Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_MMZONE_H
3#define _LINUX_MMZONE_H
4
5#ifndef __ASSEMBLY__
6#ifndef __GENERATING_BOUNDS_H
7
8#include <linux/spinlock.h>
9#include <linux/list.h>
10#include <linux/wait.h>
11#include <linux/bitops.h>
12#include <linux/cache.h>
13#include <linux/threads.h>
14#include <linux/numa.h>
15#include <linux/init.h>
16#include <linux/seqlock.h>
17#include <linux/nodemask.h>
18#include <linux/pageblock-flags.h>
19#include <linux/page-flags-layout.h>
20#include <linux/atomic.h>
21#include <linux/mm_types.h>
22#include <linux/page-flags.h>
23#include <asm/page.h>
24
25/* Free memory management - zoned buddy allocator. */
26#ifndef CONFIG_FORCE_MAX_ZONEORDER
27#define MAX_ORDER 11
28#else
29#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
30#endif
31#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
32
33/*
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
36 * coalesce naturally under reasonable reclaim pressure and those which
37 * will not.
38 */
39#define PAGE_ALLOC_COSTLY_ORDER 3
40
41enum migratetype {
42 MIGRATE_UNMOVABLE,
43 MIGRATE_MOVABLE,
44 MIGRATE_RECLAIMABLE,
45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
47#ifdef CONFIG_CMA
48 /*
49 * MIGRATE_CMA migration type is designed to mimic the way
50 * ZONE_MOVABLE works. Only movable pages can be allocated
51 * from MIGRATE_CMA pageblocks and page allocator never
52 * implicitly change migration type of MIGRATE_CMA pageblock.
53 *
54 * The way to use it is to change migratetype of a range of
55 * pageblocks to MIGRATE_CMA which can be done by
56 * __free_pageblock_cma() function. What is important though
57 * is that a range of pageblocks must be aligned to
58 * MAX_ORDER_NR_PAGES should biggest page be bigger then
59 * a single pageblock.
60 */
61 MIGRATE_CMA,
62#endif
63#ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE, /* can't allocate from here */
65#endif
66 MIGRATE_TYPES
67};
68
69/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70extern const char * const migratetype_names[MIGRATE_TYPES];
71
72#ifdef CONFIG_CMA
73# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
75#else
76# define is_migrate_cma(migratetype) false
77# define is_migrate_cma_page(_page) false
78#endif
79
80static inline bool is_migrate_movable(int mt)
81{
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
83}
84
85#define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
88
89extern int page_group_by_mobility_disabled;
90
91#define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
92
93#define get_pageblock_migratetype(page) \
94 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
95
96struct free_area {
97 struct list_head free_list[MIGRATE_TYPES];
98 unsigned long nr_free;
99};
100
101static inline struct page *get_page_from_free_area(struct free_area *area,
102 int migratetype)
103{
104 return list_first_entry_or_null(&area->free_list[migratetype],
105 struct page, lru);
106}
107
108static inline bool free_area_empty(struct free_area *area, int migratetype)
109{
110 return list_empty(&area->free_list[migratetype]);
111}
112
113struct pglist_data;
114
115/*
116 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
117 * So add a wild amount of padding here to ensure that they 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)
122struct 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
131enum 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_STAT_ITEMS
139};
140#else
141#define NR_VM_NUMA_STAT_ITEMS 0
142#endif
143
144enum 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 NR_PAGETABLE, /* used for pagetables */
156 /* Second 128 byte cacheline */
157 NR_BOUNCE,
158#if IS_ENABLED(CONFIG_ZSMALLOC)
159 NR_ZSPAGES, /* allocated in zsmalloc */
160#endif
161 NR_FREE_CMA_PAGES,
162 NR_VM_ZONE_STAT_ITEMS };
163
164enum node_stat_item {
165 NR_LRU_BASE,
166 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
167 NR_ACTIVE_ANON, /* " " " " " */
168 NR_INACTIVE_FILE, /* " " " " " */
169 NR_ACTIVE_FILE, /* " " " " " */
170 NR_UNEVICTABLE, /* " " " " " */
171 NR_SLAB_RECLAIMABLE_B,
172 NR_SLAB_UNRECLAIMABLE_B,
173 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
174 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
175 WORKINGSET_NODES,
176 WORKINGSET_REFAULT_BASE,
177 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
178 WORKINGSET_REFAULT_FILE,
179 WORKINGSET_ACTIVATE_BASE,
180 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
181 WORKINGSET_ACTIVATE_FILE,
182 WORKINGSET_RESTORE_BASE,
183 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
184 WORKINGSET_RESTORE_FILE,
185 WORKINGSET_NODERECLAIM,
186 NR_ANON_MAPPED, /* Mapped anonymous pages */
187 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
188 only modified from process context */
189 NR_FILE_PAGES,
190 NR_FILE_DIRTY,
191 NR_WRITEBACK,
192 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
193 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
194 NR_SHMEM_THPS,
195 NR_SHMEM_PMDMAPPED,
196 NR_FILE_THPS,
197 NR_FILE_PMDMAPPED,
198 NR_ANON_THPS,
199 NR_VMSCAN_WRITE,
200 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
201 NR_DIRTIED, /* page dirtyings since bootup */
202 NR_WRITTEN, /* page writings since bootup */
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_VM_NODE_STAT_ITEMS
211};
212
213/*
214 * Returns true if the value is measured in bytes (most vmstat values are
215 * measured in pages). This defines the API part, the internal representation
216 * might be different.
217 */
218static __always_inline bool vmstat_item_in_bytes(int idx)
219{
220 /*
221 * Global and per-node slab counters track slab pages.
222 * It's expected that changes are multiples of PAGE_SIZE.
223 * Internally values are stored in pages.
224 *
225 * Per-memcg and per-lruvec counters track memory, consumed
226 * by individual slab objects. These counters are actually
227 * byte-precise.
228 */
229 return (idx == NR_SLAB_RECLAIMABLE_B ||
230 idx == NR_SLAB_UNRECLAIMABLE_B);
231}
232
233/*
234 * We do arithmetic on the LRU lists in various places in the code,
235 * so it is important to keep the active lists LRU_ACTIVE higher in
236 * the array than the corresponding inactive lists, and to keep
237 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
238 *
239 * This has to be kept in sync with the statistics in zone_stat_item
240 * above and the descriptions in vmstat_text in mm/vmstat.c
241 */
242#define LRU_BASE 0
243#define LRU_ACTIVE 1
244#define LRU_FILE 2
245
246enum lru_list {
247 LRU_INACTIVE_ANON = LRU_BASE,
248 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
249 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
250 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
251 LRU_UNEVICTABLE,
252 NR_LRU_LISTS
253};
254
255#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
256
257#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
258
259static inline bool is_file_lru(enum lru_list lru)
260{
261 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
262}
263
264static inline bool is_active_lru(enum lru_list lru)
265{
266 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
267}
268
269enum lruvec_flags {
270 LRUVEC_CONGESTED, /* lruvec has many dirty pages
271 * backed by a congested BDI
272 */
273};
274
275struct lruvec {
276 struct list_head lists[NR_LRU_LISTS];
277 /*
278 * These track the cost of reclaiming one LRU - file or anon -
279 * over the other. As the observed cost of reclaiming one LRU
280 * increases, the reclaim scan balance tips toward the other.
281 */
282 unsigned long anon_cost;
283 unsigned long file_cost;
284 /* Non-resident age, driven by LRU movement */
285 atomic_long_t nonresident_age;
286 /* Refaults at the time of last reclaim cycle, anon=0, file=1 */
287 unsigned long refaults[2];
288 /* Various lruvec state flags (enum lruvec_flags) */
289 unsigned long flags;
290#ifdef CONFIG_MEMCG
291 struct pglist_data *pgdat;
292#endif
293};
294
295/* Isolate unmapped pages */
296#define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
297/* Isolate for asynchronous migration */
298#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
299/* Isolate unevictable pages */
300#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
301
302/* LRU Isolation modes. */
303typedef unsigned __bitwise isolate_mode_t;
304
305enum zone_watermarks {
306 WMARK_MIN,
307 WMARK_LOW,
308 WMARK_HIGH,
309 NR_WMARK
310};
311
312#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
313#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
314#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
315#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
316
317struct per_cpu_pages {
318 int count; /* number of pages in the list */
319 int high; /* high watermark, emptying needed */
320 int batch; /* chunk size for buddy add/remove */
321
322 /* Lists of pages, one per migrate type stored on the pcp-lists */
323 struct list_head lists[MIGRATE_PCPTYPES];
324};
325
326struct per_cpu_pageset {
327 struct per_cpu_pages pcp;
328#ifdef CONFIG_NUMA
329 s8 expire;
330 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
331#endif
332#ifdef CONFIG_SMP
333 s8 stat_threshold;
334 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
335#endif
336};
337
338struct per_cpu_nodestat {
339 s8 stat_threshold;
340 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
341};
342
343#endif /* !__GENERATING_BOUNDS.H */
344
345enum zone_type {
346 /*
347 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
348 * to DMA to all of the addressable memory (ZONE_NORMAL).
349 * On architectures where this area covers the whole 32 bit address
350 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
351 * DMA addressing constraints. This distinction is important as a 32bit
352 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
353 * platforms may need both zones as they support peripherals with
354 * different DMA addressing limitations.
355 *
356 * Some examples:
357 *
358 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
359 * rest of the lower 4G.
360 *
361 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
362 * the specific device.
363 *
364 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
365 * lower 4G.
366 *
367 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
368 * depending on the specific device.
369 *
370 * - s390 uses ZONE_DMA fixed to the lower 2G.
371 *
372 * - ia64 and riscv only use ZONE_DMA32.
373 *
374 * - parisc uses neither.
375 */
376#ifdef CONFIG_ZONE_DMA
377 ZONE_DMA,
378#endif
379#ifdef CONFIG_ZONE_DMA32
380 ZONE_DMA32,
381#endif
382 /*
383 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
384 * performed on pages in ZONE_NORMAL if the DMA devices support
385 * transfers to all addressable memory.
386 */
387 ZONE_NORMAL,
388#ifdef CONFIG_HIGHMEM
389 /*
390 * A memory area that is only addressable by the kernel through
391 * mapping portions into its own address space. This is for example
392 * used by i386 to allow the kernel to address the memory beyond
393 * 900MB. The kernel will set up special mappings (page
394 * table entries on i386) for each page that the kernel needs to
395 * access.
396 */
397 ZONE_HIGHMEM,
398#endif
399 ZONE_MOVABLE,
400#ifdef CONFIG_ZONE_DEVICE
401 ZONE_DEVICE,
402#endif
403 __MAX_NR_ZONES
404
405};
406
407#ifndef __GENERATING_BOUNDS_H
408
409struct zone {
410 /* Read-mostly fields */
411
412 /* zone watermarks, access with *_wmark_pages(zone) macros */
413 unsigned long _watermark[NR_WMARK];
414 unsigned long watermark_boost;
415
416 unsigned long nr_reserved_highatomic;
417
418 /*
419 * We don't know if the memory that we're going to allocate will be
420 * freeable or/and it will be released eventually, so to avoid totally
421 * wasting several GB of ram we must reserve some of the lower zone
422 * memory (otherwise we risk to run OOM on the lower zones despite
423 * there being tons of freeable ram on the higher zones). This array is
424 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
425 * changes.
426 */
427 long lowmem_reserve[MAX_NR_ZONES];
428
429#ifdef CONFIG_NUMA
430 int node;
431#endif
432 struct pglist_data *zone_pgdat;
433 struct per_cpu_pageset __percpu *pageset;
434
435#ifndef CONFIG_SPARSEMEM
436 /*
437 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
438 * In SPARSEMEM, this map is stored in struct mem_section
439 */
440 unsigned long *pageblock_flags;
441#endif /* CONFIG_SPARSEMEM */
442
443 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
444 unsigned long zone_start_pfn;
445
446 /*
447 * spanned_pages is the total pages spanned by the zone, including
448 * holes, which is calculated as:
449 * spanned_pages = zone_end_pfn - zone_start_pfn;
450 *
451 * present_pages is physical pages existing within the zone, which
452 * is calculated as:
453 * present_pages = spanned_pages - absent_pages(pages in holes);
454 *
455 * managed_pages is present pages managed by the buddy system, which
456 * is calculated as (reserved_pages includes pages allocated by the
457 * bootmem allocator):
458 * managed_pages = present_pages - reserved_pages;
459 *
460 * So present_pages may be used by memory hotplug or memory power
461 * management logic to figure out unmanaged pages by checking
462 * (present_pages - managed_pages). And managed_pages should be used
463 * by page allocator and vm scanner to calculate all kinds of watermarks
464 * and thresholds.
465 *
466 * Locking rules:
467 *
468 * zone_start_pfn and spanned_pages are protected by span_seqlock.
469 * It is a seqlock because it has to be read outside of zone->lock,
470 * and it is done in the main allocator path. But, it is written
471 * quite infrequently.
472 *
473 * The span_seq lock is declared along with zone->lock because it is
474 * frequently read in proximity to zone->lock. It's good to
475 * give them a chance of being in the same cacheline.
476 *
477 * Write access to present_pages at runtime should be protected by
478 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
479 * present_pages should get_online_mems() to get a stable value.
480 */
481 atomic_long_t managed_pages;
482 unsigned long spanned_pages;
483 unsigned long present_pages;
484
485 const char *name;
486
487#ifdef CONFIG_MEMORY_ISOLATION
488 /*
489 * Number of isolated pageblock. It is used to solve incorrect
490 * freepage counting problem due to racy retrieving migratetype
491 * of pageblock. Protected by zone->lock.
492 */
493 unsigned long nr_isolate_pageblock;
494#endif
495
496#ifdef CONFIG_MEMORY_HOTPLUG
497 /* see spanned/present_pages for more description */
498 seqlock_t span_seqlock;
499#endif
500
501 int initialized;
502
503 /* Write-intensive fields used from the page allocator */
504 ZONE_PADDING(_pad1_)
505
506 /* free areas of different sizes */
507 struct free_area free_area[MAX_ORDER];
508
509 /* zone flags, see below */
510 unsigned long flags;
511
512 /* Primarily protects free_area */
513 spinlock_t lock;
514
515 /* Write-intensive fields used by compaction and vmstats. */
516 ZONE_PADDING(_pad2_)
517
518 /*
519 * When free pages are below this point, additional steps are taken
520 * when reading the number of free pages to avoid per-cpu counter
521 * drift allowing watermarks to be breached
522 */
523 unsigned long percpu_drift_mark;
524
525#if defined CONFIG_COMPACTION || defined CONFIG_CMA
526 /* pfn where compaction free scanner should start */
527 unsigned long compact_cached_free_pfn;
528 /* pfn where async and sync compaction migration scanner should start */
529 unsigned long compact_cached_migrate_pfn[2];
530 unsigned long compact_init_migrate_pfn;
531 unsigned long compact_init_free_pfn;
532#endif
533
534#ifdef CONFIG_COMPACTION
535 /*
536 * On compaction failure, 1<<compact_defer_shift compactions
537 * are skipped before trying again. The number attempted since
538 * last failure is tracked with compact_considered.
539 * compact_order_failed is the minimum compaction failed order.
540 */
541 unsigned int compact_considered;
542 unsigned int compact_defer_shift;
543 int compact_order_failed;
544#endif
545
546#if defined CONFIG_COMPACTION || defined CONFIG_CMA
547 /* Set to true when the PG_migrate_skip bits should be cleared */
548 bool compact_blockskip_flush;
549#endif
550
551 bool contiguous;
552
553 ZONE_PADDING(_pad3_)
554 /* Zone statistics */
555 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
556 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
557} ____cacheline_internodealigned_in_smp;
558
559enum pgdat_flags {
560 PGDAT_DIRTY, /* reclaim scanning has recently found
561 * many dirty file pages at the tail
562 * of the LRU.
563 */
564 PGDAT_WRITEBACK, /* reclaim scanning has recently found
565 * many pages under writeback
566 */
567 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
568};
569
570enum zone_flags {
571 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
572 * Cleared when kswapd is woken.
573 */
574};
575
576static inline unsigned long zone_managed_pages(struct zone *zone)
577{
578 return (unsigned long)atomic_long_read(&zone->managed_pages);
579}
580
581static inline unsigned long zone_end_pfn(const struct zone *zone)
582{
583 return zone->zone_start_pfn + zone->spanned_pages;
584}
585
586static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
587{
588 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
589}
590
591static inline bool zone_is_initialized(struct zone *zone)
592{
593 return zone->initialized;
594}
595
596static inline bool zone_is_empty(struct zone *zone)
597{
598 return zone->spanned_pages == 0;
599}
600
601/*
602 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
603 * intersection with the given zone
604 */
605static inline bool zone_intersects(struct zone *zone,
606 unsigned long start_pfn, unsigned long nr_pages)
607{
608 if (zone_is_empty(zone))
609 return false;
610 if (start_pfn >= zone_end_pfn(zone) ||
611 start_pfn + nr_pages <= zone->zone_start_pfn)
612 return false;
613
614 return true;
615}
616
617/*
618 * The "priority" of VM scanning is how much of the queues we will scan in one
619 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
620 * queues ("queue_length >> 12") during an aging round.
621 */
622#define DEF_PRIORITY 12
623
624/* Maximum number of zones on a zonelist */
625#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
626
627enum {
628 ZONELIST_FALLBACK, /* zonelist with fallback */
629#ifdef CONFIG_NUMA
630 /*
631 * The NUMA zonelists are doubled because we need zonelists that
632 * restrict the allocations to a single node for __GFP_THISNODE.
633 */
634 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
635#endif
636 MAX_ZONELISTS
637};
638
639/*
640 * This struct contains information about a zone in a zonelist. It is stored
641 * here to avoid dereferences into large structures and lookups of tables
642 */
643struct zoneref {
644 struct zone *zone; /* Pointer to actual zone */
645 int zone_idx; /* zone_idx(zoneref->zone) */
646};
647
648/*
649 * One allocation request operates on a zonelist. A zonelist
650 * is a list of zones, the first one is the 'goal' of the
651 * allocation, the other zones are fallback zones, in decreasing
652 * priority.
653 *
654 * To speed the reading of the zonelist, the zonerefs contain the zone index
655 * of the entry being read. Helper functions to access information given
656 * a struct zoneref are
657 *
658 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
659 * zonelist_zone_idx() - Return the index of the zone for an entry
660 * zonelist_node_idx() - Return the index of the node for an entry
661 */
662struct zonelist {
663 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
664};
665
666#ifndef CONFIG_DISCONTIGMEM
667/* The array of struct pages - for discontigmem use pgdat->lmem_map */
668extern struct page *mem_map;
669#endif
670
671#ifdef CONFIG_TRANSPARENT_HUGEPAGE
672struct deferred_split {
673 spinlock_t split_queue_lock;
674 struct list_head split_queue;
675 unsigned long split_queue_len;
676};
677#endif
678
679/*
680 * On NUMA machines, each NUMA node would have a pg_data_t to describe
681 * it's memory layout. On UMA machines there is a single pglist_data which
682 * describes the whole memory.
683 *
684 * Memory statistics and page replacement data structures are maintained on a
685 * per-zone basis.
686 */
687typedef struct pglist_data {
688 /*
689 * node_zones contains just the zones for THIS node. Not all of the
690 * zones may be populated, but it is the full list. It is referenced by
691 * this node's node_zonelists as well as other node's node_zonelists.
692 */
693 struct zone node_zones[MAX_NR_ZONES];
694
695 /*
696 * node_zonelists contains references to all zones in all nodes.
697 * Generally the first zones will be references to this node's
698 * node_zones.
699 */
700 struct zonelist node_zonelists[MAX_ZONELISTS];
701
702 int nr_zones; /* number of populated zones in this node */
703#ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
704 struct page *node_mem_map;
705#ifdef CONFIG_PAGE_EXTENSION
706 struct page_ext *node_page_ext;
707#endif
708#endif
709#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
710 /*
711 * Must be held any time you expect node_start_pfn,
712 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
713 * Also synchronizes pgdat->first_deferred_pfn during deferred page
714 * init.
715 *
716 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
717 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
718 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
719 *
720 * Nests above zone->lock and zone->span_seqlock
721 */
722 spinlock_t node_size_lock;
723#endif
724 unsigned long node_start_pfn;
725 unsigned long node_present_pages; /* total number of physical pages */
726 unsigned long node_spanned_pages; /* total size of physical page
727 range, including holes */
728 int node_id;
729 wait_queue_head_t kswapd_wait;
730 wait_queue_head_t pfmemalloc_wait;
731 struct task_struct *kswapd; /* Protected by
732 mem_hotplug_begin/end() */
733 int kswapd_order;
734 enum zone_type kswapd_highest_zoneidx;
735
736 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
737
738#ifdef CONFIG_COMPACTION
739 int kcompactd_max_order;
740 enum zone_type kcompactd_highest_zoneidx;
741 wait_queue_head_t kcompactd_wait;
742 struct task_struct *kcompactd;
743#endif
744 /*
745 * This is a per-node reserve of pages that are not available
746 * to userspace allocations.
747 */
748 unsigned long totalreserve_pages;
749
750#ifdef CONFIG_NUMA
751 /*
752 * node reclaim becomes active if more unmapped pages exist.
753 */
754 unsigned long min_unmapped_pages;
755 unsigned long min_slab_pages;
756#endif /* CONFIG_NUMA */
757
758 /* Write-intensive fields used by page reclaim */
759 ZONE_PADDING(_pad1_)
760 spinlock_t lru_lock;
761
762#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
763 /*
764 * If memory initialisation on large machines is deferred then this
765 * is the first PFN that needs to be initialised.
766 */
767 unsigned long first_deferred_pfn;
768#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
769
770#ifdef CONFIG_TRANSPARENT_HUGEPAGE
771 struct deferred_split deferred_split_queue;
772#endif
773
774 /* Fields commonly accessed by the page reclaim scanner */
775
776 /*
777 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
778 *
779 * Use mem_cgroup_lruvec() to look up lruvecs.
780 */
781 struct lruvec __lruvec;
782
783 unsigned long flags;
784
785 ZONE_PADDING(_pad2_)
786
787 /* Per-node vmstats */
788 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
789 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
790} pg_data_t;
791
792#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
793#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
794#ifdef CONFIG_FLAT_NODE_MEM_MAP
795#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
796#else
797#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
798#endif
799#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
800
801#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
802#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
803
804static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
805{
806 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
807}
808
809static inline bool pgdat_is_empty(pg_data_t *pgdat)
810{
811 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
812}
813
814#include <linux/memory_hotplug.h>
815
816void build_all_zonelists(pg_data_t *pgdat);
817void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
818 enum zone_type highest_zoneidx);
819bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
820 int highest_zoneidx, unsigned int alloc_flags,
821 long free_pages);
822bool zone_watermark_ok(struct zone *z, unsigned int order,
823 unsigned long mark, int highest_zoneidx,
824 unsigned int alloc_flags);
825bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
826 unsigned long mark, int highest_zoneidx);
827/*
828 * Memory initialization context, use to differentiate memory added by
829 * the platform statically or via memory hotplug interface.
830 */
831enum meminit_context {
832 MEMINIT_EARLY,
833 MEMINIT_HOTPLUG,
834};
835
836extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
837 unsigned long size);
838
839extern void lruvec_init(struct lruvec *lruvec);
840
841static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
842{
843#ifdef CONFIG_MEMCG
844 return lruvec->pgdat;
845#else
846 return container_of(lruvec, struct pglist_data, __lruvec);
847#endif
848}
849
850extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
851
852#ifdef CONFIG_HAVE_MEMORYLESS_NODES
853int local_memory_node(int node_id);
854#else
855static inline int local_memory_node(int node_id) { return node_id; };
856#endif
857
858/*
859 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
860 */
861#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
862
863/*
864 * Returns true if a zone has pages managed by the buddy allocator.
865 * All the reclaim decisions have to use this function rather than
866 * populated_zone(). If the whole zone is reserved then we can easily
867 * end up with populated_zone() && !managed_zone().
868 */
869static inline bool managed_zone(struct zone *zone)
870{
871 return zone_managed_pages(zone);
872}
873
874/* Returns true if a zone has memory */
875static inline bool populated_zone(struct zone *zone)
876{
877 return zone->present_pages;
878}
879
880#ifdef CONFIG_NUMA
881static inline int zone_to_nid(struct zone *zone)
882{
883 return zone->node;
884}
885
886static inline void zone_set_nid(struct zone *zone, int nid)
887{
888 zone->node = nid;
889}
890#else
891static inline int zone_to_nid(struct zone *zone)
892{
893 return 0;
894}
895
896static inline void zone_set_nid(struct zone *zone, int nid) {}
897#endif
898
899extern int movable_zone;
900
901#ifdef CONFIG_HIGHMEM
902static inline int zone_movable_is_highmem(void)
903{
904#ifdef CONFIG_NEED_MULTIPLE_NODES
905 return movable_zone == ZONE_HIGHMEM;
906#else
907 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
908#endif
909}
910#endif
911
912static inline int is_highmem_idx(enum zone_type idx)
913{
914#ifdef CONFIG_HIGHMEM
915 return (idx == ZONE_HIGHMEM ||
916 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
917#else
918 return 0;
919#endif
920}
921
922/**
923 * is_highmem - helper function to quickly check if a struct zone is a
924 * highmem zone or not. This is an attempt to keep references
925 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
926 * @zone - pointer to struct zone variable
927 */
928static inline int is_highmem(struct zone *zone)
929{
930#ifdef CONFIG_HIGHMEM
931 return is_highmem_idx(zone_idx(zone));
932#else
933 return 0;
934#endif
935}
936
937/* These two functions are used to setup the per zone pages min values */
938struct ctl_table;
939
940int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
941 loff_t *);
942int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
943 size_t *, loff_t *);
944extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
945int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
946 size_t *, loff_t *);
947int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
948 void *, size_t *, loff_t *);
949int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
950 void *, size_t *, loff_t *);
951int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
952 void *, size_t *, loff_t *);
953int numa_zonelist_order_handler(struct ctl_table *, int,
954 void *, size_t *, loff_t *);
955extern int percpu_pagelist_fraction;
956extern char numa_zonelist_order[];
957#define NUMA_ZONELIST_ORDER_LEN 16
958
959#ifndef CONFIG_NEED_MULTIPLE_NODES
960
961extern struct pglist_data contig_page_data;
962#define NODE_DATA(nid) (&contig_page_data)
963#define NODE_MEM_MAP(nid) mem_map
964
965#else /* CONFIG_NEED_MULTIPLE_NODES */
966
967#include <asm/mmzone.h>
968
969#endif /* !CONFIG_NEED_MULTIPLE_NODES */
970
971extern struct pglist_data *first_online_pgdat(void);
972extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
973extern struct zone *next_zone(struct zone *zone);
974
975/**
976 * for_each_online_pgdat - helper macro to iterate over all online nodes
977 * @pgdat - pointer to a pg_data_t variable
978 */
979#define for_each_online_pgdat(pgdat) \
980 for (pgdat = first_online_pgdat(); \
981 pgdat; \
982 pgdat = next_online_pgdat(pgdat))
983/**
984 * for_each_zone - helper macro to iterate over all memory zones
985 * @zone - pointer to struct zone variable
986 *
987 * The user only needs to declare the zone variable, for_each_zone
988 * fills it in.
989 */
990#define for_each_zone(zone) \
991 for (zone = (first_online_pgdat())->node_zones; \
992 zone; \
993 zone = next_zone(zone))
994
995#define for_each_populated_zone(zone) \
996 for (zone = (first_online_pgdat())->node_zones; \
997 zone; \
998 zone = next_zone(zone)) \
999 if (!populated_zone(zone)) \
1000 ; /* do nothing */ \
1001 else
1002
1003static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1004{
1005 return zoneref->zone;
1006}
1007
1008static inline int zonelist_zone_idx(struct zoneref *zoneref)
1009{
1010 return zoneref->zone_idx;
1011}
1012
1013static inline int zonelist_node_idx(struct zoneref *zoneref)
1014{
1015 return zone_to_nid(zoneref->zone);
1016}
1017
1018struct zoneref *__next_zones_zonelist(struct zoneref *z,
1019 enum zone_type highest_zoneidx,
1020 nodemask_t *nodes);
1021
1022/**
1023 * 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
1024 * @z - The cursor used as a starting point for the search
1025 * @highest_zoneidx - The zone index of the highest zone to return
1026 * @nodes - An optional nodemask to filter the zonelist with
1027 *
1028 * This function returns the next zone at or below a given zone index that is
1029 * within the allowed nodemask using a cursor as the starting point for the
1030 * search. The zoneref returned is a cursor that represents the current zone
1031 * being examined. It should be advanced by one before calling
1032 * next_zones_zonelist again.
1033 */
1034static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1035 enum zone_type highest_zoneidx,
1036 nodemask_t *nodes)
1037{
1038 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1039 return z;
1040 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1041}
1042
1043/**
1044 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1045 * @zonelist - The zonelist to search for a suitable zone
1046 * @highest_zoneidx - The zone index of the highest zone to return
1047 * @nodes - An optional nodemask to filter the zonelist with
1048 * @return - Zoneref pointer for the first suitable zone found (see below)
1049 *
1050 * This function returns the first zone at or below a given zone index that is
1051 * within the allowed nodemask. The zoneref returned is a cursor that can be
1052 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1053 * one before calling.
1054 *
1055 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1056 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1057 * update due to cpuset modification.
1058 */
1059static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1060 enum zone_type highest_zoneidx,
1061 nodemask_t *nodes)
1062{
1063 return next_zones_zonelist(zonelist->_zonerefs,
1064 highest_zoneidx, nodes);
1065}
1066
1067/**
1068 * 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
1069 * @zone - The current zone in the iterator
1070 * @z - The current pointer within zonelist->_zonerefs being iterated
1071 * @zlist - The zonelist being iterated
1072 * @highidx - The zone index of the highest zone to return
1073 * @nodemask - Nodemask allowed by the allocator
1074 *
1075 * This iterator iterates though all zones at or below a given zone index and
1076 * within a given nodemask
1077 */
1078#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1079 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1080 zone; \
1081 z = next_zones_zonelist(++z, highidx, nodemask), \
1082 zone = zonelist_zone(z))
1083
1084#define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1085 for (zone = z->zone; \
1086 zone; \
1087 z = next_zones_zonelist(++z, highidx, nodemask), \
1088 zone = zonelist_zone(z))
1089
1090
1091/**
1092 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1093 * @zone - The current zone in the iterator
1094 * @z - The current pointer within zonelist->zones being iterated
1095 * @zlist - The zonelist being iterated
1096 * @highidx - The zone index of the highest zone to return
1097 *
1098 * This iterator iterates though all zones at or below a given zone index.
1099 */
1100#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1101 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1102
1103#ifdef CONFIG_SPARSEMEM
1104#include <asm/sparsemem.h>
1105#endif
1106
1107#ifdef CONFIG_FLATMEM
1108#define pfn_to_nid(pfn) (0)
1109#endif
1110
1111#ifdef CONFIG_SPARSEMEM
1112
1113/*
1114 * SECTION_SHIFT #bits space required to store a section #
1115 *
1116 * PA_SECTION_SHIFT physical address to/from section number
1117 * PFN_SECTION_SHIFT pfn to/from section number
1118 */
1119#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1120#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1121
1122#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1123
1124#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1125#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1126
1127#define SECTION_BLOCKFLAGS_BITS \
1128 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1129
1130#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1131#error Allocator MAX_ORDER exceeds SECTION_SIZE
1132#endif
1133
1134static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1135{
1136 return pfn >> PFN_SECTION_SHIFT;
1137}
1138static inline unsigned long section_nr_to_pfn(unsigned long sec)
1139{
1140 return sec << PFN_SECTION_SHIFT;
1141}
1142
1143#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1144#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1145
1146#define SUBSECTION_SHIFT 21
1147#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1148
1149#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1150#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1151#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1152
1153#if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1154#error Subsection size exceeds section size
1155#else
1156#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1157#endif
1158
1159#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1160#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1161
1162struct mem_section_usage {
1163#ifdef CONFIG_SPARSEMEM_VMEMMAP
1164 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1165#endif
1166 /* See declaration of similar field in struct zone */
1167 unsigned long pageblock_flags[0];
1168};
1169
1170void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1171
1172struct page;
1173struct page_ext;
1174struct mem_section {
1175 /*
1176 * This is, logically, a pointer to an array of struct
1177 * pages. However, it is stored with some other magic.
1178 * (see sparse.c::sparse_init_one_section())
1179 *
1180 * Additionally during early boot we encode node id of
1181 * the location of the section here to guide allocation.
1182 * (see sparse.c::memory_present())
1183 *
1184 * Making it a UL at least makes someone do a cast
1185 * before using it wrong.
1186 */
1187 unsigned long section_mem_map;
1188
1189 struct mem_section_usage *usage;
1190#ifdef CONFIG_PAGE_EXTENSION
1191 /*
1192 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1193 * section. (see page_ext.h about this.)
1194 */
1195 struct page_ext *page_ext;
1196 unsigned long pad;
1197#endif
1198 /*
1199 * WARNING: mem_section must be a power-of-2 in size for the
1200 * calculation and use of SECTION_ROOT_MASK to make sense.
1201 */
1202};
1203
1204#ifdef CONFIG_SPARSEMEM_EXTREME
1205#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1206#else
1207#define SECTIONS_PER_ROOT 1
1208#endif
1209
1210#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1211#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1212#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1213
1214#ifdef CONFIG_SPARSEMEM_EXTREME
1215extern struct mem_section **mem_section;
1216#else
1217extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1218#endif
1219
1220static inline unsigned long *section_to_usemap(struct mem_section *ms)
1221{
1222 return ms->usage->pageblock_flags;
1223}
1224
1225static inline struct mem_section *__nr_to_section(unsigned long nr)
1226{
1227#ifdef CONFIG_SPARSEMEM_EXTREME
1228 if (!mem_section)
1229 return NULL;
1230#endif
1231 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1232 return NULL;
1233 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1234}
1235extern unsigned long __section_nr(struct mem_section *ms);
1236extern size_t mem_section_usage_size(void);
1237
1238/*
1239 * We use the lower bits of the mem_map pointer to store
1240 * a little bit of information. The pointer is calculated
1241 * as mem_map - section_nr_to_pfn(pnum). The result is
1242 * aligned to the minimum alignment of the two values:
1243 * 1. All mem_map arrays are page-aligned.
1244 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1245 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1246 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1247 * worst combination is powerpc with 256k pages,
1248 * which results in PFN_SECTION_SHIFT equal 6.
1249 * To sum it up, at least 6 bits are available.
1250 */
1251#define SECTION_MARKED_PRESENT (1UL<<0)
1252#define SECTION_HAS_MEM_MAP (1UL<<1)
1253#define SECTION_IS_ONLINE (1UL<<2)
1254#define SECTION_IS_EARLY (1UL<<3)
1255#define SECTION_MAP_LAST_BIT (1UL<<4)
1256#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1257#define SECTION_NID_SHIFT 3
1258
1259static inline struct page *__section_mem_map_addr(struct mem_section *section)
1260{
1261 unsigned long map = section->section_mem_map;
1262 map &= SECTION_MAP_MASK;
1263 return (struct page *)map;
1264}
1265
1266static inline int present_section(struct mem_section *section)
1267{
1268 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1269}
1270
1271static inline int present_section_nr(unsigned long nr)
1272{
1273 return present_section(__nr_to_section(nr));
1274}
1275
1276static inline int valid_section(struct mem_section *section)
1277{
1278 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1279}
1280
1281static inline int early_section(struct mem_section *section)
1282{
1283 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1284}
1285
1286static inline int valid_section_nr(unsigned long nr)
1287{
1288 return valid_section(__nr_to_section(nr));
1289}
1290
1291static inline int online_section(struct mem_section *section)
1292{
1293 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1294}
1295
1296static inline int online_section_nr(unsigned long nr)
1297{
1298 return online_section(__nr_to_section(nr));
1299}
1300
1301#ifdef CONFIG_MEMORY_HOTPLUG
1302void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1303#ifdef CONFIG_MEMORY_HOTREMOVE
1304void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1305#endif
1306#endif
1307
1308static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1309{
1310 return __nr_to_section(pfn_to_section_nr(pfn));
1311}
1312
1313extern unsigned long __highest_present_section_nr;
1314
1315static inline int subsection_map_index(unsigned long pfn)
1316{
1317 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1318}
1319
1320#ifdef CONFIG_SPARSEMEM_VMEMMAP
1321static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1322{
1323 int idx = subsection_map_index(pfn);
1324
1325 return test_bit(idx, ms->usage->subsection_map);
1326}
1327#else
1328static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1329{
1330 return 1;
1331}
1332#endif
1333
1334#ifndef CONFIG_HAVE_ARCH_PFN_VALID
1335static inline int pfn_valid(unsigned long pfn)
1336{
1337 struct mem_section *ms;
1338
1339 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1340 return 0;
1341 ms = __nr_to_section(pfn_to_section_nr(pfn));
1342 if (!valid_section(ms))
1343 return 0;
1344 /*
1345 * Traditionally early sections always returned pfn_valid() for
1346 * the entire section-sized span.
1347 */
1348 return early_section(ms) || pfn_section_valid(ms, pfn);
1349}
1350#endif
1351
1352static inline int pfn_in_present_section(unsigned long pfn)
1353{
1354 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1355 return 0;
1356 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1357}
1358
1359static inline unsigned long next_present_section_nr(unsigned long section_nr)
1360{
1361 while (++section_nr <= __highest_present_section_nr) {
1362 if (present_section_nr(section_nr))
1363 return section_nr;
1364 }
1365
1366 return -1;
1367}
1368
1369/*
1370 * These are _only_ used during initialisation, therefore they
1371 * can use __initdata ... They could have names to indicate
1372 * this restriction.
1373 */
1374#ifdef CONFIG_NUMA
1375#define pfn_to_nid(pfn) \
1376({ \
1377 unsigned long __pfn_to_nid_pfn = (pfn); \
1378 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1379})
1380#else
1381#define pfn_to_nid(pfn) (0)
1382#endif
1383
1384#define early_pfn_valid(pfn) pfn_valid(pfn)
1385void sparse_init(void);
1386#else
1387#define sparse_init() do {} while (0)
1388#define sparse_index_init(_sec, _nid) do {} while (0)
1389#define pfn_in_present_section pfn_valid
1390#define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1391#endif /* CONFIG_SPARSEMEM */
1392
1393/*
1394 * During memory init memblocks map pfns to nids. The search is expensive and
1395 * this caches recent lookups. The implementation of __early_pfn_to_nid
1396 * may treat start/end as pfns or sections.
1397 */
1398struct mminit_pfnnid_cache {
1399 unsigned long last_start;
1400 unsigned long last_end;
1401 int last_nid;
1402};
1403
1404#ifndef early_pfn_valid
1405#define early_pfn_valid(pfn) (1)
1406#endif
1407
1408/*
1409 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1410 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1411 * pfn_valid_within() should be used in this case; we optimise this away
1412 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1413 */
1414#ifdef CONFIG_HOLES_IN_ZONE
1415#define pfn_valid_within(pfn) pfn_valid(pfn)
1416#else
1417#define pfn_valid_within(pfn) (1)
1418#endif
1419
1420#ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1421/*
1422 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1423 * associated with it or not. This means that a struct page exists for this
1424 * pfn. The caller cannot assume the page is fully initialized in general.
1425 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1426 * will ensure the struct page is fully online and initialized. Special pages
1427 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1428 *
1429 * In FLATMEM, it is expected that holes always have valid memmap as long as
1430 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1431 * that a valid section has a memmap for the entire section.
1432 *
1433 * However, an ARM, and maybe other embedded architectures in the future
1434 * free memmap backing holes to save memory on the assumption the memmap is
1435 * never used. The page_zone linkages are then broken even though pfn_valid()
1436 * returns true. A walker of the full memmap must then do this additional
1437 * check to ensure the memmap they are looking at is sane by making sure
1438 * the zone and PFN linkages are still valid. This is expensive, but walkers
1439 * of the full memmap are extremely rare.
1440 */
1441bool memmap_valid_within(unsigned long pfn,
1442 struct page *page, struct zone *zone);
1443#else
1444static inline bool memmap_valid_within(unsigned long pfn,
1445 struct page *page, struct zone *zone)
1446{
1447 return true;
1448}
1449#endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1450
1451#endif /* !__GENERATING_BOUNDS.H */
1452#endif /* !__ASSEMBLY__ */
1453#endif /* _LINUX_MMZONE_H */
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/list_nulls.h>
11#include <linux/wait.h>
12#include <linux/bitops.h>
13#include <linux/cache.h>
14#include <linux/threads.h>
15#include <linux/numa.h>
16#include <linux/init.h>
17#include <linux/seqlock.h>
18#include <linux/nodemask.h>
19#include <linux/pageblock-flags.h>
20#include <linux/page-flags-layout.h>
21#include <linux/atomic.h>
22#include <linux/mm_types.h>
23#include <linux/page-flags.h>
24#include <linux/local_lock.h>
25#include <linux/zswap.h>
26#include <asm/page.h>
27
28/* Free memory management - zoned buddy allocator. */
29#ifndef CONFIG_ARCH_FORCE_MAX_ORDER
30#define MAX_PAGE_ORDER 10
31#else
32#define MAX_PAGE_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
33#endif
34#define MAX_ORDER_NR_PAGES (1 << MAX_PAGE_ORDER)
35
36#define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
37
38#define NR_PAGE_ORDERS (MAX_PAGE_ORDER + 1)
39
40/*
41 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
42 * costly to service. That is between allocation orders which should
43 * coalesce naturally under reasonable reclaim pressure and those which
44 * will not.
45 */
46#define PAGE_ALLOC_COSTLY_ORDER 3
47
48enum migratetype {
49 MIGRATE_UNMOVABLE,
50 MIGRATE_MOVABLE,
51 MIGRATE_RECLAIMABLE,
52 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
53 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
54#ifdef CONFIG_CMA
55 /*
56 * MIGRATE_CMA migration type is designed to mimic the way
57 * ZONE_MOVABLE works. Only movable pages can be allocated
58 * from MIGRATE_CMA pageblocks and page allocator never
59 * implicitly change migration type of MIGRATE_CMA pageblock.
60 *
61 * The way to use it is to change migratetype of a range of
62 * pageblocks to MIGRATE_CMA which can be done by
63 * __free_pageblock_cma() function.
64 */
65 MIGRATE_CMA,
66#endif
67#ifdef CONFIG_MEMORY_ISOLATION
68 MIGRATE_ISOLATE, /* can't allocate from here */
69#endif
70 MIGRATE_TYPES
71};
72
73/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
74extern const char * const migratetype_names[MIGRATE_TYPES];
75
76#ifdef CONFIG_CMA
77# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
78# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
79# define is_migrate_cma_folio(folio, pfn) (MIGRATE_CMA == \
80 get_pfnblock_flags_mask(&folio->page, pfn, MIGRATETYPE_MASK))
81#else
82# define is_migrate_cma(migratetype) false
83# define is_migrate_cma_page(_page) false
84# define is_migrate_cma_folio(folio, pfn) false
85#endif
86
87static inline bool is_migrate_movable(int mt)
88{
89 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
90}
91
92/*
93 * Check whether a migratetype can be merged with another migratetype.
94 *
95 * It is only mergeable when it can fall back to other migratetypes for
96 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
97 */
98static inline bool migratetype_is_mergeable(int mt)
99{
100 return mt < MIGRATE_PCPTYPES;
101}
102
103#define for_each_migratetype_order(order, type) \
104 for (order = 0; order < NR_PAGE_ORDERS; order++) \
105 for (type = 0; type < MIGRATE_TYPES; type++)
106
107extern int page_group_by_mobility_disabled;
108
109#define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
110
111#define get_pageblock_migratetype(page) \
112 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
113
114#define folio_migratetype(folio) \
115 get_pfnblock_flags_mask(&folio->page, folio_pfn(folio), \
116 MIGRATETYPE_MASK)
117struct free_area {
118 struct list_head free_list[MIGRATE_TYPES];
119 unsigned long nr_free;
120};
121
122struct pglist_data;
123
124#ifdef CONFIG_NUMA
125enum numa_stat_item {
126 NUMA_HIT, /* allocated in intended node */
127 NUMA_MISS, /* allocated in non intended node */
128 NUMA_FOREIGN, /* was intended here, hit elsewhere */
129 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
130 NUMA_LOCAL, /* allocation from local node */
131 NUMA_OTHER, /* allocation from other node */
132 NR_VM_NUMA_EVENT_ITEMS
133};
134#else
135#define NR_VM_NUMA_EVENT_ITEMS 0
136#endif
137
138enum zone_stat_item {
139 /* First 128 byte cacheline (assuming 64 bit words) */
140 NR_FREE_PAGES,
141 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
142 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
143 NR_ZONE_ACTIVE_ANON,
144 NR_ZONE_INACTIVE_FILE,
145 NR_ZONE_ACTIVE_FILE,
146 NR_ZONE_UNEVICTABLE,
147 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
148 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
149 /* Second 128 byte cacheline */
150 NR_BOUNCE,
151#if IS_ENABLED(CONFIG_ZSMALLOC)
152 NR_ZSPAGES, /* allocated in zsmalloc */
153#endif
154 NR_FREE_CMA_PAGES,
155#ifdef CONFIG_UNACCEPTED_MEMORY
156 NR_UNACCEPTED,
157#endif
158 NR_VM_ZONE_STAT_ITEMS };
159
160enum node_stat_item {
161 NR_LRU_BASE,
162 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
163 NR_ACTIVE_ANON, /* " " " " " */
164 NR_INACTIVE_FILE, /* " " " " " */
165 NR_ACTIVE_FILE, /* " " " " " */
166 NR_UNEVICTABLE, /* " " " " " */
167 NR_SLAB_RECLAIMABLE_B,
168 NR_SLAB_UNRECLAIMABLE_B,
169 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
170 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
171 WORKINGSET_NODES,
172 WORKINGSET_REFAULT_BASE,
173 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
174 WORKINGSET_REFAULT_FILE,
175 WORKINGSET_ACTIVATE_BASE,
176 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
177 WORKINGSET_ACTIVATE_FILE,
178 WORKINGSET_RESTORE_BASE,
179 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
180 WORKINGSET_RESTORE_FILE,
181 WORKINGSET_NODERECLAIM,
182 NR_ANON_MAPPED, /* Mapped anonymous pages */
183 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
184 only modified from process context */
185 NR_FILE_PAGES,
186 NR_FILE_DIRTY,
187 NR_WRITEBACK,
188 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
189 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
190 NR_SHMEM_THPS,
191 NR_SHMEM_PMDMAPPED,
192 NR_FILE_THPS,
193 NR_FILE_PMDMAPPED,
194 NR_ANON_THPS,
195 NR_VMSCAN_WRITE,
196 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
197 NR_DIRTIED, /* page dirtyings since bootup */
198 NR_WRITTEN, /* page writings since bootup */
199 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
200 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
201 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
202 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
203 NR_KERNEL_STACK_KB, /* measured in KiB */
204#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
205 NR_KERNEL_SCS_KB, /* measured in KiB */
206#endif
207 NR_PAGETABLE, /* used for pagetables */
208 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
209#ifdef CONFIG_SWAP
210 NR_SWAPCACHE,
211#endif
212#ifdef CONFIG_NUMA_BALANCING
213 PGPROMOTE_SUCCESS, /* promote successfully */
214 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
215#endif
216 /* PGDEMOTE_*: pages demoted */
217 PGDEMOTE_KSWAPD,
218 PGDEMOTE_DIRECT,
219 PGDEMOTE_KHUGEPAGED,
220 NR_VM_NODE_STAT_ITEMS
221};
222
223/*
224 * Returns true if the item should be printed in THPs (/proc/vmstat
225 * currently prints number of anon, file and shmem THPs. But the item
226 * is charged in pages).
227 */
228static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
229{
230 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
231 return false;
232
233 return item == NR_ANON_THPS ||
234 item == NR_FILE_THPS ||
235 item == NR_SHMEM_THPS ||
236 item == NR_SHMEM_PMDMAPPED ||
237 item == NR_FILE_PMDMAPPED;
238}
239
240/*
241 * Returns true if the value is measured in bytes (most vmstat values are
242 * measured in pages). This defines the API part, the internal representation
243 * might be different.
244 */
245static __always_inline bool vmstat_item_in_bytes(int idx)
246{
247 /*
248 * Global and per-node slab counters track slab pages.
249 * It's expected that changes are multiples of PAGE_SIZE.
250 * Internally values are stored in pages.
251 *
252 * Per-memcg and per-lruvec counters track memory, consumed
253 * by individual slab objects. These counters are actually
254 * byte-precise.
255 */
256 return (idx == NR_SLAB_RECLAIMABLE_B ||
257 idx == NR_SLAB_UNRECLAIMABLE_B);
258}
259
260/*
261 * We do arithmetic on the LRU lists in various places in the code,
262 * so it is important to keep the active lists LRU_ACTIVE higher in
263 * the array than the corresponding inactive lists, and to keep
264 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
265 *
266 * This has to be kept in sync with the statistics in zone_stat_item
267 * above and the descriptions in vmstat_text in mm/vmstat.c
268 */
269#define LRU_BASE 0
270#define LRU_ACTIVE 1
271#define LRU_FILE 2
272
273enum lru_list {
274 LRU_INACTIVE_ANON = LRU_BASE,
275 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
276 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
277 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
278 LRU_UNEVICTABLE,
279 NR_LRU_LISTS
280};
281
282enum vmscan_throttle_state {
283 VMSCAN_THROTTLE_WRITEBACK,
284 VMSCAN_THROTTLE_ISOLATED,
285 VMSCAN_THROTTLE_NOPROGRESS,
286 VMSCAN_THROTTLE_CONGESTED,
287 NR_VMSCAN_THROTTLE,
288};
289
290#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
291
292#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
293
294static inline bool is_file_lru(enum lru_list lru)
295{
296 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
297}
298
299static inline bool is_active_lru(enum lru_list lru)
300{
301 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
302}
303
304#define WORKINGSET_ANON 0
305#define WORKINGSET_FILE 1
306#define ANON_AND_FILE 2
307
308enum lruvec_flags {
309 /*
310 * An lruvec has many dirty pages backed by a congested BDI:
311 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
312 * It can be cleared by cgroup reclaim or kswapd.
313 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
314 * It can only be cleared by kswapd.
315 *
316 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
317 * reclaim, but not vice versa. This only applies to the root cgroup.
318 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
319 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
320 * by kswapd).
321 */
322 LRUVEC_CGROUP_CONGESTED,
323 LRUVEC_NODE_CONGESTED,
324};
325
326#endif /* !__GENERATING_BOUNDS_H */
327
328/*
329 * Evictable pages are divided into multiple generations. The youngest and the
330 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
331 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
332 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
333 * corresponding generation. The gen counter in folio->flags stores gen+1 while
334 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
335 *
336 * A page is added to the youngest generation on faulting. The aging needs to
337 * check the accessed bit at least twice before handing this page over to the
338 * eviction. The first check takes care of the accessed bit set on the initial
339 * fault; the second check makes sure this page hasn't been used since then.
340 * This process, AKA second chance, requires a minimum of two generations,
341 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
342 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
343 * rest of generations, if they exist, are considered inactive. See
344 * lru_gen_is_active().
345 *
346 * PG_active is always cleared while a page is on one of lrugen->folios[] so
347 * that the aging needs not to worry about it. And it's set again when a page
348 * considered active is isolated for non-reclaiming purposes, e.g., migration.
349 * See lru_gen_add_folio() and lru_gen_del_folio().
350 *
351 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
352 * number of categories of the active/inactive LRU when keeping track of
353 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
354 * in folio->flags.
355 */
356#define MIN_NR_GENS 2U
357#define MAX_NR_GENS 4U
358
359/*
360 * Each generation is divided into multiple tiers. A page accessed N times
361 * through file descriptors is in tier order_base_2(N). A page in the first tier
362 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
363 * tables or read ahead. A page in any other tier (N>1) is marked by
364 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
365 * supported without using additional bits in folio->flags.
366 *
367 * In contrast to moving across generations which requires the LRU lock, moving
368 * across tiers only involves atomic operations on folio->flags and therefore
369 * has a negligible cost in the buffered access path. In the eviction path,
370 * comparisons of refaulted/(evicted+protected) from the first tier and the
371 * rest infer whether pages accessed multiple times through file descriptors
372 * are statistically hot and thus worth protecting.
373 *
374 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
375 * number of categories of the active/inactive LRU when keeping track of
376 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
377 * folio->flags.
378 */
379#define MAX_NR_TIERS 4U
380
381#ifndef __GENERATING_BOUNDS_H
382
383struct lruvec;
384struct page_vma_mapped_walk;
385
386#define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
387#define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
388
389#ifdef CONFIG_LRU_GEN
390
391enum {
392 LRU_GEN_ANON,
393 LRU_GEN_FILE,
394};
395
396enum {
397 LRU_GEN_CORE,
398 LRU_GEN_MM_WALK,
399 LRU_GEN_NONLEAF_YOUNG,
400 NR_LRU_GEN_CAPS
401};
402
403#define MIN_LRU_BATCH BITS_PER_LONG
404#define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
405
406/* whether to keep historical stats from evicted generations */
407#ifdef CONFIG_LRU_GEN_STATS
408#define NR_HIST_GENS MAX_NR_GENS
409#else
410#define NR_HIST_GENS 1U
411#endif
412
413/*
414 * The youngest generation number is stored in max_seq for both anon and file
415 * types as they are aged on an equal footing. The oldest generation numbers are
416 * stored in min_seq[] separately for anon and file types as clean file pages
417 * can be evicted regardless of swap constraints.
418 *
419 * Normally anon and file min_seq are in sync. But if swapping is constrained,
420 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
421 * min_seq behind.
422 *
423 * The number of pages in each generation is eventually consistent and therefore
424 * can be transiently negative when reset_batch_size() is pending.
425 */
426struct lru_gen_folio {
427 /* the aging increments the youngest generation number */
428 unsigned long max_seq;
429 /* the eviction increments the oldest generation numbers */
430 unsigned long min_seq[ANON_AND_FILE];
431 /* the birth time of each generation in jiffies */
432 unsigned long timestamps[MAX_NR_GENS];
433 /* the multi-gen LRU lists, lazily sorted on eviction */
434 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
435 /* the multi-gen LRU sizes, eventually consistent */
436 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
437 /* the exponential moving average of refaulted */
438 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
439 /* the exponential moving average of evicted+protected */
440 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
441 /* the first tier doesn't need protection, hence the minus one */
442 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
443 /* can be modified without holding the LRU lock */
444 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
445 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
446 /* whether the multi-gen LRU is enabled */
447 bool enabled;
448 /* the memcg generation this lru_gen_folio belongs to */
449 u8 gen;
450 /* the list segment this lru_gen_folio belongs to */
451 u8 seg;
452 /* per-node lru_gen_folio list for global reclaim */
453 struct hlist_nulls_node list;
454};
455
456enum {
457 MM_LEAF_TOTAL, /* total leaf entries */
458 MM_LEAF_OLD, /* old leaf entries */
459 MM_LEAF_YOUNG, /* young leaf entries */
460 MM_NONLEAF_TOTAL, /* total non-leaf entries */
461 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
462 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
463 NR_MM_STATS
464};
465
466/* double-buffering Bloom filters */
467#define NR_BLOOM_FILTERS 2
468
469struct lru_gen_mm_state {
470 /* synced with max_seq after each iteration */
471 unsigned long seq;
472 /* where the current iteration continues after */
473 struct list_head *head;
474 /* where the last iteration ended before */
475 struct list_head *tail;
476 /* Bloom filters flip after each iteration */
477 unsigned long *filters[NR_BLOOM_FILTERS];
478 /* the mm stats for debugging */
479 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
480};
481
482struct lru_gen_mm_walk {
483 /* the lruvec under reclaim */
484 struct lruvec *lruvec;
485 /* max_seq from lru_gen_folio: can be out of date */
486 unsigned long seq;
487 /* the next address within an mm to scan */
488 unsigned long next_addr;
489 /* to batch promoted pages */
490 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
491 /* to batch the mm stats */
492 int mm_stats[NR_MM_STATS];
493 /* total batched items */
494 int batched;
495 bool can_swap;
496 bool force_scan;
497};
498
499/*
500 * For each node, memcgs are divided into two generations: the old and the
501 * young. For each generation, memcgs are randomly sharded into multiple bins
502 * to improve scalability. For each bin, the hlist_nulls is virtually divided
503 * into three segments: the head, the tail and the default.
504 *
505 * An onlining memcg is added to the tail of a random bin in the old generation.
506 * The eviction starts at the head of a random bin in the old generation. The
507 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
508 * the old generation, is incremented when all its bins become empty.
509 *
510 * There are four operations:
511 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
512 * current generation (old or young) and updates its "seg" to "head";
513 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
514 * current generation (old or young) and updates its "seg" to "tail";
515 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
516 * generation, updates its "gen" to "old" and resets its "seg" to "default";
517 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
518 * young generation, updates its "gen" to "young" and resets its "seg" to
519 * "default".
520 *
521 * The events that trigger the above operations are:
522 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
523 * 2. The first attempt to reclaim a memcg below low, which triggers
524 * MEMCG_LRU_TAIL;
525 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
526 * threshold, which triggers MEMCG_LRU_TAIL;
527 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
528 * threshold, which triggers MEMCG_LRU_YOUNG;
529 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
530 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
531 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
532 *
533 * Notes:
534 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
535 * of their max_seq counters ensures the eventual fairness to all eligible
536 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
537 * 2. There are only two valid generations: old (seq) and young (seq+1).
538 * MEMCG_NR_GENS is set to three so that when reading the generation counter
539 * locklessly, a stale value (seq-1) does not wraparound to young.
540 */
541#define MEMCG_NR_GENS 3
542#define MEMCG_NR_BINS 8
543
544struct lru_gen_memcg {
545 /* the per-node memcg generation counter */
546 unsigned long seq;
547 /* each memcg has one lru_gen_folio per node */
548 unsigned long nr_memcgs[MEMCG_NR_GENS];
549 /* per-node lru_gen_folio list for global reclaim */
550 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
551 /* protects the above */
552 spinlock_t lock;
553};
554
555void lru_gen_init_pgdat(struct pglist_data *pgdat);
556void lru_gen_init_lruvec(struct lruvec *lruvec);
557void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
558
559void lru_gen_init_memcg(struct mem_cgroup *memcg);
560void lru_gen_exit_memcg(struct mem_cgroup *memcg);
561void lru_gen_online_memcg(struct mem_cgroup *memcg);
562void lru_gen_offline_memcg(struct mem_cgroup *memcg);
563void lru_gen_release_memcg(struct mem_cgroup *memcg);
564void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
565
566#else /* !CONFIG_LRU_GEN */
567
568static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
569{
570}
571
572static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
573{
574}
575
576static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
577{
578}
579
580static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
581{
582}
583
584static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
585{
586}
587
588static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
589{
590}
591
592static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
593{
594}
595
596static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
597{
598}
599
600static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
601{
602}
603
604#endif /* CONFIG_LRU_GEN */
605
606struct lruvec {
607 struct list_head lists[NR_LRU_LISTS];
608 /* per lruvec lru_lock for memcg */
609 spinlock_t lru_lock;
610 /*
611 * These track the cost of reclaiming one LRU - file or anon -
612 * over the other. As the observed cost of reclaiming one LRU
613 * increases, the reclaim scan balance tips toward the other.
614 */
615 unsigned long anon_cost;
616 unsigned long file_cost;
617 /* Non-resident age, driven by LRU movement */
618 atomic_long_t nonresident_age;
619 /* Refaults at the time of last reclaim cycle */
620 unsigned long refaults[ANON_AND_FILE];
621 /* Various lruvec state flags (enum lruvec_flags) */
622 unsigned long flags;
623#ifdef CONFIG_LRU_GEN
624 /* evictable pages divided into generations */
625 struct lru_gen_folio lrugen;
626#ifdef CONFIG_LRU_GEN_WALKS_MMU
627 /* to concurrently iterate lru_gen_mm_list */
628 struct lru_gen_mm_state mm_state;
629#endif
630#endif /* CONFIG_LRU_GEN */
631#ifdef CONFIG_MEMCG
632 struct pglist_data *pgdat;
633#endif
634 struct zswap_lruvec_state zswap_lruvec_state;
635};
636
637/* Isolate for asynchronous migration */
638#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
639/* Isolate unevictable pages */
640#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
641
642/* LRU Isolation modes. */
643typedef unsigned __bitwise isolate_mode_t;
644
645enum zone_watermarks {
646 WMARK_MIN,
647 WMARK_LOW,
648 WMARK_HIGH,
649 WMARK_PROMO,
650 NR_WMARK
651};
652
653/*
654 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
655 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
656 * it should not contribute to serious fragmentation causing THP allocation
657 * failures.
658 */
659#ifdef CONFIG_TRANSPARENT_HUGEPAGE
660#define NR_PCP_THP 1
661#else
662#define NR_PCP_THP 0
663#endif
664#define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
665#define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
666
667#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
668#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
669#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
670#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
671
672/*
673 * Flags used in pcp->flags field.
674 *
675 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
676 * previous page freeing. To avoid to drain PCP for an accident
677 * high-order page freeing.
678 *
679 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
680 * draining PCP for consecutive high-order pages freeing without
681 * allocation if data cache slice of CPU is large enough. To reduce
682 * zone lock contention and keep cache-hot pages reusing.
683 */
684#define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
685#define PCPF_FREE_HIGH_BATCH BIT(1)
686
687struct per_cpu_pages {
688 spinlock_t lock; /* Protects lists field */
689 int count; /* number of pages in the list */
690 int high; /* high watermark, emptying needed */
691 int high_min; /* min high watermark */
692 int high_max; /* max high watermark */
693 int batch; /* chunk size for buddy add/remove */
694 u8 flags; /* protected by pcp->lock */
695 u8 alloc_factor; /* batch scaling factor during allocate */
696#ifdef CONFIG_NUMA
697 u8 expire; /* When 0, remote pagesets are drained */
698#endif
699 short free_count; /* consecutive free count */
700
701 /* Lists of pages, one per migrate type stored on the pcp-lists */
702 struct list_head lists[NR_PCP_LISTS];
703} ____cacheline_aligned_in_smp;
704
705struct per_cpu_zonestat {
706#ifdef CONFIG_SMP
707 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
708 s8 stat_threshold;
709#endif
710#ifdef CONFIG_NUMA
711 /*
712 * Low priority inaccurate counters that are only folded
713 * on demand. Use a large type to avoid the overhead of
714 * folding during refresh_cpu_vm_stats.
715 */
716 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
717#endif
718};
719
720struct per_cpu_nodestat {
721 s8 stat_threshold;
722 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
723};
724
725#endif /* !__GENERATING_BOUNDS.H */
726
727enum zone_type {
728 /*
729 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
730 * to DMA to all of the addressable memory (ZONE_NORMAL).
731 * On architectures where this area covers the whole 32 bit address
732 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
733 * DMA addressing constraints. This distinction is important as a 32bit
734 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
735 * platforms may need both zones as they support peripherals with
736 * different DMA addressing limitations.
737 */
738#ifdef CONFIG_ZONE_DMA
739 ZONE_DMA,
740#endif
741#ifdef CONFIG_ZONE_DMA32
742 ZONE_DMA32,
743#endif
744 /*
745 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
746 * performed on pages in ZONE_NORMAL if the DMA devices support
747 * transfers to all addressable memory.
748 */
749 ZONE_NORMAL,
750#ifdef CONFIG_HIGHMEM
751 /*
752 * A memory area that is only addressable by the kernel through
753 * mapping portions into its own address space. This is for example
754 * used by i386 to allow the kernel to address the memory beyond
755 * 900MB. The kernel will set up special mappings (page
756 * table entries on i386) for each page that the kernel needs to
757 * access.
758 */
759 ZONE_HIGHMEM,
760#endif
761 /*
762 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
763 * movable pages with few exceptional cases described below. Main use
764 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
765 * likely to succeed, and to locally limit unmovable allocations - e.g.,
766 * to increase the number of THP/huge pages. Notable special cases are:
767 *
768 * 1. Pinned pages: (long-term) pinning of movable pages might
769 * essentially turn such pages unmovable. Therefore, we do not allow
770 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
771 * faulted, they come from the right zone right away. However, it is
772 * still possible that address space already has pages in
773 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
774 * touches that memory before pinning). In such case we migrate them
775 * to a different zone. When migration fails - pinning fails.
776 * 2. memblock allocations: kernelcore/movablecore setups might create
777 * situations where ZONE_MOVABLE contains unmovable allocations
778 * after boot. Memory offlining and allocations fail early.
779 * 3. Memory holes: kernelcore/movablecore setups might create very rare
780 * situations where ZONE_MOVABLE contains memory holes after boot,
781 * for example, if we have sections that are only partially
782 * populated. Memory offlining and allocations fail early.
783 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
784 * memory offlining, such pages cannot be allocated.
785 * 5. Unmovable PG_offline pages: in paravirtualized environments,
786 * hotplugged memory blocks might only partially be managed by the
787 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
788 * parts not manged by the buddy are unmovable PG_offline pages. In
789 * some cases (virtio-mem), such pages can be skipped during
790 * memory offlining, however, cannot be moved/allocated. These
791 * techniques might use alloc_contig_range() to hide previously
792 * exposed pages from the buddy again (e.g., to implement some sort
793 * of memory unplug in virtio-mem).
794 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
795 * situations where ZERO_PAGE(0) which is allocated differently
796 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
797 * cannot be migrated.
798 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
799 * memory to the MOVABLE zone, the vmemmap pages are also placed in
800 * such zone. Such pages cannot be really moved around as they are
801 * self-stored in the range, but they are treated as movable when
802 * the range they describe is about to be offlined.
803 *
804 * In general, no unmovable allocations that degrade memory offlining
805 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
806 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
807 * if has_unmovable_pages() states that there are no unmovable pages,
808 * there can be false negatives).
809 */
810 ZONE_MOVABLE,
811#ifdef CONFIG_ZONE_DEVICE
812 ZONE_DEVICE,
813#endif
814 __MAX_NR_ZONES
815
816};
817
818#ifndef __GENERATING_BOUNDS_H
819
820#define ASYNC_AND_SYNC 2
821
822struct zone {
823 /* Read-mostly fields */
824
825 /* zone watermarks, access with *_wmark_pages(zone) macros */
826 unsigned long _watermark[NR_WMARK];
827 unsigned long watermark_boost;
828
829 unsigned long nr_reserved_highatomic;
830
831 /*
832 * We don't know if the memory that we're going to allocate will be
833 * freeable or/and it will be released eventually, so to avoid totally
834 * wasting several GB of ram we must reserve some of the lower zone
835 * memory (otherwise we risk to run OOM on the lower zones despite
836 * there being tons of freeable ram on the higher zones). This array is
837 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
838 * changes.
839 */
840 long lowmem_reserve[MAX_NR_ZONES];
841
842#ifdef CONFIG_NUMA
843 int node;
844#endif
845 struct pglist_data *zone_pgdat;
846 struct per_cpu_pages __percpu *per_cpu_pageset;
847 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
848 /*
849 * the high and batch values are copied to individual pagesets for
850 * faster access
851 */
852 int pageset_high_min;
853 int pageset_high_max;
854 int pageset_batch;
855
856#ifndef CONFIG_SPARSEMEM
857 /*
858 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
859 * In SPARSEMEM, this map is stored in struct mem_section
860 */
861 unsigned long *pageblock_flags;
862#endif /* CONFIG_SPARSEMEM */
863
864 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
865 unsigned long zone_start_pfn;
866
867 /*
868 * spanned_pages is the total pages spanned by the zone, including
869 * holes, which is calculated as:
870 * spanned_pages = zone_end_pfn - zone_start_pfn;
871 *
872 * present_pages is physical pages existing within the zone, which
873 * is calculated as:
874 * present_pages = spanned_pages - absent_pages(pages in holes);
875 *
876 * present_early_pages is present pages existing within the zone
877 * located on memory available since early boot, excluding hotplugged
878 * memory.
879 *
880 * managed_pages is present pages managed by the buddy system, which
881 * is calculated as (reserved_pages includes pages allocated by the
882 * bootmem allocator):
883 * managed_pages = present_pages - reserved_pages;
884 *
885 * cma pages is present pages that are assigned for CMA use
886 * (MIGRATE_CMA).
887 *
888 * So present_pages may be used by memory hotplug or memory power
889 * management logic to figure out unmanaged pages by checking
890 * (present_pages - managed_pages). And managed_pages should be used
891 * by page allocator and vm scanner to calculate all kinds of watermarks
892 * and thresholds.
893 *
894 * Locking rules:
895 *
896 * zone_start_pfn and spanned_pages are protected by span_seqlock.
897 * It is a seqlock because it has to be read outside of zone->lock,
898 * and it is done in the main allocator path. But, it is written
899 * quite infrequently.
900 *
901 * The span_seq lock is declared along with zone->lock because it is
902 * frequently read in proximity to zone->lock. It's good to
903 * give them a chance of being in the same cacheline.
904 *
905 * Write access to present_pages at runtime should be protected by
906 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
907 * present_pages should use get_online_mems() to get a stable value.
908 */
909 atomic_long_t managed_pages;
910 unsigned long spanned_pages;
911 unsigned long present_pages;
912#if defined(CONFIG_MEMORY_HOTPLUG)
913 unsigned long present_early_pages;
914#endif
915#ifdef CONFIG_CMA
916 unsigned long cma_pages;
917#endif
918
919 const char *name;
920
921#ifdef CONFIG_MEMORY_ISOLATION
922 /*
923 * Number of isolated pageblock. It is used to solve incorrect
924 * freepage counting problem due to racy retrieving migratetype
925 * of pageblock. Protected by zone->lock.
926 */
927 unsigned long nr_isolate_pageblock;
928#endif
929
930#ifdef CONFIG_MEMORY_HOTPLUG
931 /* see spanned/present_pages for more description */
932 seqlock_t span_seqlock;
933#endif
934
935 int initialized;
936
937 /* Write-intensive fields used from the page allocator */
938 CACHELINE_PADDING(_pad1_);
939
940 /* free areas of different sizes */
941 struct free_area free_area[NR_PAGE_ORDERS];
942
943#ifdef CONFIG_UNACCEPTED_MEMORY
944 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
945 struct list_head unaccepted_pages;
946#endif
947
948 /* zone flags, see below */
949 unsigned long flags;
950
951 /* Primarily protects free_area */
952 spinlock_t lock;
953
954 /* Write-intensive fields used by compaction and vmstats. */
955 CACHELINE_PADDING(_pad2_);
956
957 /*
958 * When free pages are below this point, additional steps are taken
959 * when reading the number of free pages to avoid per-cpu counter
960 * drift allowing watermarks to be breached
961 */
962 unsigned long percpu_drift_mark;
963
964#if defined CONFIG_COMPACTION || defined CONFIG_CMA
965 /* pfn where compaction free scanner should start */
966 unsigned long compact_cached_free_pfn;
967 /* pfn where compaction migration scanner should start */
968 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
969 unsigned long compact_init_migrate_pfn;
970 unsigned long compact_init_free_pfn;
971#endif
972
973#ifdef CONFIG_COMPACTION
974 /*
975 * On compaction failure, 1<<compact_defer_shift compactions
976 * are skipped before trying again. The number attempted since
977 * last failure is tracked with compact_considered.
978 * compact_order_failed is the minimum compaction failed order.
979 */
980 unsigned int compact_considered;
981 unsigned int compact_defer_shift;
982 int compact_order_failed;
983#endif
984
985#if defined CONFIG_COMPACTION || defined CONFIG_CMA
986 /* Set to true when the PG_migrate_skip bits should be cleared */
987 bool compact_blockskip_flush;
988#endif
989
990 bool contiguous;
991
992 CACHELINE_PADDING(_pad3_);
993 /* Zone statistics */
994 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
995 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
996} ____cacheline_internodealigned_in_smp;
997
998enum pgdat_flags {
999 PGDAT_DIRTY, /* reclaim scanning has recently found
1000 * many dirty file pages at the tail
1001 * of the LRU.
1002 */
1003 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1004 * many pages under writeback
1005 */
1006 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1007};
1008
1009enum zone_flags {
1010 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1011 * Cleared when kswapd is woken.
1012 */
1013 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1014 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1015};
1016
1017static inline unsigned long zone_managed_pages(struct zone *zone)
1018{
1019 return (unsigned long)atomic_long_read(&zone->managed_pages);
1020}
1021
1022static inline unsigned long zone_cma_pages(struct zone *zone)
1023{
1024#ifdef CONFIG_CMA
1025 return zone->cma_pages;
1026#else
1027 return 0;
1028#endif
1029}
1030
1031static inline unsigned long zone_end_pfn(const struct zone *zone)
1032{
1033 return zone->zone_start_pfn + zone->spanned_pages;
1034}
1035
1036static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1037{
1038 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1039}
1040
1041static inline bool zone_is_initialized(struct zone *zone)
1042{
1043 return zone->initialized;
1044}
1045
1046static inline bool zone_is_empty(struct zone *zone)
1047{
1048 return zone->spanned_pages == 0;
1049}
1050
1051#ifndef BUILD_VDSO32_64
1052/*
1053 * The zone field is never updated after free_area_init_core()
1054 * sets it, so none of the operations on it need to be atomic.
1055 */
1056
1057/* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1058#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1059#define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1060#define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1061#define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1062#define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1063#define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1064#define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1065
1066/*
1067 * Define the bit shifts to access each section. For non-existent
1068 * sections we define the shift as 0; that plus a 0 mask ensures
1069 * the compiler will optimise away reference to them.
1070 */
1071#define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1072#define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1073#define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1074#define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1075#define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1076
1077/* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1078#ifdef NODE_NOT_IN_PAGE_FLAGS
1079#define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1080#define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1081 SECTIONS_PGOFF : ZONES_PGOFF)
1082#else
1083#define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1084#define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1085 NODES_PGOFF : ZONES_PGOFF)
1086#endif
1087
1088#define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1089
1090#define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1091#define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1092#define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1093#define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1094#define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1095#define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1096
1097static inline enum zone_type page_zonenum(const struct page *page)
1098{
1099 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1100 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1101}
1102
1103static inline enum zone_type folio_zonenum(const struct folio *folio)
1104{
1105 return page_zonenum(&folio->page);
1106}
1107
1108#ifdef CONFIG_ZONE_DEVICE
1109static inline bool is_zone_device_page(const struct page *page)
1110{
1111 return page_zonenum(page) == ZONE_DEVICE;
1112}
1113
1114/*
1115 * Consecutive zone device pages should not be merged into the same sgl
1116 * or bvec segment with other types of pages or if they belong to different
1117 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1118 * without scanning the entire segment. This helper returns true either if
1119 * both pages are not zone device pages or both pages are zone device pages
1120 * with the same pgmap.
1121 */
1122static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1123 const struct page *b)
1124{
1125 if (is_zone_device_page(a) != is_zone_device_page(b))
1126 return false;
1127 if (!is_zone_device_page(a))
1128 return true;
1129 return a->pgmap == b->pgmap;
1130}
1131
1132extern void memmap_init_zone_device(struct zone *, unsigned long,
1133 unsigned long, struct dev_pagemap *);
1134#else
1135static inline bool is_zone_device_page(const struct page *page)
1136{
1137 return false;
1138}
1139static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1140 const struct page *b)
1141{
1142 return true;
1143}
1144#endif
1145
1146static inline bool folio_is_zone_device(const struct folio *folio)
1147{
1148 return is_zone_device_page(&folio->page);
1149}
1150
1151static inline bool is_zone_movable_page(const struct page *page)
1152{
1153 return page_zonenum(page) == ZONE_MOVABLE;
1154}
1155
1156static inline bool folio_is_zone_movable(const struct folio *folio)
1157{
1158 return folio_zonenum(folio) == ZONE_MOVABLE;
1159}
1160#endif
1161
1162/*
1163 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1164 * intersection with the given zone
1165 */
1166static inline bool zone_intersects(struct zone *zone,
1167 unsigned long start_pfn, unsigned long nr_pages)
1168{
1169 if (zone_is_empty(zone))
1170 return false;
1171 if (start_pfn >= zone_end_pfn(zone) ||
1172 start_pfn + nr_pages <= zone->zone_start_pfn)
1173 return false;
1174
1175 return true;
1176}
1177
1178/*
1179 * The "priority" of VM scanning is how much of the queues we will scan in one
1180 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1181 * queues ("queue_length >> 12") during an aging round.
1182 */
1183#define DEF_PRIORITY 12
1184
1185/* Maximum number of zones on a zonelist */
1186#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1187
1188enum {
1189 ZONELIST_FALLBACK, /* zonelist with fallback */
1190#ifdef CONFIG_NUMA
1191 /*
1192 * The NUMA zonelists are doubled because we need zonelists that
1193 * restrict the allocations to a single node for __GFP_THISNODE.
1194 */
1195 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1196#endif
1197 MAX_ZONELISTS
1198};
1199
1200/*
1201 * This struct contains information about a zone in a zonelist. It is stored
1202 * here to avoid dereferences into large structures and lookups of tables
1203 */
1204struct zoneref {
1205 struct zone *zone; /* Pointer to actual zone */
1206 int zone_idx; /* zone_idx(zoneref->zone) */
1207};
1208
1209/*
1210 * One allocation request operates on a zonelist. A zonelist
1211 * is a list of zones, the first one is the 'goal' of the
1212 * allocation, the other zones are fallback zones, in decreasing
1213 * priority.
1214 *
1215 * To speed the reading of the zonelist, the zonerefs contain the zone index
1216 * of the entry being read. Helper functions to access information given
1217 * a struct zoneref are
1218 *
1219 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1220 * zonelist_zone_idx() - Return the index of the zone for an entry
1221 * zonelist_node_idx() - Return the index of the node for an entry
1222 */
1223struct zonelist {
1224 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1225};
1226
1227/*
1228 * The array of struct pages for flatmem.
1229 * It must be declared for SPARSEMEM as well because there are configurations
1230 * that rely on that.
1231 */
1232extern struct page *mem_map;
1233
1234#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1235struct deferred_split {
1236 spinlock_t split_queue_lock;
1237 struct list_head split_queue;
1238 unsigned long split_queue_len;
1239};
1240#endif
1241
1242#ifdef CONFIG_MEMORY_FAILURE
1243/*
1244 * Per NUMA node memory failure handling statistics.
1245 */
1246struct memory_failure_stats {
1247 /*
1248 * Number of raw pages poisoned.
1249 * Cases not accounted: memory outside kernel control, offline page,
1250 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1251 * error events, and unpoison actions from hwpoison_unpoison.
1252 */
1253 unsigned long total;
1254 /*
1255 * Recovery results of poisoned raw pages handled by memory_failure,
1256 * in sync with mf_result.
1257 * total = ignored + failed + delayed + recovered.
1258 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1259 */
1260 unsigned long ignored;
1261 unsigned long failed;
1262 unsigned long delayed;
1263 unsigned long recovered;
1264};
1265#endif
1266
1267/*
1268 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1269 * it's memory layout. On UMA machines there is a single pglist_data which
1270 * describes the whole memory.
1271 *
1272 * Memory statistics and page replacement data structures are maintained on a
1273 * per-zone basis.
1274 */
1275typedef struct pglist_data {
1276 /*
1277 * node_zones contains just the zones for THIS node. Not all of the
1278 * zones may be populated, but it is the full list. It is referenced by
1279 * this node's node_zonelists as well as other node's node_zonelists.
1280 */
1281 struct zone node_zones[MAX_NR_ZONES];
1282
1283 /*
1284 * node_zonelists contains references to all zones in all nodes.
1285 * Generally the first zones will be references to this node's
1286 * node_zones.
1287 */
1288 struct zonelist node_zonelists[MAX_ZONELISTS];
1289
1290 int nr_zones; /* number of populated zones in this node */
1291#ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1292 struct page *node_mem_map;
1293#ifdef CONFIG_PAGE_EXTENSION
1294 struct page_ext *node_page_ext;
1295#endif
1296#endif
1297#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1298 /*
1299 * Must be held any time you expect node_start_pfn,
1300 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1301 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1302 * init.
1303 *
1304 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1305 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1306 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1307 *
1308 * Nests above zone->lock and zone->span_seqlock
1309 */
1310 spinlock_t node_size_lock;
1311#endif
1312 unsigned long node_start_pfn;
1313 unsigned long node_present_pages; /* total number of physical pages */
1314 unsigned long node_spanned_pages; /* total size of physical page
1315 range, including holes */
1316 int node_id;
1317 wait_queue_head_t kswapd_wait;
1318 wait_queue_head_t pfmemalloc_wait;
1319
1320 /* workqueues for throttling reclaim for different reasons. */
1321 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1322
1323 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1324 unsigned long nr_reclaim_start; /* nr pages written while throttled
1325 * when throttling started. */
1326#ifdef CONFIG_MEMORY_HOTPLUG
1327 struct mutex kswapd_lock;
1328#endif
1329 struct task_struct *kswapd; /* Protected by kswapd_lock */
1330 int kswapd_order;
1331 enum zone_type kswapd_highest_zoneidx;
1332
1333 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1334
1335#ifdef CONFIG_COMPACTION
1336 int kcompactd_max_order;
1337 enum zone_type kcompactd_highest_zoneidx;
1338 wait_queue_head_t kcompactd_wait;
1339 struct task_struct *kcompactd;
1340 bool proactive_compact_trigger;
1341#endif
1342 /*
1343 * This is a per-node reserve of pages that are not available
1344 * to userspace allocations.
1345 */
1346 unsigned long totalreserve_pages;
1347
1348#ifdef CONFIG_NUMA
1349 /*
1350 * node reclaim becomes active if more unmapped pages exist.
1351 */
1352 unsigned long min_unmapped_pages;
1353 unsigned long min_slab_pages;
1354#endif /* CONFIG_NUMA */
1355
1356 /* Write-intensive fields used by page reclaim */
1357 CACHELINE_PADDING(_pad1_);
1358
1359#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1360 /*
1361 * If memory initialisation on large machines is deferred then this
1362 * is the first PFN that needs to be initialised.
1363 */
1364 unsigned long first_deferred_pfn;
1365#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1366
1367#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1368 struct deferred_split deferred_split_queue;
1369#endif
1370
1371#ifdef CONFIG_NUMA_BALANCING
1372 /* start time in ms of current promote rate limit period */
1373 unsigned int nbp_rl_start;
1374 /* number of promote candidate pages at start time of current rate limit period */
1375 unsigned long nbp_rl_nr_cand;
1376 /* promote threshold in ms */
1377 unsigned int nbp_threshold;
1378 /* start time in ms of current promote threshold adjustment period */
1379 unsigned int nbp_th_start;
1380 /*
1381 * number of promote candidate pages at start time of current promote
1382 * threshold adjustment period
1383 */
1384 unsigned long nbp_th_nr_cand;
1385#endif
1386 /* Fields commonly accessed by the page reclaim scanner */
1387
1388 /*
1389 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1390 *
1391 * Use mem_cgroup_lruvec() to look up lruvecs.
1392 */
1393 struct lruvec __lruvec;
1394
1395 unsigned long flags;
1396
1397#ifdef CONFIG_LRU_GEN
1398 /* kswap mm walk data */
1399 struct lru_gen_mm_walk mm_walk;
1400 /* lru_gen_folio list */
1401 struct lru_gen_memcg memcg_lru;
1402#endif
1403
1404 CACHELINE_PADDING(_pad2_);
1405
1406 /* Per-node vmstats */
1407 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1408 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1409#ifdef CONFIG_NUMA
1410 struct memory_tier __rcu *memtier;
1411#endif
1412#ifdef CONFIG_MEMORY_FAILURE
1413 struct memory_failure_stats mf_stats;
1414#endif
1415} pg_data_t;
1416
1417#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1418#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1419
1420#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1421#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1422
1423static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1424{
1425 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1426}
1427
1428#include <linux/memory_hotplug.h>
1429
1430void build_all_zonelists(pg_data_t *pgdat);
1431void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1432 enum zone_type highest_zoneidx);
1433bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1434 int highest_zoneidx, unsigned int alloc_flags,
1435 long free_pages);
1436bool zone_watermark_ok(struct zone *z, unsigned int order,
1437 unsigned long mark, int highest_zoneidx,
1438 unsigned int alloc_flags);
1439bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1440 unsigned long mark, int highest_zoneidx);
1441/*
1442 * Memory initialization context, use to differentiate memory added by
1443 * the platform statically or via memory hotplug interface.
1444 */
1445enum meminit_context {
1446 MEMINIT_EARLY,
1447 MEMINIT_HOTPLUG,
1448};
1449
1450extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1451 unsigned long size);
1452
1453extern void lruvec_init(struct lruvec *lruvec);
1454
1455static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1456{
1457#ifdef CONFIG_MEMCG
1458 return lruvec->pgdat;
1459#else
1460 return container_of(lruvec, struct pglist_data, __lruvec);
1461#endif
1462}
1463
1464#ifdef CONFIG_HAVE_MEMORYLESS_NODES
1465int local_memory_node(int node_id);
1466#else
1467static inline int local_memory_node(int node_id) { return node_id; };
1468#endif
1469
1470/*
1471 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1472 */
1473#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1474
1475#ifdef CONFIG_ZONE_DEVICE
1476static inline bool zone_is_zone_device(struct zone *zone)
1477{
1478 return zone_idx(zone) == ZONE_DEVICE;
1479}
1480#else
1481static inline bool zone_is_zone_device(struct zone *zone)
1482{
1483 return false;
1484}
1485#endif
1486
1487/*
1488 * Returns true if a zone has pages managed by the buddy allocator.
1489 * All the reclaim decisions have to use this function rather than
1490 * populated_zone(). If the whole zone is reserved then we can easily
1491 * end up with populated_zone() && !managed_zone().
1492 */
1493static inline bool managed_zone(struct zone *zone)
1494{
1495 return zone_managed_pages(zone);
1496}
1497
1498/* Returns true if a zone has memory */
1499static inline bool populated_zone(struct zone *zone)
1500{
1501 return zone->present_pages;
1502}
1503
1504#ifdef CONFIG_NUMA
1505static inline int zone_to_nid(struct zone *zone)
1506{
1507 return zone->node;
1508}
1509
1510static inline void zone_set_nid(struct zone *zone, int nid)
1511{
1512 zone->node = nid;
1513}
1514#else
1515static inline int zone_to_nid(struct zone *zone)
1516{
1517 return 0;
1518}
1519
1520static inline void zone_set_nid(struct zone *zone, int nid) {}
1521#endif
1522
1523extern int movable_zone;
1524
1525static inline int is_highmem_idx(enum zone_type idx)
1526{
1527#ifdef CONFIG_HIGHMEM
1528 return (idx == ZONE_HIGHMEM ||
1529 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1530#else
1531 return 0;
1532#endif
1533}
1534
1535/**
1536 * is_highmem - helper function to quickly check if a struct zone is a
1537 * highmem zone or not. This is an attempt to keep references
1538 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1539 * @zone: pointer to struct zone variable
1540 * Return: 1 for a highmem zone, 0 otherwise
1541 */
1542static inline int is_highmem(struct zone *zone)
1543{
1544 return is_highmem_idx(zone_idx(zone));
1545}
1546
1547#ifdef CONFIG_ZONE_DMA
1548bool has_managed_dma(void);
1549#else
1550static inline bool has_managed_dma(void)
1551{
1552 return false;
1553}
1554#endif
1555
1556
1557#ifndef CONFIG_NUMA
1558
1559extern struct pglist_data contig_page_data;
1560static inline struct pglist_data *NODE_DATA(int nid)
1561{
1562 return &contig_page_data;
1563}
1564
1565#else /* CONFIG_NUMA */
1566
1567#include <asm/mmzone.h>
1568
1569#endif /* !CONFIG_NUMA */
1570
1571extern struct pglist_data *first_online_pgdat(void);
1572extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1573extern struct zone *next_zone(struct zone *zone);
1574
1575/**
1576 * for_each_online_pgdat - helper macro to iterate over all online nodes
1577 * @pgdat: pointer to a pg_data_t variable
1578 */
1579#define for_each_online_pgdat(pgdat) \
1580 for (pgdat = first_online_pgdat(); \
1581 pgdat; \
1582 pgdat = next_online_pgdat(pgdat))
1583/**
1584 * for_each_zone - helper macro to iterate over all memory zones
1585 * @zone: pointer to struct zone variable
1586 *
1587 * The user only needs to declare the zone variable, for_each_zone
1588 * fills it in.
1589 */
1590#define for_each_zone(zone) \
1591 for (zone = (first_online_pgdat())->node_zones; \
1592 zone; \
1593 zone = next_zone(zone))
1594
1595#define for_each_populated_zone(zone) \
1596 for (zone = (first_online_pgdat())->node_zones; \
1597 zone; \
1598 zone = next_zone(zone)) \
1599 if (!populated_zone(zone)) \
1600 ; /* do nothing */ \
1601 else
1602
1603static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1604{
1605 return zoneref->zone;
1606}
1607
1608static inline int zonelist_zone_idx(struct zoneref *zoneref)
1609{
1610 return zoneref->zone_idx;
1611}
1612
1613static inline int zonelist_node_idx(struct zoneref *zoneref)
1614{
1615 return zone_to_nid(zoneref->zone);
1616}
1617
1618struct zoneref *__next_zones_zonelist(struct zoneref *z,
1619 enum zone_type highest_zoneidx,
1620 nodemask_t *nodes);
1621
1622/**
1623 * 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
1624 * @z: The cursor used as a starting point for the search
1625 * @highest_zoneidx: The zone index of the highest zone to return
1626 * @nodes: An optional nodemask to filter the zonelist with
1627 *
1628 * This function returns the next zone at or below a given zone index that is
1629 * within the allowed nodemask using a cursor as the starting point for the
1630 * search. The zoneref returned is a cursor that represents the current zone
1631 * being examined. It should be advanced by one before calling
1632 * next_zones_zonelist again.
1633 *
1634 * Return: the next zone at or below highest_zoneidx within the allowed
1635 * nodemask using a cursor within a zonelist as a starting point
1636 */
1637static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1638 enum zone_type highest_zoneidx,
1639 nodemask_t *nodes)
1640{
1641 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1642 return z;
1643 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1644}
1645
1646/**
1647 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1648 * @zonelist: The zonelist to search for a suitable zone
1649 * @highest_zoneidx: The zone index of the highest zone to return
1650 * @nodes: An optional nodemask to filter the zonelist with
1651 *
1652 * This function returns the first zone at or below a given zone index that is
1653 * within the allowed nodemask. The zoneref returned is a cursor that can be
1654 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1655 * one before calling.
1656 *
1657 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1658 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1659 * update due to cpuset modification.
1660 *
1661 * Return: Zoneref pointer for the first suitable zone found
1662 */
1663static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1664 enum zone_type highest_zoneidx,
1665 nodemask_t *nodes)
1666{
1667 return next_zones_zonelist(zonelist->_zonerefs,
1668 highest_zoneidx, nodes);
1669}
1670
1671/**
1672 * 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
1673 * @zone: The current zone in the iterator
1674 * @z: The current pointer within zonelist->_zonerefs being iterated
1675 * @zlist: The zonelist being iterated
1676 * @highidx: The zone index of the highest zone to return
1677 * @nodemask: Nodemask allowed by the allocator
1678 *
1679 * This iterator iterates though all zones at or below a given zone index and
1680 * within a given nodemask
1681 */
1682#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1683 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1684 zone; \
1685 z = next_zones_zonelist(++z, highidx, nodemask), \
1686 zone = zonelist_zone(z))
1687
1688#define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1689 for (zone = z->zone; \
1690 zone; \
1691 z = next_zones_zonelist(++z, highidx, nodemask), \
1692 zone = zonelist_zone(z))
1693
1694
1695/**
1696 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1697 * @zone: The current zone in the iterator
1698 * @z: The current pointer within zonelist->zones being iterated
1699 * @zlist: The zonelist being iterated
1700 * @highidx: The zone index of the highest zone to return
1701 *
1702 * This iterator iterates though all zones at or below a given zone index.
1703 */
1704#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1705 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1706
1707/* Whether the 'nodes' are all movable nodes */
1708static inline bool movable_only_nodes(nodemask_t *nodes)
1709{
1710 struct zonelist *zonelist;
1711 struct zoneref *z;
1712 int nid;
1713
1714 if (nodes_empty(*nodes))
1715 return false;
1716
1717 /*
1718 * We can chose arbitrary node from the nodemask to get a
1719 * zonelist as they are interlinked. We just need to find
1720 * at least one zone that can satisfy kernel allocations.
1721 */
1722 nid = first_node(*nodes);
1723 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1724 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1725 return (!z->zone) ? true : false;
1726}
1727
1728
1729#ifdef CONFIG_SPARSEMEM
1730#include <asm/sparsemem.h>
1731#endif
1732
1733#ifdef CONFIG_FLATMEM
1734#define pfn_to_nid(pfn) (0)
1735#endif
1736
1737#ifdef CONFIG_SPARSEMEM
1738
1739/*
1740 * PA_SECTION_SHIFT physical address to/from section number
1741 * PFN_SECTION_SHIFT pfn to/from section number
1742 */
1743#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1744#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1745
1746#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1747
1748#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1749#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1750
1751#define SECTION_BLOCKFLAGS_BITS \
1752 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1753
1754#if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1755#error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1756#endif
1757
1758static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1759{
1760 return pfn >> PFN_SECTION_SHIFT;
1761}
1762static inline unsigned long section_nr_to_pfn(unsigned long sec)
1763{
1764 return sec << PFN_SECTION_SHIFT;
1765}
1766
1767#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1768#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1769
1770#define SUBSECTION_SHIFT 21
1771#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1772
1773#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1774#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1775#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1776
1777#if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1778#error Subsection size exceeds section size
1779#else
1780#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1781#endif
1782
1783#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1784#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1785
1786struct mem_section_usage {
1787 struct rcu_head rcu;
1788#ifdef CONFIG_SPARSEMEM_VMEMMAP
1789 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1790#endif
1791 /* See declaration of similar field in struct zone */
1792 unsigned long pageblock_flags[0];
1793};
1794
1795void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1796
1797struct page;
1798struct page_ext;
1799struct mem_section {
1800 /*
1801 * This is, logically, a pointer to an array of struct
1802 * pages. However, it is stored with some other magic.
1803 * (see sparse.c::sparse_init_one_section())
1804 *
1805 * Additionally during early boot we encode node id of
1806 * the location of the section here to guide allocation.
1807 * (see sparse.c::memory_present())
1808 *
1809 * Making it a UL at least makes someone do a cast
1810 * before using it wrong.
1811 */
1812 unsigned long section_mem_map;
1813
1814 struct mem_section_usage *usage;
1815#ifdef CONFIG_PAGE_EXTENSION
1816 /*
1817 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1818 * section. (see page_ext.h about this.)
1819 */
1820 struct page_ext *page_ext;
1821 unsigned long pad;
1822#endif
1823 /*
1824 * WARNING: mem_section must be a power-of-2 in size for the
1825 * calculation and use of SECTION_ROOT_MASK to make sense.
1826 */
1827};
1828
1829#ifdef CONFIG_SPARSEMEM_EXTREME
1830#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1831#else
1832#define SECTIONS_PER_ROOT 1
1833#endif
1834
1835#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1836#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1837#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1838
1839#ifdef CONFIG_SPARSEMEM_EXTREME
1840extern struct mem_section **mem_section;
1841#else
1842extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1843#endif
1844
1845static inline unsigned long *section_to_usemap(struct mem_section *ms)
1846{
1847 return ms->usage->pageblock_flags;
1848}
1849
1850static inline struct mem_section *__nr_to_section(unsigned long nr)
1851{
1852 unsigned long root = SECTION_NR_TO_ROOT(nr);
1853
1854 if (unlikely(root >= NR_SECTION_ROOTS))
1855 return NULL;
1856
1857#ifdef CONFIG_SPARSEMEM_EXTREME
1858 if (!mem_section || !mem_section[root])
1859 return NULL;
1860#endif
1861 return &mem_section[root][nr & SECTION_ROOT_MASK];
1862}
1863extern size_t mem_section_usage_size(void);
1864
1865/*
1866 * We use the lower bits of the mem_map pointer to store
1867 * a little bit of information. The pointer is calculated
1868 * as mem_map - section_nr_to_pfn(pnum). The result is
1869 * aligned to the minimum alignment of the two values:
1870 * 1. All mem_map arrays are page-aligned.
1871 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1872 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1873 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1874 * worst combination is powerpc with 256k pages,
1875 * which results in PFN_SECTION_SHIFT equal 6.
1876 * To sum it up, at least 6 bits are available on all architectures.
1877 * However, we can exceed 6 bits on some other architectures except
1878 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1879 * with the worst case of 64K pages on arm64) if we make sure the
1880 * exceeded bit is not applicable to powerpc.
1881 */
1882enum {
1883 SECTION_MARKED_PRESENT_BIT,
1884 SECTION_HAS_MEM_MAP_BIT,
1885 SECTION_IS_ONLINE_BIT,
1886 SECTION_IS_EARLY_BIT,
1887#ifdef CONFIG_ZONE_DEVICE
1888 SECTION_TAINT_ZONE_DEVICE_BIT,
1889#endif
1890 SECTION_MAP_LAST_BIT,
1891};
1892
1893#define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1894#define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1895#define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1896#define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1897#ifdef CONFIG_ZONE_DEVICE
1898#define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1899#endif
1900#define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1901#define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1902
1903static inline struct page *__section_mem_map_addr(struct mem_section *section)
1904{
1905 unsigned long map = section->section_mem_map;
1906 map &= SECTION_MAP_MASK;
1907 return (struct page *)map;
1908}
1909
1910static inline int present_section(struct mem_section *section)
1911{
1912 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1913}
1914
1915static inline int present_section_nr(unsigned long nr)
1916{
1917 return present_section(__nr_to_section(nr));
1918}
1919
1920static inline int valid_section(struct mem_section *section)
1921{
1922 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1923}
1924
1925static inline int early_section(struct mem_section *section)
1926{
1927 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1928}
1929
1930static inline int valid_section_nr(unsigned long nr)
1931{
1932 return valid_section(__nr_to_section(nr));
1933}
1934
1935static inline int online_section(struct mem_section *section)
1936{
1937 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1938}
1939
1940#ifdef CONFIG_ZONE_DEVICE
1941static inline int online_device_section(struct mem_section *section)
1942{
1943 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1944
1945 return section && ((section->section_mem_map & flags) == flags);
1946}
1947#else
1948static inline int online_device_section(struct mem_section *section)
1949{
1950 return 0;
1951}
1952#endif
1953
1954static inline int online_section_nr(unsigned long nr)
1955{
1956 return online_section(__nr_to_section(nr));
1957}
1958
1959#ifdef CONFIG_MEMORY_HOTPLUG
1960void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1961void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1962#endif
1963
1964static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1965{
1966 return __nr_to_section(pfn_to_section_nr(pfn));
1967}
1968
1969extern unsigned long __highest_present_section_nr;
1970
1971static inline int subsection_map_index(unsigned long pfn)
1972{
1973 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1974}
1975
1976#ifdef CONFIG_SPARSEMEM_VMEMMAP
1977static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1978{
1979 int idx = subsection_map_index(pfn);
1980
1981 return test_bit(idx, READ_ONCE(ms->usage)->subsection_map);
1982}
1983#else
1984static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1985{
1986 return 1;
1987}
1988#endif
1989
1990#ifndef CONFIG_HAVE_ARCH_PFN_VALID
1991/**
1992 * pfn_valid - check if there is a valid memory map entry for a PFN
1993 * @pfn: the page frame number to check
1994 *
1995 * Check if there is a valid memory map entry aka struct page for the @pfn.
1996 * Note, that availability of the memory map entry does not imply that
1997 * there is actual usable memory at that @pfn. The struct page may
1998 * represent a hole or an unusable page frame.
1999 *
2000 * Return: 1 for PFNs that have memory map entries and 0 otherwise
2001 */
2002static inline int pfn_valid(unsigned long pfn)
2003{
2004 struct mem_section *ms;
2005 int ret;
2006
2007 /*
2008 * Ensure the upper PAGE_SHIFT bits are clear in the
2009 * pfn. Else it might lead to false positives when
2010 * some of the upper bits are set, but the lower bits
2011 * match a valid pfn.
2012 */
2013 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2014 return 0;
2015
2016 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2017 return 0;
2018 ms = __pfn_to_section(pfn);
2019 rcu_read_lock_sched();
2020 if (!valid_section(ms)) {
2021 rcu_read_unlock_sched();
2022 return 0;
2023 }
2024 /*
2025 * Traditionally early sections always returned pfn_valid() for
2026 * the entire section-sized span.
2027 */
2028 ret = early_section(ms) || pfn_section_valid(ms, pfn);
2029 rcu_read_unlock_sched();
2030
2031 return ret;
2032}
2033#endif
2034
2035static inline int pfn_in_present_section(unsigned long pfn)
2036{
2037 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2038 return 0;
2039 return present_section(__pfn_to_section(pfn));
2040}
2041
2042static inline unsigned long next_present_section_nr(unsigned long section_nr)
2043{
2044 while (++section_nr <= __highest_present_section_nr) {
2045 if (present_section_nr(section_nr))
2046 return section_nr;
2047 }
2048
2049 return -1;
2050}
2051
2052/*
2053 * These are _only_ used during initialisation, therefore they
2054 * can use __initdata ... They could have names to indicate
2055 * this restriction.
2056 */
2057#ifdef CONFIG_NUMA
2058#define pfn_to_nid(pfn) \
2059({ \
2060 unsigned long __pfn_to_nid_pfn = (pfn); \
2061 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2062})
2063#else
2064#define pfn_to_nid(pfn) (0)
2065#endif
2066
2067void sparse_init(void);
2068#else
2069#define sparse_init() do {} while (0)
2070#define sparse_index_init(_sec, _nid) do {} while (0)
2071#define pfn_in_present_section pfn_valid
2072#define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2073#endif /* CONFIG_SPARSEMEM */
2074
2075#endif /* !__GENERATING_BOUNDS.H */
2076#endif /* !__ASSEMBLY__ */
2077#endif /* _LINUX_MMZONE_H */