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
42enum 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.
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 */
68extern const 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
78static inline bool is_migrate_movable(int mt)
79{
80 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
81}
82
83/*
84 * Check whether a migratetype can be merged with another migratetype.
85 *
86 * It is only mergeable when it can fall back to other migratetypes for
87 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
88 */
89static inline bool migratetype_is_mergeable(int mt)
90{
91 return mt < MIGRATE_PCPTYPES;
92}
93
94#define for_each_migratetype_order(order, type) \
95 for (order = 0; order < MAX_ORDER; order++) \
96 for (type = 0; type < MIGRATE_TYPES; type++)
97
98extern int page_group_by_mobility_disabled;
99
100#define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
101
102#define get_pageblock_migratetype(page) \
103 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
104
105struct free_area {
106 struct list_head free_list[MIGRATE_TYPES];
107 unsigned long nr_free;
108};
109
110static inline struct page *get_page_from_free_area(struct free_area *area,
111 int migratetype)
112{
113 return list_first_entry_or_null(&area->free_list[migratetype],
114 struct page, lru);
115}
116
117static inline bool free_area_empty(struct free_area *area, int migratetype)
118{
119 return list_empty(&area->free_list[migratetype]);
120}
121
122struct pglist_data;
123
124/*
125 * Add a wild amount of padding here to ensure data fall into separate
126 * cachelines. There are very few zone structures in the machine, so space
127 * consumption is not a concern here.
128 */
129#if defined(CONFIG_SMP)
130struct zone_padding {
131 char x[0];
132} ____cacheline_internodealigned_in_smp;
133#define ZONE_PADDING(name) struct zone_padding name;
134#else
135#define ZONE_PADDING(name)
136#endif
137
138#ifdef CONFIG_NUMA
139enum numa_stat_item {
140 NUMA_HIT, /* allocated in intended node */
141 NUMA_MISS, /* allocated in non intended node */
142 NUMA_FOREIGN, /* was intended here, hit elsewhere */
143 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
144 NUMA_LOCAL, /* allocation from local node */
145 NUMA_OTHER, /* allocation from other node */
146 NR_VM_NUMA_EVENT_ITEMS
147};
148#else
149#define NR_VM_NUMA_EVENT_ITEMS 0
150#endif
151
152enum zone_stat_item {
153 /* First 128 byte cacheline (assuming 64 bit words) */
154 NR_FREE_PAGES,
155 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
156 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
157 NR_ZONE_ACTIVE_ANON,
158 NR_ZONE_INACTIVE_FILE,
159 NR_ZONE_ACTIVE_FILE,
160 NR_ZONE_UNEVICTABLE,
161 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
162 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
163 /* Second 128 byte cacheline */
164 NR_BOUNCE,
165#if IS_ENABLED(CONFIG_ZSMALLOC)
166 NR_ZSPAGES, /* allocated in zsmalloc */
167#endif
168 NR_FREE_CMA_PAGES,
169 NR_VM_ZONE_STAT_ITEMS };
170
171enum node_stat_item {
172 NR_LRU_BASE,
173 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
174 NR_ACTIVE_ANON, /* " " " " " */
175 NR_INACTIVE_FILE, /* " " " " " */
176 NR_ACTIVE_FILE, /* " " " " " */
177 NR_UNEVICTABLE, /* " " " " " */
178 NR_SLAB_RECLAIMABLE_B,
179 NR_SLAB_UNRECLAIMABLE_B,
180 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
181 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
182 WORKINGSET_NODES,
183 WORKINGSET_REFAULT_BASE,
184 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
185 WORKINGSET_REFAULT_FILE,
186 WORKINGSET_ACTIVATE_BASE,
187 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
188 WORKINGSET_ACTIVATE_FILE,
189 WORKINGSET_RESTORE_BASE,
190 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
191 WORKINGSET_RESTORE_FILE,
192 WORKINGSET_NODERECLAIM,
193 NR_ANON_MAPPED, /* Mapped anonymous pages */
194 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
195 only modified from process context */
196 NR_FILE_PAGES,
197 NR_FILE_DIRTY,
198 NR_WRITEBACK,
199 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
200 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
201 NR_SHMEM_THPS,
202 NR_SHMEM_PMDMAPPED,
203 NR_FILE_THPS,
204 NR_FILE_PMDMAPPED,
205 NR_ANON_THPS,
206 NR_VMSCAN_WRITE,
207 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
208 NR_DIRTIED, /* page dirtyings since bootup */
209 NR_WRITTEN, /* page writings since bootup */
210 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
211 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
212 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
213 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
214 NR_KERNEL_STACK_KB, /* measured in KiB */
215#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
216 NR_KERNEL_SCS_KB, /* measured in KiB */
217#endif
218 NR_PAGETABLE, /* used for pagetables */
219#ifdef CONFIG_SWAP
220 NR_SWAPCACHE,
221#endif
222#ifdef CONFIG_NUMA_BALANCING
223 PGPROMOTE_SUCCESS, /* promote successfully */
224#endif
225 NR_VM_NODE_STAT_ITEMS
226};
227
228/*
229 * Returns true if the item should be printed in THPs (/proc/vmstat
230 * currently prints number of anon, file and shmem THPs. But the item
231 * is charged in pages).
232 */
233static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
234{
235 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
236 return false;
237
238 return item == NR_ANON_THPS ||
239 item == NR_FILE_THPS ||
240 item == NR_SHMEM_THPS ||
241 item == NR_SHMEM_PMDMAPPED ||
242 item == NR_FILE_PMDMAPPED;
243}
244
245/*
246 * Returns true if the value is measured in bytes (most vmstat values are
247 * measured in pages). This defines the API part, the internal representation
248 * might be different.
249 */
250static __always_inline bool vmstat_item_in_bytes(int idx)
251{
252 /*
253 * Global and per-node slab counters track slab pages.
254 * It's expected that changes are multiples of PAGE_SIZE.
255 * Internally values are stored in pages.
256 *
257 * Per-memcg and per-lruvec counters track memory, consumed
258 * by individual slab objects. These counters are actually
259 * byte-precise.
260 */
261 return (idx == NR_SLAB_RECLAIMABLE_B ||
262 idx == NR_SLAB_UNRECLAIMABLE_B);
263}
264
265/*
266 * We do arithmetic on the LRU lists in various places in the code,
267 * so it is important to keep the active lists LRU_ACTIVE higher in
268 * the array than the corresponding inactive lists, and to keep
269 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
270 *
271 * This has to be kept in sync with the statistics in zone_stat_item
272 * above and the descriptions in vmstat_text in mm/vmstat.c
273 */
274#define LRU_BASE 0
275#define LRU_ACTIVE 1
276#define LRU_FILE 2
277
278enum lru_list {
279 LRU_INACTIVE_ANON = LRU_BASE,
280 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
281 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
282 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
283 LRU_UNEVICTABLE,
284 NR_LRU_LISTS
285};
286
287enum vmscan_throttle_state {
288 VMSCAN_THROTTLE_WRITEBACK,
289 VMSCAN_THROTTLE_ISOLATED,
290 VMSCAN_THROTTLE_NOPROGRESS,
291 VMSCAN_THROTTLE_CONGESTED,
292 NR_VMSCAN_THROTTLE,
293};
294
295#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
296
297#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
298
299static inline bool is_file_lru(enum lru_list lru)
300{
301 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
302}
303
304static inline bool is_active_lru(enum lru_list lru)
305{
306 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
307}
308
309#define ANON_AND_FILE 2
310
311enum lruvec_flags {
312 LRUVEC_CONGESTED, /* lruvec has many dirty pages
313 * backed by a congested BDI
314 */
315};
316
317struct lruvec {
318 struct list_head lists[NR_LRU_LISTS];
319 /* per lruvec lru_lock for memcg */
320 spinlock_t lru_lock;
321 /*
322 * These track the cost of reclaiming one LRU - file or anon -
323 * over the other. As the observed cost of reclaiming one LRU
324 * increases, the reclaim scan balance tips toward the other.
325 */
326 unsigned long anon_cost;
327 unsigned long file_cost;
328 /* Non-resident age, driven by LRU movement */
329 atomic_long_t nonresident_age;
330 /* Refaults at the time of last reclaim cycle */
331 unsigned long refaults[ANON_AND_FILE];
332 /* Various lruvec state flags (enum lruvec_flags) */
333 unsigned long flags;
334#ifdef CONFIG_MEMCG
335 struct pglist_data *pgdat;
336#endif
337};
338
339/* Isolate unmapped pages */
340#define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
341/* Isolate for asynchronous migration */
342#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
343/* Isolate unevictable pages */
344#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
345
346/* LRU Isolation modes. */
347typedef unsigned __bitwise isolate_mode_t;
348
349enum zone_watermarks {
350 WMARK_MIN,
351 WMARK_LOW,
352 WMARK_HIGH,
353 WMARK_PROMO,
354 NR_WMARK
355};
356
357/*
358 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
359 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
360 * it should not contribute to serious fragmentation causing THP allocation
361 * failures.
362 */
363#ifdef CONFIG_TRANSPARENT_HUGEPAGE
364#define NR_PCP_THP 1
365#else
366#define NR_PCP_THP 0
367#endif
368#define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
369#define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
370
371/*
372 * Shift to encode migratetype and order in the same integer, with order
373 * in the least significant bits.
374 */
375#define NR_PCP_ORDER_WIDTH 8
376#define NR_PCP_ORDER_MASK ((1<<NR_PCP_ORDER_WIDTH) - 1)
377
378#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
379#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
380#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
381#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
382
383/* Fields and list protected by pagesets local_lock in page_alloc.c */
384struct per_cpu_pages {
385 spinlock_t lock; /* Protects lists field */
386 int count; /* number of pages in the list */
387 int high; /* high watermark, emptying needed */
388 int batch; /* chunk size for buddy add/remove */
389 short free_factor; /* batch scaling factor during free */
390#ifdef CONFIG_NUMA
391 short expire; /* When 0, remote pagesets are drained */
392#endif
393
394 /* Lists of pages, one per migrate type stored on the pcp-lists */
395 struct list_head lists[NR_PCP_LISTS];
396} ____cacheline_aligned_in_smp;
397
398struct per_cpu_zonestat {
399#ifdef CONFIG_SMP
400 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
401 s8 stat_threshold;
402#endif
403#ifdef CONFIG_NUMA
404 /*
405 * Low priority inaccurate counters that are only folded
406 * on demand. Use a large type to avoid the overhead of
407 * folding during refresh_cpu_vm_stats.
408 */
409 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
410#endif
411};
412
413struct per_cpu_nodestat {
414 s8 stat_threshold;
415 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
416};
417
418#endif /* !__GENERATING_BOUNDS.H */
419
420enum zone_type {
421 /*
422 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
423 * to DMA to all of the addressable memory (ZONE_NORMAL).
424 * On architectures where this area covers the whole 32 bit address
425 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
426 * DMA addressing constraints. This distinction is important as a 32bit
427 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
428 * platforms may need both zones as they support peripherals with
429 * different DMA addressing limitations.
430 */
431#ifdef CONFIG_ZONE_DMA
432 ZONE_DMA,
433#endif
434#ifdef CONFIG_ZONE_DMA32
435 ZONE_DMA32,
436#endif
437 /*
438 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
439 * performed on pages in ZONE_NORMAL if the DMA devices support
440 * transfers to all addressable memory.
441 */
442 ZONE_NORMAL,
443#ifdef CONFIG_HIGHMEM
444 /*
445 * A memory area that is only addressable by the kernel through
446 * mapping portions into its own address space. This is for example
447 * used by i386 to allow the kernel to address the memory beyond
448 * 900MB. The kernel will set up special mappings (page
449 * table entries on i386) for each page that the kernel needs to
450 * access.
451 */
452 ZONE_HIGHMEM,
453#endif
454 /*
455 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
456 * movable pages with few exceptional cases described below. Main use
457 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
458 * likely to succeed, and to locally limit unmovable allocations - e.g.,
459 * to increase the number of THP/huge pages. Notable special cases are:
460 *
461 * 1. Pinned pages: (long-term) pinning of movable pages might
462 * essentially turn such pages unmovable. Therefore, we do not allow
463 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
464 * faulted, they come from the right zone right away. However, it is
465 * still possible that address space already has pages in
466 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
467 * touches that memory before pinning). In such case we migrate them
468 * to a different zone. When migration fails - pinning fails.
469 * 2. memblock allocations: kernelcore/movablecore setups might create
470 * situations where ZONE_MOVABLE contains unmovable allocations
471 * after boot. Memory offlining and allocations fail early.
472 * 3. Memory holes: kernelcore/movablecore setups might create very rare
473 * situations where ZONE_MOVABLE contains memory holes after boot,
474 * for example, if we have sections that are only partially
475 * populated. Memory offlining and allocations fail early.
476 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
477 * memory offlining, such pages cannot be allocated.
478 * 5. Unmovable PG_offline pages: in paravirtualized environments,
479 * hotplugged memory blocks might only partially be managed by the
480 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
481 * parts not manged by the buddy are unmovable PG_offline pages. In
482 * some cases (virtio-mem), such pages can be skipped during
483 * memory offlining, however, cannot be moved/allocated. These
484 * techniques might use alloc_contig_range() to hide previously
485 * exposed pages from the buddy again (e.g., to implement some sort
486 * of memory unplug in virtio-mem).
487 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
488 * situations where ZERO_PAGE(0) which is allocated differently
489 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
490 * cannot be migrated.
491 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
492 * memory to the MOVABLE zone, the vmemmap pages are also placed in
493 * such zone. Such pages cannot be really moved around as they are
494 * self-stored in the range, but they are treated as movable when
495 * the range they describe is about to be offlined.
496 *
497 * In general, no unmovable allocations that degrade memory offlining
498 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
499 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
500 * if has_unmovable_pages() states that there are no unmovable pages,
501 * there can be false negatives).
502 */
503 ZONE_MOVABLE,
504#ifdef CONFIG_ZONE_DEVICE
505 ZONE_DEVICE,
506#endif
507 __MAX_NR_ZONES
508
509};
510
511#ifndef __GENERATING_BOUNDS_H
512
513#define ASYNC_AND_SYNC 2
514
515struct zone {
516 /* Read-mostly fields */
517
518 /* zone watermarks, access with *_wmark_pages(zone) macros */
519 unsigned long _watermark[NR_WMARK];
520 unsigned long watermark_boost;
521
522 unsigned long nr_reserved_highatomic;
523
524 /*
525 * We don't know if the memory that we're going to allocate will be
526 * freeable or/and it will be released eventually, so to avoid totally
527 * wasting several GB of ram we must reserve some of the lower zone
528 * memory (otherwise we risk to run OOM on the lower zones despite
529 * there being tons of freeable ram on the higher zones). This array is
530 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
531 * changes.
532 */
533 long lowmem_reserve[MAX_NR_ZONES];
534
535#ifdef CONFIG_NUMA
536 int node;
537#endif
538 struct pglist_data *zone_pgdat;
539 struct per_cpu_pages __percpu *per_cpu_pageset;
540 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
541 /*
542 * the high and batch values are copied to individual pagesets for
543 * faster access
544 */
545 int pageset_high;
546 int pageset_batch;
547
548#ifndef CONFIG_SPARSEMEM
549 /*
550 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
551 * In SPARSEMEM, this map is stored in struct mem_section
552 */
553 unsigned long *pageblock_flags;
554#endif /* CONFIG_SPARSEMEM */
555
556 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
557 unsigned long zone_start_pfn;
558
559 /*
560 * spanned_pages is the total pages spanned by the zone, including
561 * holes, which is calculated as:
562 * spanned_pages = zone_end_pfn - zone_start_pfn;
563 *
564 * present_pages is physical pages existing within the zone, which
565 * is calculated as:
566 * present_pages = spanned_pages - absent_pages(pages in holes);
567 *
568 * present_early_pages is present pages existing within the zone
569 * located on memory available since early boot, excluding hotplugged
570 * memory.
571 *
572 * managed_pages is present pages managed by the buddy system, which
573 * is calculated as (reserved_pages includes pages allocated by the
574 * bootmem allocator):
575 * managed_pages = present_pages - reserved_pages;
576 *
577 * cma pages is present pages that are assigned for CMA use
578 * (MIGRATE_CMA).
579 *
580 * So present_pages may be used by memory hotplug or memory power
581 * management logic to figure out unmanaged pages by checking
582 * (present_pages - managed_pages). And managed_pages should be used
583 * by page allocator and vm scanner to calculate all kinds of watermarks
584 * and thresholds.
585 *
586 * Locking rules:
587 *
588 * zone_start_pfn and spanned_pages are protected by span_seqlock.
589 * It is a seqlock because it has to be read outside of zone->lock,
590 * and it is done in the main allocator path. But, it is written
591 * quite infrequently.
592 *
593 * The span_seq lock is declared along with zone->lock because it is
594 * frequently read in proximity to zone->lock. It's good to
595 * give them a chance of being in the same cacheline.
596 *
597 * Write access to present_pages at runtime should be protected by
598 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
599 * present_pages should use get_online_mems() to get a stable value.
600 */
601 atomic_long_t managed_pages;
602 unsigned long spanned_pages;
603 unsigned long present_pages;
604#if defined(CONFIG_MEMORY_HOTPLUG)
605 unsigned long present_early_pages;
606#endif
607#ifdef CONFIG_CMA
608 unsigned long cma_pages;
609#endif
610
611 const char *name;
612
613#ifdef CONFIG_MEMORY_ISOLATION
614 /*
615 * Number of isolated pageblock. It is used to solve incorrect
616 * freepage counting problem due to racy retrieving migratetype
617 * of pageblock. Protected by zone->lock.
618 */
619 unsigned long nr_isolate_pageblock;
620#endif
621
622#ifdef CONFIG_MEMORY_HOTPLUG
623 /* see spanned/present_pages for more description */
624 seqlock_t span_seqlock;
625#endif
626
627 int initialized;
628
629 /* Write-intensive fields used from the page allocator */
630 ZONE_PADDING(_pad1_)
631
632 /* free areas of different sizes */
633 struct free_area free_area[MAX_ORDER];
634
635 /* zone flags, see below */
636 unsigned long flags;
637
638 /* Primarily protects free_area */
639 spinlock_t lock;
640
641 /* Write-intensive fields used by compaction and vmstats. */
642 ZONE_PADDING(_pad2_)
643
644 /*
645 * When free pages are below this point, additional steps are taken
646 * when reading the number of free pages to avoid per-cpu counter
647 * drift allowing watermarks to be breached
648 */
649 unsigned long percpu_drift_mark;
650
651#if defined CONFIG_COMPACTION || defined CONFIG_CMA
652 /* pfn where compaction free scanner should start */
653 unsigned long compact_cached_free_pfn;
654 /* pfn where compaction migration scanner should start */
655 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
656 unsigned long compact_init_migrate_pfn;
657 unsigned long compact_init_free_pfn;
658#endif
659
660#ifdef CONFIG_COMPACTION
661 /*
662 * On compaction failure, 1<<compact_defer_shift compactions
663 * are skipped before trying again. The number attempted since
664 * last failure is tracked with compact_considered.
665 * compact_order_failed is the minimum compaction failed order.
666 */
667 unsigned int compact_considered;
668 unsigned int compact_defer_shift;
669 int compact_order_failed;
670#endif
671
672#if defined CONFIG_COMPACTION || defined CONFIG_CMA
673 /* Set to true when the PG_migrate_skip bits should be cleared */
674 bool compact_blockskip_flush;
675#endif
676
677 bool contiguous;
678
679 ZONE_PADDING(_pad3_)
680 /* Zone statistics */
681 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
682 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
683} ____cacheline_internodealigned_in_smp;
684
685enum pgdat_flags {
686 PGDAT_DIRTY, /* reclaim scanning has recently found
687 * many dirty file pages at the tail
688 * of the LRU.
689 */
690 PGDAT_WRITEBACK, /* reclaim scanning has recently found
691 * many pages under writeback
692 */
693 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
694};
695
696enum zone_flags {
697 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
698 * Cleared when kswapd is woken.
699 */
700 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
701};
702
703static inline unsigned long zone_managed_pages(struct zone *zone)
704{
705 return (unsigned long)atomic_long_read(&zone->managed_pages);
706}
707
708static inline unsigned long zone_cma_pages(struct zone *zone)
709{
710#ifdef CONFIG_CMA
711 return zone->cma_pages;
712#else
713 return 0;
714#endif
715}
716
717static inline unsigned long zone_end_pfn(const struct zone *zone)
718{
719 return zone->zone_start_pfn + zone->spanned_pages;
720}
721
722static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
723{
724 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
725}
726
727static inline bool zone_is_initialized(struct zone *zone)
728{
729 return zone->initialized;
730}
731
732static inline bool zone_is_empty(struct zone *zone)
733{
734 return zone->spanned_pages == 0;
735}
736
737#ifndef BUILD_VDSO32_64
738/*
739 * The zone field is never updated after free_area_init_core()
740 * sets it, so none of the operations on it need to be atomic.
741 */
742
743/* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
744#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
745#define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
746#define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
747#define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
748#define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
749
750/*
751 * Define the bit shifts to access each section. For non-existent
752 * sections we define the shift as 0; that plus a 0 mask ensures
753 * the compiler will optimise away reference to them.
754 */
755#define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
756#define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
757#define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
758#define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
759#define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
760
761/* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
762#ifdef NODE_NOT_IN_PAGE_FLAGS
763#define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
764#define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
765 SECTIONS_PGOFF : ZONES_PGOFF)
766#else
767#define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
768#define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
769 NODES_PGOFF : ZONES_PGOFF)
770#endif
771
772#define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
773
774#define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
775#define NODES_MASK ((1UL << NODES_WIDTH) - 1)
776#define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
777#define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
778#define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
779#define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
780
781static inline enum zone_type page_zonenum(const struct page *page)
782{
783 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
784 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
785}
786
787static inline enum zone_type folio_zonenum(const struct folio *folio)
788{
789 return page_zonenum(&folio->page);
790}
791
792#ifdef CONFIG_ZONE_DEVICE
793static inline bool is_zone_device_page(const struct page *page)
794{
795 return page_zonenum(page) == ZONE_DEVICE;
796}
797extern void memmap_init_zone_device(struct zone *, unsigned long,
798 unsigned long, struct dev_pagemap *);
799#else
800static inline bool is_zone_device_page(const struct page *page)
801{
802 return false;
803}
804#endif
805
806static inline bool folio_is_zone_device(const struct folio *folio)
807{
808 return is_zone_device_page(&folio->page);
809}
810
811static inline bool is_zone_movable_page(const struct page *page)
812{
813 return page_zonenum(page) == ZONE_MOVABLE;
814}
815#endif
816
817/*
818 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
819 * intersection with the given zone
820 */
821static inline bool zone_intersects(struct zone *zone,
822 unsigned long start_pfn, unsigned long nr_pages)
823{
824 if (zone_is_empty(zone))
825 return false;
826 if (start_pfn >= zone_end_pfn(zone) ||
827 start_pfn + nr_pages <= zone->zone_start_pfn)
828 return false;
829
830 return true;
831}
832
833/*
834 * The "priority" of VM scanning is how much of the queues we will scan in one
835 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
836 * queues ("queue_length >> 12") during an aging round.
837 */
838#define DEF_PRIORITY 12
839
840/* Maximum number of zones on a zonelist */
841#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
842
843enum {
844 ZONELIST_FALLBACK, /* zonelist with fallback */
845#ifdef CONFIG_NUMA
846 /*
847 * The NUMA zonelists are doubled because we need zonelists that
848 * restrict the allocations to a single node for __GFP_THISNODE.
849 */
850 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
851#endif
852 MAX_ZONELISTS
853};
854
855/*
856 * This struct contains information about a zone in a zonelist. It is stored
857 * here to avoid dereferences into large structures and lookups of tables
858 */
859struct zoneref {
860 struct zone *zone; /* Pointer to actual zone */
861 int zone_idx; /* zone_idx(zoneref->zone) */
862};
863
864/*
865 * One allocation request operates on a zonelist. A zonelist
866 * is a list of zones, the first one is the 'goal' of the
867 * allocation, the other zones are fallback zones, in decreasing
868 * priority.
869 *
870 * To speed the reading of the zonelist, the zonerefs contain the zone index
871 * of the entry being read. Helper functions to access information given
872 * a struct zoneref are
873 *
874 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
875 * zonelist_zone_idx() - Return the index of the zone for an entry
876 * zonelist_node_idx() - Return the index of the node for an entry
877 */
878struct zonelist {
879 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
880};
881
882/*
883 * The array of struct pages for flatmem.
884 * It must be declared for SPARSEMEM as well because there are configurations
885 * that rely on that.
886 */
887extern struct page *mem_map;
888
889#ifdef CONFIG_TRANSPARENT_HUGEPAGE
890struct deferred_split {
891 spinlock_t split_queue_lock;
892 struct list_head split_queue;
893 unsigned long split_queue_len;
894};
895#endif
896
897/*
898 * On NUMA machines, each NUMA node would have a pg_data_t to describe
899 * it's memory layout. On UMA machines there is a single pglist_data which
900 * describes the whole memory.
901 *
902 * Memory statistics and page replacement data structures are maintained on a
903 * per-zone basis.
904 */
905typedef struct pglist_data {
906 /*
907 * node_zones contains just the zones for THIS node. Not all of the
908 * zones may be populated, but it is the full list. It is referenced by
909 * this node's node_zonelists as well as other node's node_zonelists.
910 */
911 struct zone node_zones[MAX_NR_ZONES];
912
913 /*
914 * node_zonelists contains references to all zones in all nodes.
915 * Generally the first zones will be references to this node's
916 * node_zones.
917 */
918 struct zonelist node_zonelists[MAX_ZONELISTS];
919
920 int nr_zones; /* number of populated zones in this node */
921#ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
922 struct page *node_mem_map;
923#ifdef CONFIG_PAGE_EXTENSION
924 struct page_ext *node_page_ext;
925#endif
926#endif
927#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
928 /*
929 * Must be held any time you expect node_start_pfn,
930 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
931 * Also synchronizes pgdat->first_deferred_pfn during deferred page
932 * init.
933 *
934 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
935 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
936 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
937 *
938 * Nests above zone->lock and zone->span_seqlock
939 */
940 spinlock_t node_size_lock;
941#endif
942 unsigned long node_start_pfn;
943 unsigned long node_present_pages; /* total number of physical pages */
944 unsigned long node_spanned_pages; /* total size of physical page
945 range, including holes */
946 int node_id;
947 wait_queue_head_t kswapd_wait;
948 wait_queue_head_t pfmemalloc_wait;
949
950 /* workqueues for throttling reclaim for different reasons. */
951 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
952
953 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
954 unsigned long nr_reclaim_start; /* nr pages written while throttled
955 * when throttling started. */
956 struct task_struct *kswapd; /* Protected by
957 mem_hotplug_begin/done() */
958 int kswapd_order;
959 enum zone_type kswapd_highest_zoneidx;
960
961 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
962
963#ifdef CONFIG_COMPACTION
964 int kcompactd_max_order;
965 enum zone_type kcompactd_highest_zoneidx;
966 wait_queue_head_t kcompactd_wait;
967 struct task_struct *kcompactd;
968 bool proactive_compact_trigger;
969#endif
970 /*
971 * This is a per-node reserve of pages that are not available
972 * to userspace allocations.
973 */
974 unsigned long totalreserve_pages;
975
976#ifdef CONFIG_NUMA
977 /*
978 * node reclaim becomes active if more unmapped pages exist.
979 */
980 unsigned long min_unmapped_pages;
981 unsigned long min_slab_pages;
982#endif /* CONFIG_NUMA */
983
984 /* Write-intensive fields used by page reclaim */
985 ZONE_PADDING(_pad1_)
986
987#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
988 /*
989 * If memory initialisation on large machines is deferred then this
990 * is the first PFN that needs to be initialised.
991 */
992 unsigned long first_deferred_pfn;
993#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
994
995#ifdef CONFIG_TRANSPARENT_HUGEPAGE
996 struct deferred_split deferred_split_queue;
997#endif
998
999 /* Fields commonly accessed by the page reclaim scanner */
1000
1001 /*
1002 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1003 *
1004 * Use mem_cgroup_lruvec() to look up lruvecs.
1005 */
1006 struct lruvec __lruvec;
1007
1008 unsigned long flags;
1009
1010 ZONE_PADDING(_pad2_)
1011
1012 /* Per-node vmstats */
1013 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1014 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1015} pg_data_t;
1016
1017#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1018#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1019
1020#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1021#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1022
1023static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1024{
1025 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1026}
1027
1028static inline bool pgdat_is_empty(pg_data_t *pgdat)
1029{
1030 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
1031}
1032
1033#include <linux/memory_hotplug.h>
1034
1035void build_all_zonelists(pg_data_t *pgdat);
1036void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1037 enum zone_type highest_zoneidx);
1038bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1039 int highest_zoneidx, unsigned int alloc_flags,
1040 long free_pages);
1041bool zone_watermark_ok(struct zone *z, unsigned int order,
1042 unsigned long mark, int highest_zoneidx,
1043 unsigned int alloc_flags);
1044bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1045 unsigned long mark, int highest_zoneidx);
1046/*
1047 * Memory initialization context, use to differentiate memory added by
1048 * the platform statically or via memory hotplug interface.
1049 */
1050enum meminit_context {
1051 MEMINIT_EARLY,
1052 MEMINIT_HOTPLUG,
1053};
1054
1055extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1056 unsigned long size);
1057
1058extern void lruvec_init(struct lruvec *lruvec);
1059
1060static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1061{
1062#ifdef CONFIG_MEMCG
1063 return lruvec->pgdat;
1064#else
1065 return container_of(lruvec, struct pglist_data, __lruvec);
1066#endif
1067}
1068
1069#ifdef CONFIG_HAVE_MEMORYLESS_NODES
1070int local_memory_node(int node_id);
1071#else
1072static inline int local_memory_node(int node_id) { return node_id; };
1073#endif
1074
1075/*
1076 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1077 */
1078#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1079
1080#ifdef CONFIG_ZONE_DEVICE
1081static inline bool zone_is_zone_device(struct zone *zone)
1082{
1083 return zone_idx(zone) == ZONE_DEVICE;
1084}
1085#else
1086static inline bool zone_is_zone_device(struct zone *zone)
1087{
1088 return false;
1089}
1090#endif
1091
1092/*
1093 * Returns true if a zone has pages managed by the buddy allocator.
1094 * All the reclaim decisions have to use this function rather than
1095 * populated_zone(). If the whole zone is reserved then we can easily
1096 * end up with populated_zone() && !managed_zone().
1097 */
1098static inline bool managed_zone(struct zone *zone)
1099{
1100 return zone_managed_pages(zone);
1101}
1102
1103/* Returns true if a zone has memory */
1104static inline bool populated_zone(struct zone *zone)
1105{
1106 return zone->present_pages;
1107}
1108
1109#ifdef CONFIG_NUMA
1110static inline int zone_to_nid(struct zone *zone)
1111{
1112 return zone->node;
1113}
1114
1115static inline void zone_set_nid(struct zone *zone, int nid)
1116{
1117 zone->node = nid;
1118}
1119#else
1120static inline int zone_to_nid(struct zone *zone)
1121{
1122 return 0;
1123}
1124
1125static inline void zone_set_nid(struct zone *zone, int nid) {}
1126#endif
1127
1128extern int movable_zone;
1129
1130static inline int is_highmem_idx(enum zone_type idx)
1131{
1132#ifdef CONFIG_HIGHMEM
1133 return (idx == ZONE_HIGHMEM ||
1134 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1135#else
1136 return 0;
1137#endif
1138}
1139
1140/**
1141 * is_highmem - helper function to quickly check if a struct zone is a
1142 * highmem zone or not. This is an attempt to keep references
1143 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1144 * @zone: pointer to struct zone variable
1145 * Return: 1 for a highmem zone, 0 otherwise
1146 */
1147static inline int is_highmem(struct zone *zone)
1148{
1149 return is_highmem_idx(zone_idx(zone));
1150}
1151
1152#ifdef CONFIG_ZONE_DMA
1153bool has_managed_dma(void);
1154#else
1155static inline bool has_managed_dma(void)
1156{
1157 return false;
1158}
1159#endif
1160
1161/* These two functions are used to setup the per zone pages min values */
1162struct ctl_table;
1163
1164int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1165 loff_t *);
1166int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1167 size_t *, loff_t *);
1168extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1169int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1170 size_t *, loff_t *);
1171int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1172 void *, size_t *, loff_t *);
1173int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1174 void *, size_t *, loff_t *);
1175int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1176 void *, size_t *, loff_t *);
1177int numa_zonelist_order_handler(struct ctl_table *, int,
1178 void *, size_t *, loff_t *);
1179extern int percpu_pagelist_high_fraction;
1180extern char numa_zonelist_order[];
1181#define NUMA_ZONELIST_ORDER_LEN 16
1182
1183#ifndef CONFIG_NUMA
1184
1185extern struct pglist_data contig_page_data;
1186static inline struct pglist_data *NODE_DATA(int nid)
1187{
1188 return &contig_page_data;
1189}
1190
1191#else /* CONFIG_NUMA */
1192
1193#include <asm/mmzone.h>
1194
1195#endif /* !CONFIG_NUMA */
1196
1197extern struct pglist_data *first_online_pgdat(void);
1198extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1199extern struct zone *next_zone(struct zone *zone);
1200
1201/**
1202 * for_each_online_pgdat - helper macro to iterate over all online nodes
1203 * @pgdat: pointer to a pg_data_t variable
1204 */
1205#define for_each_online_pgdat(pgdat) \
1206 for (pgdat = first_online_pgdat(); \
1207 pgdat; \
1208 pgdat = next_online_pgdat(pgdat))
1209/**
1210 * for_each_zone - helper macro to iterate over all memory zones
1211 * @zone: pointer to struct zone variable
1212 *
1213 * The user only needs to declare the zone variable, for_each_zone
1214 * fills it in.
1215 */
1216#define for_each_zone(zone) \
1217 for (zone = (first_online_pgdat())->node_zones; \
1218 zone; \
1219 zone = next_zone(zone))
1220
1221#define for_each_populated_zone(zone) \
1222 for (zone = (first_online_pgdat())->node_zones; \
1223 zone; \
1224 zone = next_zone(zone)) \
1225 if (!populated_zone(zone)) \
1226 ; /* do nothing */ \
1227 else
1228
1229static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1230{
1231 return zoneref->zone;
1232}
1233
1234static inline int zonelist_zone_idx(struct zoneref *zoneref)
1235{
1236 return zoneref->zone_idx;
1237}
1238
1239static inline int zonelist_node_idx(struct zoneref *zoneref)
1240{
1241 return zone_to_nid(zoneref->zone);
1242}
1243
1244struct zoneref *__next_zones_zonelist(struct zoneref *z,
1245 enum zone_type highest_zoneidx,
1246 nodemask_t *nodes);
1247
1248/**
1249 * 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
1250 * @z: The cursor used as a starting point for the search
1251 * @highest_zoneidx: The zone index of the highest zone to return
1252 * @nodes: An optional nodemask to filter the zonelist with
1253 *
1254 * This function returns the next zone at or below a given zone index that is
1255 * within the allowed nodemask using a cursor as the starting point for the
1256 * search. The zoneref returned is a cursor that represents the current zone
1257 * being examined. It should be advanced by one before calling
1258 * next_zones_zonelist again.
1259 *
1260 * Return: the next zone at or below highest_zoneidx within the allowed
1261 * nodemask using a cursor within a zonelist as a starting point
1262 */
1263static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1264 enum zone_type highest_zoneidx,
1265 nodemask_t *nodes)
1266{
1267 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1268 return z;
1269 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1270}
1271
1272/**
1273 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1274 * @zonelist: The zonelist to search for a suitable zone
1275 * @highest_zoneidx: The zone index of the highest zone to return
1276 * @nodes: An optional nodemask to filter the zonelist with
1277 *
1278 * This function returns the first zone at or below a given zone index that is
1279 * within the allowed nodemask. The zoneref returned is a cursor that can be
1280 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1281 * one before calling.
1282 *
1283 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1284 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1285 * update due to cpuset modification.
1286 *
1287 * Return: Zoneref pointer for the first suitable zone found
1288 */
1289static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1290 enum zone_type highest_zoneidx,
1291 nodemask_t *nodes)
1292{
1293 return next_zones_zonelist(zonelist->_zonerefs,
1294 highest_zoneidx, nodes);
1295}
1296
1297/**
1298 * 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
1299 * @zone: The current zone in the iterator
1300 * @z: The current pointer within zonelist->_zonerefs being iterated
1301 * @zlist: The zonelist being iterated
1302 * @highidx: The zone index of the highest zone to return
1303 * @nodemask: Nodemask allowed by the allocator
1304 *
1305 * This iterator iterates though all zones at or below a given zone index and
1306 * within a given nodemask
1307 */
1308#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1309 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1310 zone; \
1311 z = next_zones_zonelist(++z, highidx, nodemask), \
1312 zone = zonelist_zone(z))
1313
1314#define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1315 for (zone = z->zone; \
1316 zone; \
1317 z = next_zones_zonelist(++z, highidx, nodemask), \
1318 zone = zonelist_zone(z))
1319
1320
1321/**
1322 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1323 * @zone: The current zone in the iterator
1324 * @z: The current pointer within zonelist->zones being iterated
1325 * @zlist: The zonelist being iterated
1326 * @highidx: The zone index of the highest zone to return
1327 *
1328 * This iterator iterates though all zones at or below a given zone index.
1329 */
1330#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1331 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1332
1333/* Whether the 'nodes' are all movable nodes */
1334static inline bool movable_only_nodes(nodemask_t *nodes)
1335{
1336 struct zonelist *zonelist;
1337 struct zoneref *z;
1338 int nid;
1339
1340 if (nodes_empty(*nodes))
1341 return false;
1342
1343 /*
1344 * We can chose arbitrary node from the nodemask to get a
1345 * zonelist as they are interlinked. We just need to find
1346 * at least one zone that can satisfy kernel allocations.
1347 */
1348 nid = first_node(*nodes);
1349 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1350 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1351 return (!z->zone) ? true : false;
1352}
1353
1354
1355#ifdef CONFIG_SPARSEMEM
1356#include <asm/sparsemem.h>
1357#endif
1358
1359#ifdef CONFIG_FLATMEM
1360#define pfn_to_nid(pfn) (0)
1361#endif
1362
1363#ifdef CONFIG_SPARSEMEM
1364
1365/*
1366 * PA_SECTION_SHIFT physical address to/from section number
1367 * PFN_SECTION_SHIFT pfn to/from section number
1368 */
1369#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1370#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1371
1372#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1373
1374#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1375#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1376
1377#define SECTION_BLOCKFLAGS_BITS \
1378 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1379
1380#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1381#error Allocator MAX_ORDER exceeds SECTION_SIZE
1382#endif
1383
1384static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1385{
1386 return pfn >> PFN_SECTION_SHIFT;
1387}
1388static inline unsigned long section_nr_to_pfn(unsigned long sec)
1389{
1390 return sec << PFN_SECTION_SHIFT;
1391}
1392
1393#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1394#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1395
1396#define SUBSECTION_SHIFT 21
1397#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1398
1399#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1400#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1401#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1402
1403#if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1404#error Subsection size exceeds section size
1405#else
1406#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1407#endif
1408
1409#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1410#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1411
1412struct mem_section_usage {
1413#ifdef CONFIG_SPARSEMEM_VMEMMAP
1414 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1415#endif
1416 /* See declaration of similar field in struct zone */
1417 unsigned long pageblock_flags[0];
1418};
1419
1420void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1421
1422struct page;
1423struct page_ext;
1424struct mem_section {
1425 /*
1426 * This is, logically, a pointer to an array of struct
1427 * pages. However, it is stored with some other magic.
1428 * (see sparse.c::sparse_init_one_section())
1429 *
1430 * Additionally during early boot we encode node id of
1431 * the location of the section here to guide allocation.
1432 * (see sparse.c::memory_present())
1433 *
1434 * Making it a UL at least makes someone do a cast
1435 * before using it wrong.
1436 */
1437 unsigned long section_mem_map;
1438
1439 struct mem_section_usage *usage;
1440#ifdef CONFIG_PAGE_EXTENSION
1441 /*
1442 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1443 * section. (see page_ext.h about this.)
1444 */
1445 struct page_ext *page_ext;
1446 unsigned long pad;
1447#endif
1448 /*
1449 * WARNING: mem_section must be a power-of-2 in size for the
1450 * calculation and use of SECTION_ROOT_MASK to make sense.
1451 */
1452};
1453
1454#ifdef CONFIG_SPARSEMEM_EXTREME
1455#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1456#else
1457#define SECTIONS_PER_ROOT 1
1458#endif
1459
1460#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1461#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1462#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1463
1464#ifdef CONFIG_SPARSEMEM_EXTREME
1465extern struct mem_section **mem_section;
1466#else
1467extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1468#endif
1469
1470static inline unsigned long *section_to_usemap(struct mem_section *ms)
1471{
1472 return ms->usage->pageblock_flags;
1473}
1474
1475static inline struct mem_section *__nr_to_section(unsigned long nr)
1476{
1477 unsigned long root = SECTION_NR_TO_ROOT(nr);
1478
1479 if (unlikely(root >= NR_SECTION_ROOTS))
1480 return NULL;
1481
1482#ifdef CONFIG_SPARSEMEM_EXTREME
1483 if (!mem_section || !mem_section[root])
1484 return NULL;
1485#endif
1486 return &mem_section[root][nr & SECTION_ROOT_MASK];
1487}
1488extern size_t mem_section_usage_size(void);
1489
1490/*
1491 * We use the lower bits of the mem_map pointer to store
1492 * a little bit of information. The pointer is calculated
1493 * as mem_map - section_nr_to_pfn(pnum). The result is
1494 * aligned to the minimum alignment of the two values:
1495 * 1. All mem_map arrays are page-aligned.
1496 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1497 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1498 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1499 * worst combination is powerpc with 256k pages,
1500 * which results in PFN_SECTION_SHIFT equal 6.
1501 * To sum it up, at least 6 bits are available on all architectures.
1502 * However, we can exceed 6 bits on some other architectures except
1503 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1504 * with the worst case of 64K pages on arm64) if we make sure the
1505 * exceeded bit is not applicable to powerpc.
1506 */
1507enum {
1508 SECTION_MARKED_PRESENT_BIT,
1509 SECTION_HAS_MEM_MAP_BIT,
1510 SECTION_IS_ONLINE_BIT,
1511 SECTION_IS_EARLY_BIT,
1512#ifdef CONFIG_ZONE_DEVICE
1513 SECTION_TAINT_ZONE_DEVICE_BIT,
1514#endif
1515 SECTION_MAP_LAST_BIT,
1516};
1517
1518#define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1519#define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1520#define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1521#define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1522#ifdef CONFIG_ZONE_DEVICE
1523#define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1524#endif
1525#define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1526#define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1527
1528static inline struct page *__section_mem_map_addr(struct mem_section *section)
1529{
1530 unsigned long map = section->section_mem_map;
1531 map &= SECTION_MAP_MASK;
1532 return (struct page *)map;
1533}
1534
1535static inline int present_section(struct mem_section *section)
1536{
1537 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1538}
1539
1540static inline int present_section_nr(unsigned long nr)
1541{
1542 return present_section(__nr_to_section(nr));
1543}
1544
1545static inline int valid_section(struct mem_section *section)
1546{
1547 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1548}
1549
1550static inline int early_section(struct mem_section *section)
1551{
1552 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1553}
1554
1555static inline int valid_section_nr(unsigned long nr)
1556{
1557 return valid_section(__nr_to_section(nr));
1558}
1559
1560static inline int online_section(struct mem_section *section)
1561{
1562 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1563}
1564
1565#ifdef CONFIG_ZONE_DEVICE
1566static inline int online_device_section(struct mem_section *section)
1567{
1568 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1569
1570 return section && ((section->section_mem_map & flags) == flags);
1571}
1572#else
1573static inline int online_device_section(struct mem_section *section)
1574{
1575 return 0;
1576}
1577#endif
1578
1579static inline int online_section_nr(unsigned long nr)
1580{
1581 return online_section(__nr_to_section(nr));
1582}
1583
1584#ifdef CONFIG_MEMORY_HOTPLUG
1585void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1586void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1587#endif
1588
1589static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1590{
1591 return __nr_to_section(pfn_to_section_nr(pfn));
1592}
1593
1594extern unsigned long __highest_present_section_nr;
1595
1596static inline int subsection_map_index(unsigned long pfn)
1597{
1598 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1599}
1600
1601#ifdef CONFIG_SPARSEMEM_VMEMMAP
1602static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1603{
1604 int idx = subsection_map_index(pfn);
1605
1606 return test_bit(idx, ms->usage->subsection_map);
1607}
1608#else
1609static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1610{
1611 return 1;
1612}
1613#endif
1614
1615#ifndef CONFIG_HAVE_ARCH_PFN_VALID
1616/**
1617 * pfn_valid - check if there is a valid memory map entry for a PFN
1618 * @pfn: the page frame number to check
1619 *
1620 * Check if there is a valid memory map entry aka struct page for the @pfn.
1621 * Note, that availability of the memory map entry does not imply that
1622 * there is actual usable memory at that @pfn. The struct page may
1623 * represent a hole or an unusable page frame.
1624 *
1625 * Return: 1 for PFNs that have memory map entries and 0 otherwise
1626 */
1627static inline int pfn_valid(unsigned long pfn)
1628{
1629 struct mem_section *ms;
1630
1631 /*
1632 * Ensure the upper PAGE_SHIFT bits are clear in the
1633 * pfn. Else it might lead to false positives when
1634 * some of the upper bits are set, but the lower bits
1635 * match a valid pfn.
1636 */
1637 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
1638 return 0;
1639
1640 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1641 return 0;
1642 ms = __pfn_to_section(pfn);
1643 if (!valid_section(ms))
1644 return 0;
1645 /*
1646 * Traditionally early sections always returned pfn_valid() for
1647 * the entire section-sized span.
1648 */
1649 return early_section(ms) || pfn_section_valid(ms, pfn);
1650}
1651#endif
1652
1653static inline int pfn_in_present_section(unsigned long pfn)
1654{
1655 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1656 return 0;
1657 return present_section(__pfn_to_section(pfn));
1658}
1659
1660static inline unsigned long next_present_section_nr(unsigned long section_nr)
1661{
1662 while (++section_nr <= __highest_present_section_nr) {
1663 if (present_section_nr(section_nr))
1664 return section_nr;
1665 }
1666
1667 return -1;
1668}
1669
1670/*
1671 * These are _only_ used during initialisation, therefore they
1672 * can use __initdata ... They could have names to indicate
1673 * this restriction.
1674 */
1675#ifdef CONFIG_NUMA
1676#define pfn_to_nid(pfn) \
1677({ \
1678 unsigned long __pfn_to_nid_pfn = (pfn); \
1679 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1680})
1681#else
1682#define pfn_to_nid(pfn) (0)
1683#endif
1684
1685void sparse_init(void);
1686#else
1687#define sparse_init() do {} while (0)
1688#define sparse_index_init(_sec, _nid) do {} while (0)
1689#define pfn_in_present_section pfn_valid
1690#define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1691#endif /* CONFIG_SPARSEMEM */
1692
1693#endif /* !__GENERATING_BOUNDS.H */
1694#endif /* !__ASSEMBLY__ */
1695#endif /* _LINUX_MMZONE_H */
1696

source code of linux/include/linux/mmzone.h