xref: /linux-6.15/include/linux/mmzone.h (revision ea8a163e)
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 <linux/local_lock.h>
24 #include <asm/page.h>
25 
26 /* Free memory management - zoned buddy allocator.  */
27 #ifndef CONFIG_FORCE_MAX_ZONEORDER
28 #define MAX_ORDER 11
29 #else
30 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
31 #endif
32 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
33 
34 /*
35  * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
36  * costly to service.  That is between allocation orders which should
37  * coalesce naturally under reasonable reclaim pressure and those which
38  * will not.
39  */
40 #define PAGE_ALLOC_COSTLY_ORDER 3
41 
42 enum migratetype {
43 	MIGRATE_UNMOVABLE,
44 	MIGRATE_MOVABLE,
45 	MIGRATE_RECLAIMABLE,
46 	MIGRATE_PCPTYPES,	/* the number of types on the pcp lists */
47 	MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
48 #ifdef CONFIG_CMA
49 	/*
50 	 * MIGRATE_CMA migration type is designed to mimic the way
51 	 * ZONE_MOVABLE works.  Only movable pages can be allocated
52 	 * from MIGRATE_CMA pageblocks and page allocator never
53 	 * implicitly change migration type of MIGRATE_CMA pageblock.
54 	 *
55 	 * The way to use it is to change migratetype of a range of
56 	 * pageblocks to MIGRATE_CMA which can be done by
57 	 * __free_pageblock_cma() function.  What is important though
58 	 * is that a range of pageblocks must be aligned to
59 	 * MAX_ORDER_NR_PAGES should biggest page be bigger than
60 	 * a single pageblock.
61 	 */
62 	MIGRATE_CMA,
63 #endif
64 #ifdef CONFIG_MEMORY_ISOLATION
65 	MIGRATE_ISOLATE,	/* can't allocate from here */
66 #endif
67 	MIGRATE_TYPES
68 };
69 
70 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
71 extern const char * const migratetype_names[MIGRATE_TYPES];
72 
73 #ifdef CONFIG_CMA
74 #  define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
75 #  define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76 #else
77 #  define is_migrate_cma(migratetype) false
78 #  define is_migrate_cma_page(_page) false
79 #endif
80 
81 static inline bool is_migrate_movable(int mt)
82 {
83 	return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
84 }
85 
86 #define for_each_migratetype_order(order, type) \
87 	for (order = 0; order < MAX_ORDER; order++) \
88 		for (type = 0; type < MIGRATE_TYPES; type++)
89 
90 extern int page_group_by_mobility_disabled;
91 
92 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
93 
94 #define get_pageblock_migratetype(page)					\
95 	get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
96 
97 struct free_area {
98 	struct list_head	free_list[MIGRATE_TYPES];
99 	unsigned long		nr_free;
100 };
101 
102 static inline struct page *get_page_from_free_area(struct free_area *area,
103 					    int migratetype)
104 {
105 	return list_first_entry_or_null(&area->free_list[migratetype],
106 					struct page, lru);
107 }
108 
109 static inline bool free_area_empty(struct free_area *area, int migratetype)
110 {
111 	return list_empty(&area->free_list[migratetype]);
112 }
113 
114 struct pglist_data;
115 
116 /*
117  * Add a wild amount of padding here to ensure data fall into separate
118  * cachelines.  There are very few zone structures in the machine, so space
119  * consumption is not a concern here.
120  */
121 #if defined(CONFIG_SMP)
122 struct zone_padding {
123 	char x[0];
124 } ____cacheline_internodealigned_in_smp;
125 #define ZONE_PADDING(name)	struct zone_padding name;
126 #else
127 #define ZONE_PADDING(name)
128 #endif
129 
130 #ifdef CONFIG_NUMA
131 enum numa_stat_item {
132 	NUMA_HIT,		/* allocated in intended node */
133 	NUMA_MISS,		/* allocated in non intended node */
134 	NUMA_FOREIGN,		/* was intended here, hit elsewhere */
135 	NUMA_INTERLEAVE_HIT,	/* interleaver preferred this zone */
136 	NUMA_LOCAL,		/* allocation from local node */
137 	NUMA_OTHER,		/* allocation from other node */
138 	NR_VM_NUMA_EVENT_ITEMS
139 };
140 #else
141 #define NR_VM_NUMA_EVENT_ITEMS 0
142 #endif
143 
144 enum zone_stat_item {
145 	/* First 128 byte cacheline (assuming 64 bit words) */
146 	NR_FREE_PAGES,
147 	NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
148 	NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
149 	NR_ZONE_ACTIVE_ANON,
150 	NR_ZONE_INACTIVE_FILE,
151 	NR_ZONE_ACTIVE_FILE,
152 	NR_ZONE_UNEVICTABLE,
153 	NR_ZONE_WRITE_PENDING,	/* Count of dirty, writeback and unstable pages */
154 	NR_MLOCK,		/* mlock()ed pages found and moved off LRU */
155 	/* Second 128 byte cacheline */
156 	NR_BOUNCE,
157 #if IS_ENABLED(CONFIG_ZSMALLOC)
158 	NR_ZSPAGES,		/* allocated in zsmalloc */
159 #endif
160 	NR_FREE_CMA_PAGES,
161 	NR_VM_ZONE_STAT_ITEMS };
162 
163 enum node_stat_item {
164 	NR_LRU_BASE,
165 	NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
166 	NR_ACTIVE_ANON,		/*  "     "     "   "       "         */
167 	NR_INACTIVE_FILE,	/*  "     "     "   "       "         */
168 	NR_ACTIVE_FILE,		/*  "     "     "   "       "         */
169 	NR_UNEVICTABLE,		/*  "     "     "   "       "         */
170 	NR_SLAB_RECLAIMABLE_B,
171 	NR_SLAB_UNRECLAIMABLE_B,
172 	NR_ISOLATED_ANON,	/* Temporary isolated pages from anon lru */
173 	NR_ISOLATED_FILE,	/* Temporary isolated pages from file lru */
174 	WORKINGSET_NODES,
175 	WORKINGSET_REFAULT_BASE,
176 	WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
177 	WORKINGSET_REFAULT_FILE,
178 	WORKINGSET_ACTIVATE_BASE,
179 	WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
180 	WORKINGSET_ACTIVATE_FILE,
181 	WORKINGSET_RESTORE_BASE,
182 	WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
183 	WORKINGSET_RESTORE_FILE,
184 	WORKINGSET_NODERECLAIM,
185 	NR_ANON_MAPPED,	/* Mapped anonymous pages */
186 	NR_FILE_MAPPED,	/* pagecache pages mapped into pagetables.
187 			   only modified from process context */
188 	NR_FILE_PAGES,
189 	NR_FILE_DIRTY,
190 	NR_WRITEBACK,
191 	NR_WRITEBACK_TEMP,	/* Writeback using temporary buffers */
192 	NR_SHMEM,		/* shmem pages (included tmpfs/GEM pages) */
193 	NR_SHMEM_THPS,
194 	NR_SHMEM_PMDMAPPED,
195 	NR_FILE_THPS,
196 	NR_FILE_PMDMAPPED,
197 	NR_ANON_THPS,
198 	NR_VMSCAN_WRITE,
199 	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
200 	NR_DIRTIED,		/* page dirtyings since bootup */
201 	NR_WRITTEN,		/* page writings since bootup */
202 	NR_THROTTLED_WRITTEN,	/* NR_WRITTEN while reclaim throttled */
203 	NR_KERNEL_MISC_RECLAIMABLE,	/* reclaimable non-slab kernel pages */
204 	NR_FOLL_PIN_ACQUIRED,	/* via: pin_user_page(), gup flag: FOLL_PIN */
205 	NR_FOLL_PIN_RELEASED,	/* pages returned via unpin_user_page() */
206 	NR_KERNEL_STACK_KB,	/* measured in KiB */
207 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
208 	NR_KERNEL_SCS_KB,	/* measured in KiB */
209 #endif
210 	NR_PAGETABLE,		/* used for pagetables */
211 #ifdef CONFIG_SWAP
212 	NR_SWAPCACHE,
213 #endif
214 	NR_VM_NODE_STAT_ITEMS
215 };
216 
217 /*
218  * Returns true if the item should be printed in THPs (/proc/vmstat
219  * currently prints number of anon, file and shmem THPs. But the item
220  * is charged in pages).
221  */
222 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
223 {
224 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
225 		return false;
226 
227 	return item == NR_ANON_THPS ||
228 	       item == NR_FILE_THPS ||
229 	       item == NR_SHMEM_THPS ||
230 	       item == NR_SHMEM_PMDMAPPED ||
231 	       item == NR_FILE_PMDMAPPED;
232 }
233 
234 /*
235  * Returns true if the value is measured in bytes (most vmstat values are
236  * measured in pages). This defines the API part, the internal representation
237  * might be different.
238  */
239 static __always_inline bool vmstat_item_in_bytes(int idx)
240 {
241 	/*
242 	 * Global and per-node slab counters track slab pages.
243 	 * It's expected that changes are multiples of PAGE_SIZE.
244 	 * Internally values are stored in pages.
245 	 *
246 	 * Per-memcg and per-lruvec counters track memory, consumed
247 	 * by individual slab objects. These counters are actually
248 	 * byte-precise.
249 	 */
250 	return (idx == NR_SLAB_RECLAIMABLE_B ||
251 		idx == NR_SLAB_UNRECLAIMABLE_B);
252 }
253 
254 /*
255  * We do arithmetic on the LRU lists in various places in the code,
256  * so it is important to keep the active lists LRU_ACTIVE higher in
257  * the array than the corresponding inactive lists, and to keep
258  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
259  *
260  * This has to be kept in sync with the statistics in zone_stat_item
261  * above and the descriptions in vmstat_text in mm/vmstat.c
262  */
263 #define LRU_BASE 0
264 #define LRU_ACTIVE 1
265 #define LRU_FILE 2
266 
267 enum lru_list {
268 	LRU_INACTIVE_ANON = LRU_BASE,
269 	LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
270 	LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
271 	LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
272 	LRU_UNEVICTABLE,
273 	NR_LRU_LISTS
274 };
275 
276 enum vmscan_throttle_state {
277 	VMSCAN_THROTTLE_WRITEBACK,
278 	VMSCAN_THROTTLE_ISOLATED,
279 	VMSCAN_THROTTLE_NOPROGRESS,
280 	NR_VMSCAN_THROTTLE,
281 };
282 
283 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
284 
285 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
286 
287 static inline bool is_file_lru(enum lru_list lru)
288 {
289 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
290 }
291 
292 static inline bool is_active_lru(enum lru_list lru)
293 {
294 	return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
295 }
296 
297 #define ANON_AND_FILE 2
298 
299 enum lruvec_flags {
300 	LRUVEC_CONGESTED,		/* lruvec has many dirty pages
301 					 * backed by a congested BDI
302 					 */
303 };
304 
305 struct lruvec {
306 	struct list_head		lists[NR_LRU_LISTS];
307 	/* per lruvec lru_lock for memcg */
308 	spinlock_t			lru_lock;
309 	/*
310 	 * These track the cost of reclaiming one LRU - file or anon -
311 	 * over the other. As the observed cost of reclaiming one LRU
312 	 * increases, the reclaim scan balance tips toward the other.
313 	 */
314 	unsigned long			anon_cost;
315 	unsigned long			file_cost;
316 	/* Non-resident age, driven by LRU movement */
317 	atomic_long_t			nonresident_age;
318 	/* Refaults at the time of last reclaim cycle */
319 	unsigned long			refaults[ANON_AND_FILE];
320 	/* Various lruvec state flags (enum lruvec_flags) */
321 	unsigned long			flags;
322 #ifdef CONFIG_MEMCG
323 	struct pglist_data *pgdat;
324 #endif
325 };
326 
327 /* Isolate unmapped pages */
328 #define ISOLATE_UNMAPPED	((__force isolate_mode_t)0x2)
329 /* Isolate for asynchronous migration */
330 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
331 /* Isolate unevictable pages */
332 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
333 
334 /* LRU Isolation modes. */
335 typedef unsigned __bitwise isolate_mode_t;
336 
337 enum zone_watermarks {
338 	WMARK_MIN,
339 	WMARK_LOW,
340 	WMARK_HIGH,
341 	NR_WMARK
342 };
343 
344 /*
345  * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER plus one additional
346  * for pageblock size for THP if configured.
347  */
348 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
349 #define NR_PCP_THP 1
350 #else
351 #define NR_PCP_THP 0
352 #endif
353 #define NR_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1 + NR_PCP_THP))
354 
355 /*
356  * Shift to encode migratetype and order in the same integer, with order
357  * in the least significant bits.
358  */
359 #define NR_PCP_ORDER_WIDTH 8
360 #define NR_PCP_ORDER_MASK ((1<<NR_PCP_ORDER_WIDTH) - 1)
361 
362 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
363 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
364 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
365 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
366 
367 /* Fields and list protected by pagesets local_lock in page_alloc.c */
368 struct per_cpu_pages {
369 	int count;		/* number of pages in the list */
370 	int high;		/* high watermark, emptying needed */
371 	int batch;		/* chunk size for buddy add/remove */
372 	short free_factor;	/* batch scaling factor during free */
373 #ifdef CONFIG_NUMA
374 	short expire;		/* When 0, remote pagesets are drained */
375 #endif
376 
377 	/* Lists of pages, one per migrate type stored on the pcp-lists */
378 	struct list_head lists[NR_PCP_LISTS];
379 };
380 
381 struct per_cpu_zonestat {
382 #ifdef CONFIG_SMP
383 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
384 	s8 stat_threshold;
385 #endif
386 #ifdef CONFIG_NUMA
387 	/*
388 	 * Low priority inaccurate counters that are only folded
389 	 * on demand. Use a large type to avoid the overhead of
390 	 * folding during refresh_cpu_vm_stats.
391 	 */
392 	unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
393 #endif
394 };
395 
396 struct per_cpu_nodestat {
397 	s8 stat_threshold;
398 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
399 };
400 
401 #endif /* !__GENERATING_BOUNDS.H */
402 
403 enum zone_type {
404 	/*
405 	 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
406 	 * to DMA to all of the addressable memory (ZONE_NORMAL).
407 	 * On architectures where this area covers the whole 32 bit address
408 	 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
409 	 * DMA addressing constraints. This distinction is important as a 32bit
410 	 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
411 	 * platforms may need both zones as they support peripherals with
412 	 * different DMA addressing limitations.
413 	 */
414 #ifdef CONFIG_ZONE_DMA
415 	ZONE_DMA,
416 #endif
417 #ifdef CONFIG_ZONE_DMA32
418 	ZONE_DMA32,
419 #endif
420 	/*
421 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
422 	 * performed on pages in ZONE_NORMAL if the DMA devices support
423 	 * transfers to all addressable memory.
424 	 */
425 	ZONE_NORMAL,
426 #ifdef CONFIG_HIGHMEM
427 	/*
428 	 * A memory area that is only addressable by the kernel through
429 	 * mapping portions into its own address space. This is for example
430 	 * used by i386 to allow the kernel to address the memory beyond
431 	 * 900MB. The kernel will set up special mappings (page
432 	 * table entries on i386) for each page that the kernel needs to
433 	 * access.
434 	 */
435 	ZONE_HIGHMEM,
436 #endif
437 	/*
438 	 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
439 	 * movable pages with few exceptional cases described below. Main use
440 	 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
441 	 * likely to succeed, and to locally limit unmovable allocations - e.g.,
442 	 * to increase the number of THP/huge pages. Notable special cases are:
443 	 *
444 	 * 1. Pinned pages: (long-term) pinning of movable pages might
445 	 *    essentially turn such pages unmovable. Therefore, we do not allow
446 	 *    pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
447 	 *    faulted, they come from the right zone right away. However, it is
448 	 *    still possible that address space already has pages in
449 	 *    ZONE_MOVABLE at the time when pages are pinned (i.e. user has
450 	 *    touches that memory before pinning). In such case we migrate them
451 	 *    to a different zone. When migration fails - pinning fails.
452 	 * 2. memblock allocations: kernelcore/movablecore setups might create
453 	 *    situations where ZONE_MOVABLE contains unmovable allocations
454 	 *    after boot. Memory offlining and allocations fail early.
455 	 * 3. Memory holes: kernelcore/movablecore setups might create very rare
456 	 *    situations where ZONE_MOVABLE contains memory holes after boot,
457 	 *    for example, if we have sections that are only partially
458 	 *    populated. Memory offlining and allocations fail early.
459 	 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
460 	 *    memory offlining, such pages cannot be allocated.
461 	 * 5. Unmovable PG_offline pages: in paravirtualized environments,
462 	 *    hotplugged memory blocks might only partially be managed by the
463 	 *    buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
464 	 *    parts not manged by the buddy are unmovable PG_offline pages. In
465 	 *    some cases (virtio-mem), such pages can be skipped during
466 	 *    memory offlining, however, cannot be moved/allocated. These
467 	 *    techniques might use alloc_contig_range() to hide previously
468 	 *    exposed pages from the buddy again (e.g., to implement some sort
469 	 *    of memory unplug in virtio-mem).
470 	 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
471 	 *    situations where ZERO_PAGE(0) which is allocated differently
472 	 *    on different platforms may end up in a movable zone. ZERO_PAGE(0)
473 	 *    cannot be migrated.
474 	 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
475 	 *    memory to the MOVABLE zone, the vmemmap pages are also placed in
476 	 *    such zone. Such pages cannot be really moved around as they are
477 	 *    self-stored in the range, but they are treated as movable when
478 	 *    the range they describe is about to be offlined.
479 	 *
480 	 * In general, no unmovable allocations that degrade memory offlining
481 	 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
482 	 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
483 	 * if has_unmovable_pages() states that there are no unmovable pages,
484 	 * there can be false negatives).
485 	 */
486 	ZONE_MOVABLE,
487 #ifdef CONFIG_ZONE_DEVICE
488 	ZONE_DEVICE,
489 #endif
490 	__MAX_NR_ZONES
491 
492 };
493 
494 #ifndef __GENERATING_BOUNDS_H
495 
496 #define ASYNC_AND_SYNC 2
497 
498 struct zone {
499 	/* Read-mostly fields */
500 
501 	/* zone watermarks, access with *_wmark_pages(zone) macros */
502 	unsigned long _watermark[NR_WMARK];
503 	unsigned long watermark_boost;
504 
505 	unsigned long nr_reserved_highatomic;
506 
507 	/*
508 	 * We don't know if the memory that we're going to allocate will be
509 	 * freeable or/and it will be released eventually, so to avoid totally
510 	 * wasting several GB of ram we must reserve some of the lower zone
511 	 * memory (otherwise we risk to run OOM on the lower zones despite
512 	 * there being tons of freeable ram on the higher zones).  This array is
513 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
514 	 * changes.
515 	 */
516 	long lowmem_reserve[MAX_NR_ZONES];
517 
518 #ifdef CONFIG_NUMA
519 	int node;
520 #endif
521 	struct pglist_data	*zone_pgdat;
522 	struct per_cpu_pages	__percpu *per_cpu_pageset;
523 	struct per_cpu_zonestat	__percpu *per_cpu_zonestats;
524 	/*
525 	 * the high and batch values are copied to individual pagesets for
526 	 * faster access
527 	 */
528 	int pageset_high;
529 	int pageset_batch;
530 
531 #ifndef CONFIG_SPARSEMEM
532 	/*
533 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
534 	 * In SPARSEMEM, this map is stored in struct mem_section
535 	 */
536 	unsigned long		*pageblock_flags;
537 #endif /* CONFIG_SPARSEMEM */
538 
539 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
540 	unsigned long		zone_start_pfn;
541 
542 	/*
543 	 * spanned_pages is the total pages spanned by the zone, including
544 	 * holes, which is calculated as:
545 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
546 	 *
547 	 * present_pages is physical pages existing within the zone, which
548 	 * is calculated as:
549 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
550 	 *
551 	 * present_early_pages is present pages existing within the zone
552 	 * located on memory available since early boot, excluding hotplugged
553 	 * memory.
554 	 *
555 	 * managed_pages is present pages managed by the buddy system, which
556 	 * is calculated as (reserved_pages includes pages allocated by the
557 	 * bootmem allocator):
558 	 *	managed_pages = present_pages - reserved_pages;
559 	 *
560 	 * cma pages is present pages that are assigned for CMA use
561 	 * (MIGRATE_CMA).
562 	 *
563 	 * So present_pages may be used by memory hotplug or memory power
564 	 * management logic to figure out unmanaged pages by checking
565 	 * (present_pages - managed_pages). And managed_pages should be used
566 	 * by page allocator and vm scanner to calculate all kinds of watermarks
567 	 * and thresholds.
568 	 *
569 	 * Locking rules:
570 	 *
571 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
572 	 * It is a seqlock because it has to be read outside of zone->lock,
573 	 * and it is done in the main allocator path.  But, it is written
574 	 * quite infrequently.
575 	 *
576 	 * The span_seq lock is declared along with zone->lock because it is
577 	 * frequently read in proximity to zone->lock.  It's good to
578 	 * give them a chance of being in the same cacheline.
579 	 *
580 	 * Write access to present_pages at runtime should be protected by
581 	 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
582 	 * present_pages should get_online_mems() to get a stable value.
583 	 */
584 	atomic_long_t		managed_pages;
585 	unsigned long		spanned_pages;
586 	unsigned long		present_pages;
587 #if defined(CONFIG_MEMORY_HOTPLUG)
588 	unsigned long		present_early_pages;
589 #endif
590 #ifdef CONFIG_CMA
591 	unsigned long		cma_pages;
592 #endif
593 
594 	const char		*name;
595 
596 #ifdef CONFIG_MEMORY_ISOLATION
597 	/*
598 	 * Number of isolated pageblock. It is used to solve incorrect
599 	 * freepage counting problem due to racy retrieving migratetype
600 	 * of pageblock. Protected by zone->lock.
601 	 */
602 	unsigned long		nr_isolate_pageblock;
603 #endif
604 
605 #ifdef CONFIG_MEMORY_HOTPLUG
606 	/* see spanned/present_pages for more description */
607 	seqlock_t		span_seqlock;
608 #endif
609 
610 	int initialized;
611 
612 	/* Write-intensive fields used from the page allocator */
613 	ZONE_PADDING(_pad1_)
614 
615 	/* free areas of different sizes */
616 	struct free_area	free_area[MAX_ORDER];
617 
618 	/* zone flags, see below */
619 	unsigned long		flags;
620 
621 	/* Primarily protects free_area */
622 	spinlock_t		lock;
623 
624 	/* Write-intensive fields used by compaction and vmstats. */
625 	ZONE_PADDING(_pad2_)
626 
627 	/*
628 	 * When free pages are below this point, additional steps are taken
629 	 * when reading the number of free pages to avoid per-cpu counter
630 	 * drift allowing watermarks to be breached
631 	 */
632 	unsigned long percpu_drift_mark;
633 
634 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
635 	/* pfn where compaction free scanner should start */
636 	unsigned long		compact_cached_free_pfn;
637 	/* pfn where compaction migration scanner should start */
638 	unsigned long		compact_cached_migrate_pfn[ASYNC_AND_SYNC];
639 	unsigned long		compact_init_migrate_pfn;
640 	unsigned long		compact_init_free_pfn;
641 #endif
642 
643 #ifdef CONFIG_COMPACTION
644 	/*
645 	 * On compaction failure, 1<<compact_defer_shift compactions
646 	 * are skipped before trying again. The number attempted since
647 	 * last failure is tracked with compact_considered.
648 	 * compact_order_failed is the minimum compaction failed order.
649 	 */
650 	unsigned int		compact_considered;
651 	unsigned int		compact_defer_shift;
652 	int			compact_order_failed;
653 #endif
654 
655 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
656 	/* Set to true when the PG_migrate_skip bits should be cleared */
657 	bool			compact_blockskip_flush;
658 #endif
659 
660 	bool			contiguous;
661 
662 	ZONE_PADDING(_pad3_)
663 	/* Zone statistics */
664 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
665 	atomic_long_t		vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
666 } ____cacheline_internodealigned_in_smp;
667 
668 enum pgdat_flags {
669 	PGDAT_DIRTY,			/* reclaim scanning has recently found
670 					 * many dirty file pages at the tail
671 					 * of the LRU.
672 					 */
673 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
674 					 * many pages under writeback
675 					 */
676 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
677 };
678 
679 enum zone_flags {
680 	ZONE_BOOSTED_WATERMARK,		/* zone recently boosted watermarks.
681 					 * Cleared when kswapd is woken.
682 					 */
683 	ZONE_RECLAIM_ACTIVE,		/* kswapd may be scanning the zone. */
684 };
685 
686 static inline unsigned long zone_managed_pages(struct zone *zone)
687 {
688 	return (unsigned long)atomic_long_read(&zone->managed_pages);
689 }
690 
691 static inline unsigned long zone_cma_pages(struct zone *zone)
692 {
693 #ifdef CONFIG_CMA
694 	return zone->cma_pages;
695 #else
696 	return 0;
697 #endif
698 }
699 
700 static inline unsigned long zone_end_pfn(const struct zone *zone)
701 {
702 	return zone->zone_start_pfn + zone->spanned_pages;
703 }
704 
705 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
706 {
707 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
708 }
709 
710 static inline bool zone_is_initialized(struct zone *zone)
711 {
712 	return zone->initialized;
713 }
714 
715 static inline bool zone_is_empty(struct zone *zone)
716 {
717 	return zone->spanned_pages == 0;
718 }
719 
720 /*
721  * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
722  * intersection with the given zone
723  */
724 static inline bool zone_intersects(struct zone *zone,
725 		unsigned long start_pfn, unsigned long nr_pages)
726 {
727 	if (zone_is_empty(zone))
728 		return false;
729 	if (start_pfn >= zone_end_pfn(zone) ||
730 	    start_pfn + nr_pages <= zone->zone_start_pfn)
731 		return false;
732 
733 	return true;
734 }
735 
736 /*
737  * The "priority" of VM scanning is how much of the queues we will scan in one
738  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
739  * queues ("queue_length >> 12") during an aging round.
740  */
741 #define DEF_PRIORITY 12
742 
743 /* Maximum number of zones on a zonelist */
744 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
745 
746 enum {
747 	ZONELIST_FALLBACK,	/* zonelist with fallback */
748 #ifdef CONFIG_NUMA
749 	/*
750 	 * The NUMA zonelists are doubled because we need zonelists that
751 	 * restrict the allocations to a single node for __GFP_THISNODE.
752 	 */
753 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
754 #endif
755 	MAX_ZONELISTS
756 };
757 
758 /*
759  * This struct contains information about a zone in a zonelist. It is stored
760  * here to avoid dereferences into large structures and lookups of tables
761  */
762 struct zoneref {
763 	struct zone *zone;	/* Pointer to actual zone */
764 	int zone_idx;		/* zone_idx(zoneref->zone) */
765 };
766 
767 /*
768  * One allocation request operates on a zonelist. A zonelist
769  * is a list of zones, the first one is the 'goal' of the
770  * allocation, the other zones are fallback zones, in decreasing
771  * priority.
772  *
773  * To speed the reading of the zonelist, the zonerefs contain the zone index
774  * of the entry being read. Helper functions to access information given
775  * a struct zoneref are
776  *
777  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
778  * zonelist_zone_idx()	- Return the index of the zone for an entry
779  * zonelist_node_idx()	- Return the index of the node for an entry
780  */
781 struct zonelist {
782 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
783 };
784 
785 /*
786  * The array of struct pages for flatmem.
787  * It must be declared for SPARSEMEM as well because there are configurations
788  * that rely on that.
789  */
790 extern struct page *mem_map;
791 
792 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
793 struct deferred_split {
794 	spinlock_t split_queue_lock;
795 	struct list_head split_queue;
796 	unsigned long split_queue_len;
797 };
798 #endif
799 
800 /*
801  * On NUMA machines, each NUMA node would have a pg_data_t to describe
802  * it's memory layout. On UMA machines there is a single pglist_data which
803  * describes the whole memory.
804  *
805  * Memory statistics and page replacement data structures are maintained on a
806  * per-zone basis.
807  */
808 typedef struct pglist_data {
809 	/*
810 	 * node_zones contains just the zones for THIS node. Not all of the
811 	 * zones may be populated, but it is the full list. It is referenced by
812 	 * this node's node_zonelists as well as other node's node_zonelists.
813 	 */
814 	struct zone node_zones[MAX_NR_ZONES];
815 
816 	/*
817 	 * node_zonelists contains references to all zones in all nodes.
818 	 * Generally the first zones will be references to this node's
819 	 * node_zones.
820 	 */
821 	struct zonelist node_zonelists[MAX_ZONELISTS];
822 
823 	int nr_zones; /* number of populated zones in this node */
824 #ifdef CONFIG_FLATMEM	/* means !SPARSEMEM */
825 	struct page *node_mem_map;
826 #ifdef CONFIG_PAGE_EXTENSION
827 	struct page_ext *node_page_ext;
828 #endif
829 #endif
830 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
831 	/*
832 	 * Must be held any time you expect node_start_pfn,
833 	 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
834 	 * Also synchronizes pgdat->first_deferred_pfn during deferred page
835 	 * init.
836 	 *
837 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
838 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
839 	 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
840 	 *
841 	 * Nests above zone->lock and zone->span_seqlock
842 	 */
843 	spinlock_t node_size_lock;
844 #endif
845 	unsigned long node_start_pfn;
846 	unsigned long node_present_pages; /* total number of physical pages */
847 	unsigned long node_spanned_pages; /* total size of physical page
848 					     range, including holes */
849 	int node_id;
850 	wait_queue_head_t kswapd_wait;
851 	wait_queue_head_t pfmemalloc_wait;
852 
853 	/* workqueues for throttling reclaim for different reasons. */
854 	wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
855 
856 	atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
857 	unsigned long nr_reclaim_start;	/* nr pages written while throttled
858 					 * when throttling started. */
859 	struct task_struct *kswapd;	/* Protected by
860 					   mem_hotplug_begin/end() */
861 	int kswapd_order;
862 	enum zone_type kswapd_highest_zoneidx;
863 
864 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
865 
866 #ifdef CONFIG_COMPACTION
867 	int kcompactd_max_order;
868 	enum zone_type kcompactd_highest_zoneidx;
869 	wait_queue_head_t kcompactd_wait;
870 	struct task_struct *kcompactd;
871 	bool proactive_compact_trigger;
872 #endif
873 	/*
874 	 * This is a per-node reserve of pages that are not available
875 	 * to userspace allocations.
876 	 */
877 	unsigned long		totalreserve_pages;
878 
879 #ifdef CONFIG_NUMA
880 	/*
881 	 * node reclaim becomes active if more unmapped pages exist.
882 	 */
883 	unsigned long		min_unmapped_pages;
884 	unsigned long		min_slab_pages;
885 #endif /* CONFIG_NUMA */
886 
887 	/* Write-intensive fields used by page reclaim */
888 	ZONE_PADDING(_pad1_)
889 
890 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
891 	/*
892 	 * If memory initialisation on large machines is deferred then this
893 	 * is the first PFN that needs to be initialised.
894 	 */
895 	unsigned long first_deferred_pfn;
896 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
897 
898 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
899 	struct deferred_split deferred_split_queue;
900 #endif
901 
902 	/* Fields commonly accessed by the page reclaim scanner */
903 
904 	/*
905 	 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
906 	 *
907 	 * Use mem_cgroup_lruvec() to look up lruvecs.
908 	 */
909 	struct lruvec		__lruvec;
910 
911 	unsigned long		flags;
912 
913 	ZONE_PADDING(_pad2_)
914 
915 	/* Per-node vmstats */
916 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
917 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
918 } pg_data_t;
919 
920 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
921 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
922 #ifdef CONFIG_FLATMEM
923 #define pgdat_page_nr(pgdat, pagenr)	((pgdat)->node_mem_map + (pagenr))
924 #else
925 #define pgdat_page_nr(pgdat, pagenr)	pfn_to_page((pgdat)->node_start_pfn + (pagenr))
926 #endif
927 #define nid_page_nr(nid, pagenr) 	pgdat_page_nr(NODE_DATA(nid),(pagenr))
928 
929 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
930 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
931 
932 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
933 {
934 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
935 }
936 
937 static inline bool pgdat_is_empty(pg_data_t *pgdat)
938 {
939 	return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
940 }
941 
942 #include <linux/memory_hotplug.h>
943 
944 void build_all_zonelists(pg_data_t *pgdat);
945 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
946 		   enum zone_type highest_zoneidx);
947 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
948 			 int highest_zoneidx, unsigned int alloc_flags,
949 			 long free_pages);
950 bool zone_watermark_ok(struct zone *z, unsigned int order,
951 		unsigned long mark, int highest_zoneidx,
952 		unsigned int alloc_flags);
953 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
954 		unsigned long mark, int highest_zoneidx);
955 /*
956  * Memory initialization context, use to differentiate memory added by
957  * the platform statically or via memory hotplug interface.
958  */
959 enum meminit_context {
960 	MEMINIT_EARLY,
961 	MEMINIT_HOTPLUG,
962 };
963 
964 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
965 				     unsigned long size);
966 
967 extern void lruvec_init(struct lruvec *lruvec);
968 
969 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
970 {
971 #ifdef CONFIG_MEMCG
972 	return lruvec->pgdat;
973 #else
974 	return container_of(lruvec, struct pglist_data, __lruvec);
975 #endif
976 }
977 
978 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
979 int local_memory_node(int node_id);
980 #else
981 static inline int local_memory_node(int node_id) { return node_id; };
982 #endif
983 
984 /*
985  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
986  */
987 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
988 
989 #ifdef CONFIG_ZONE_DEVICE
990 static inline bool zone_is_zone_device(struct zone *zone)
991 {
992 	return zone_idx(zone) == ZONE_DEVICE;
993 }
994 #else
995 static inline bool zone_is_zone_device(struct zone *zone)
996 {
997 	return false;
998 }
999 #endif
1000 
1001 /*
1002  * Returns true if a zone has pages managed by the buddy allocator.
1003  * All the reclaim decisions have to use this function rather than
1004  * populated_zone(). If the whole zone is reserved then we can easily
1005  * end up with populated_zone() && !managed_zone().
1006  */
1007 static inline bool managed_zone(struct zone *zone)
1008 {
1009 	return zone_managed_pages(zone);
1010 }
1011 
1012 /* Returns true if a zone has memory */
1013 static inline bool populated_zone(struct zone *zone)
1014 {
1015 	return zone->present_pages;
1016 }
1017 
1018 #ifdef CONFIG_NUMA
1019 static inline int zone_to_nid(struct zone *zone)
1020 {
1021 	return zone->node;
1022 }
1023 
1024 static inline void zone_set_nid(struct zone *zone, int nid)
1025 {
1026 	zone->node = nid;
1027 }
1028 #else
1029 static inline int zone_to_nid(struct zone *zone)
1030 {
1031 	return 0;
1032 }
1033 
1034 static inline void zone_set_nid(struct zone *zone, int nid) {}
1035 #endif
1036 
1037 extern int movable_zone;
1038 
1039 static inline int is_highmem_idx(enum zone_type idx)
1040 {
1041 #ifdef CONFIG_HIGHMEM
1042 	return (idx == ZONE_HIGHMEM ||
1043 		(idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1044 #else
1045 	return 0;
1046 #endif
1047 }
1048 
1049 /**
1050  * is_highmem - helper function to quickly check if a struct zone is a
1051  *              highmem zone or not.  This is an attempt to keep references
1052  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1053  * @zone: pointer to struct zone variable
1054  * Return: 1 for a highmem zone, 0 otherwise
1055  */
1056 static inline int is_highmem(struct zone *zone)
1057 {
1058 #ifdef CONFIG_HIGHMEM
1059 	return is_highmem_idx(zone_idx(zone));
1060 #else
1061 	return 0;
1062 #endif
1063 }
1064 
1065 /* These two functions are used to setup the per zone pages min values */
1066 struct ctl_table;
1067 
1068 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1069 		loff_t *);
1070 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1071 		size_t *, loff_t *);
1072 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1073 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1074 		size_t *, loff_t *);
1075 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1076 		void *, size_t *, loff_t *);
1077 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1078 		void *, size_t *, loff_t *);
1079 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1080 		void *, size_t *, loff_t *);
1081 int numa_zonelist_order_handler(struct ctl_table *, int,
1082 		void *, size_t *, loff_t *);
1083 extern int percpu_pagelist_high_fraction;
1084 extern char numa_zonelist_order[];
1085 #define NUMA_ZONELIST_ORDER_LEN	16
1086 
1087 #ifndef CONFIG_NUMA
1088 
1089 extern struct pglist_data contig_page_data;
1090 static inline struct pglist_data *NODE_DATA(int nid)
1091 {
1092 	return &contig_page_data;
1093 }
1094 #define NODE_MEM_MAP(nid)	mem_map
1095 
1096 #else /* CONFIG_NUMA */
1097 
1098 #include <asm/mmzone.h>
1099 
1100 #endif /* !CONFIG_NUMA */
1101 
1102 extern struct pglist_data *first_online_pgdat(void);
1103 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1104 extern struct zone *next_zone(struct zone *zone);
1105 
1106 /**
1107  * for_each_online_pgdat - helper macro to iterate over all online nodes
1108  * @pgdat: pointer to a pg_data_t variable
1109  */
1110 #define for_each_online_pgdat(pgdat)			\
1111 	for (pgdat = first_online_pgdat();		\
1112 	     pgdat;					\
1113 	     pgdat = next_online_pgdat(pgdat))
1114 /**
1115  * for_each_zone - helper macro to iterate over all memory zones
1116  * @zone: pointer to struct zone variable
1117  *
1118  * The user only needs to declare the zone variable, for_each_zone
1119  * fills it in.
1120  */
1121 #define for_each_zone(zone)			        \
1122 	for (zone = (first_online_pgdat())->node_zones; \
1123 	     zone;					\
1124 	     zone = next_zone(zone))
1125 
1126 #define for_each_populated_zone(zone)		        \
1127 	for (zone = (first_online_pgdat())->node_zones; \
1128 	     zone;					\
1129 	     zone = next_zone(zone))			\
1130 		if (!populated_zone(zone))		\
1131 			; /* do nothing */		\
1132 		else
1133 
1134 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1135 {
1136 	return zoneref->zone;
1137 }
1138 
1139 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1140 {
1141 	return zoneref->zone_idx;
1142 }
1143 
1144 static inline int zonelist_node_idx(struct zoneref *zoneref)
1145 {
1146 	return zone_to_nid(zoneref->zone);
1147 }
1148 
1149 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1150 					enum zone_type highest_zoneidx,
1151 					nodemask_t *nodes);
1152 
1153 /**
1154  * 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
1155  * @z: The cursor used as a starting point for the search
1156  * @highest_zoneidx: The zone index of the highest zone to return
1157  * @nodes: An optional nodemask to filter the zonelist with
1158  *
1159  * This function returns the next zone at or below a given zone index that is
1160  * within the allowed nodemask using a cursor as the starting point for the
1161  * search. The zoneref returned is a cursor that represents the current zone
1162  * being examined. It should be advanced by one before calling
1163  * next_zones_zonelist again.
1164  *
1165  * Return: the next zone at or below highest_zoneidx within the allowed
1166  * nodemask using a cursor within a zonelist as a starting point
1167  */
1168 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1169 					enum zone_type highest_zoneidx,
1170 					nodemask_t *nodes)
1171 {
1172 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1173 		return z;
1174 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
1175 }
1176 
1177 /**
1178  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1179  * @zonelist: The zonelist to search for a suitable zone
1180  * @highest_zoneidx: The zone index of the highest zone to return
1181  * @nodes: An optional nodemask to filter the zonelist with
1182  *
1183  * This function returns the first zone at or below a given zone index that is
1184  * within the allowed nodemask. The zoneref returned is a cursor that can be
1185  * used to iterate the zonelist with next_zones_zonelist by advancing it by
1186  * one before calling.
1187  *
1188  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1189  * never NULL). This may happen either genuinely, or due to concurrent nodemask
1190  * update due to cpuset modification.
1191  *
1192  * Return: Zoneref pointer for the first suitable zone found
1193  */
1194 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1195 					enum zone_type highest_zoneidx,
1196 					nodemask_t *nodes)
1197 {
1198 	return next_zones_zonelist(zonelist->_zonerefs,
1199 							highest_zoneidx, nodes);
1200 }
1201 
1202 /**
1203  * 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
1204  * @zone: The current zone in the iterator
1205  * @z: The current pointer within zonelist->_zonerefs being iterated
1206  * @zlist: The zonelist being iterated
1207  * @highidx: The zone index of the highest zone to return
1208  * @nodemask: Nodemask allowed by the allocator
1209  *
1210  * This iterator iterates though all zones at or below a given zone index and
1211  * within a given nodemask
1212  */
1213 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1214 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1215 		zone;							\
1216 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1217 			zone = zonelist_zone(z))
1218 
1219 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1220 	for (zone = z->zone;	\
1221 		zone;							\
1222 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1223 			zone = zonelist_zone(z))
1224 
1225 
1226 /**
1227  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1228  * @zone: The current zone in the iterator
1229  * @z: The current pointer within zonelist->zones being iterated
1230  * @zlist: The zonelist being iterated
1231  * @highidx: The zone index of the highest zone to return
1232  *
1233  * This iterator iterates though all zones at or below a given zone index.
1234  */
1235 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1236 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1237 
1238 /* Whether the 'nodes' are all movable nodes */
1239 static inline bool movable_only_nodes(nodemask_t *nodes)
1240 {
1241 	struct zonelist *zonelist;
1242 	struct zoneref *z;
1243 	int nid;
1244 
1245 	if (nodes_empty(*nodes))
1246 		return false;
1247 
1248 	/*
1249 	 * We can chose arbitrary node from the nodemask to get a
1250 	 * zonelist as they are interlinked. We just need to find
1251 	 * at least one zone that can satisfy kernel allocations.
1252 	 */
1253 	nid = first_node(*nodes);
1254 	zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1255 	z = first_zones_zonelist(zonelist, ZONE_NORMAL,	nodes);
1256 	return (!z->zone) ? true : false;
1257 }
1258 
1259 
1260 #ifdef CONFIG_SPARSEMEM
1261 #include <asm/sparsemem.h>
1262 #endif
1263 
1264 #ifdef CONFIG_FLATMEM
1265 #define pfn_to_nid(pfn)		(0)
1266 #endif
1267 
1268 #ifdef CONFIG_SPARSEMEM
1269 
1270 /*
1271  * PA_SECTION_SHIFT		physical address to/from section number
1272  * PFN_SECTION_SHIFT		pfn to/from section number
1273  */
1274 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1275 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1276 
1277 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1278 
1279 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1280 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1281 
1282 #define SECTION_BLOCKFLAGS_BITS \
1283 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1284 
1285 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1286 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1287 #endif
1288 
1289 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1290 {
1291 	return pfn >> PFN_SECTION_SHIFT;
1292 }
1293 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1294 {
1295 	return sec << PFN_SECTION_SHIFT;
1296 }
1297 
1298 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1299 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1300 
1301 #define SUBSECTION_SHIFT 21
1302 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1303 
1304 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1305 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1306 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1307 
1308 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1309 #error Subsection size exceeds section size
1310 #else
1311 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1312 #endif
1313 
1314 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1315 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1316 
1317 struct mem_section_usage {
1318 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1319 	DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1320 #endif
1321 	/* See declaration of similar field in struct zone */
1322 	unsigned long pageblock_flags[0];
1323 };
1324 
1325 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1326 
1327 struct page;
1328 struct page_ext;
1329 struct mem_section {
1330 	/*
1331 	 * This is, logically, a pointer to an array of struct
1332 	 * pages.  However, it is stored with some other magic.
1333 	 * (see sparse.c::sparse_init_one_section())
1334 	 *
1335 	 * Additionally during early boot we encode node id of
1336 	 * the location of the section here to guide allocation.
1337 	 * (see sparse.c::memory_present())
1338 	 *
1339 	 * Making it a UL at least makes someone do a cast
1340 	 * before using it wrong.
1341 	 */
1342 	unsigned long section_mem_map;
1343 
1344 	struct mem_section_usage *usage;
1345 #ifdef CONFIG_PAGE_EXTENSION
1346 	/*
1347 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1348 	 * section. (see page_ext.h about this.)
1349 	 */
1350 	struct page_ext *page_ext;
1351 	unsigned long pad;
1352 #endif
1353 	/*
1354 	 * WARNING: mem_section must be a power-of-2 in size for the
1355 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1356 	 */
1357 };
1358 
1359 #ifdef CONFIG_SPARSEMEM_EXTREME
1360 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1361 #else
1362 #define SECTIONS_PER_ROOT	1
1363 #endif
1364 
1365 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1366 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1367 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1368 
1369 #ifdef CONFIG_SPARSEMEM_EXTREME
1370 extern struct mem_section **mem_section;
1371 #else
1372 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1373 #endif
1374 
1375 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1376 {
1377 	return ms->usage->pageblock_flags;
1378 }
1379 
1380 static inline struct mem_section *__nr_to_section(unsigned long nr)
1381 {
1382 #ifdef CONFIG_SPARSEMEM_EXTREME
1383 	if (!mem_section)
1384 		return NULL;
1385 #endif
1386 	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1387 		return NULL;
1388 	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1389 }
1390 extern size_t mem_section_usage_size(void);
1391 
1392 /*
1393  * We use the lower bits of the mem_map pointer to store
1394  * a little bit of information.  The pointer is calculated
1395  * as mem_map - section_nr_to_pfn(pnum).  The result is
1396  * aligned to the minimum alignment of the two values:
1397  *   1. All mem_map arrays are page-aligned.
1398  *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1399  *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1400  *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1401  *      worst combination is powerpc with 256k pages,
1402  *      which results in PFN_SECTION_SHIFT equal 6.
1403  * To sum it up, at least 6 bits are available.
1404  */
1405 #define SECTION_MARKED_PRESENT		(1UL<<0)
1406 #define SECTION_HAS_MEM_MAP		(1UL<<1)
1407 #define SECTION_IS_ONLINE		(1UL<<2)
1408 #define SECTION_IS_EARLY		(1UL<<3)
1409 #define SECTION_TAINT_ZONE_DEVICE	(1UL<<4)
1410 #define SECTION_MAP_LAST_BIT		(1UL<<5)
1411 #define SECTION_MAP_MASK		(~(SECTION_MAP_LAST_BIT-1))
1412 #define SECTION_NID_SHIFT		6
1413 
1414 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1415 {
1416 	unsigned long map = section->section_mem_map;
1417 	map &= SECTION_MAP_MASK;
1418 	return (struct page *)map;
1419 }
1420 
1421 static inline int present_section(struct mem_section *section)
1422 {
1423 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1424 }
1425 
1426 static inline int present_section_nr(unsigned long nr)
1427 {
1428 	return present_section(__nr_to_section(nr));
1429 }
1430 
1431 static inline int valid_section(struct mem_section *section)
1432 {
1433 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1434 }
1435 
1436 static inline int early_section(struct mem_section *section)
1437 {
1438 	return (section && (section->section_mem_map & SECTION_IS_EARLY));
1439 }
1440 
1441 static inline int valid_section_nr(unsigned long nr)
1442 {
1443 	return valid_section(__nr_to_section(nr));
1444 }
1445 
1446 static inline int online_section(struct mem_section *section)
1447 {
1448 	return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1449 }
1450 
1451 static inline int online_device_section(struct mem_section *section)
1452 {
1453 	unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1454 
1455 	return section && ((section->section_mem_map & flags) == flags);
1456 }
1457 
1458 static inline int online_section_nr(unsigned long nr)
1459 {
1460 	return online_section(__nr_to_section(nr));
1461 }
1462 
1463 #ifdef CONFIG_MEMORY_HOTPLUG
1464 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1465 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1466 #endif
1467 
1468 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1469 {
1470 	return __nr_to_section(pfn_to_section_nr(pfn));
1471 }
1472 
1473 extern unsigned long __highest_present_section_nr;
1474 
1475 static inline int subsection_map_index(unsigned long pfn)
1476 {
1477 	return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1478 }
1479 
1480 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1481 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1482 {
1483 	int idx = subsection_map_index(pfn);
1484 
1485 	return test_bit(idx, ms->usage->subsection_map);
1486 }
1487 #else
1488 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1489 {
1490 	return 1;
1491 }
1492 #endif
1493 
1494 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1495 /**
1496  * pfn_valid - check if there is a valid memory map entry for a PFN
1497  * @pfn: the page frame number to check
1498  *
1499  * Check if there is a valid memory map entry aka struct page for the @pfn.
1500  * Note, that availability of the memory map entry does not imply that
1501  * there is actual usable memory at that @pfn. The struct page may
1502  * represent a hole or an unusable page frame.
1503  *
1504  * Return: 1 for PFNs that have memory map entries and 0 otherwise
1505  */
1506 static inline int pfn_valid(unsigned long pfn)
1507 {
1508 	struct mem_section *ms;
1509 
1510 	/*
1511 	 * Ensure the upper PAGE_SHIFT bits are clear in the
1512 	 * pfn. Else it might lead to false positives when
1513 	 * some of the upper bits are set, but the lower bits
1514 	 * match a valid pfn.
1515 	 */
1516 	if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
1517 		return 0;
1518 
1519 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1520 		return 0;
1521 	ms = __pfn_to_section(pfn);
1522 	if (!valid_section(ms))
1523 		return 0;
1524 	/*
1525 	 * Traditionally early sections always returned pfn_valid() for
1526 	 * the entire section-sized span.
1527 	 */
1528 	return early_section(ms) || pfn_section_valid(ms, pfn);
1529 }
1530 #endif
1531 
1532 static inline int pfn_in_present_section(unsigned long pfn)
1533 {
1534 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1535 		return 0;
1536 	return present_section(__pfn_to_section(pfn));
1537 }
1538 
1539 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1540 {
1541 	while (++section_nr <= __highest_present_section_nr) {
1542 		if (present_section_nr(section_nr))
1543 			return section_nr;
1544 	}
1545 
1546 	return -1;
1547 }
1548 
1549 /*
1550  * These are _only_ used during initialisation, therefore they
1551  * can use __initdata ...  They could have names to indicate
1552  * this restriction.
1553  */
1554 #ifdef CONFIG_NUMA
1555 #define pfn_to_nid(pfn)							\
1556 ({									\
1557 	unsigned long __pfn_to_nid_pfn = (pfn);				\
1558 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
1559 })
1560 #else
1561 #define pfn_to_nid(pfn)		(0)
1562 #endif
1563 
1564 void sparse_init(void);
1565 #else
1566 #define sparse_init()	do {} while (0)
1567 #define sparse_index_init(_sec, _nid)  do {} while (0)
1568 #define pfn_in_present_section pfn_valid
1569 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1570 #endif /* CONFIG_SPARSEMEM */
1571 
1572 #endif /* !__GENERATING_BOUNDS.H */
1573 #endif /* !__ASSEMBLY__ */
1574 #endif /* _LINUX_MMZONE_H */
1575