif /* CONFIG_NUMA */
944 }
945 
946 struct zoneref *__next_zones_zonelist(struct zoneref *z,
947 					enum zone_type highest_zoneidx,
948 					nodemask_t *nodes);
949 
950 /**
951  * 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
952  * @z - The cursor used as a starting point for the search
953  * @highest_zoneidx - The zone index of the highest zone to return
954  * @nodes - An optional nodemask to filter the zonelist with
955  *
956  * This function returns the next zone at or below a given zone index that is
957  * within the allowed nodemask using a cursor as the starting point for the
958  * search. The zoneref returned is a cursor that represents the current zone
959  * being examined. It should be advanced by one before calling
960  * next_zones_zonelist again.
961  */
962 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
963 					enum zone_type highest_zoneidx,
964 					nodemask_t *nodes)
965 {
966 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
967 		return z;
968 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
969 }
970 
971 /**
972  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
973  * @zonelist - The zonelist to search for a suitable zone
974  * @highest_zoneidx - The zone index of the highest zone to return
975  * @nodes - An optional nodemask to filter the zonelist with
976  * @return - Zoneref pointer for the first suitable zone found (see below)
977  *
978  * This function returns the first zone at or below a given zone index that is
979  * within the allowed nodemask. The zoneref returned is a cursor that can be
980  * used to iterate the zonelist with next_zones_zonelist by advancing it by
981  * one before calling.
982  *
983  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
984  * never NULL). This may happen either genuinely, or due to concurrent nodemask
985  * update due to cpuset modification.
986  */
987 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
988 					enum zone_type highest_zoneidx,
989 					nodemask_t *nodes)
990 {
991 	return next_zones_zonelist(zonelist->_zonerefs,
992 							highest_zoneidx, nodes);
993 }
994 
995 /**
996  * 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
997  * @zone - The current zone in the iterator
998  * @z - The current pointer within zonelist->zones being iterated
999  * @zlist - The zonelist being iterated
1000  * @highidx - The zone index of the highest zone to return
1001  * @nodemask - Nodemask allowed by the allocator
1002  *
1003  * This iterator iterates though all zones at or below a given zone index and
1004  * within a given nodemask
1005  */
1006 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1007 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1008 		zone;							\
1009 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1010 			zone = zonelist_zone(z))
1011 
1012 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1013 	for (zone = z->zone;	\
1014 		zone;							\
1015 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1016 			zone = zonelist_zone(z))
1017 
1018 
1019 /**
1020  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1021  * @zone - The current zone in the iterator
1022  * @z - The current pointer within zonelist->zones being iterated
1023  * @zlist - The zonelist being iterated
1024  * @highidx - The zone index of the highest zone to return
1025  *
1026  * This iterator iterates though all zones at or below a given zone index.
1027  */
1028 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1029 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1030 
1031 #ifdef CONFIG_SPARSEMEM
1032 #include <asm/sparsemem.h>
1033 #endif
1034 
1035 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1036 	!defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1037 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1038 {
1039 	return 0;
1040 }
1041 #endif
1042 
1043 #ifdef CONFIG_FLATMEM
1044 #define pfn_to_nid(pfn)		(0)
1045 #endif
1046 
1047 #ifdef CONFIG_SPARSEMEM
1048 
1049 /*
1050  * SECTION_SHIFT    		#bits space required to store a section #
1051  *
1052  * PA_SECTION_SHIFT		physical address to/from section number
1053  * PFN_SECTION_SHIFT		pfn to/from section number
1054  */
1055 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1056 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1057 
1058 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1059 
1060 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1061 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1062 
1063 #define SECTION_BLOCKFLAGS_BITS \
1064 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1065 
1066 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1067 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1068 #endif
1069 
1070 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1071 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1072 
1073 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1074 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1075 
1076 struct page;
1077 struct page_ext;
1078 struct mem_section {
1079 	/*
1080 	 * This is, logically, a pointer to an array of struct
1081 	 * pages.  However, it is stored with some other magic.
1082 	 * (see sparse.c::sparse_init_one_section())
1083 	 *
1084 	 * Additionally during early boot we encode node id of
1085 	 * the location of the section here to guide allocation.
1086 	 * (see sparse.c::memory_present())
1087 	 *
1088 	 * Making it a UL at least makes someone do a cast
1089 	 * before using it wrong.
1090 	 */
1091 	unsigned long section_mem_map;
1092 
1093 	/* See declaration of similar field in struct zone */
1094 	unsigned long *pageblock_flags;
1095 #ifdef CONFIG_PAGE_EXTENSION
1096 	/*
1097 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1098 	 * section. (see page_ext.h about this.)
1099 	 */
1100 	struct page_ext *page_ext;
1101 	unsigned long pad;
1102 #endif
1103 	/*
1104 	 * WARNING: mem_section must be a power-of-2 in size for the
1105 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1106 	 */
1107 };
1108 
1109 #ifdef CONFIG_SPARSEMEM_EXTREME
1110 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1111 #else
1112 #define SECTIONS_PER_ROOT	1
1113 #endif
1114 
1115 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1116 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1117 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1118 
1119 #ifdef CONFIG_SPARSEMEM_EXTREME
1120 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1121 #else
1122 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1123 #endif
1124 
1125 static inline struct mem_section *__nr_to_section(unsigned long nr)
1126 {
1127 	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1128 		return NULL;
1129 	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1130 }
1131 extern int __section_nr(struct mem_section* ms);
1132 extern unsigned long usemap_size(void);
1133 
1134 /*
1135  * We use the lower bits of the mem_map pointer to store
1136  * a little bit of information.  There should be at least
1137  * 3 bits here due to 32-bit alignment.
1138  */
1139 #define	SECTION_MARKED_PRESENT	(1UL<<0)
1140 #define SECTION_HAS_MEM_MAP	(1UL<<1)
1141 #define SECTION_MAP_LAST_BIT	(1UL<<2)
1142 #define SECTION_MAP_MASK	(~(SECTION_MAP_LAST_BIT-1))
1143 #define SECTION_NID_SHIFT	2
1144 
1145 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1146 {
1147 	unsigned long map = section->section_mem_map;
1148 	map &= SECTION_MAP_MASK;
1149 	return (struct page *)map;
1150 }
1151 
1152 static inline int present_section(struct mem_section *section)
1153 {
1154 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1155 }
1156 
1157 static inline int present_section_nr(unsigned long nr)
1158 {
1159 	return present_section(__nr_to_section(nr));
1160 }
1161 
1162 static inline int valid_section(struct mem_section *section)
1163 {
1164 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1165 }
1166 
1167 static inline int valid_section_nr(unsigned long nr)
1168 {
1169 	return valid_section(__nr_to_section(nr));
1170 }
1171 
1172 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1173 {
1174 	return __nr_to_section(pfn_to_section_nr(pfn));
1175 }
1176 
1177 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1178 static inline int pfn_valid(unsigned long pfn)
1179 {
1180 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1181 		return 0;
1182 	return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1183 }
1184 #endif
1185 
1186 static inline int pfn_present(unsigned long pfn)
1187 {
1188 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1189 		return 0;
1190 	return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1191 }
1192 
1193 /*
1194  * These are _only_ used during initialisation, therefore they
1195  * can use __initdata ...  They could have names to indicate
1196  * this restriction.
1197  */
1198 #ifdef CONFIG_NUMA
1199 #define pfn_to_nid(pfn)							\
1200 ({									\
1201 	unsigned long __pfn_to_nid_pfn = (pfn);				\
1202 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
1203 })
1204 #else
1205 #define pfn_to_nid(pfn)		(0)
1206 #endif
1207 
1208 #define early_pfn_valid(pfn)	pfn_valid(pfn)
1209 void sparse_init(void);
1210 #else
1211 #define sparse_init()	do {} while (0)
1212 #define sparse_index_init(_sec, _nid)  do {} while (0)
1213 #endif /* CONFIG_SPARSEMEM */
1214 
1215 /*
1216  * During memory init memblocks map pfns to nids. The search is expensive and
1217  * this caches recent lookups. The implementation of __early_pfn_to_nid
1218  * may treat start/end as pfns or sections.
1219  */
1220 struct mminit_pfnnid_cache {
1221 	unsigned long last_start;
1222 	unsigned long last_end;
1223 	int last_nid;
1224 };
1225 
1226 #ifndef early_pfn_valid
1227 #define early_pfn_valid(pfn)	(1)
1228 #endif
1229 
1230 void memory_present(int nid, unsigned long start, unsigned long end);
1231 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1232 
1233 /*
1234  * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1235  * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1236  * pfn_valid_within() should be used in this case; we optimise this away
1237  * when we have no holes within a MAX_ORDER_NR_PAGES block.
1238  */
1239 #ifdef CONFIG_HOLES_IN_ZONE
1240 #define pfn_valid_within(pfn) pfn_valid(pfn)
1241 #else
1242 #define pfn_valid_within(pfn) (1)
1243 #endif
1244 
1245 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1246 /*
1247  * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1248  * associated with it or not. In FLATMEM, it is expected that holes always
1249  * have valid memmap as long as there is valid PFNs either side of the hole.
1250  * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1251  * entire section.
1252  *
1253  * However, an ARM, and maybe other embedded architectures in the future
1254  * free memmap backing holes to save memory on the assumption the memmap is
1255  * never used. The page_zone linkages are then broken even though pfn_valid()
1256  * returns true. A walker of the full memmap must then do this additional
1257  * check to ensure the memmap they are looking at is sane by making sure
1258  * the zone and PFN linkages are still valid. This is expensive, but walkers
1259  * of the full memmap are extremely rare.
1260  */
1261 bool memmap_valid_within(unsigned long pfn,
1262 					struct page *page, struct zone *zone);
1263 #else
1264 static inline bool memmap_valid_within(unsigned long pfn,
1265 					struct page *page, struct zone *zone)
1266 {
1267 	return true;
1268 }
1269 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1270 
1271 #endif /* !__GENERATING_BOUNDS.H */
1272 #endif /* !__ASSEMBLY__ */
1273 #endif /* _LINUX_MMZONE_H */
1274