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