xref: /linux-6.15/include/linux/mmzone.h (revision 79b6bb73)
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_FILE_THPS,
239 	NR_FILE_PMDMAPPED,
240 	NR_ANON_THPS,
241 	NR_UNSTABLE_NFS,	/* NFS unstable pages */
242 	NR_VMSCAN_WRITE,
243 	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
244 	NR_DIRTIED,		/* page dirtyings since bootup */
245 	NR_WRITTEN,		/* page writings since bootup */
246 	NR_KERNEL_MISC_RECLAIMABLE,	/* reclaimable non-slab kernel pages */
247 	NR_VM_NODE_STAT_ITEMS
248 };
249 
250 /*
251  * We do arithmetic on the LRU lists in various places in the code,
252  * so it is important to keep the active lists LRU_ACTIVE higher in
253  * the array than the corresponding inactive lists, and to keep
254  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
255  *
256  * This has to be kept in sync with the statistics in zone_stat_item
257  * above and the descriptions in vmstat_text in mm/vmstat.c
258  */
259 #define LRU_BASE 0
260 #define LRU_ACTIVE 1
261 #define LRU_FILE 2
262 
263 enum lru_list {
264 	LRU_INACTIVE_ANON = LRU_BASE,
265 	LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
266 	LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
267 	LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
268 	LRU_UNEVICTABLE,
269 	NR_LRU_LISTS
270 };
271 
272 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
273 
274 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
275 
276 static inline int is_file_lru(enum lru_list lru)
277 {
278 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
279 }
280 
281 static inline int is_active_lru(enum lru_list lru)
282 {
283 	return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
284 }
285 
286 struct zone_reclaim_stat {
287 	/*
288 	 * The pageout code in vmscan.c keeps track of how many of the
289 	 * mem/swap backed and file backed pages are referenced.
290 	 * The higher the rotated/scanned ratio, the more valuable
291 	 * that cache is.
292 	 *
293 	 * The anon LRU stats live in [0], file LRU stats in [1]
294 	 */
295 	unsigned long		recent_rotated[2];
296 	unsigned long		recent_scanned[2];
297 };
298 
299 struct lruvec {
300 	struct list_head		lists[NR_LRU_LISTS];
301 	struct zone_reclaim_stat	reclaim_stat;
302 	/* Evictions & activations on the inactive file list */
303 	atomic_long_t			inactive_age;
304 	/* Refaults at the time of last reclaim cycle */
305 	unsigned long			refaults;
306 #ifdef CONFIG_MEMCG
307 	struct pglist_data *pgdat;
308 #endif
309 };
310 
311 /* Isolate unmapped file */
312 #define ISOLATE_UNMAPPED	((__force isolate_mode_t)0x2)
313 /* Isolate for asynchronous migration */
314 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
315 /* Isolate unevictable pages */
316 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
317 
318 /* LRU Isolation modes. */
319 typedef unsigned __bitwise isolate_mode_t;
320 
321 enum zone_watermarks {
322 	WMARK_MIN,
323 	WMARK_LOW,
324 	WMARK_HIGH,
325 	NR_WMARK
326 };
327 
328 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
329 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
330 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
331 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
332 
333 struct per_cpu_pages {
334 	int count;		/* number of pages in the list */
335 	int high;		/* high watermark, emptying needed */
336 	int batch;		/* chunk size for buddy add/remove */
337 
338 	/* Lists of pages, one per migrate type stored on the pcp-lists */
339 	struct list_head lists[MIGRATE_PCPTYPES];
340 };
341 
342 struct per_cpu_pageset {
343 	struct per_cpu_pages pcp;
344 #ifdef CONFIG_NUMA
345 	s8 expire;
346 	u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
347 #endif
348 #ifdef CONFIG_SMP
349 	s8 stat_threshold;
350 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
351 #endif
352 };
353 
354 struct per_cpu_nodestat {
355 	s8 stat_threshold;
356 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
357 };
358 
359 #endif /* !__GENERATING_BOUNDS.H */
360 
361 enum zone_type {
362 	/*
363 	 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
364 	 * to DMA to all of the addressable memory (ZONE_NORMAL).
365 	 * On architectures where this area covers the whole 32 bit address
366 	 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
367 	 * DMA addressing constraints. This distinction is important as a 32bit
368 	 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
369 	 * platforms may need both zones as they support peripherals with
370 	 * different DMA addressing limitations.
371 	 *
372 	 * Some examples:
373 	 *
374 	 *  - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
375 	 *    rest of the lower 4G.
376 	 *
377 	 *  - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
378 	 *    the specific device.
379 	 *
380 	 *  - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
381 	 *    lower 4G.
382 	 *
383 	 *  - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
384 	 *    depending on the specific device.
385 	 *
386 	 *  - s390 uses ZONE_DMA fixed to the lower 2G.
387 	 *
388 	 *  - ia64 and riscv only use ZONE_DMA32.
389 	 *
390 	 *  - parisc uses neither.
391 	 */
392 #ifdef CONFIG_ZONE_DMA
393 	ZONE_DMA,
394 #endif
395 #ifdef CONFIG_ZONE_DMA32
396 	ZONE_DMA32,
397 #endif
398 	/*
399 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
400 	 * performed on pages in ZONE_NORMAL if the DMA devices support
401 	 * transfers to all addressable memory.
402 	 */
403 	ZONE_NORMAL,
404 #ifdef CONFIG_HIGHMEM
405 	/*
406 	 * A memory area that is only addressable by the kernel through
407 	 * mapping portions into its own address space. This is for example
408 	 * used by i386 to allow the kernel to address the memory beyond
409 	 * 900MB. The kernel will set up special mappings (page
410 	 * table entries on i386) for each page that the kernel needs to
411 	 * access.
412 	 */
413 	ZONE_HIGHMEM,
414 #endif
415 	ZONE_MOVABLE,
416 #ifdef CONFIG_ZONE_DEVICE
417 	ZONE_DEVICE,
418 #endif
419 	__MAX_NR_ZONES
420 
421 };
422 
423 #ifndef __GENERATING_BOUNDS_H
424 
425 struct zone {
426 	/* Read-mostly fields */
427 
428 	/* zone watermarks, access with *_wmark_pages(zone) macros */
429 	unsigned long _watermark[NR_WMARK];
430 	unsigned long watermark_boost;
431 
432 	unsigned long nr_reserved_highatomic;
433 
434 	/*
435 	 * We don't know if the memory that we're going to allocate will be
436 	 * freeable or/and it will be released eventually, so to avoid totally
437 	 * wasting several GB of ram we must reserve some of the lower zone
438 	 * memory (otherwise we risk to run OOM on the lower zones despite
439 	 * there being tons of freeable ram on the higher zones).  This array is
440 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
441 	 * changes.
442 	 */
443 	long lowmem_reserve[MAX_NR_ZONES];
444 
445 #ifdef CONFIG_NUMA
446 	int node;
447 #endif
448 	struct pglist_data	*zone_pgdat;
449 	struct per_cpu_pageset __percpu *pageset;
450 
451 #ifndef CONFIG_SPARSEMEM
452 	/*
453 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
454 	 * In SPARSEMEM, this map is stored in struct mem_section
455 	 */
456 	unsigned long		*pageblock_flags;
457 #endif /* CONFIG_SPARSEMEM */
458 
459 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
460 	unsigned long		zone_start_pfn;
461 
462 	/*
463 	 * spanned_pages is the total pages spanned by the zone, including
464 	 * holes, which is calculated as:
465 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
466 	 *
467 	 * present_pages is physical pages existing within the zone, which
468 	 * is calculated as:
469 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
470 	 *
471 	 * managed_pages is present pages managed by the buddy system, which
472 	 * is calculated as (reserved_pages includes pages allocated by the
473 	 * bootmem allocator):
474 	 *	managed_pages = present_pages - reserved_pages;
475 	 *
476 	 * So present_pages may be used by memory hotplug or memory power
477 	 * management logic to figure out unmanaged pages by checking
478 	 * (present_pages - managed_pages). And managed_pages should be used
479 	 * by page allocator and vm scanner to calculate all kinds of watermarks
480 	 * and thresholds.
481 	 *
482 	 * Locking rules:
483 	 *
484 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
485 	 * It is a seqlock because it has to be read outside of zone->lock,
486 	 * and it is done in the main allocator path.  But, it is written
487 	 * quite infrequently.
488 	 *
489 	 * The span_seq lock is declared along with zone->lock because it is
490 	 * frequently read in proximity to zone->lock.  It's good to
491 	 * give them a chance of being in the same cacheline.
492 	 *
493 	 * Write access to present_pages at runtime should be protected by
494 	 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
495 	 * present_pages should get_online_mems() to get a stable value.
496 	 */
497 	atomic_long_t		managed_pages;
498 	unsigned long		spanned_pages;
499 	unsigned long		present_pages;
500 
501 	const char		*name;
502 
503 #ifdef CONFIG_MEMORY_ISOLATION
504 	/*
505 	 * Number of isolated pageblock. It is used to solve incorrect
506 	 * freepage counting problem due to racy retrieving migratetype
507 	 * of pageblock. Protected by zone->lock.
508 	 */
509 	unsigned long		nr_isolate_pageblock;
510 #endif
511 
512 #ifdef CONFIG_MEMORY_HOTPLUG
513 	/* see spanned/present_pages for more description */
514 	seqlock_t		span_seqlock;
515 #endif
516 
517 	int initialized;
518 
519 	/* Write-intensive fields used from the page allocator */
520 	ZONE_PADDING(_pad1_)
521 
522 	/* free areas of different sizes */
523 	struct free_area	free_area[MAX_ORDER];
524 
525 	/* zone flags, see below */
526 	unsigned long		flags;
527 
528 	/* Primarily protects free_area */
529 	spinlock_t		lock;
530 
531 	/* Write-intensive fields used by compaction and vmstats. */
532 	ZONE_PADDING(_pad2_)
533 
534 	/*
535 	 * When free pages are below this point, additional steps are taken
536 	 * when reading the number of free pages to avoid per-cpu counter
537 	 * drift allowing watermarks to be breached
538 	 */
539 	unsigned long percpu_drift_mark;
540 
541 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
542 	/* pfn where compaction free scanner should start */
543 	unsigned long		compact_cached_free_pfn;
544 	/* pfn where async and sync compaction migration scanner should start */
545 	unsigned long		compact_cached_migrate_pfn[2];
546 	unsigned long		compact_init_migrate_pfn;
547 	unsigned long		compact_init_free_pfn;
548 #endif
549 
550 #ifdef CONFIG_COMPACTION
551 	/*
552 	 * On compaction failure, 1<<compact_defer_shift compactions
553 	 * are skipped before trying again. The number attempted since
554 	 * last failure is tracked with compact_considered.
555 	 */
556 	unsigned int		compact_considered;
557 	unsigned int		compact_defer_shift;
558 	int			compact_order_failed;
559 #endif
560 
561 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
562 	/* Set to true when the PG_migrate_skip bits should be cleared */
563 	bool			compact_blockskip_flush;
564 #endif
565 
566 	bool			contiguous;
567 
568 	ZONE_PADDING(_pad3_)
569 	/* Zone statistics */
570 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
571 	atomic_long_t		vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
572 } ____cacheline_internodealigned_in_smp;
573 
574 enum pgdat_flags {
575 	PGDAT_CONGESTED,		/* pgdat has many dirty pages backed by
576 					 * a congested BDI
577 					 */
578 	PGDAT_DIRTY,			/* reclaim scanning has recently found
579 					 * many dirty file pages at the tail
580 					 * of the LRU.
581 					 */
582 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
583 					 * many pages under writeback
584 					 */
585 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
586 };
587 
588 enum zone_flags {
589 	ZONE_BOOSTED_WATERMARK,		/* zone recently boosted watermarks.
590 					 * Cleared when kswapd is woken.
591 					 */
592 };
593 
594 static inline unsigned long zone_managed_pages(struct zone *zone)
595 {
596 	return (unsigned long)atomic_long_read(&zone->managed_pages);
597 }
598 
599 static inline unsigned long zone_end_pfn(const struct zone *zone)
600 {
601 	return zone->zone_start_pfn + zone->spanned_pages;
602 }
603 
604 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
605 {
606 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
607 }
608 
609 static inline bool zone_is_initialized(struct zone *zone)
610 {
611 	return zone->initialized;
612 }
613 
614 static inline bool zone_is_empty(struct zone *zone)
615 {
616 	return zone->spanned_pages == 0;
617 }
618 
619 /*
620  * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
621  * intersection with the given zone
622  */
623 static inline bool zone_intersects(struct zone *zone,
624 		unsigned long start_pfn, unsigned long nr_pages)
625 {
626 	if (zone_is_empty(zone))
627 		return false;
628 	if (start_pfn >= zone_end_pfn(zone) ||
629 	    start_pfn + nr_pages <= zone->zone_start_pfn)
630 		return false;
631 
632 	return true;
633 }
634 
635 /*
636  * The "priority" of VM scanning is how much of the queues we will scan in one
637  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
638  * queues ("queue_length >> 12") during an aging round.
639  */
640 #define DEF_PRIORITY 12
641 
642 /* Maximum number of zones on a zonelist */
643 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
644 
645 enum {
646 	ZONELIST_FALLBACK,	/* zonelist with fallback */
647 #ifdef CONFIG_NUMA
648 	/*
649 	 * The NUMA zonelists are doubled because we need zonelists that
650 	 * restrict the allocations to a single node for __GFP_THISNODE.
651 	 */
652 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
653 #endif
654 	MAX_ZONELISTS
655 };
656 
657 /*
658  * This struct contains information about a zone in a zonelist. It is stored
659  * here to avoid dereferences into large structures and lookups of tables
660  */
661 struct zoneref {
662 	struct zone *zone;	/* Pointer to actual zone */
663 	int zone_idx;		/* zone_idx(zoneref->zone) */
664 };
665 
666 /*
667  * One allocation request operates on a zonelist. A zonelist
668  * is a list of zones, the first one is the 'goal' of the
669  * allocation, the other zones are fallback zones, in decreasing
670  * priority.
671  *
672  * To speed the reading of the zonelist, the zonerefs contain the zone index
673  * of the entry being read. Helper functions to access information given
674  * a struct zoneref are
675  *
676  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
677  * zonelist_zone_idx()	- Return the index of the zone for an entry
678  * zonelist_node_idx()	- Return the index of the node for an entry
679  */
680 struct zonelist {
681 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
682 };
683 
684 #ifndef CONFIG_DISCONTIGMEM
685 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
686 extern struct page *mem_map;
687 #endif
688 
689 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
690 struct deferred_split {
691 	spinlock_t split_queue_lock;
692 	struct list_head split_queue;
693 	unsigned long split_queue_len;
694 };
695 #endif
696 
697 /*
698  * On NUMA machines, each NUMA node would have a pg_data_t to describe
699  * it's memory layout. On UMA machines there is a single pglist_data which
700  * describes the whole memory.
701  *
702  * Memory statistics and page replacement data structures are maintained on a
703  * per-zone basis.
704  */
705 struct bootmem_data;
706 typedef struct pglist_data {
707 	struct zone node_zones[MAX_NR_ZONES];
708 	struct zonelist node_zonelists[MAX_ZONELISTS];
709 	int nr_zones;
710 #ifdef CONFIG_FLAT_NODE_MEM_MAP	/* means !SPARSEMEM */
711 	struct page *node_mem_map;
712 #ifdef CONFIG_PAGE_EXTENSION
713 	struct page_ext *node_page_ext;
714 #endif
715 #endif
716 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
717 	/*
718 	 * Must be held any time you expect node_start_pfn,
719 	 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
720 	 *
721 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
722 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
723 	 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
724 	 *
725 	 * Nests above zone->lock and zone->span_seqlock
726 	 */
727 	spinlock_t node_size_lock;
728 #endif
729 	unsigned long node_start_pfn;
730 	unsigned long node_present_pages; /* total number of physical pages */
731 	unsigned long node_spanned_pages; /* total size of physical page
732 					     range, including holes */
733 	int node_id;
734 	wait_queue_head_t kswapd_wait;
735 	wait_queue_head_t pfmemalloc_wait;
736 	struct task_struct *kswapd;	/* Protected by
737 					   mem_hotplug_begin/end() */
738 	int kswapd_order;
739 	enum zone_type kswapd_classzone_idx;
740 
741 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
742 
743 #ifdef CONFIG_COMPACTION
744 	int kcompactd_max_order;
745 	enum zone_type kcompactd_classzone_idx;
746 	wait_queue_head_t kcompactd_wait;
747 	struct task_struct *kcompactd;
748 #endif
749 	/*
750 	 * This is a per-node reserve of pages that are not available
751 	 * to userspace allocations.
752 	 */
753 	unsigned long		totalreserve_pages;
754 
755 #ifdef CONFIG_NUMA
756 	/*
757 	 * zone reclaim becomes active if more unmapped pages exist.
758 	 */
759 	unsigned long		min_unmapped_pages;
760 	unsigned long		min_slab_pages;
761 #endif /* CONFIG_NUMA */
762 
763 	/* Write-intensive fields used by page reclaim */
764 	ZONE_PADDING(_pad1_)
765 	spinlock_t		lru_lock;
766 
767 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
768 	/*
769 	 * If memory initialisation on large machines is deferred then this
770 	 * is the first PFN that needs to be initialised.
771 	 */
772 	unsigned long first_deferred_pfn;
773 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
774 
775 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
776 	struct deferred_split deferred_split_queue;
777 #endif
778 
779 	/* Fields commonly accessed by the page reclaim scanner */
780 	struct lruvec		lruvec;
781 
782 	unsigned long		flags;
783 
784 	ZONE_PADDING(_pad2_)
785 
786 	/* Per-node vmstats */
787 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
788 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
789 } pg_data_t;
790 
791 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
792 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
793 #ifdef CONFIG_FLAT_NODE_MEM_MAP
794 #define pgdat_page_nr(pgdat, pagenr)	((pgdat)->node_mem_map + (pagenr))
795 #else
796 #define pgdat_page_nr(pgdat, pagenr)	pfn_to_page((pgdat)->node_start_pfn + (pagenr))
797 #endif
798 #define nid_page_nr(nid, pagenr) 	pgdat_page_nr(NODE_DATA(nid),(pagenr))
799 
800 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
801 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
802 
803 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
804 {
805 	return &pgdat->lruvec;
806 }
807 
808 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
809 {
810 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
811 }
812 
813 static inline bool pgdat_is_empty(pg_data_t *pgdat)
814 {
815 	return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
816 }
817 
818 #include <linux/memory_hotplug.h>
819 
820 void build_all_zonelists(pg_data_t *pgdat);
821 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
822 		   enum zone_type classzone_idx);
823 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
824 			 int classzone_idx, unsigned int alloc_flags,
825 			 long free_pages);
826 bool zone_watermark_ok(struct zone *z, unsigned int order,
827 		unsigned long mark, int classzone_idx,
828 		unsigned int alloc_flags);
829 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
830 		unsigned long mark, int classzone_idx);
831 enum memmap_context {
832 	MEMMAP_EARLY,
833 	MEMMAP_HOTPLUG,
834 };
835 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
836 				     unsigned long size);
837 
838 extern void lruvec_init(struct lruvec *lruvec);
839 
840 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
841 {
842 #ifdef CONFIG_MEMCG
843 	return lruvec->pgdat;
844 #else
845 	return container_of(lruvec, struct pglist_data, lruvec);
846 #endif
847 }
848 
849 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
850 
851 #ifdef CONFIG_HAVE_MEMORY_PRESENT
852 void memory_present(int nid, unsigned long start, unsigned long end);
853 #else
854 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
855 #endif
856 
857 #if defined(CONFIG_SPARSEMEM)
858 void memblocks_present(void);
859 #else
860 static inline void memblocks_present(void) {}
861 #endif
862 
863 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
864 int local_memory_node(int node_id);
865 #else
866 static inline int local_memory_node(int node_id) { return node_id; };
867 #endif
868 
869 /*
870  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
871  */
872 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
873 
874 /*
875  * Returns true if a zone has pages managed by the buddy allocator.
876  * All the reclaim decisions have to use this function rather than
877  * populated_zone(). If the whole zone is reserved then we can easily
878  * end up with populated_zone() && !managed_zone().
879  */
880 static inline bool managed_zone(struct zone *zone)
881 {
882 	return zone_managed_pages(zone);
883 }
884 
885 /* Returns true if a zone has memory */
886 static inline bool populated_zone(struct zone *zone)
887 {
888 	return zone->present_pages;
889 }
890 
891 #ifdef CONFIG_NUMA
892 static inline int zone_to_nid(struct zone *zone)
893 {
894 	return zone->node;
895 }
896 
897 static inline void zone_set_nid(struct zone *zone, int nid)
898 {
899 	zone->node = nid;
900 }
901 #else
902 static inline int zone_to_nid(struct zone *zone)
903 {
904 	return 0;
905 }
906 
907 static inline void zone_set_nid(struct zone *zone, int nid) {}
908 #endif
909 
910 extern int movable_zone;
911 
912 #ifdef CONFIG_HIGHMEM
913 static inline int zone_movable_is_highmem(void)
914 {
915 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
916 	return movable_zone == ZONE_HIGHMEM;
917 #else
918 	return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
919 #endif
920 }
921 #endif
922 
923 static inline int is_highmem_idx(enum zone_type idx)
924 {
925 #ifdef CONFIG_HIGHMEM
926 	return (idx == ZONE_HIGHMEM ||
927 		(idx == ZONE_MOVABLE && zone_movable_is_highmem()));
928 #else
929 	return 0;
930 #endif
931 }
932 
933 /**
934  * is_highmem - helper function to quickly check if a struct zone is a
935  *              highmem zone or not.  This is an attempt to keep references
936  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
937  * @zone - pointer to struct zone variable
938  */
939 static inline int is_highmem(struct zone *zone)
940 {
941 #ifdef CONFIG_HIGHMEM
942 	return is_highmem_idx(zone_idx(zone));
943 #else
944 	return 0;
945 #endif
946 }
947 
948 /* These two functions are used to setup the per zone pages min values */
949 struct ctl_table;
950 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
951 					void __user *, size_t *, loff_t *);
952 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
953 					void __user *, size_t *, loff_t *);
954 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
955 					void __user *, size_t *, loff_t *);
956 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
957 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
958 					void __user *, size_t *, loff_t *);
959 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
960 					void __user *, size_t *, loff_t *);
961 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
962 			void __user *, size_t *, loff_t *);
963 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
964 			void __user *, size_t *, loff_t *);
965 
966 extern int numa_zonelist_order_handler(struct ctl_table *, int,
967 			void __user *, size_t *, loff_t *);
968 extern char numa_zonelist_order[];
969 #define NUMA_ZONELIST_ORDER_LEN	16
970 
971 #ifndef CONFIG_NEED_MULTIPLE_NODES
972 
973 extern struct pglist_data contig_page_data;
974 #define NODE_DATA(nid)		(&contig_page_data)
975 #define NODE_MEM_MAP(nid)	mem_map
976 
977 #else /* CONFIG_NEED_MULTIPLE_NODES */
978 
979 #include <asm/mmzone.h>
980 
981 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
982 
983 extern struct pglist_data *first_online_pgdat(void);
984 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
985 extern struct zone *next_zone(struct zone *zone);
986 
987 /**
988  * for_each_online_pgdat - helper macro to iterate over all online nodes
989  * @pgdat - pointer to a pg_data_t variable
990  */
991 #define for_each_online_pgdat(pgdat)			\
992 	for (pgdat = first_online_pgdat();		\
993 	     pgdat;					\
994 	     pgdat = next_online_pgdat(pgdat))
995 /**
996  * for_each_zone - helper macro to iterate over all memory zones
997  * @zone - pointer to struct zone variable
998  *
999  * The user only needs to declare the zone variable, for_each_zone
1000  * fills it in.
1001  */
1002 #define for_each_zone(zone)			        \
1003 	for (zone = (first_online_pgdat())->node_zones; \
1004 	     zone;					\
1005 	     zone = next_zone(zone))
1006 
1007 #define for_each_populated_zone(zone)		        \
1008 	for (zone = (first_online_pgdat())->node_zones; \
1009 	     zone;					\
1010 	     zone = next_zone(zone))			\
1011 		if (!populated_zone(zone))		\
1012 			; /* do nothing */		\
1013 		else
1014 
1015 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1016 {
1017 	return zoneref->zone;
1018 }
1019 
1020 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1021 {
1022 	return zoneref->zone_idx;
1023 }
1024 
1025 static inline int zonelist_node_idx(struct zoneref *zoneref)
1026 {
1027 	return zone_to_nid(zoneref->zone);
1028 }
1029 
1030 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1031 					enum zone_type highest_zoneidx,
1032 					nodemask_t *nodes);
1033 
1034 /**
1035  * 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
1036  * @z - The cursor used as a starting point for the search
1037  * @highest_zoneidx - The zone index of the highest zone to return
1038  * @nodes - An optional nodemask to filter the zonelist with
1039  *
1040  * This function returns the next zone at or below a given zone index that is
1041  * within the allowed nodemask using a cursor as the starting point for the
1042  * search. The zoneref returned is a cursor that represents the current zone
1043  * being examined. It should be advanced by one before calling
1044  * next_zones_zonelist again.
1045  */
1046 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1047 					enum zone_type highest_zoneidx,
1048 					nodemask_t *nodes)
1049 {
1050 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1051 		return z;
1052 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
1053 }
1054 
1055 /**
1056  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1057  * @zonelist - The zonelist to search for a suitable zone
1058  * @highest_zoneidx - The zone index of the highest zone to return
1059  * @nodes - An optional nodemask to filter the zonelist with
1060  * @return - Zoneref pointer for the first suitable zone found (see below)
1061  *
1062  * This function returns the first zone at or below a given zone index that is
1063  * within the allowed nodemask. The zoneref returned is a cursor that can be
1064  * used to iterate the zonelist with next_zones_zonelist by advancing it by
1065  * one before calling.
1066  *
1067  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1068  * never NULL). This may happen either genuinely, or due to concurrent nodemask
1069  * update due to cpuset modification.
1070  */
1071 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1072 					enum zone_type highest_zoneidx,
1073 					nodemask_t *nodes)
1074 {
1075 	return next_zones_zonelist(zonelist->_zonerefs,
1076 							highest_zoneidx, nodes);
1077 }
1078 
1079 /**
1080  * 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
1081  * @zone - The current zone in the iterator
1082  * @z - The current pointer within zonelist->zones being iterated
1083  * @zlist - The zonelist being iterated
1084  * @highidx - The zone index of the highest zone to return
1085  * @nodemask - Nodemask allowed by the allocator
1086  *
1087  * This iterator iterates though all zones at or below a given zone index and
1088  * within a given nodemask
1089  */
1090 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1091 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1092 		zone;							\
1093 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1094 			zone = zonelist_zone(z))
1095 
1096 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1097 	for (zone = z->zone;	\
1098 		zone;							\
1099 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1100 			zone = zonelist_zone(z))
1101 
1102 
1103 /**
1104  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1105  * @zone - The current zone in the iterator
1106  * @z - The current pointer within zonelist->zones being iterated
1107  * @zlist - The zonelist being iterated
1108  * @highidx - The zone index of the highest zone to return
1109  *
1110  * This iterator iterates though all zones at or below a given zone index.
1111  */
1112 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1113 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1114 
1115 #ifdef CONFIG_SPARSEMEM
1116 #include <asm/sparsemem.h>
1117 #endif
1118 
1119 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1120 	!defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1121 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1122 {
1123 	BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1124 	return 0;
1125 }
1126 #endif
1127 
1128 #ifdef CONFIG_FLATMEM
1129 #define pfn_to_nid(pfn)		(0)
1130 #endif
1131 
1132 #ifdef CONFIG_SPARSEMEM
1133 
1134 /*
1135  * SECTION_SHIFT    		#bits space required to store a section #
1136  *
1137  * PA_SECTION_SHIFT		physical address to/from section number
1138  * PFN_SECTION_SHIFT		pfn to/from section number
1139  */
1140 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1141 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1142 
1143 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1144 
1145 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1146 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1147 
1148 #define SECTION_BLOCKFLAGS_BITS \
1149 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1150 
1151 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1152 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1153 #endif
1154 
1155 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1156 {
1157 	return pfn >> PFN_SECTION_SHIFT;
1158 }
1159 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1160 {
1161 	return sec << PFN_SECTION_SHIFT;
1162 }
1163 
1164 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1165 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1166 
1167 #define SUBSECTION_SHIFT 21
1168 
1169 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1170 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1171 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1172 
1173 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1174 #error Subsection size exceeds section size
1175 #else
1176 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1177 #endif
1178 
1179 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1180 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1181 
1182 struct mem_section_usage {
1183 	DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1184 	/* See declaration of similar field in struct zone */
1185 	unsigned long pageblock_flags[0];
1186 };
1187 
1188 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1189 
1190 struct page;
1191 struct page_ext;
1192 struct mem_section {
1193 	/*
1194 	 * This is, logically, a pointer to an array of struct
1195 	 * pages.  However, it is stored with some other magic.
1196 	 * (see sparse.c::sparse_init_one_section())
1197 	 *
1198 	 * Additionally during early boot we encode node id of
1199 	 * the location of the section here to guide allocation.
1200 	 * (see sparse.c::memory_present())
1201 	 *
1202 	 * Making it a UL at least makes someone do a cast
1203 	 * before using it wrong.
1204 	 */
1205 	unsigned long section_mem_map;
1206 
1207 	struct mem_section_usage *usage;
1208 #ifdef CONFIG_PAGE_EXTENSION
1209 	/*
1210 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1211 	 * section. (see page_ext.h about this.)
1212 	 */
1213 	struct page_ext *page_ext;
1214 	unsigned long pad;
1215 #endif
1216 	/*
1217 	 * WARNING: mem_section must be a power-of-2 in size for the
1218 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1219 	 */
1220 };
1221 
1222 #ifdef CONFIG_SPARSEMEM_EXTREME
1223 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1224 #else
1225 #define SECTIONS_PER_ROOT	1
1226 #endif
1227 
1228 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1229 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1230 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1231 
1232 #ifdef CONFIG_SPARSEMEM_EXTREME
1233 extern struct mem_section **mem_section;
1234 #else
1235 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1236 #endif
1237 
1238 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1239 {
1240 	return ms->usage->pageblock_flags;
1241 }
1242 
1243 static inline struct mem_section *__nr_to_section(unsigned long nr)
1244 {
1245 #ifdef CONFIG_SPARSEMEM_EXTREME
1246 	if (!mem_section)
1247 		return NULL;
1248 #endif
1249 	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1250 		return NULL;
1251 	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1252 }
1253 extern unsigned long __section_nr(struct mem_section *ms);
1254 extern size_t mem_section_usage_size(void);
1255 
1256 /*
1257  * We use the lower bits of the mem_map pointer to store
1258  * a little bit of information.  The pointer is calculated
1259  * as mem_map - section_nr_to_pfn(pnum).  The result is
1260  * aligned to the minimum alignment of the two values:
1261  *   1. All mem_map arrays are page-aligned.
1262  *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1263  *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1264  *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1265  *      worst combination is powerpc with 256k pages,
1266  *      which results in PFN_SECTION_SHIFT equal 6.
1267  * To sum it up, at least 6 bits are available.
1268  */
1269 #define	SECTION_MARKED_PRESENT	(1UL<<0)
1270 #define SECTION_HAS_MEM_MAP	(1UL<<1)
1271 #define SECTION_IS_ONLINE	(1UL<<2)
1272 #define SECTION_IS_EARLY	(1UL<<3)
1273 #define SECTION_MAP_LAST_BIT	(1UL<<4)
1274 #define SECTION_MAP_MASK	(~(SECTION_MAP_LAST_BIT-1))
1275 #define SECTION_NID_SHIFT	3
1276 
1277 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1278 {
1279 	unsigned long map = section->section_mem_map;
1280 	map &= SECTION_MAP_MASK;
1281 	return (struct page *)map;
1282 }
1283 
1284 static inline int present_section(struct mem_section *section)
1285 {
1286 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1287 }
1288 
1289 static inline int present_section_nr(unsigned long nr)
1290 {
1291 	return present_section(__nr_to_section(nr));
1292 }
1293 
1294 static inline int valid_section(struct mem_section *section)
1295 {
1296 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1297 }
1298 
1299 static inline int early_section(struct mem_section *section)
1300 {
1301 	return (section && (section->section_mem_map & SECTION_IS_EARLY));
1302 }
1303 
1304 static inline int valid_section_nr(unsigned long nr)
1305 {
1306 	return valid_section(__nr_to_section(nr));
1307 }
1308 
1309 static inline int online_section(struct mem_section *section)
1310 {
1311 	return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1312 }
1313 
1314 static inline int online_section_nr(unsigned long nr)
1315 {
1316 	return online_section(__nr_to_section(nr));
1317 }
1318 
1319 #ifdef CONFIG_MEMORY_HOTPLUG
1320 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1321 #ifdef CONFIG_MEMORY_HOTREMOVE
1322 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1323 #endif
1324 #endif
1325 
1326 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1327 {
1328 	return __nr_to_section(pfn_to_section_nr(pfn));
1329 }
1330 
1331 extern unsigned long __highest_present_section_nr;
1332 
1333 static inline int subsection_map_index(unsigned long pfn)
1334 {
1335 	return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1336 }
1337 
1338 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1339 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1340 {
1341 	int idx = subsection_map_index(pfn);
1342 
1343 	return test_bit(idx, ms->usage->subsection_map);
1344 }
1345 #else
1346 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1347 {
1348 	return 1;
1349 }
1350 #endif
1351 
1352 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1353 static inline int pfn_valid(unsigned long pfn)
1354 {
1355 	struct mem_section *ms;
1356 
1357 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1358 		return 0;
1359 	ms = __nr_to_section(pfn_to_section_nr(pfn));
1360 	if (!valid_section(ms))
1361 		return 0;
1362 	/*
1363 	 * Traditionally early sections always returned pfn_valid() for
1364 	 * the entire section-sized span.
1365 	 */
1366 	return early_section(ms) || pfn_section_valid(ms, pfn);
1367 }
1368 #endif
1369 
1370 static inline int pfn_present(unsigned long pfn)
1371 {
1372 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1373 		return 0;
1374 	return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1375 }
1376 
1377 /*
1378  * These are _only_ used during initialisation, therefore they
1379  * can use __initdata ...  They could have names to indicate
1380  * this restriction.
1381  */
1382 #ifdef CONFIG_NUMA
1383 #define pfn_to_nid(pfn)							\
1384 ({									\
1385 	unsigned long __pfn_to_nid_pfn = (pfn);				\
1386 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
1387 })
1388 #else
1389 #define pfn_to_nid(pfn)		(0)
1390 #endif
1391 
1392 #define early_pfn_valid(pfn)	pfn_valid(pfn)
1393 void sparse_init(void);
1394 #else
1395 #define sparse_init()	do {} while (0)
1396 #define sparse_index_init(_sec, _nid)  do {} while (0)
1397 #define pfn_present pfn_valid
1398 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1399 #endif /* CONFIG_SPARSEMEM */
1400 
1401 /*
1402  * During memory init memblocks map pfns to nids. The search is expensive and
1403  * this caches recent lookups. The implementation of __early_pfn_to_nid
1404  * may treat start/end as pfns or sections.
1405  */
1406 struct mminit_pfnnid_cache {
1407 	unsigned long last_start;
1408 	unsigned long last_end;
1409 	int last_nid;
1410 };
1411 
1412 #ifndef early_pfn_valid
1413 #define early_pfn_valid(pfn)	(1)
1414 #endif
1415 
1416 void memory_present(int nid, unsigned long start, unsigned long end);
1417 
1418 /*
1419  * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1420  * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1421  * pfn_valid_within() should be used in this case; we optimise this away
1422  * when we have no holes within a MAX_ORDER_NR_PAGES block.
1423  */
1424 #ifdef CONFIG_HOLES_IN_ZONE
1425 #define pfn_valid_within(pfn) pfn_valid(pfn)
1426 #else
1427 #define pfn_valid_within(pfn) (1)
1428 #endif
1429 
1430 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1431 /*
1432  * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1433  * associated with it or not. This means that a struct page exists for this
1434  * pfn. The caller cannot assume the page is fully initialized in general.
1435  * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1436  * will ensure the struct page is fully online and initialized. Special pages
1437  * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1438  *
1439  * In FLATMEM, it is expected that holes always have valid memmap as long as
1440  * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1441  * that a valid section has a memmap for the entire section.
1442  *
1443  * However, an ARM, and maybe other embedded architectures in the future
1444  * free memmap backing holes to save memory on the assumption the memmap is
1445  * never used. The page_zone linkages are then broken even though pfn_valid()
1446  * returns true. A walker of the full memmap must then do this additional
1447  * check to ensure the memmap they are looking at is sane by making sure
1448  * the zone and PFN linkages are still valid. This is expensive, but walkers
1449  * of the full memmap are extremely rare.
1450  */
1451 bool memmap_valid_within(unsigned long pfn,
1452 					struct page *page, struct zone *zone);
1453 #else
1454 static inline bool memmap_valid_within(unsigned long pfn,
1455 					struct page *page, struct zone *zone)
1456 {
1457 	return true;
1458 }
1459 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1460 
1461 #endif /* !__GENERATING_BOUNDS.H */
1462 #endif /* !__ASSEMBLY__ */
1463 #endif /* _LINUX_MMZONE_H */
1464