xref: /linux-6.15/include/linux/mmzone.h (revision bb7e5ce7)
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 static inline bool is_migrate_movable(int mt)
78 {
79 	return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
80 }
81 
82 #define for_each_migratetype_order(order, type) \
83 	for (order = 0; order < MAX_ORDER; order++) \
84 		for (type = 0; type < MIGRATE_TYPES; type++)
85 
86 extern int page_group_by_mobility_disabled;
87 
88 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
89 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
90 
91 #define get_pageblock_migratetype(page)					\
92 	get_pfnblock_flags_mask(page, page_to_pfn(page),		\
93 			PB_migrate_end, MIGRATETYPE_MASK)
94 
95 struct free_area {
96 	struct list_head	free_list[MIGRATE_TYPES];
97 	unsigned long		nr_free;
98 };
99 
100 struct pglist_data;
101 
102 /*
103  * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
104  * So add a wild amount of padding here to ensure that they fall into separate
105  * cachelines.  There are very few zone structures in the machine, so space
106  * consumption is not a concern here.
107  */
108 #if defined(CONFIG_SMP)
109 struct zone_padding {
110 	char x[0];
111 } ____cacheline_internodealigned_in_smp;
112 #define ZONE_PADDING(name)	struct zone_padding name;
113 #else
114 #define ZONE_PADDING(name)
115 #endif
116 
117 #ifdef CONFIG_NUMA
118 enum numa_stat_item {
119 	NUMA_HIT,		/* allocated in intended node */
120 	NUMA_MISS,		/* allocated in non intended node */
121 	NUMA_FOREIGN,		/* was intended here, hit elsewhere */
122 	NUMA_INTERLEAVE_HIT,	/* interleaver preferred this zone */
123 	NUMA_LOCAL,		/* allocation from local node */
124 	NUMA_OTHER,		/* allocation from other node */
125 	NR_VM_NUMA_STAT_ITEMS
126 };
127 #else
128 #define NR_VM_NUMA_STAT_ITEMS 0
129 #endif
130 
131 enum zone_stat_item {
132 	/* First 128 byte cacheline (assuming 64 bit words) */
133 	NR_FREE_PAGES,
134 	NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
135 	NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
136 	NR_ZONE_ACTIVE_ANON,
137 	NR_ZONE_INACTIVE_FILE,
138 	NR_ZONE_ACTIVE_FILE,
139 	NR_ZONE_UNEVICTABLE,
140 	NR_ZONE_WRITE_PENDING,	/* Count of dirty, writeback and unstable pages */
141 	NR_MLOCK,		/* mlock()ed pages found and moved off LRU */
142 	NR_PAGETABLE,		/* used for pagetables */
143 	NR_KERNEL_STACK_KB,	/* measured in KiB */
144 	/* Second 128 byte cacheline */
145 	NR_BOUNCE,
146 #if IS_ENABLED(CONFIG_ZSMALLOC)
147 	NR_ZSPAGES,		/* allocated in zsmalloc */
148 #endif
149 	NR_FREE_CMA_PAGES,
150 	NR_VM_ZONE_STAT_ITEMS };
151 
152 enum node_stat_item {
153 	NR_LRU_BASE,
154 	NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
155 	NR_ACTIVE_ANON,		/*  "     "     "   "       "         */
156 	NR_INACTIVE_FILE,	/*  "     "     "   "       "         */
157 	NR_ACTIVE_FILE,		/*  "     "     "   "       "         */
158 	NR_UNEVICTABLE,		/*  "     "     "   "       "         */
159 	NR_SLAB_RECLAIMABLE,
160 	NR_SLAB_UNRECLAIMABLE,
161 	NR_ISOLATED_ANON,	/* Temporary isolated pages from anon lru */
162 	NR_ISOLATED_FILE,	/* Temporary isolated pages from file lru */
163 	WORKINGSET_REFAULT,
164 	WORKINGSET_ACTIVATE,
165 	WORKINGSET_NODERECLAIM,
166 	NR_ANON_MAPPED,	/* Mapped anonymous pages */
167 	NR_FILE_MAPPED,	/* pagecache pages mapped into pagetables.
168 			   only modified from process context */
169 	NR_FILE_PAGES,
170 	NR_FILE_DIRTY,
171 	NR_WRITEBACK,
172 	NR_WRITEBACK_TEMP,	/* Writeback using temporary buffers */
173 	NR_SHMEM,		/* shmem pages (included tmpfs/GEM pages) */
174 	NR_SHMEM_THPS,
175 	NR_SHMEM_PMDMAPPED,
176 	NR_ANON_THPS,
177 	NR_UNSTABLE_NFS,	/* NFS unstable pages */
178 	NR_VMSCAN_WRITE,
179 	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
180 	NR_DIRTIED,		/* page dirtyings since bootup */
181 	NR_WRITTEN,		/* page writings since bootup */
182 	NR_VM_NODE_STAT_ITEMS
183 };
184 
185 /*
186  * We do arithmetic on the LRU lists in various places in the code,
187  * so it is important to keep the active lists LRU_ACTIVE higher in
188  * the array than the corresponding inactive lists, and to keep
189  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
190  *
191  * This has to be kept in sync with the statistics in zone_stat_item
192  * above and the descriptions in vmstat_text in mm/vmstat.c
193  */
194 #define LRU_BASE 0
195 #define LRU_ACTIVE 1
196 #define LRU_FILE 2
197 
198 enum lru_list {
199 	LRU_INACTIVE_ANON = LRU_BASE,
200 	LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
201 	LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
202 	LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
203 	LRU_UNEVICTABLE,
204 	NR_LRU_LISTS
205 };
206 
207 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
208 
209 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
210 
211 static inline int is_file_lru(enum lru_list lru)
212 {
213 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
214 }
215 
216 static inline int is_active_lru(enum lru_list lru)
217 {
218 	return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
219 }
220 
221 struct zone_reclaim_stat {
222 	/*
223 	 * The pageout code in vmscan.c keeps track of how many of the
224 	 * mem/swap backed and file backed pages are referenced.
225 	 * The higher the rotated/scanned ratio, the more valuable
226 	 * that cache is.
227 	 *
228 	 * The anon LRU stats live in [0], file LRU stats in [1]
229 	 */
230 	unsigned long		recent_rotated[2];
231 	unsigned long		recent_scanned[2];
232 };
233 
234 struct lruvec {
235 	struct list_head		lists[NR_LRU_LISTS];
236 	struct zone_reclaim_stat	reclaim_stat;
237 	/* Evictions & activations on the inactive file list */
238 	atomic_long_t			inactive_age;
239 	/* Refaults at the time of last reclaim cycle */
240 	unsigned long			refaults;
241 #ifdef CONFIG_MEMCG
242 	struct pglist_data *pgdat;
243 #endif
244 };
245 
246 /* Mask used at gathering information at once (see memcontrol.c) */
247 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
248 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
249 #define LRU_ALL	     ((1 << NR_LRU_LISTS) - 1)
250 
251 /* Isolate unmapped file */
252 #define ISOLATE_UNMAPPED	((__force isolate_mode_t)0x2)
253 /* Isolate for asynchronous migration */
254 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
255 /* Isolate unevictable pages */
256 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
257 
258 /* LRU Isolation modes. */
259 typedef unsigned __bitwise isolate_mode_t;
260 
261 enum zone_watermarks {
262 	WMARK_MIN,
263 	WMARK_LOW,
264 	WMARK_HIGH,
265 	NR_WMARK
266 };
267 
268 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
269 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
270 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
271 
272 struct per_cpu_pages {
273 	int count;		/* number of pages in the list */
274 	int high;		/* high watermark, emptying needed */
275 	int batch;		/* chunk size for buddy add/remove */
276 
277 	/* Lists of pages, one per migrate type stored on the pcp-lists */
278 	struct list_head lists[MIGRATE_PCPTYPES];
279 };
280 
281 struct per_cpu_pageset {
282 	struct per_cpu_pages pcp;
283 #ifdef CONFIG_NUMA
284 	s8 expire;
285 	u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
286 #endif
287 #ifdef CONFIG_SMP
288 	s8 stat_threshold;
289 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
290 #endif
291 };
292 
293 struct per_cpu_nodestat {
294 	s8 stat_threshold;
295 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
296 };
297 
298 #endif /* !__GENERATING_BOUNDS.H */
299 
300 enum zone_type {
301 #ifdef CONFIG_ZONE_DMA
302 	/*
303 	 * ZONE_DMA is used when there are devices that are not able
304 	 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
305 	 * carve out the portion of memory that is needed for these devices.
306 	 * The range is arch specific.
307 	 *
308 	 * Some examples
309 	 *
310 	 * Architecture		Limit
311 	 * ---------------------------
312 	 * parisc, ia64, sparc	<4G
313 	 * s390			<2G
314 	 * arm			Various
315 	 * alpha		Unlimited or 0-16MB.
316 	 *
317 	 * i386, x86_64 and multiple other arches
318 	 * 			<16M.
319 	 */
320 	ZONE_DMA,
321 #endif
322 #ifdef CONFIG_ZONE_DMA32
323 	/*
324 	 * x86_64 needs two ZONE_DMAs because it supports devices that are
325 	 * only able to do DMA to the lower 16M but also 32 bit devices that
326 	 * can only do DMA areas below 4G.
327 	 */
328 	ZONE_DMA32,
329 #endif
330 	/*
331 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
332 	 * performed on pages in ZONE_NORMAL if the DMA devices support
333 	 * transfers to all addressable memory.
334 	 */
335 	ZONE_NORMAL,
336 #ifdef CONFIG_HIGHMEM
337 	/*
338 	 * A memory area that is only addressable by the kernel through
339 	 * mapping portions into its own address space. This is for example
340 	 * used by i386 to allow the kernel to address the memory beyond
341 	 * 900MB. The kernel will set up special mappings (page
342 	 * table entries on i386) for each page that the kernel needs to
343 	 * access.
344 	 */
345 	ZONE_HIGHMEM,
346 #endif
347 	ZONE_MOVABLE,
348 #ifdef CONFIG_ZONE_DEVICE
349 	ZONE_DEVICE,
350 #endif
351 	__MAX_NR_ZONES
352 
353 };
354 
355 #ifndef __GENERATING_BOUNDS_H
356 
357 struct zone {
358 	/* Read-mostly fields */
359 
360 	/* zone watermarks, access with *_wmark_pages(zone) macros */
361 	unsigned long watermark[NR_WMARK];
362 
363 	unsigned long nr_reserved_highatomic;
364 
365 	/*
366 	 * We don't know if the memory that we're going to allocate will be
367 	 * freeable or/and it will be released eventually, so to avoid totally
368 	 * wasting several GB of ram we must reserve some of the lower zone
369 	 * memory (otherwise we risk to run OOM on the lower zones despite
370 	 * there being tons of freeable ram on the higher zones).  This array is
371 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
372 	 * changes.
373 	 */
374 	long lowmem_reserve[MAX_NR_ZONES];
375 
376 #ifdef CONFIG_NUMA
377 	int node;
378 #endif
379 	struct pglist_data	*zone_pgdat;
380 	struct per_cpu_pageset __percpu *pageset;
381 
382 #ifndef CONFIG_SPARSEMEM
383 	/*
384 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
385 	 * In SPARSEMEM, this map is stored in struct mem_section
386 	 */
387 	unsigned long		*pageblock_flags;
388 #endif /* CONFIG_SPARSEMEM */
389 
390 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
391 	unsigned long		zone_start_pfn;
392 
393 	/*
394 	 * spanned_pages is the total pages spanned by the zone, including
395 	 * holes, which is calculated as:
396 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
397 	 *
398 	 * present_pages is physical pages existing within the zone, which
399 	 * is calculated as:
400 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
401 	 *
402 	 * managed_pages is present pages managed by the buddy system, which
403 	 * is calculated as (reserved_pages includes pages allocated by the
404 	 * bootmem allocator):
405 	 *	managed_pages = present_pages - reserved_pages;
406 	 *
407 	 * So present_pages may be used by memory hotplug or memory power
408 	 * management logic to figure out unmanaged pages by checking
409 	 * (present_pages - managed_pages). And managed_pages should be used
410 	 * by page allocator and vm scanner to calculate all kinds of watermarks
411 	 * and thresholds.
412 	 *
413 	 * Locking rules:
414 	 *
415 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
416 	 * It is a seqlock because it has to be read outside of zone->lock,
417 	 * and it is done in the main allocator path.  But, it is written
418 	 * quite infrequently.
419 	 *
420 	 * The span_seq lock is declared along with zone->lock because it is
421 	 * frequently read in proximity to zone->lock.  It's good to
422 	 * give them a chance of being in the same cacheline.
423 	 *
424 	 * Write access to present_pages at runtime should be protected by
425 	 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
426 	 * present_pages should get_online_mems() to get a stable value.
427 	 *
428 	 * Read access to managed_pages should be safe because it's unsigned
429 	 * long. Write access to zone->managed_pages and totalram_pages are
430 	 * protected by managed_page_count_lock at runtime. Idealy only
431 	 * adjust_managed_page_count() should be used instead of directly
432 	 * touching zone->managed_pages and totalram_pages.
433 	 */
434 	unsigned long		managed_pages;
435 	unsigned long		spanned_pages;
436 	unsigned long		present_pages;
437 
438 	const char		*name;
439 
440 #ifdef CONFIG_MEMORY_ISOLATION
441 	/*
442 	 * Number of isolated pageblock. It is used to solve incorrect
443 	 * freepage counting problem due to racy retrieving migratetype
444 	 * of pageblock. Protected by zone->lock.
445 	 */
446 	unsigned long		nr_isolate_pageblock;
447 #endif
448 
449 #ifdef CONFIG_MEMORY_HOTPLUG
450 	/* see spanned/present_pages for more description */
451 	seqlock_t		span_seqlock;
452 #endif
453 
454 	int initialized;
455 
456 	/* Write-intensive fields used from the page allocator */
457 	ZONE_PADDING(_pad1_)
458 
459 	/* free areas of different sizes */
460 	struct free_area	free_area[MAX_ORDER];
461 
462 	/* zone flags, see below */
463 	unsigned long		flags;
464 
465 	/* Primarily protects free_area */
466 	spinlock_t		lock;
467 
468 	/* Write-intensive fields used by compaction and vmstats. */
469 	ZONE_PADDING(_pad2_)
470 
471 	/*
472 	 * When free pages are below this point, additional steps are taken
473 	 * when reading the number of free pages to avoid per-cpu counter
474 	 * drift allowing watermarks to be breached
475 	 */
476 	unsigned long percpu_drift_mark;
477 
478 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
479 	/* pfn where compaction free scanner should start */
480 	unsigned long		compact_cached_free_pfn;
481 	/* pfn where async and sync compaction migration scanner should start */
482 	unsigned long		compact_cached_migrate_pfn[2];
483 #endif
484 
485 #ifdef CONFIG_COMPACTION
486 	/*
487 	 * On compaction failure, 1<<compact_defer_shift compactions
488 	 * are skipped before trying again. The number attempted since
489 	 * last failure is tracked with compact_considered.
490 	 */
491 	unsigned int		compact_considered;
492 	unsigned int		compact_defer_shift;
493 	int			compact_order_failed;
494 #endif
495 
496 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
497 	/* Set to true when the PG_migrate_skip bits should be cleared */
498 	bool			compact_blockskip_flush;
499 #endif
500 
501 	bool			contiguous;
502 
503 	ZONE_PADDING(_pad3_)
504 	/* Zone statistics */
505 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
506 	atomic_long_t		vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
507 } ____cacheline_internodealigned_in_smp;
508 
509 enum pgdat_flags {
510 	PGDAT_CONGESTED,		/* pgdat has many dirty pages backed by
511 					 * a congested BDI
512 					 */
513 	PGDAT_DIRTY,			/* reclaim scanning has recently found
514 					 * many dirty file pages at the tail
515 					 * of the LRU.
516 					 */
517 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
518 					 * many pages under writeback
519 					 */
520 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
521 };
522 
523 static inline unsigned long zone_end_pfn(const struct zone *zone)
524 {
525 	return zone->zone_start_pfn + zone->spanned_pages;
526 }
527 
528 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
529 {
530 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
531 }
532 
533 static inline bool zone_is_initialized(struct zone *zone)
534 {
535 	return zone->initialized;
536 }
537 
538 static inline bool zone_is_empty(struct zone *zone)
539 {
540 	return zone->spanned_pages == 0;
541 }
542 
543 /*
544  * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
545  * intersection with the given zone
546  */
547 static inline bool zone_intersects(struct zone *zone,
548 		unsigned long start_pfn, unsigned long nr_pages)
549 {
550 	if (zone_is_empty(zone))
551 		return false;
552 	if (start_pfn >= zone_end_pfn(zone) ||
553 	    start_pfn + nr_pages <= zone->zone_start_pfn)
554 		return false;
555 
556 	return true;
557 }
558 
559 /*
560  * The "priority" of VM scanning is how much of the queues we will scan in one
561  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
562  * queues ("queue_length >> 12") during an aging round.
563  */
564 #define DEF_PRIORITY 12
565 
566 /* Maximum number of zones on a zonelist */
567 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
568 
569 enum {
570 	ZONELIST_FALLBACK,	/* zonelist with fallback */
571 #ifdef CONFIG_NUMA
572 	/*
573 	 * The NUMA zonelists are doubled because we need zonelists that
574 	 * restrict the allocations to a single node for __GFP_THISNODE.
575 	 */
576 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
577 #endif
578 	MAX_ZONELISTS
579 };
580 
581 /*
582  * This struct contains information about a zone in a zonelist. It is stored
583  * here to avoid dereferences into large structures and lookups of tables
584  */
585 struct zoneref {
586 	struct zone *zone;	/* Pointer to actual zone */
587 	int zone_idx;		/* zone_idx(zoneref->zone) */
588 };
589 
590 /*
591  * One allocation request operates on a zonelist. A zonelist
592  * is a list of zones, the first one is the 'goal' of the
593  * allocation, the other zones are fallback zones, in decreasing
594  * priority.
595  *
596  * To speed the reading of the zonelist, the zonerefs contain the zone index
597  * of the entry being read. Helper functions to access information given
598  * a struct zoneref are
599  *
600  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
601  * zonelist_zone_idx()	- Return the index of the zone for an entry
602  * zonelist_node_idx()	- Return the index of the node for an entry
603  */
604 struct zonelist {
605 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
606 };
607 
608 #ifndef CONFIG_DISCONTIGMEM
609 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
610 extern struct page *mem_map;
611 #endif
612 
613 /*
614  * On NUMA machines, each NUMA node would have a pg_data_t to describe
615  * it's memory layout. On UMA machines there is a single pglist_data which
616  * describes the whole memory.
617  *
618  * Memory statistics and page replacement data structures are maintained on a
619  * per-zone basis.
620  */
621 struct bootmem_data;
622 typedef struct pglist_data {
623 	struct zone node_zones[MAX_NR_ZONES];
624 	struct zonelist node_zonelists[MAX_ZONELISTS];
625 	int nr_zones;
626 #ifdef CONFIG_FLAT_NODE_MEM_MAP	/* means !SPARSEMEM */
627 	struct page *node_mem_map;
628 #ifdef CONFIG_PAGE_EXTENSION
629 	struct page_ext *node_page_ext;
630 #endif
631 #endif
632 #ifndef CONFIG_NO_BOOTMEM
633 	struct bootmem_data *bdata;
634 #endif
635 #ifdef CONFIG_MEMORY_HOTPLUG
636 	/*
637 	 * Must be held any time you expect node_start_pfn, node_present_pages
638 	 * or node_spanned_pages stay constant.  Holding this will also
639 	 * guarantee that any pfn_valid() stays that way.
640 	 *
641 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
642 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
643 	 *
644 	 * Nests above zone->lock and zone->span_seqlock
645 	 */
646 	spinlock_t node_size_lock;
647 #endif
648 	unsigned long node_start_pfn;
649 	unsigned long node_present_pages; /* total number of physical pages */
650 	unsigned long node_spanned_pages; /* total size of physical page
651 					     range, including holes */
652 	int node_id;
653 	wait_queue_head_t kswapd_wait;
654 	wait_queue_head_t pfmemalloc_wait;
655 	struct task_struct *kswapd;	/* Protected by
656 					   mem_hotplug_begin/end() */
657 	int kswapd_order;
658 	enum zone_type kswapd_classzone_idx;
659 
660 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
661 
662 #ifdef CONFIG_COMPACTION
663 	int kcompactd_max_order;
664 	enum zone_type kcompactd_classzone_idx;
665 	wait_queue_head_t kcompactd_wait;
666 	struct task_struct *kcompactd;
667 #endif
668 #ifdef CONFIG_NUMA_BALANCING
669 	/* Lock serializing the migrate rate limiting window */
670 	spinlock_t numabalancing_migrate_lock;
671 
672 	/* Rate limiting time interval */
673 	unsigned long numabalancing_migrate_next_window;
674 
675 	/* Number of pages migrated during the rate limiting time interval */
676 	unsigned long numabalancing_migrate_nr_pages;
677 #endif
678 	/*
679 	 * This is a per-node reserve of pages that are not available
680 	 * to userspace allocations.
681 	 */
682 	unsigned long		totalreserve_pages;
683 
684 #ifdef CONFIG_NUMA
685 	/*
686 	 * zone reclaim becomes active if more unmapped pages exist.
687 	 */
688 	unsigned long		min_unmapped_pages;
689 	unsigned long		min_slab_pages;
690 #endif /* CONFIG_NUMA */
691 
692 	/* Write-intensive fields used by page reclaim */
693 	ZONE_PADDING(_pad1_)
694 	spinlock_t		lru_lock;
695 
696 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
697 	/*
698 	 * If memory initialisation on large machines is deferred then this
699 	 * is the first PFN that needs to be initialised.
700 	 */
701 	unsigned long first_deferred_pfn;
702 	unsigned long static_init_size;
703 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
704 
705 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
706 	spinlock_t split_queue_lock;
707 	struct list_head split_queue;
708 	unsigned long split_queue_len;
709 #endif
710 
711 	/* Fields commonly accessed by the page reclaim scanner */
712 	struct lruvec		lruvec;
713 
714 	/*
715 	 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
716 	 * this node's LRU.  Maintained by the pageout code.
717 	 */
718 	unsigned int inactive_ratio;
719 
720 	unsigned long		flags;
721 
722 	ZONE_PADDING(_pad2_)
723 
724 	/* Per-node vmstats */
725 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
726 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
727 } pg_data_t;
728 
729 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
730 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
731 #ifdef CONFIG_FLAT_NODE_MEM_MAP
732 #define pgdat_page_nr(pgdat, pagenr)	((pgdat)->node_mem_map + (pagenr))
733 #else
734 #define pgdat_page_nr(pgdat, pagenr)	pfn_to_page((pgdat)->node_start_pfn + (pagenr))
735 #endif
736 #define nid_page_nr(nid, pagenr) 	pgdat_page_nr(NODE_DATA(nid),(pagenr))
737 
738 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
739 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
740 static inline spinlock_t *zone_lru_lock(struct zone *zone)
741 {
742 	return &zone->zone_pgdat->lru_lock;
743 }
744 
745 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
746 {
747 	return &pgdat->lruvec;
748 }
749 
750 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
751 {
752 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
753 }
754 
755 static inline bool pgdat_is_empty(pg_data_t *pgdat)
756 {
757 	return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
758 }
759 
760 static inline int zone_id(const struct zone *zone)
761 {
762 	struct pglist_data *pgdat = zone->zone_pgdat;
763 
764 	return zone - pgdat->node_zones;
765 }
766 
767 #ifdef CONFIG_ZONE_DEVICE
768 static inline bool is_dev_zone(const struct zone *zone)
769 {
770 	return zone_id(zone) == ZONE_DEVICE;
771 }
772 #else
773 static inline bool is_dev_zone(const struct zone *zone)
774 {
775 	return false;
776 }
777 #endif
778 
779 #include <linux/memory_hotplug.h>
780 
781 void build_all_zonelists(pg_data_t *pgdat);
782 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
783 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
784 			 int classzone_idx, unsigned int alloc_flags,
785 			 long free_pages);
786 bool zone_watermark_ok(struct zone *z, unsigned int order,
787 		unsigned long mark, int classzone_idx,
788 		unsigned int alloc_flags);
789 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
790 		unsigned long mark, int classzone_idx);
791 enum memmap_context {
792 	MEMMAP_EARLY,
793 	MEMMAP_HOTPLUG,
794 };
795 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
796 				     unsigned long size);
797 
798 extern void lruvec_init(struct lruvec *lruvec);
799 
800 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
801 {
802 #ifdef CONFIG_MEMCG
803 	return lruvec->pgdat;
804 #else
805 	return container_of(lruvec, struct pglist_data, lruvec);
806 #endif
807 }
808 
809 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
810 
811 #ifdef CONFIG_HAVE_MEMORY_PRESENT
812 void memory_present(int nid, unsigned long start, unsigned long end);
813 #else
814 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
815 #endif
816 
817 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
818 int local_memory_node(int node_id);
819 #else
820 static inline int local_memory_node(int node_id) { return node_id; };
821 #endif
822 
823 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
824 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
825 #endif
826 
827 /*
828  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
829  */
830 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
831 
832 /*
833  * Returns true if a zone has pages managed by the buddy allocator.
834  * All the reclaim decisions have to use this function rather than
835  * populated_zone(). If the whole zone is reserved then we can easily
836  * end up with populated_zone() && !managed_zone().
837  */
838 static inline bool managed_zone(struct zone *zone)
839 {
840 	return zone->managed_pages;
841 }
842 
843 /* Returns true if a zone has memory */
844 static inline bool populated_zone(struct zone *zone)
845 {
846 	return zone->present_pages;
847 }
848 
849 extern int movable_zone;
850 
851 #ifdef CONFIG_HIGHMEM
852 static inline int zone_movable_is_highmem(void)
853 {
854 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
855 	return movable_zone == ZONE_HIGHMEM;
856 #else
857 	return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
858 #endif
859 }
860 #endif
861 
862 static inline int is_highmem_idx(enum zone_type idx)
863 {
864 #ifdef CONFIG_HIGHMEM
865 	return (idx == ZONE_HIGHMEM ||
866 		(idx == ZONE_MOVABLE && zone_movable_is_highmem()));
867 #else
868 	return 0;
869 #endif
870 }
871 
872 /**
873  * is_highmem - helper function to quickly check if a struct zone is a
874  *              highmem zone or not.  This is an attempt to keep references
875  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
876  * @zone - pointer to struct zone variable
877  */
878 static inline int is_highmem(struct zone *zone)
879 {
880 #ifdef CONFIG_HIGHMEM
881 	return is_highmem_idx(zone_idx(zone));
882 #else
883 	return 0;
884 #endif
885 }
886 
887 /* These two functions are used to setup the per zone pages min values */
888 struct ctl_table;
889 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
890 					void __user *, size_t *, loff_t *);
891 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
892 					void __user *, size_t *, loff_t *);
893 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
894 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
895 					void __user *, size_t *, loff_t *);
896 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
897 					void __user *, size_t *, loff_t *);
898 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
899 			void __user *, size_t *, loff_t *);
900 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
901 			void __user *, size_t *, loff_t *);
902 
903 extern int numa_zonelist_order_handler(struct ctl_table *, int,
904 			void __user *, size_t *, loff_t *);
905 extern char numa_zonelist_order[];
906 #define NUMA_ZONELIST_ORDER_LEN	16
907 
908 #ifndef CONFIG_NEED_MULTIPLE_NODES
909 
910 extern struct pglist_data contig_page_data;
911 #define NODE_DATA(nid)		(&contig_page_data)
912 #define NODE_MEM_MAP(nid)	mem_map
913 
914 #else /* CONFIG_NEED_MULTIPLE_NODES */
915 
916 #include <asm/mmzone.h>
917 
918 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
919 
920 extern struct pglist_data *first_online_pgdat(void);
921 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
922 extern struct zone *next_zone(struct zone *zone);
923 
924 /**
925  * for_each_online_pgdat - helper macro to iterate over all online nodes
926  * @pgdat - pointer to a pg_data_t variable
927  */
928 #define for_each_online_pgdat(pgdat)			\
929 	for (pgdat = first_online_pgdat();		\
930 	     pgdat;					\
931 	     pgdat = next_online_pgdat(pgdat))
932 /**
933  * for_each_zone - helper macro to iterate over all memory zones
934  * @zone - pointer to struct zone variable
935  *
936  * The user only needs to declare the zone variable, for_each_zone
937  * fills it in.
938  */
939 #define for_each_zone(zone)			        \
940 	for (zone = (first_online_pgdat())->node_zones; \
941 	     zone;					\
942 	     zone = next_zone(zone))
943 
944 #define for_each_populated_zone(zone)		        \
945 	for (zone = (first_online_pgdat())->node_zones; \
946 	     zone;					\
947 	     zone = next_zone(zone))			\
948 		if (!populated_zone(zone))		\
949 			; /* do nothing */		\
950 		else
951 
952 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
953 {
954 	return zoneref->zone;
955 }
956 
957 static inline int zonelist_zone_idx(struct zoneref *zoneref)
958 {
959 	return zoneref->zone_idx;
960 }
961 
962 static inline int zonelist_node_idx(struct zoneref *zoneref)
963 {
964 #ifdef CONFIG_NUMA
965 	/* zone_to_nid not available in this context */
966 	return zoneref->zone->node;
967 #else
968 	return 0;
969 #endif /* CONFIG_NUMA */
970 }
971 
972 struct zoneref *__next_zones_zonelist(struct zoneref *z,
973 					enum zone_type highest_zoneidx,
974 					nodemask_t *nodes);
975 
976 /**
977  * 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
978  * @z - The cursor used as a starting point for the search
979  * @highest_zoneidx - The zone index of the highest zone to return
980  * @nodes - An optional nodemask to filter the zonelist with
981  *
982  * This function returns the next zone at or below a given zone index that is
983  * within the allowed nodemask using a cursor as the starting point for the
984  * search. The zoneref returned is a cursor that represents the current zone
985  * being examined. It should be advanced by one before calling
986  * next_zones_zonelist again.
987  */
988 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
989 					enum zone_type highest_zoneidx,
990 					nodemask_t *nodes)
991 {
992 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
993 		return z;
994 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
995 }
996 
997 /**
998  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
999  * @zonelist - The zonelist to search for a suitable zone
1000  * @highest_zoneidx - The zone index of the highest zone to return
1001  * @nodes - An optional nodemask to filter the zonelist with
1002  * @return - Zoneref pointer for the first suitable zone found (see below)
1003  *
1004  * This function returns the first zone at or below a given zone index that is
1005  * within the allowed nodemask. The zoneref returned is a cursor that can be
1006  * used to iterate the zonelist with next_zones_zonelist by advancing it by
1007  * one before calling.
1008  *
1009  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1010  * never NULL). This may happen either genuinely, or due to concurrent nodemask
1011  * update due to cpuset modification.
1012  */
1013 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1014 					enum zone_type highest_zoneidx,
1015 					nodemask_t *nodes)
1016 {
1017 	return next_zones_zonelist(zonelist->_zonerefs,
1018 							highest_zoneidx, nodes);
1019 }
1020 
1021 /**
1022  * 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
1023  * @zone - The current zone in the iterator
1024  * @z - The current pointer within zonelist->zones being iterated
1025  * @zlist - The zonelist being iterated
1026  * @highidx - The zone index of the highest zone to return
1027  * @nodemask - Nodemask allowed by the allocator
1028  *
1029  * This iterator iterates though all zones at or below a given zone index and
1030  * within a given nodemask
1031  */
1032 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1033 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1034 		zone;							\
1035 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1036 			zone = zonelist_zone(z))
1037 
1038 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1039 	for (zone = z->zone;	\
1040 		zone;							\
1041 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1042 			zone = zonelist_zone(z))
1043 
1044 
1045 /**
1046  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1047  * @zone - The current zone in the iterator
1048  * @z - The current pointer within zonelist->zones being iterated
1049  * @zlist - The zonelist being iterated
1050  * @highidx - The zone index of the highest zone to return
1051  *
1052  * This iterator iterates though all zones at or below a given zone index.
1053  */
1054 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1055 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1056 
1057 #ifdef CONFIG_SPARSEMEM
1058 #include <asm/sparsemem.h>
1059 #endif
1060 
1061 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1062 	!defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1063 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1064 {
1065 	BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1066 	return 0;
1067 }
1068 #endif
1069 
1070 #ifdef CONFIG_FLATMEM
1071 #define pfn_to_nid(pfn)		(0)
1072 #endif
1073 
1074 #ifdef CONFIG_SPARSEMEM
1075 
1076 /*
1077  * SECTION_SHIFT    		#bits space required to store a section #
1078  *
1079  * PA_SECTION_SHIFT		physical address to/from section number
1080  * PFN_SECTION_SHIFT		pfn to/from section number
1081  */
1082 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1083 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1084 
1085 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1086 
1087 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1088 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1089 
1090 #define SECTION_BLOCKFLAGS_BITS \
1091 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1092 
1093 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1094 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1095 #endif
1096 
1097 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1098 {
1099 	return pfn >> PFN_SECTION_SHIFT;
1100 }
1101 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1102 {
1103 	return sec << PFN_SECTION_SHIFT;
1104 }
1105 
1106 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1107 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1108 
1109 struct page;
1110 struct page_ext;
1111 struct mem_section {
1112 	/*
1113 	 * This is, logically, a pointer to an array of struct
1114 	 * pages.  However, it is stored with some other magic.
1115 	 * (see sparse.c::sparse_init_one_section())
1116 	 *
1117 	 * Additionally during early boot we encode node id of
1118 	 * the location of the section here to guide allocation.
1119 	 * (see sparse.c::memory_present())
1120 	 *
1121 	 * Making it a UL at least makes someone do a cast
1122 	 * before using it wrong.
1123 	 */
1124 	unsigned long section_mem_map;
1125 
1126 	/* See declaration of similar field in struct zone */
1127 	unsigned long *pageblock_flags;
1128 #ifdef CONFIG_PAGE_EXTENSION
1129 	/*
1130 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1131 	 * section. (see page_ext.h about this.)
1132 	 */
1133 	struct page_ext *page_ext;
1134 	unsigned long pad;
1135 #endif
1136 	/*
1137 	 * WARNING: mem_section must be a power-of-2 in size for the
1138 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1139 	 */
1140 };
1141 
1142 #ifdef CONFIG_SPARSEMEM_EXTREME
1143 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1144 #else
1145 #define SECTIONS_PER_ROOT	1
1146 #endif
1147 
1148 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1149 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1150 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1151 
1152 #ifdef CONFIG_SPARSEMEM_EXTREME
1153 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1154 #else
1155 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1156 #endif
1157 
1158 static inline struct mem_section *__nr_to_section(unsigned long nr)
1159 {
1160 	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1161 		return NULL;
1162 	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1163 }
1164 extern int __section_nr(struct mem_section* ms);
1165 extern unsigned long usemap_size(void);
1166 
1167 /*
1168  * We use the lower bits of the mem_map pointer to store
1169  * a little bit of information.  There should be at least
1170  * 3 bits here due to 32-bit alignment.
1171  */
1172 #define	SECTION_MARKED_PRESENT	(1UL<<0)
1173 #define SECTION_HAS_MEM_MAP	(1UL<<1)
1174 #define SECTION_IS_ONLINE	(1UL<<2)
1175 #define SECTION_MAP_LAST_BIT	(1UL<<3)
1176 #define SECTION_MAP_MASK	(~(SECTION_MAP_LAST_BIT-1))
1177 #define SECTION_NID_SHIFT	3
1178 
1179 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1180 {
1181 	unsigned long map = section->section_mem_map;
1182 	map &= SECTION_MAP_MASK;
1183 	return (struct page *)map;
1184 }
1185 
1186 static inline int present_section(struct mem_section *section)
1187 {
1188 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1189 }
1190 
1191 static inline int present_section_nr(unsigned long nr)
1192 {
1193 	return present_section(__nr_to_section(nr));
1194 }
1195 
1196 static inline int valid_section(struct mem_section *section)
1197 {
1198 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1199 }
1200 
1201 static inline int valid_section_nr(unsigned long nr)
1202 {
1203 	return valid_section(__nr_to_section(nr));
1204 }
1205 
1206 static inline int online_section(struct mem_section *section)
1207 {
1208 	return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1209 }
1210 
1211 static inline int online_section_nr(unsigned long nr)
1212 {
1213 	return online_section(__nr_to_section(nr));
1214 }
1215 
1216 #ifdef CONFIG_MEMORY_HOTPLUG
1217 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1218 #ifdef CONFIG_MEMORY_HOTREMOVE
1219 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1220 #endif
1221 #endif
1222 
1223 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1224 {
1225 	return __nr_to_section(pfn_to_section_nr(pfn));
1226 }
1227 
1228 extern int __highest_present_section_nr;
1229 
1230 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1231 static inline int pfn_valid(unsigned long pfn)
1232 {
1233 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1234 		return 0;
1235 	return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1236 }
1237 #endif
1238 
1239 static inline int pfn_present(unsigned long pfn)
1240 {
1241 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1242 		return 0;
1243 	return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1244 }
1245 
1246 /*
1247  * These are _only_ used during initialisation, therefore they
1248  * can use __initdata ...  They could have names to indicate
1249  * this restriction.
1250  */
1251 #ifdef CONFIG_NUMA
1252 #define pfn_to_nid(pfn)							\
1253 ({									\
1254 	unsigned long __pfn_to_nid_pfn = (pfn);				\
1255 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
1256 })
1257 #else
1258 #define pfn_to_nid(pfn)		(0)
1259 #endif
1260 
1261 #define early_pfn_valid(pfn)	pfn_valid(pfn)
1262 void sparse_init(void);
1263 #else
1264 #define sparse_init()	do {} while (0)
1265 #define sparse_index_init(_sec, _nid)  do {} while (0)
1266 #endif /* CONFIG_SPARSEMEM */
1267 
1268 /*
1269  * During memory init memblocks map pfns to nids. The search is expensive and
1270  * this caches recent lookups. The implementation of __early_pfn_to_nid
1271  * may treat start/end as pfns or sections.
1272  */
1273 struct mminit_pfnnid_cache {
1274 	unsigned long last_start;
1275 	unsigned long last_end;
1276 	int last_nid;
1277 };
1278 
1279 #ifndef early_pfn_valid
1280 #define early_pfn_valid(pfn)	(1)
1281 #endif
1282 
1283 void memory_present(int nid, unsigned long start, unsigned long end);
1284 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1285 
1286 /*
1287  * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1288  * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1289  * pfn_valid_within() should be used in this case; we optimise this away
1290  * when we have no holes within a MAX_ORDER_NR_PAGES block.
1291  */
1292 #ifdef CONFIG_HOLES_IN_ZONE
1293 #define pfn_valid_within(pfn) pfn_valid(pfn)
1294 #else
1295 #define pfn_valid_within(pfn) (1)
1296 #endif
1297 
1298 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1299 /*
1300  * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1301  * associated with it or not. This means that a struct page exists for this
1302  * pfn. The caller cannot assume the page is fully initialized in general.
1303  * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1304  * will ensure the struct page is fully online and initialized. Special pages
1305  * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1306  *
1307  * In FLATMEM, it is expected that holes always have valid memmap as long as
1308  * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1309  * that a valid section has a memmap for the entire section.
1310  *
1311  * However, an ARM, and maybe other embedded architectures in the future
1312  * free memmap backing holes to save memory on the assumption the memmap is
1313  * never used. The page_zone linkages are then broken even though pfn_valid()
1314  * returns true. A walker of the full memmap must then do this additional
1315  * check to ensure the memmap they are looking at is sane by making sure
1316  * the zone and PFN linkages are still valid. This is expensive, but walkers
1317  * of the full memmap are extremely rare.
1318  */
1319 bool memmap_valid_within(unsigned long pfn,
1320 					struct page *page, struct zone *zone);
1321 #else
1322 static inline bool memmap_valid_within(unsigned long pfn,
1323 					struct page *page, struct zone *zone)
1324 {
1325 	return true;
1326 }
1327 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1328 
1329 #endif /* !__GENERATING_BOUNDS.H */
1330 #endif /* !__ASSEMBLY__ */
1331 #endif /* _LINUX_MMZONE_H */
1332