xref: /linux-6.15/include/linux/mmzone.h (revision ea49e01c)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MMZONE_H
3 #define _LINUX_MMZONE_H
4 
5 #ifndef __ASSEMBLY__
6 #ifndef __GENERATING_BOUNDS_H
7 
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <linux/cache.h>
13 #include <linux/threads.h>
14 #include <linux/numa.h>
15 #include <linux/init.h>
16 #include <linux/seqlock.h>
17 #include <linux/nodemask.h>
18 #include <linux/pageblock-flags.h>
19 #include <linux/page-flags-layout.h>
20 #include <linux/atomic.h>
21 #include <linux/mm_types.h>
22 #include <linux/page-flags.h>
23 #include <asm/page.h>
24 
25 /* Free memory management - zoned buddy allocator.  */
26 #ifndef CONFIG_FORCE_MAX_ZONEORDER
27 #define MAX_ORDER 11
28 #else
29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
30 #endif
31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
32 
33 /*
34  * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35  * costly to service.  That is between allocation orders which should
36  * coalesce naturally under reasonable reclaim pressure and those which
37  * will not.
38  */
39 #define PAGE_ALLOC_COSTLY_ORDER 3
40 
41 enum migratetype {
42 	MIGRATE_UNMOVABLE,
43 	MIGRATE_MOVABLE,
44 	MIGRATE_RECLAIMABLE,
45 	MIGRATE_PCPTYPES,	/* the number of types on the pcp lists */
46 	MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
47 #ifdef CONFIG_CMA
48 	/*
49 	 * MIGRATE_CMA migration type is designed to mimic the way
50 	 * ZONE_MOVABLE works.  Only movable pages can be allocated
51 	 * from MIGRATE_CMA pageblocks and page allocator never
52 	 * implicitly change migration type of MIGRATE_CMA pageblock.
53 	 *
54 	 * The way to use it is to change migratetype of a range of
55 	 * pageblocks to MIGRATE_CMA which can be done by
56 	 * __free_pageblock_cma() function.  What is important though
57 	 * is that a range of pageblocks must be aligned to
58 	 * MAX_ORDER_NR_PAGES should biggest page be bigger then
59 	 * a single pageblock.
60 	 */
61 	MIGRATE_CMA,
62 #endif
63 #ifdef CONFIG_MEMORY_ISOLATION
64 	MIGRATE_ISOLATE,	/* can't allocate from here */
65 #endif
66 	MIGRATE_TYPES
67 };
68 
69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70 extern const char * const migratetype_names[MIGRATE_TYPES];
71 
72 #ifdef CONFIG_CMA
73 #  define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74 #  define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
75 #else
76 #  define is_migrate_cma(migratetype) false
77 #  define is_migrate_cma_page(_page) false
78 #endif
79 
80 static inline bool is_migrate_movable(int mt)
81 {
82 	return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
83 }
84 
85 #define for_each_migratetype_order(order, type) \
86 	for (order = 0; order < MAX_ORDER; order++) \
87 		for (type = 0; type < MIGRATE_TYPES; type++)
88 
89 extern int page_group_by_mobility_disabled;
90 
91 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
92 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
93 
94 #define get_pageblock_migratetype(page)					\
95 	get_pfnblock_flags_mask(page, page_to_pfn(page),		\
96 			PB_migrate_end, MIGRATETYPE_MASK)
97 
98 struct free_area {
99 	struct list_head	free_list[MIGRATE_TYPES];
100 	unsigned long		nr_free;
101 };
102 
103 /* Used for pages not on another list */
104 static inline void add_to_free_area(struct page *page, struct free_area *area,
105 			     int migratetype)
106 {
107 	list_add(&page->lru, &area->free_list[migratetype]);
108 	area->nr_free++;
109 }
110 
111 /* Used for pages not on another list */
112 static inline void add_to_free_area_tail(struct page *page, struct free_area *area,
113 				  int migratetype)
114 {
115 	list_add_tail(&page->lru, &area->free_list[migratetype]);
116 	area->nr_free++;
117 }
118 
119 #ifdef CONFIG_SHUFFLE_PAGE_ALLOCATOR
120 /* Used to preserve page allocation order entropy */
121 void add_to_free_area_random(struct page *page, struct free_area *area,
122 		int migratetype);
123 #else
124 static inline void add_to_free_area_random(struct page *page,
125 		struct free_area *area, int migratetype)
126 {
127 	add_to_free_area(page, area, migratetype);
128 }
129 #endif
130 
131 /* Used for pages which are on another list */
132 static inline void move_to_free_area(struct page *page, struct free_area *area,
133 			     int migratetype)
134 {
135 	list_move(&page->lru, &area->free_list[migratetype]);
136 }
137 
138 static inline struct page *get_page_from_free_area(struct free_area *area,
139 					    int migratetype)
140 {
141 	return list_first_entry_or_null(&area->free_list[migratetype],
142 					struct page, lru);
143 }
144 
145 static inline void del_page_from_free_area(struct page *page,
146 		struct free_area *area)
147 {
148 	list_del(&page->lru);
149 	__ClearPageBuddy(page);
150 	set_page_private(page, 0);
151 	area->nr_free--;
152 }
153 
154 static inline bool free_area_empty(struct free_area *area, int migratetype)
155 {
156 	return list_empty(&area->free_list[migratetype]);
157 }
158 
159 struct pglist_data;
160 
161 /*
162  * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
163  * So add a wild amount of padding here to ensure that they fall into separate
164  * cachelines.  There are very few zone structures in the machine, so space
165  * consumption is not a concern here.
166  */
167 #if defined(CONFIG_SMP)
168 struct zone_padding {
169 	char x[0];
170 } ____cacheline_internodealigned_in_smp;
171 #define ZONE_PADDING(name)	struct zone_padding name;
172 #else
173 #define ZONE_PADDING(name)
174 #endif
175 
176 #ifdef CONFIG_NUMA
177 enum numa_stat_item {
178 	NUMA_HIT,		/* allocated in intended node */
179 	NUMA_MISS,		/* allocated in non intended node */
180 	NUMA_FOREIGN,		/* was intended here, hit elsewhere */
181 	NUMA_INTERLEAVE_HIT,	/* interleaver preferred this zone */
182 	NUMA_LOCAL,		/* allocation from local node */
183 	NUMA_OTHER,		/* allocation from other node */
184 	NR_VM_NUMA_STAT_ITEMS
185 };
186 #else
187 #define NR_VM_NUMA_STAT_ITEMS 0
188 #endif
189 
190 enum zone_stat_item {
191 	/* First 128 byte cacheline (assuming 64 bit words) */
192 	NR_FREE_PAGES,
193 	NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
194 	NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
195 	NR_ZONE_ACTIVE_ANON,
196 	NR_ZONE_INACTIVE_FILE,
197 	NR_ZONE_ACTIVE_FILE,
198 	NR_ZONE_UNEVICTABLE,
199 	NR_ZONE_WRITE_PENDING,	/* Count of dirty, writeback and unstable pages */
200 	NR_MLOCK,		/* mlock()ed pages found and moved off LRU */
201 	NR_PAGETABLE,		/* used for pagetables */
202 	NR_KERNEL_STACK_KB,	/* measured in KiB */
203 	/* Second 128 byte cacheline */
204 	NR_BOUNCE,
205 #if IS_ENABLED(CONFIG_ZSMALLOC)
206 	NR_ZSPAGES,		/* allocated in zsmalloc */
207 #endif
208 	NR_FREE_CMA_PAGES,
209 	NR_VM_ZONE_STAT_ITEMS };
210 
211 enum node_stat_item {
212 	NR_LRU_BASE,
213 	NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
214 	NR_ACTIVE_ANON,		/*  "     "     "   "       "         */
215 	NR_INACTIVE_FILE,	/*  "     "     "   "       "         */
216 	NR_ACTIVE_FILE,		/*  "     "     "   "       "         */
217 	NR_UNEVICTABLE,		/*  "     "     "   "       "         */
218 	NR_SLAB_RECLAIMABLE,	/* Please do not reorder this item */
219 	NR_SLAB_UNRECLAIMABLE,	/* and this one without looking at
220 				 * memcg_flush_percpu_vmstats() first. */
221 	NR_ISOLATED_ANON,	/* Temporary isolated pages from anon lru */
222 	NR_ISOLATED_FILE,	/* Temporary isolated pages from file lru */
223 	WORKINGSET_NODES,
224 	WORKINGSET_REFAULT,
225 	WORKINGSET_ACTIVATE,
226 	WORKINGSET_RESTORE,
227 	WORKINGSET_NODERECLAIM,
228 	NR_ANON_MAPPED,	/* Mapped anonymous pages */
229 	NR_FILE_MAPPED,	/* pagecache pages mapped into pagetables.
230 			   only modified from process context */
231 	NR_FILE_PAGES,
232 	NR_FILE_DIRTY,
233 	NR_WRITEBACK,
234 	NR_WRITEBACK_TEMP,	/* Writeback using temporary buffers */
235 	NR_SHMEM,		/* shmem pages (included tmpfs/GEM pages) */
236 	NR_SHMEM_THPS,
237 	NR_SHMEM_PMDMAPPED,
238 	NR_ANON_THPS,
239 	NR_UNSTABLE_NFS,	/* NFS unstable pages */
240 	NR_VMSCAN_WRITE,
241 	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
242 	NR_DIRTIED,		/* page dirtyings since bootup */
243 	NR_WRITTEN,		/* page writings since bootup */
244 	NR_KERNEL_MISC_RECLAIMABLE,	/* reclaimable non-slab kernel pages */
245 	NR_VM_NODE_STAT_ITEMS
246 };
247 
248 /*
249  * We do arithmetic on the LRU lists in various places in the code,
250  * so it is important to keep the active lists LRU_ACTIVE higher in
251  * the array than the corresponding inactive lists, and to keep
252  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
253  *
254  * This has to be kept in sync with the statistics in zone_stat_item
255  * above and the descriptions in vmstat_text in mm/vmstat.c
256  */
257 #define LRU_BASE 0
258 #define LRU_ACTIVE 1
259 #define LRU_FILE 2
260 
261 enum lru_list {
262 	LRU_INACTIVE_ANON = LRU_BASE,
263 	LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
264 	LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
265 	LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
266 	LRU_UNEVICTABLE,
267 	NR_LRU_LISTS
268 };
269 
270 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
271 
272 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
273 
274 static inline int is_file_lru(enum lru_list lru)
275 {
276 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
277 }
278 
279 static inline int is_active_lru(enum lru_list lru)
280 {
281 	return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
282 }
283 
284 struct zone_reclaim_stat {
285 	/*
286 	 * The pageout code in vmscan.c keeps track of how many of the
287 	 * mem/swap backed and file backed pages are referenced.
288 	 * The higher the rotated/scanned ratio, the more valuable
289 	 * that cache is.
290 	 *
291 	 * The anon LRU stats live in [0], file LRU stats in [1]
292 	 */
293 	unsigned long		recent_rotated[2];
294 	unsigned long		recent_scanned[2];
295 };
296 
297 struct lruvec {
298 	struct list_head		lists[NR_LRU_LISTS];
299 	struct zone_reclaim_stat	reclaim_stat;
300 	/* Evictions & activations on the inactive file list */
301 	atomic_long_t			inactive_age;
302 	/* Refaults at the time of last reclaim cycle */
303 	unsigned long			refaults;
304 #ifdef CONFIG_MEMCG
305 	struct pglist_data *pgdat;
306 #endif
307 };
308 
309 /* Isolate unmapped file */
310 #define ISOLATE_UNMAPPED	((__force isolate_mode_t)0x2)
311 /* Isolate for asynchronous migration */
312 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
313 /* Isolate unevictable pages */
314 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
315 
316 /* LRU Isolation modes. */
317 typedef unsigned __bitwise isolate_mode_t;
318 
319 enum zone_watermarks {
320 	WMARK_MIN,
321 	WMARK_LOW,
322 	WMARK_HIGH,
323 	NR_WMARK
324 };
325 
326 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
327 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
328 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
329 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
330 
331 struct per_cpu_pages {
332 	int count;		/* number of pages in the list */
333 	int high;		/* high watermark, emptying needed */
334 	int batch;		/* chunk size for buddy add/remove */
335 
336 	/* Lists of pages, one per migrate type stored on the pcp-lists */
337 	struct list_head lists[MIGRATE_PCPTYPES];
338 };
339 
340 struct per_cpu_pageset {
341 	struct per_cpu_pages pcp;
342 #ifdef CONFIG_NUMA
343 	s8 expire;
344 	u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
345 #endif
346 #ifdef CONFIG_SMP
347 	s8 stat_threshold;
348 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
349 #endif
350 };
351 
352 struct per_cpu_nodestat {
353 	s8 stat_threshold;
354 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
355 };
356 
357 #endif /* !__GENERATING_BOUNDS.H */
358 
359 enum zone_type {
360 #ifdef CONFIG_ZONE_DMA
361 	/*
362 	 * ZONE_DMA is used when there are devices that are not able
363 	 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
364 	 * carve out the portion of memory that is needed for these devices.
365 	 * The range is arch specific.
366 	 *
367 	 * Some examples
368 	 *
369 	 * Architecture		Limit
370 	 * ---------------------------
371 	 * parisc, ia64, sparc	<4G
372 	 * s390, powerpc	<2G
373 	 * arm			Various
374 	 * alpha		Unlimited or 0-16MB.
375 	 *
376 	 * i386, x86_64 and multiple other arches
377 	 * 			<16M.
378 	 */
379 	ZONE_DMA,
380 #endif
381 #ifdef CONFIG_ZONE_DMA32
382 	/*
383 	 * x86_64 needs two ZONE_DMAs because it supports devices that are
384 	 * only able to do DMA to the lower 16M but also 32 bit devices that
385 	 * can only do DMA areas below 4G.
386 	 */
387 	ZONE_DMA32,
388 #endif
389 	/*
390 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
391 	 * performed on pages in ZONE_NORMAL if the DMA devices support
392 	 * transfers to all addressable memory.
393 	 */
394 	ZONE_NORMAL,
395 #ifdef CONFIG_HIGHMEM
396 	/*
397 	 * A memory area that is only addressable by the kernel through
398 	 * mapping portions into its own address space. This is for example
399 	 * used by i386 to allow the kernel to address the memory beyond
400 	 * 900MB. The kernel will set up special mappings (page
401 	 * table entries on i386) for each page that the kernel needs to
402 	 * access.
403 	 */
404 	ZONE_HIGHMEM,
405 #endif
406 	ZONE_MOVABLE,
407 #ifdef CONFIG_ZONE_DEVICE
408 	ZONE_DEVICE,
409 #endif
410 	__MAX_NR_ZONES
411 
412 };
413 
414 #ifndef __GENERATING_BOUNDS_H
415 
416 struct zone {
417 	/* Read-mostly fields */
418 
419 	/* zone watermarks, access with *_wmark_pages(zone) macros */
420 	unsigned long _watermark[NR_WMARK];
421 	unsigned long watermark_boost;
422 
423 	unsigned long nr_reserved_highatomic;
424 
425 	/*
426 	 * We don't know if the memory that we're going to allocate will be
427 	 * freeable or/and it will be released eventually, so to avoid totally
428 	 * wasting several GB of ram we must reserve some of the lower zone
429 	 * memory (otherwise we risk to run OOM on the lower zones despite
430 	 * there being tons of freeable ram on the higher zones).  This array is
431 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
432 	 * changes.
433 	 */
434 	long lowmem_reserve[MAX_NR_ZONES];
435 
436 #ifdef CONFIG_NUMA
437 	int node;
438 #endif
439 	struct pglist_data	*zone_pgdat;
440 	struct per_cpu_pageset __percpu *pageset;
441 
442 #ifndef CONFIG_SPARSEMEM
443 	/*
444 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
445 	 * In SPARSEMEM, this map is stored in struct mem_section
446 	 */
447 	unsigned long		*pageblock_flags;
448 #endif /* CONFIG_SPARSEMEM */
449 
450 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
451 	unsigned long		zone_start_pfn;
452 
453 	/*
454 	 * spanned_pages is the total pages spanned by the zone, including
455 	 * holes, which is calculated as:
456 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
457 	 *
458 	 * present_pages is physical pages existing within the zone, which
459 	 * is calculated as:
460 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
461 	 *
462 	 * managed_pages is present pages managed by the buddy system, which
463 	 * is calculated as (reserved_pages includes pages allocated by the
464 	 * bootmem allocator):
465 	 *	managed_pages = present_pages - reserved_pages;
466 	 *
467 	 * So present_pages may be used by memory hotplug or memory power
468 	 * management logic to figure out unmanaged pages by checking
469 	 * (present_pages - managed_pages). And managed_pages should be used
470 	 * by page allocator and vm scanner to calculate all kinds of watermarks
471 	 * and thresholds.
472 	 *
473 	 * Locking rules:
474 	 *
475 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
476 	 * It is a seqlock because it has to be read outside of zone->lock,
477 	 * and it is done in the main allocator path.  But, it is written
478 	 * quite infrequently.
479 	 *
480 	 * The span_seq lock is declared along with zone->lock because it is
481 	 * frequently read in proximity to zone->lock.  It's good to
482 	 * give them a chance of being in the same cacheline.
483 	 *
484 	 * Write access to present_pages at runtime should be protected by
485 	 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
486 	 * present_pages should get_online_mems() to get a stable value.
487 	 */
488 	atomic_long_t		managed_pages;
489 	unsigned long		spanned_pages;
490 	unsigned long		present_pages;
491 
492 	const char		*name;
493 
494 #ifdef CONFIG_MEMORY_ISOLATION
495 	/*
496 	 * Number of isolated pageblock. It is used to solve incorrect
497 	 * freepage counting problem due to racy retrieving migratetype
498 	 * of pageblock. Protected by zone->lock.
499 	 */
500 	unsigned long		nr_isolate_pageblock;
501 #endif
502 
503 #ifdef CONFIG_MEMORY_HOTPLUG
504 	/* see spanned/present_pages for more description */
505 	seqlock_t		span_seqlock;
506 #endif
507 
508 	int initialized;
509 
510 	/* Write-intensive fields used from the page allocator */
511 	ZONE_PADDING(_pad1_)
512 
513 	/* free areas of different sizes */
514 	struct free_area	free_area[MAX_ORDER];
515 
516 	/* zone flags, see below */
517 	unsigned long		flags;
518 
519 	/* Primarily protects free_area */
520 	spinlock_t		lock;
521 
522 	/* Write-intensive fields used by compaction and vmstats. */
523 	ZONE_PADDING(_pad2_)
524 
525 	/*
526 	 * When free pages are below this point, additional steps are taken
527 	 * when reading the number of free pages to avoid per-cpu counter
528 	 * drift allowing watermarks to be breached
529 	 */
530 	unsigned long percpu_drift_mark;
531 
532 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
533 	/* pfn where compaction free scanner should start */
534 	unsigned long		compact_cached_free_pfn;
535 	/* pfn where async and sync compaction migration scanner should start */
536 	unsigned long		compact_cached_migrate_pfn[2];
537 	unsigned long		compact_init_migrate_pfn;
538 	unsigned long		compact_init_free_pfn;
539 #endif
540 
541 #ifdef CONFIG_COMPACTION
542 	/*
543 	 * On compaction failure, 1<<compact_defer_shift compactions
544 	 * are skipped before trying again. The number attempted since
545 	 * last failure is tracked with compact_considered.
546 	 */
547 	unsigned int		compact_considered;
548 	unsigned int		compact_defer_shift;
549 	int			compact_order_failed;
550 #endif
551 
552 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
553 	/* Set to true when the PG_migrate_skip bits should be cleared */
554 	bool			compact_blockskip_flush;
555 #endif
556 
557 	bool			contiguous;
558 
559 	ZONE_PADDING(_pad3_)
560 	/* Zone statistics */
561 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
562 	atomic_long_t		vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
563 } ____cacheline_internodealigned_in_smp;
564 
565 enum pgdat_flags {
566 	PGDAT_CONGESTED,		/* pgdat has many dirty pages backed by
567 					 * a congested BDI
568 					 */
569 	PGDAT_DIRTY,			/* reclaim scanning has recently found
570 					 * many dirty file pages at the tail
571 					 * of the LRU.
572 					 */
573 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
574 					 * many pages under writeback
575 					 */
576 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
577 };
578 
579 enum zone_flags {
580 	ZONE_BOOSTED_WATERMARK,		/* zone recently boosted watermarks.
581 					 * Cleared when kswapd is woken.
582 					 */
583 };
584 
585 static inline unsigned long zone_managed_pages(struct zone *zone)
586 {
587 	return (unsigned long)atomic_long_read(&zone->managed_pages);
588 }
589 
590 static inline unsigned long zone_end_pfn(const struct zone *zone)
591 {
592 	return zone->zone_start_pfn + zone->spanned_pages;
593 }
594 
595 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
596 {
597 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
598 }
599 
600 static inline bool zone_is_initialized(struct zone *zone)
601 {
602 	return zone->initialized;
603 }
604 
605 static inline bool zone_is_empty(struct zone *zone)
606 {
607 	return zone->spanned_pages == 0;
608 }
609 
610 /*
611  * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
612  * intersection with the given zone
613  */
614 static inline bool zone_intersects(struct zone *zone,
615 		unsigned long start_pfn, unsigned long nr_pages)
616 {
617 	if (zone_is_empty(zone))
618 		return false;
619 	if (start_pfn >= zone_end_pfn(zone) ||
620 	    start_pfn + nr_pages <= zone->zone_start_pfn)
621 		return false;
622 
623 	return true;
624 }
625 
626 /*
627  * The "priority" of VM scanning is how much of the queues we will scan in one
628  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
629  * queues ("queue_length >> 12") during an aging round.
630  */
631 #define DEF_PRIORITY 12
632 
633 /* Maximum number of zones on a zonelist */
634 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
635 
636 enum {
637 	ZONELIST_FALLBACK,	/* zonelist with fallback */
638 #ifdef CONFIG_NUMA
639 	/*
640 	 * The NUMA zonelists are doubled because we need zonelists that
641 	 * restrict the allocations to a single node for __GFP_THISNODE.
642 	 */
643 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
644 #endif
645 	MAX_ZONELISTS
646 };
647 
648 /*
649  * This struct contains information about a zone in a zonelist. It is stored
650  * here to avoid dereferences into large structures and lookups of tables
651  */
652 struct zoneref {
653 	struct zone *zone;	/* Pointer to actual zone */
654 	int zone_idx;		/* zone_idx(zoneref->zone) */
655 };
656 
657 /*
658  * One allocation request operates on a zonelist. A zonelist
659  * is a list of zones, the first one is the 'goal' of the
660  * allocation, the other zones are fallback zones, in decreasing
661  * priority.
662  *
663  * To speed the reading of the zonelist, the zonerefs contain the zone index
664  * of the entry being read. Helper functions to access information given
665  * a struct zoneref are
666  *
667  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
668  * zonelist_zone_idx()	- Return the index of the zone for an entry
669  * zonelist_node_idx()	- Return the index of the node for an entry
670  */
671 struct zonelist {
672 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
673 };
674 
675 #ifndef CONFIG_DISCONTIGMEM
676 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
677 extern struct page *mem_map;
678 #endif
679 
680 /*
681  * On NUMA machines, each NUMA node would have a pg_data_t to describe
682  * it's memory layout. On UMA machines there is a single pglist_data which
683  * describes the whole memory.
684  *
685  * Memory statistics and page replacement data structures are maintained on a
686  * per-zone basis.
687  */
688 struct bootmem_data;
689 typedef struct pglist_data {
690 	struct zone node_zones[MAX_NR_ZONES];
691 	struct zonelist node_zonelists[MAX_ZONELISTS];
692 	int nr_zones;
693 #ifdef CONFIG_FLAT_NODE_MEM_MAP	/* means !SPARSEMEM */
694 	struct page *node_mem_map;
695 #ifdef CONFIG_PAGE_EXTENSION
696 	struct page_ext *node_page_ext;
697 #endif
698 #endif
699 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
700 	/*
701 	 * Must be held any time you expect node_start_pfn,
702 	 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
703 	 *
704 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
705 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
706 	 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
707 	 *
708 	 * Nests above zone->lock and zone->span_seqlock
709 	 */
710 	spinlock_t node_size_lock;
711 #endif
712 	unsigned long node_start_pfn;
713 	unsigned long node_present_pages; /* total number of physical pages */
714 	unsigned long node_spanned_pages; /* total size of physical page
715 					     range, including holes */
716 	int node_id;
717 	wait_queue_head_t kswapd_wait;
718 	wait_queue_head_t pfmemalloc_wait;
719 	struct task_struct *kswapd;	/* Protected by
720 					   mem_hotplug_begin/end() */
721 	int kswapd_order;
722 	enum zone_type kswapd_classzone_idx;
723 
724 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
725 
726 #ifdef CONFIG_COMPACTION
727 	int kcompactd_max_order;
728 	enum zone_type kcompactd_classzone_idx;
729 	wait_queue_head_t kcompactd_wait;
730 	struct task_struct *kcompactd;
731 #endif
732 	/*
733 	 * This is a per-node reserve of pages that are not available
734 	 * to userspace allocations.
735 	 */
736 	unsigned long		totalreserve_pages;
737 
738 #ifdef CONFIG_NUMA
739 	/*
740 	 * zone reclaim becomes active if more unmapped pages exist.
741 	 */
742 	unsigned long		min_unmapped_pages;
743 	unsigned long		min_slab_pages;
744 #endif /* CONFIG_NUMA */
745 
746 	/* Write-intensive fields used by page reclaim */
747 	ZONE_PADDING(_pad1_)
748 	spinlock_t		lru_lock;
749 
750 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
751 	/*
752 	 * If memory initialisation on large machines is deferred then this
753 	 * is the first PFN that needs to be initialised.
754 	 */
755 	unsigned long first_deferred_pfn;
756 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
757 
758 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
759 	spinlock_t split_queue_lock;
760 	struct list_head split_queue;
761 	unsigned long split_queue_len;
762 #endif
763 
764 	/* Fields commonly accessed by the page reclaim scanner */
765 	struct lruvec		lruvec;
766 
767 	unsigned long		flags;
768 
769 	ZONE_PADDING(_pad2_)
770 
771 	/* Per-node vmstats */
772 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
773 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
774 } pg_data_t;
775 
776 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
777 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
778 #ifdef CONFIG_FLAT_NODE_MEM_MAP
779 #define pgdat_page_nr(pgdat, pagenr)	((pgdat)->node_mem_map + (pagenr))
780 #else
781 #define pgdat_page_nr(pgdat, pagenr)	pfn_to_page((pgdat)->node_start_pfn + (pagenr))
782 #endif
783 #define nid_page_nr(nid, pagenr) 	pgdat_page_nr(NODE_DATA(nid),(pagenr))
784 
785 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
786 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
787 
788 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
789 {
790 	return &pgdat->lruvec;
791 }
792 
793 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
794 {
795 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
796 }
797 
798 static inline bool pgdat_is_empty(pg_data_t *pgdat)
799 {
800 	return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
801 }
802 
803 #include <linux/memory_hotplug.h>
804 
805 void build_all_zonelists(pg_data_t *pgdat);
806 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
807 		   enum zone_type classzone_idx);
808 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
809 			 int classzone_idx, unsigned int alloc_flags,
810 			 long free_pages);
811 bool zone_watermark_ok(struct zone *z, unsigned int order,
812 		unsigned long mark, int classzone_idx,
813 		unsigned int alloc_flags);
814 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
815 		unsigned long mark, int classzone_idx);
816 enum memmap_context {
817 	MEMMAP_EARLY,
818 	MEMMAP_HOTPLUG,
819 };
820 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
821 				     unsigned long size);
822 
823 extern void lruvec_init(struct lruvec *lruvec);
824 
825 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
826 {
827 #ifdef CONFIG_MEMCG
828 	return lruvec->pgdat;
829 #else
830 	return container_of(lruvec, struct pglist_data, lruvec);
831 #endif
832 }
833 
834 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
835 
836 #ifdef CONFIG_HAVE_MEMORY_PRESENT
837 void memory_present(int nid, unsigned long start, unsigned long end);
838 #else
839 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
840 #endif
841 
842 #if defined(CONFIG_SPARSEMEM)
843 void memblocks_present(void);
844 #else
845 static inline void memblocks_present(void) {}
846 #endif
847 
848 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
849 int local_memory_node(int node_id);
850 #else
851 static inline int local_memory_node(int node_id) { return node_id; };
852 #endif
853 
854 /*
855  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
856  */
857 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
858 
859 /*
860  * Returns true if a zone has pages managed by the buddy allocator.
861  * All the reclaim decisions have to use this function rather than
862  * populated_zone(). If the whole zone is reserved then we can easily
863  * end up with populated_zone() && !managed_zone().
864  */
865 static inline bool managed_zone(struct zone *zone)
866 {
867 	return zone_managed_pages(zone);
868 }
869 
870 /* Returns true if a zone has memory */
871 static inline bool populated_zone(struct zone *zone)
872 {
873 	return zone->present_pages;
874 }
875 
876 #ifdef CONFIG_NUMA
877 static inline int zone_to_nid(struct zone *zone)
878 {
879 	return zone->node;
880 }
881 
882 static inline void zone_set_nid(struct zone *zone, int nid)
883 {
884 	zone->node = nid;
885 }
886 #else
887 static inline int zone_to_nid(struct zone *zone)
888 {
889 	return 0;
890 }
891 
892 static inline void zone_set_nid(struct zone *zone, int nid) {}
893 #endif
894 
895 extern int movable_zone;
896 
897 #ifdef CONFIG_HIGHMEM
898 static inline int zone_movable_is_highmem(void)
899 {
900 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
901 	return movable_zone == ZONE_HIGHMEM;
902 #else
903 	return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
904 #endif
905 }
906 #endif
907 
908 static inline int is_highmem_idx(enum zone_type idx)
909 {
910 #ifdef CONFIG_HIGHMEM
911 	return (idx == ZONE_HIGHMEM ||
912 		(idx == ZONE_MOVABLE && zone_movable_is_highmem()));
913 #else
914 	return 0;
915 #endif
916 }
917 
918 /**
919  * is_highmem - helper function to quickly check if a struct zone is a
920  *              highmem zone or not.  This is an attempt to keep references
921  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
922  * @zone - pointer to struct zone variable
923  */
924 static inline int is_highmem(struct zone *zone)
925 {
926 #ifdef CONFIG_HIGHMEM
927 	return is_highmem_idx(zone_idx(zone));
928 #else
929 	return 0;
930 #endif
931 }
932 
933 /* These two functions are used to setup the per zone pages min values */
934 struct ctl_table;
935 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
936 					void __user *, size_t *, loff_t *);
937 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
938 					void __user *, size_t *, loff_t *);
939 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
940 					void __user *, size_t *, loff_t *);
941 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
942 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
943 					void __user *, size_t *, loff_t *);
944 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
945 					void __user *, size_t *, loff_t *);
946 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
947 			void __user *, size_t *, loff_t *);
948 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
949 			void __user *, size_t *, loff_t *);
950 
951 extern int numa_zonelist_order_handler(struct ctl_table *, int,
952 			void __user *, size_t *, loff_t *);
953 extern char numa_zonelist_order[];
954 #define NUMA_ZONELIST_ORDER_LEN	16
955 
956 #ifndef CONFIG_NEED_MULTIPLE_NODES
957 
958 extern struct pglist_data contig_page_data;
959 #define NODE_DATA(nid)		(&contig_page_data)
960 #define NODE_MEM_MAP(nid)	mem_map
961 
962 #else /* CONFIG_NEED_MULTIPLE_NODES */
963 
964 #include <asm/mmzone.h>
965 
966 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
967 
968 extern struct pglist_data *first_online_pgdat(void);
969 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
970 extern struct zone *next_zone(struct zone *zone);
971 
972 /**
973  * for_each_online_pgdat - helper macro to iterate over all online nodes
974  * @pgdat - pointer to a pg_data_t variable
975  */
976 #define for_each_online_pgdat(pgdat)			\
977 	for (pgdat = first_online_pgdat();		\
978 	     pgdat;					\
979 	     pgdat = next_online_pgdat(pgdat))
980 /**
981  * for_each_zone - helper macro to iterate over all memory zones
982  * @zone - pointer to struct zone variable
983  *
984  * The user only needs to declare the zone variable, for_each_zone
985  * fills it in.
986  */
987 #define for_each_zone(zone)			        \
988 	for (zone = (first_online_pgdat())->node_zones; \
989 	     zone;					\
990 	     zone = next_zone(zone))
991 
992 #define for_each_populated_zone(zone)		        \
993 	for (zone = (first_online_pgdat())->node_zones; \
994 	     zone;					\
995 	     zone = next_zone(zone))			\
996 		if (!populated_zone(zone))		\
997 			; /* do nothing */		\
998 		else
999 
1000 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1001 {
1002 	return zoneref->zone;
1003 }
1004 
1005 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1006 {
1007 	return zoneref->zone_idx;
1008 }
1009 
1010 static inline int zonelist_node_idx(struct zoneref *zoneref)
1011 {
1012 	return zone_to_nid(zoneref->zone);
1013 }
1014 
1015 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1016 					enum zone_type highest_zoneidx,
1017 					nodemask_t *nodes);
1018 
1019 /**
1020  * 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
1021  * @z - The cursor used as a starting point for the search
1022  * @highest_zoneidx - The zone index of the highest zone to return
1023  * @nodes - An optional nodemask to filter the zonelist with
1024  *
1025  * This function returns the next zone at or below a given zone index that is
1026  * within the allowed nodemask using a cursor as the starting point for the
1027  * search. The zoneref returned is a cursor that represents the current zone
1028  * being examined. It should be advanced by one before calling
1029  * next_zones_zonelist again.
1030  */
1031 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1032 					enum zone_type highest_zoneidx,
1033 					nodemask_t *nodes)
1034 {
1035 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1036 		return z;
1037 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
1038 }
1039 
1040 /**
1041  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1042  * @zonelist - The zonelist to search for a suitable zone
1043  * @highest_zoneidx - The zone index of the highest zone to return
1044  * @nodes - An optional nodemask to filter the zonelist with
1045  * @return - Zoneref pointer for the first suitable zone found (see below)
1046  *
1047  * This function returns the first zone at or below a given zone index that is
1048  * within the allowed nodemask. The zoneref returned is a cursor that can be
1049  * used to iterate the zonelist with next_zones_zonelist by advancing it by
1050  * one before calling.
1051  *
1052  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1053  * never NULL). This may happen either genuinely, or due to concurrent nodemask
1054  * update due to cpuset modification.
1055  */
1056 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1057 					enum zone_type highest_zoneidx,
1058 					nodemask_t *nodes)
1059 {
1060 	return next_zones_zonelist(zonelist->_zonerefs,
1061 							highest_zoneidx, nodes);
1062 }
1063 
1064 /**
1065  * 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
1066  * @zone - The current zone in the iterator
1067  * @z - The current pointer within zonelist->zones being iterated
1068  * @zlist - The zonelist being iterated
1069  * @highidx - The zone index of the highest zone to return
1070  * @nodemask - Nodemask allowed by the allocator
1071  *
1072  * This iterator iterates though all zones at or below a given zone index and
1073  * within a given nodemask
1074  */
1075 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1076 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1077 		zone;							\
1078 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1079 			zone = zonelist_zone(z))
1080 
1081 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1082 	for (zone = z->zone;	\
1083 		zone;							\
1084 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1085 			zone = zonelist_zone(z))
1086 
1087 
1088 /**
1089  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1090  * @zone - The current zone in the iterator
1091  * @z - The current pointer within zonelist->zones being iterated
1092  * @zlist - The zonelist being iterated
1093  * @highidx - The zone index of the highest zone to return
1094  *
1095  * This iterator iterates though all zones at or below a given zone index.
1096  */
1097 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1098 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1099 
1100 #ifdef CONFIG_SPARSEMEM
1101 #include <asm/sparsemem.h>
1102 #endif
1103 
1104 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1105 	!defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1106 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1107 {
1108 	BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1109 	return 0;
1110 }
1111 #endif
1112 
1113 #ifdef CONFIG_FLATMEM
1114 #define pfn_to_nid(pfn)		(0)
1115 #endif
1116 
1117 #ifdef CONFIG_SPARSEMEM
1118 
1119 /*
1120  * SECTION_SHIFT    		#bits space required to store a section #
1121  *
1122  * PA_SECTION_SHIFT		physical address to/from section number
1123  * PFN_SECTION_SHIFT		pfn to/from section number
1124  */
1125 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1126 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1127 
1128 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1129 
1130 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1131 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1132 
1133 #define SECTION_BLOCKFLAGS_BITS \
1134 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1135 
1136 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1137 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1138 #endif
1139 
1140 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1141 {
1142 	return pfn >> PFN_SECTION_SHIFT;
1143 }
1144 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1145 {
1146 	return sec << PFN_SECTION_SHIFT;
1147 }
1148 
1149 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1150 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1151 
1152 #define SUBSECTION_SHIFT 21
1153 
1154 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1155 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1156 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1157 
1158 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1159 #error Subsection size exceeds section size
1160 #else
1161 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1162 #endif
1163 
1164 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1165 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1166 
1167 struct mem_section_usage {
1168 	DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1169 	/* See declaration of similar field in struct zone */
1170 	unsigned long pageblock_flags[0];
1171 };
1172 
1173 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1174 
1175 struct page;
1176 struct page_ext;
1177 struct mem_section {
1178 	/*
1179 	 * This is, logically, a pointer to an array of struct
1180 	 * pages.  However, it is stored with some other magic.
1181 	 * (see sparse.c::sparse_init_one_section())
1182 	 *
1183 	 * Additionally during early boot we encode node id of
1184 	 * the location of the section here to guide allocation.
1185 	 * (see sparse.c::memory_present())
1186 	 *
1187 	 * Making it a UL at least makes someone do a cast
1188 	 * before using it wrong.
1189 	 */
1190 	unsigned long section_mem_map;
1191 
1192 	struct mem_section_usage *usage;
1193 #ifdef CONFIG_PAGE_EXTENSION
1194 	/*
1195 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1196 	 * section. (see page_ext.h about this.)
1197 	 */
1198 	struct page_ext *page_ext;
1199 	unsigned long pad;
1200 #endif
1201 	/*
1202 	 * WARNING: mem_section must be a power-of-2 in size for the
1203 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1204 	 */
1205 };
1206 
1207 #ifdef CONFIG_SPARSEMEM_EXTREME
1208 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1209 #else
1210 #define SECTIONS_PER_ROOT	1
1211 #endif
1212 
1213 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1214 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1215 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1216 
1217 #ifdef CONFIG_SPARSEMEM_EXTREME
1218 extern struct mem_section **mem_section;
1219 #else
1220 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1221 #endif
1222 
1223 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1224 {
1225 	return ms->usage->pageblock_flags;
1226 }
1227 
1228 static inline struct mem_section *__nr_to_section(unsigned long nr)
1229 {
1230 #ifdef CONFIG_SPARSEMEM_EXTREME
1231 	if (!mem_section)
1232 		return NULL;
1233 #endif
1234 	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1235 		return NULL;
1236 	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1237 }
1238 extern unsigned long __section_nr(struct mem_section *ms);
1239 extern size_t mem_section_usage_size(void);
1240 
1241 /*
1242  * We use the lower bits of the mem_map pointer to store
1243  * a little bit of information.  The pointer is calculated
1244  * as mem_map - section_nr_to_pfn(pnum).  The result is
1245  * aligned to the minimum alignment of the two values:
1246  *   1. All mem_map arrays are page-aligned.
1247  *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1248  *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1249  *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1250  *      worst combination is powerpc with 256k pages,
1251  *      which results in PFN_SECTION_SHIFT equal 6.
1252  * To sum it up, at least 6 bits are available.
1253  */
1254 #define	SECTION_MARKED_PRESENT	(1UL<<0)
1255 #define SECTION_HAS_MEM_MAP	(1UL<<1)
1256 #define SECTION_IS_ONLINE	(1UL<<2)
1257 #define SECTION_IS_EARLY	(1UL<<3)
1258 #define SECTION_MAP_LAST_BIT	(1UL<<4)
1259 #define SECTION_MAP_MASK	(~(SECTION_MAP_LAST_BIT-1))
1260 #define SECTION_NID_SHIFT	3
1261 
1262 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1263 {
1264 	unsigned long map = section->section_mem_map;
1265 	map &= SECTION_MAP_MASK;
1266 	return (struct page *)map;
1267 }
1268 
1269 static inline int present_section(struct mem_section *section)
1270 {
1271 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1272 }
1273 
1274 static inline int present_section_nr(unsigned long nr)
1275 {
1276 	return present_section(__nr_to_section(nr));
1277 }
1278 
1279 static inline int valid_section(struct mem_section *section)
1280 {
1281 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1282 }
1283 
1284 static inline int early_section(struct mem_section *section)
1285 {
1286 	return (section && (section->section_mem_map & SECTION_IS_EARLY));
1287 }
1288 
1289 static inline int valid_section_nr(unsigned long nr)
1290 {
1291 	return valid_section(__nr_to_section(nr));
1292 }
1293 
1294 static inline int online_section(struct mem_section *section)
1295 {
1296 	return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1297 }
1298 
1299 static inline int online_section_nr(unsigned long nr)
1300 {
1301 	return online_section(__nr_to_section(nr));
1302 }
1303 
1304 #ifdef CONFIG_MEMORY_HOTPLUG
1305 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1306 #ifdef CONFIG_MEMORY_HOTREMOVE
1307 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1308 #endif
1309 #endif
1310 
1311 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1312 {
1313 	return __nr_to_section(pfn_to_section_nr(pfn));
1314 }
1315 
1316 extern unsigned long __highest_present_section_nr;
1317 
1318 static inline int subsection_map_index(unsigned long pfn)
1319 {
1320 	return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1321 }
1322 
1323 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1324 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1325 {
1326 	int idx = subsection_map_index(pfn);
1327 
1328 	return test_bit(idx, ms->usage->subsection_map);
1329 }
1330 #else
1331 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1332 {
1333 	return 1;
1334 }
1335 #endif
1336 
1337 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1338 static inline int pfn_valid(unsigned long pfn)
1339 {
1340 	struct mem_section *ms;
1341 
1342 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1343 		return 0;
1344 	ms = __nr_to_section(pfn_to_section_nr(pfn));
1345 	if (!valid_section(ms))
1346 		return 0;
1347 	/*
1348 	 * Traditionally early sections always returned pfn_valid() for
1349 	 * the entire section-sized span.
1350 	 */
1351 	return early_section(ms) || pfn_section_valid(ms, pfn);
1352 }
1353 #endif
1354 
1355 static inline int pfn_present(unsigned long pfn)
1356 {
1357 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1358 		return 0;
1359 	return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1360 }
1361 
1362 /*
1363  * These are _only_ used during initialisation, therefore they
1364  * can use __initdata ...  They could have names to indicate
1365  * this restriction.
1366  */
1367 #ifdef CONFIG_NUMA
1368 #define pfn_to_nid(pfn)							\
1369 ({									\
1370 	unsigned long __pfn_to_nid_pfn = (pfn);				\
1371 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
1372 })
1373 #else
1374 #define pfn_to_nid(pfn)		(0)
1375 #endif
1376 
1377 #define early_pfn_valid(pfn)	pfn_valid(pfn)
1378 void sparse_init(void);
1379 #else
1380 #define sparse_init()	do {} while (0)
1381 #define sparse_index_init(_sec, _nid)  do {} while (0)
1382 #define pfn_present pfn_valid
1383 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1384 #endif /* CONFIG_SPARSEMEM */
1385 
1386 /*
1387  * During memory init memblocks map pfns to nids. The search is expensive and
1388  * this caches recent lookups. The implementation of __early_pfn_to_nid
1389  * may treat start/end as pfns or sections.
1390  */
1391 struct mminit_pfnnid_cache {
1392 	unsigned long last_start;
1393 	unsigned long last_end;
1394 	int last_nid;
1395 };
1396 
1397 #ifndef early_pfn_valid
1398 #define early_pfn_valid(pfn)	(1)
1399 #endif
1400 
1401 void memory_present(int nid, unsigned long start, unsigned long end);
1402 
1403 /*
1404  * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1405  * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1406  * pfn_valid_within() should be used in this case; we optimise this away
1407  * when we have no holes within a MAX_ORDER_NR_PAGES block.
1408  */
1409 #ifdef CONFIG_HOLES_IN_ZONE
1410 #define pfn_valid_within(pfn) pfn_valid(pfn)
1411 #else
1412 #define pfn_valid_within(pfn) (1)
1413 #endif
1414 
1415 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1416 /*
1417  * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1418  * associated with it or not. This means that a struct page exists for this
1419  * pfn. The caller cannot assume the page is fully initialized in general.
1420  * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1421  * will ensure the struct page is fully online and initialized. Special pages
1422  * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1423  *
1424  * In FLATMEM, it is expected that holes always have valid memmap as long as
1425  * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1426  * that a valid section has a memmap for the entire section.
1427  *
1428  * However, an ARM, and maybe other embedded architectures in the future
1429  * free memmap backing holes to save memory on the assumption the memmap is
1430  * never used. The page_zone linkages are then broken even though pfn_valid()
1431  * returns true. A walker of the full memmap must then do this additional
1432  * check to ensure the memmap they are looking at is sane by making sure
1433  * the zone and PFN linkages are still valid. This is expensive, but walkers
1434  * of the full memmap are extremely rare.
1435  */
1436 bool memmap_valid_within(unsigned long pfn,
1437 					struct page *page, struct zone *zone);
1438 #else
1439 static inline bool memmap_valid_within(unsigned long pfn,
1440 					struct page *page, struct zone *zone)
1441 {
1442 	return true;
1443 }
1444 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1445 
1446 #endif /* !__GENERATING_BOUNDS.H */
1447 #endif /* !__ASSEMBLY__ */
1448 #endif /* _LINUX_MMZONE_H */
1449