1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MMZONE_H
3 #define _LINUX_MMZONE_H
6 #ifndef __GENERATING_BOUNDS_H
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <linux/cache.h>
13 #include <linux/threads.h>
14 #include <linux/numa.h>
15 #include <linux/init.h>
16 #include <linux/seqlock.h>
17 #include <linux/nodemask.h>
18 #include <linux/pageblock-flags.h>
19 #include <linux/page-flags-layout.h>
20 #include <linux/atomic.h>
21 #include <linux/mm_types.h>
22 #include <linux/page-flags.h>
23 #include <linux/local_lock.h>
26 /* Free memory management - zoned buddy allocator. */
27 #ifndef CONFIG_ARCH_FORCE_MAX_ORDER
30 #define MAX_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
32 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
35 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
36 * costly to service. That is between allocation orders which should
37 * coalesce naturally under reasonable reclaim pressure and those which
40 #define PAGE_ALLOC_COSTLY_ORDER 3
46 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
47 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
50 * MIGRATE_CMA migration type is designed to mimic the way
51 * ZONE_MOVABLE works. Only movable pages can be allocated
52 * from MIGRATE_CMA pageblocks and page allocator never
53 * implicitly change migration type of MIGRATE_CMA pageblock.
55 * The way to use it is to change migratetype of a range of
56 * pageblocks to MIGRATE_CMA which can be done by
57 * __free_pageblock_cma() function.
61 #ifdef CONFIG_MEMORY_ISOLATION
62 MIGRATE_ISOLATE, /* can't allocate from here */
67 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
68 extern const char * const migratetype_names[MIGRATE_TYPES];
71 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
72 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
74 # define is_migrate_cma(migratetype) false
75 # define is_migrate_cma_page(_page) false
78 static inline bool is_migrate_movable(int mt)
80 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
84 * Check whether a migratetype can be merged with another migratetype.
86 * It is only mergeable when it can fall back to other migratetypes for
87 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
89 static inline bool migratetype_is_mergeable(int mt)
91 return mt < MIGRATE_PCPTYPES;
94 #define for_each_migratetype_order(order, type) \
95 for (order = 0; order < MAX_ORDER; order++) \
96 for (type = 0; type < MIGRATE_TYPES; type++)
98 extern int page_group_by_mobility_disabled;
100 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
102 #define get_pageblock_migratetype(page) \
103 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
106 struct list_head free_list[MIGRATE_TYPES];
107 unsigned long nr_free;
110 static inline struct page *get_page_from_free_area(struct free_area *area,
113 return list_first_entry_or_null(&area->free_list[migratetype],
117 static inline bool free_area_empty(struct free_area *area, int migratetype)
119 return list_empty(&area->free_list[migratetype]);
125 * Add a wild amount of padding here to ensure data fall into separate
126 * cachelines. There are very few zone structures in the machine, so space
127 * consumption is not a concern here.
129 #if defined(CONFIG_SMP)
130 struct zone_padding {
132 } ____cacheline_internodealigned_in_smp;
133 #define ZONE_PADDING(name) struct zone_padding name;
135 #define ZONE_PADDING(name)
139 enum numa_stat_item {
140 NUMA_HIT, /* allocated in intended node */
141 NUMA_MISS, /* allocated in non intended node */
142 NUMA_FOREIGN, /* was intended here, hit elsewhere */
143 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
144 NUMA_LOCAL, /* allocation from local node */
145 NUMA_OTHER, /* allocation from other node */
146 NR_VM_NUMA_EVENT_ITEMS
149 #define NR_VM_NUMA_EVENT_ITEMS 0
152 enum zone_stat_item {
153 /* First 128 byte cacheline (assuming 64 bit words) */
155 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
156 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
158 NR_ZONE_INACTIVE_FILE,
161 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
162 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
163 /* Second 128 byte cacheline */
165 #if IS_ENABLED(CONFIG_ZSMALLOC)
166 NR_ZSPAGES, /* allocated in zsmalloc */
169 NR_VM_ZONE_STAT_ITEMS };
171 enum node_stat_item {
173 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
174 NR_ACTIVE_ANON, /* " " " " " */
175 NR_INACTIVE_FILE, /* " " " " " */
176 NR_ACTIVE_FILE, /* " " " " " */
177 NR_UNEVICTABLE, /* " " " " " */
178 NR_SLAB_RECLAIMABLE_B,
179 NR_SLAB_UNRECLAIMABLE_B,
180 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
181 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
183 WORKINGSET_REFAULT_BASE,
184 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
185 WORKINGSET_REFAULT_FILE,
186 WORKINGSET_ACTIVATE_BASE,
187 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
188 WORKINGSET_ACTIVATE_FILE,
189 WORKINGSET_RESTORE_BASE,
190 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
191 WORKINGSET_RESTORE_FILE,
192 WORKINGSET_NODERECLAIM,
193 NR_ANON_MAPPED, /* Mapped anonymous pages */
194 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
195 only modified from process context */
199 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
200 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
207 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
208 NR_DIRTIED, /* page dirtyings since bootup */
209 NR_WRITTEN, /* page writings since bootup */
210 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
211 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
212 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
213 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
214 NR_KERNEL_STACK_KB, /* measured in KiB */
215 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
216 NR_KERNEL_SCS_KB, /* measured in KiB */
218 NR_PAGETABLE, /* used for pagetables */
222 #ifdef CONFIG_NUMA_BALANCING
223 PGPROMOTE_SUCCESS, /* promote successfully */
224 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
226 NR_VM_NODE_STAT_ITEMS
230 * Returns true if the item should be printed in THPs (/proc/vmstat
231 * currently prints number of anon, file and shmem THPs. But the item
232 * is charged in pages).
234 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
236 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
239 return item == NR_ANON_THPS ||
240 item == NR_FILE_THPS ||
241 item == NR_SHMEM_THPS ||
242 item == NR_SHMEM_PMDMAPPED ||
243 item == NR_FILE_PMDMAPPED;
247 * Returns true if the value is measured in bytes (most vmstat values are
248 * measured in pages). This defines the API part, the internal representation
249 * might be different.
251 static __always_inline bool vmstat_item_in_bytes(int idx)
254 * Global and per-node slab counters track slab pages.
255 * It's expected that changes are multiples of PAGE_SIZE.
256 * Internally values are stored in pages.
258 * Per-memcg and per-lruvec counters track memory, consumed
259 * by individual slab objects. These counters are actually
262 return (idx == NR_SLAB_RECLAIMABLE_B ||
263 idx == NR_SLAB_UNRECLAIMABLE_B);
267 * We do arithmetic on the LRU lists in various places in the code,
268 * so it is important to keep the active lists LRU_ACTIVE higher in
269 * the array than the corresponding inactive lists, and to keep
270 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
272 * This has to be kept in sync with the statistics in zone_stat_item
273 * above and the descriptions in vmstat_text in mm/vmstat.c
280 LRU_INACTIVE_ANON = LRU_BASE,
281 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
282 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
283 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
288 enum vmscan_throttle_state {
289 VMSCAN_THROTTLE_WRITEBACK,
290 VMSCAN_THROTTLE_ISOLATED,
291 VMSCAN_THROTTLE_NOPROGRESS,
292 VMSCAN_THROTTLE_CONGESTED,
296 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
298 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
300 static inline bool is_file_lru(enum lru_list lru)
302 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
305 static inline bool is_active_lru(enum lru_list lru)
307 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
310 #define WORKINGSET_ANON 0
311 #define WORKINGSET_FILE 1
312 #define ANON_AND_FILE 2
315 LRUVEC_CONGESTED, /* lruvec has many dirty pages
316 * backed by a congested BDI
321 struct list_head lists[NR_LRU_LISTS];
322 /* per lruvec lru_lock for memcg */
325 * These track the cost of reclaiming one LRU - file or anon -
326 * over the other. As the observed cost of reclaiming one LRU
327 * increases, the reclaim scan balance tips toward the other.
329 unsigned long anon_cost;
330 unsigned long file_cost;
331 /* Non-resident age, driven by LRU movement */
332 atomic_long_t nonresident_age;
333 /* Refaults at the time of last reclaim cycle */
334 unsigned long refaults[ANON_AND_FILE];
335 /* Various lruvec state flags (enum lruvec_flags) */
338 struct pglist_data *pgdat;
342 /* Isolate unmapped pages */
343 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
344 /* Isolate for asynchronous migration */
345 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
346 /* Isolate unevictable pages */
347 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
349 /* LRU Isolation modes. */
350 typedef unsigned __bitwise isolate_mode_t;
352 enum zone_watermarks {
361 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
362 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
363 * it should not contribute to serious fragmentation causing THP allocation
366 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
371 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
372 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
375 * Shift to encode migratetype and order in the same integer, with order
376 * in the least significant bits.
378 #define NR_PCP_ORDER_WIDTH 8
379 #define NR_PCP_ORDER_MASK ((1<<NR_PCP_ORDER_WIDTH) - 1)
381 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
382 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
383 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
384 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
386 /* Fields and list protected by pagesets local_lock in page_alloc.c */
387 struct per_cpu_pages {
388 spinlock_t lock; /* Protects lists field */
389 int count; /* number of pages in the list */
390 int high; /* high watermark, emptying needed */
391 int batch; /* chunk size for buddy add/remove */
392 short free_factor; /* batch scaling factor during free */
394 short expire; /* When 0, remote pagesets are drained */
397 /* Lists of pages, one per migrate type stored on the pcp-lists */
398 struct list_head lists[NR_PCP_LISTS];
399 } ____cacheline_aligned_in_smp;
401 struct per_cpu_zonestat {
403 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
408 * Low priority inaccurate counters that are only folded
409 * on demand. Use a large type to avoid the overhead of
410 * folding during refresh_cpu_vm_stats.
412 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
416 struct per_cpu_nodestat {
418 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
421 #endif /* !__GENERATING_BOUNDS.H */
425 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
426 * to DMA to all of the addressable memory (ZONE_NORMAL).
427 * On architectures where this area covers the whole 32 bit address
428 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
429 * DMA addressing constraints. This distinction is important as a 32bit
430 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
431 * platforms may need both zones as they support peripherals with
432 * different DMA addressing limitations.
434 #ifdef CONFIG_ZONE_DMA
437 #ifdef CONFIG_ZONE_DMA32
441 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
442 * performed on pages in ZONE_NORMAL if the DMA devices support
443 * transfers to all addressable memory.
446 #ifdef CONFIG_HIGHMEM
448 * A memory area that is only addressable by the kernel through
449 * mapping portions into its own address space. This is for example
450 * used by i386 to allow the kernel to address the memory beyond
451 * 900MB. The kernel will set up special mappings (page
452 * table entries on i386) for each page that the kernel needs to
458 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
459 * movable pages with few exceptional cases described below. Main use
460 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
461 * likely to succeed, and to locally limit unmovable allocations - e.g.,
462 * to increase the number of THP/huge pages. Notable special cases are:
464 * 1. Pinned pages: (long-term) pinning of movable pages might
465 * essentially turn such pages unmovable. Therefore, we do not allow
466 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
467 * faulted, they come from the right zone right away. However, it is
468 * still possible that address space already has pages in
469 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
470 * touches that memory before pinning). In such case we migrate them
471 * to a different zone. When migration fails - pinning fails.
472 * 2. memblock allocations: kernelcore/movablecore setups might create
473 * situations where ZONE_MOVABLE contains unmovable allocations
474 * after boot. Memory offlining and allocations fail early.
475 * 3. Memory holes: kernelcore/movablecore setups might create very rare
476 * situations where ZONE_MOVABLE contains memory holes after boot,
477 * for example, if we have sections that are only partially
478 * populated. Memory offlining and allocations fail early.
479 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
480 * memory offlining, such pages cannot be allocated.
481 * 5. Unmovable PG_offline pages: in paravirtualized environments,
482 * hotplugged memory blocks might only partially be managed by the
483 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
484 * parts not manged by the buddy are unmovable PG_offline pages. In
485 * some cases (virtio-mem), such pages can be skipped during
486 * memory offlining, however, cannot be moved/allocated. These
487 * techniques might use alloc_contig_range() to hide previously
488 * exposed pages from the buddy again (e.g., to implement some sort
489 * of memory unplug in virtio-mem).
490 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
491 * situations where ZERO_PAGE(0) which is allocated differently
492 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
493 * cannot be migrated.
494 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
495 * memory to the MOVABLE zone, the vmemmap pages are also placed in
496 * such zone. Such pages cannot be really moved around as they are
497 * self-stored in the range, but they are treated as movable when
498 * the range they describe is about to be offlined.
500 * In general, no unmovable allocations that degrade memory offlining
501 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
502 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
503 * if has_unmovable_pages() states that there are no unmovable pages,
504 * there can be false negatives).
507 #ifdef CONFIG_ZONE_DEVICE
514 #ifndef __GENERATING_BOUNDS_H
516 #define ASYNC_AND_SYNC 2
519 /* Read-mostly fields */
521 /* zone watermarks, access with *_wmark_pages(zone) macros */
522 unsigned long _watermark[NR_WMARK];
523 unsigned long watermark_boost;
525 unsigned long nr_reserved_highatomic;
528 * We don't know if the memory that we're going to allocate will be
529 * freeable or/and it will be released eventually, so to avoid totally
530 * wasting several GB of ram we must reserve some of the lower zone
531 * memory (otherwise we risk to run OOM on the lower zones despite
532 * there being tons of freeable ram on the higher zones). This array is
533 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
536 long lowmem_reserve[MAX_NR_ZONES];
541 struct pglist_data *zone_pgdat;
542 struct per_cpu_pages __percpu *per_cpu_pageset;
543 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
545 * the high and batch values are copied to individual pagesets for
551 #ifndef CONFIG_SPARSEMEM
553 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
554 * In SPARSEMEM, this map is stored in struct mem_section
556 unsigned long *pageblock_flags;
557 #endif /* CONFIG_SPARSEMEM */
559 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
560 unsigned long zone_start_pfn;
563 * spanned_pages is the total pages spanned by the zone, including
564 * holes, which is calculated as:
565 * spanned_pages = zone_end_pfn - zone_start_pfn;
567 * present_pages is physical pages existing within the zone, which
569 * present_pages = spanned_pages - absent_pages(pages in holes);
571 * present_early_pages is present pages existing within the zone
572 * located on memory available since early boot, excluding hotplugged
575 * managed_pages is present pages managed by the buddy system, which
576 * is calculated as (reserved_pages includes pages allocated by the
577 * bootmem allocator):
578 * managed_pages = present_pages - reserved_pages;
580 * cma pages is present pages that are assigned for CMA use
583 * So present_pages may be used by memory hotplug or memory power
584 * management logic to figure out unmanaged pages by checking
585 * (present_pages - managed_pages). And managed_pages should be used
586 * by page allocator and vm scanner to calculate all kinds of watermarks
591 * zone_start_pfn and spanned_pages are protected by span_seqlock.
592 * It is a seqlock because it has to be read outside of zone->lock,
593 * and it is done in the main allocator path. But, it is written
594 * quite infrequently.
596 * The span_seq lock is declared along with zone->lock because it is
597 * frequently read in proximity to zone->lock. It's good to
598 * give them a chance of being in the same cacheline.
600 * Write access to present_pages at runtime should be protected by
601 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
602 * present_pages should use get_online_mems() to get a stable value.
604 atomic_long_t managed_pages;
605 unsigned long spanned_pages;
606 unsigned long present_pages;
607 #if defined(CONFIG_MEMORY_HOTPLUG)
608 unsigned long present_early_pages;
611 unsigned long cma_pages;
616 #ifdef CONFIG_MEMORY_ISOLATION
618 * Number of isolated pageblock. It is used to solve incorrect
619 * freepage counting problem due to racy retrieving migratetype
620 * of pageblock. Protected by zone->lock.
622 unsigned long nr_isolate_pageblock;
625 #ifdef CONFIG_MEMORY_HOTPLUG
626 /* see spanned/present_pages for more description */
627 seqlock_t span_seqlock;
632 /* Write-intensive fields used from the page allocator */
635 /* free areas of different sizes */
636 struct free_area free_area[MAX_ORDER];
638 /* zone flags, see below */
641 /* Primarily protects free_area */
644 /* Write-intensive fields used by compaction and vmstats. */
648 * When free pages are below this point, additional steps are taken
649 * when reading the number of free pages to avoid per-cpu counter
650 * drift allowing watermarks to be breached
652 unsigned long percpu_drift_mark;
654 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
655 /* pfn where compaction free scanner should start */
656 unsigned long compact_cached_free_pfn;
657 /* pfn where compaction migration scanner should start */
658 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
659 unsigned long compact_init_migrate_pfn;
660 unsigned long compact_init_free_pfn;
663 #ifdef CONFIG_COMPACTION
665 * On compaction failure, 1<<compact_defer_shift compactions
666 * are skipped before trying again. The number attempted since
667 * last failure is tracked with compact_considered.
668 * compact_order_failed is the minimum compaction failed order.
670 unsigned int compact_considered;
671 unsigned int compact_defer_shift;
672 int compact_order_failed;
675 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
676 /* Set to true when the PG_migrate_skip bits should be cleared */
677 bool compact_blockskip_flush;
683 /* Zone statistics */
684 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
685 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
686 } ____cacheline_internodealigned_in_smp;
689 PGDAT_DIRTY, /* reclaim scanning has recently found
690 * many dirty file pages at the tail
693 PGDAT_WRITEBACK, /* reclaim scanning has recently found
694 * many pages under writeback
696 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
700 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
701 * Cleared when kswapd is woken.
703 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
706 static inline unsigned long zone_managed_pages(struct zone *zone)
708 return (unsigned long)atomic_long_read(&zone->managed_pages);
711 static inline unsigned long zone_cma_pages(struct zone *zone)
714 return zone->cma_pages;
720 static inline unsigned long zone_end_pfn(const struct zone *zone)
722 return zone->zone_start_pfn + zone->spanned_pages;
725 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
727 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
730 static inline bool zone_is_initialized(struct zone *zone)
732 return zone->initialized;
735 static inline bool zone_is_empty(struct zone *zone)
737 return zone->spanned_pages == 0;
740 #ifndef BUILD_VDSO32_64
742 * The zone field is never updated after free_area_init_core()
743 * sets it, so none of the operations on it need to be atomic.
746 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
747 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
748 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
749 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
750 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
751 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
754 * Define the bit shifts to access each section. For non-existent
755 * sections we define the shift as 0; that plus a 0 mask ensures
756 * the compiler will optimise away reference to them.
758 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
759 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
760 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
761 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
762 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
764 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
765 #ifdef NODE_NOT_IN_PAGE_FLAGS
766 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
767 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
768 SECTIONS_PGOFF : ZONES_PGOFF)
770 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
771 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
772 NODES_PGOFF : ZONES_PGOFF)
775 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
777 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
778 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
779 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
780 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
781 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
782 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
784 static inline enum zone_type page_zonenum(const struct page *page)
786 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
787 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
790 static inline enum zone_type folio_zonenum(const struct folio *folio)
792 return page_zonenum(&folio->page);
795 #ifdef CONFIG_ZONE_DEVICE
796 static inline bool is_zone_device_page(const struct page *page)
798 return page_zonenum(page) == ZONE_DEVICE;
800 extern void memmap_init_zone_device(struct zone *, unsigned long,
801 unsigned long, struct dev_pagemap *);
803 static inline bool is_zone_device_page(const struct page *page)
809 static inline bool folio_is_zone_device(const struct folio *folio)
811 return is_zone_device_page(&folio->page);
814 static inline bool is_zone_movable_page(const struct page *page)
816 return page_zonenum(page) == ZONE_MOVABLE;
821 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
822 * intersection with the given zone
824 static inline bool zone_intersects(struct zone *zone,
825 unsigned long start_pfn, unsigned long nr_pages)
827 if (zone_is_empty(zone))
829 if (start_pfn >= zone_end_pfn(zone) ||
830 start_pfn + nr_pages <= zone->zone_start_pfn)
837 * The "priority" of VM scanning is how much of the queues we will scan in one
838 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
839 * queues ("queue_length >> 12") during an aging round.
841 #define DEF_PRIORITY 12
843 /* Maximum number of zones on a zonelist */
844 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
847 ZONELIST_FALLBACK, /* zonelist with fallback */
850 * The NUMA zonelists are doubled because we need zonelists that
851 * restrict the allocations to a single node for __GFP_THISNODE.
853 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
859 * This struct contains information about a zone in a zonelist. It is stored
860 * here to avoid dereferences into large structures and lookups of tables
863 struct zone *zone; /* Pointer to actual zone */
864 int zone_idx; /* zone_idx(zoneref->zone) */
868 * One allocation request operates on a zonelist. A zonelist
869 * is a list of zones, the first one is the 'goal' of the
870 * allocation, the other zones are fallback zones, in decreasing
873 * To speed the reading of the zonelist, the zonerefs contain the zone index
874 * of the entry being read. Helper functions to access information given
875 * a struct zoneref are
877 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
878 * zonelist_zone_idx() - Return the index of the zone for an entry
879 * zonelist_node_idx() - Return the index of the node for an entry
882 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
886 * The array of struct pages for flatmem.
887 * It must be declared for SPARSEMEM as well because there are configurations
890 extern struct page *mem_map;
892 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
893 struct deferred_split {
894 spinlock_t split_queue_lock;
895 struct list_head split_queue;
896 unsigned long split_queue_len;
901 * On NUMA machines, each NUMA node would have a pg_data_t to describe
902 * it's memory layout. On UMA machines there is a single pglist_data which
903 * describes the whole memory.
905 * Memory statistics and page replacement data structures are maintained on a
908 typedef struct pglist_data {
910 * node_zones contains just the zones for THIS node. Not all of the
911 * zones may be populated, but it is the full list. It is referenced by
912 * this node's node_zonelists as well as other node's node_zonelists.
914 struct zone node_zones[MAX_NR_ZONES];
917 * node_zonelists contains references to all zones in all nodes.
918 * Generally the first zones will be references to this node's
921 struct zonelist node_zonelists[MAX_ZONELISTS];
923 int nr_zones; /* number of populated zones in this node */
924 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
925 struct page *node_mem_map;
926 #ifdef CONFIG_PAGE_EXTENSION
927 struct page_ext *node_page_ext;
930 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
932 * Must be held any time you expect node_start_pfn,
933 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
934 * Also synchronizes pgdat->first_deferred_pfn during deferred page
937 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
938 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
939 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
941 * Nests above zone->lock and zone->span_seqlock
943 spinlock_t node_size_lock;
945 unsigned long node_start_pfn;
946 unsigned long node_present_pages; /* total number of physical pages */
947 unsigned long node_spanned_pages; /* total size of physical page
948 range, including holes */
950 wait_queue_head_t kswapd_wait;
951 wait_queue_head_t pfmemalloc_wait;
953 /* workqueues for throttling reclaim for different reasons. */
954 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
956 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
957 unsigned long nr_reclaim_start; /* nr pages written while throttled
958 * when throttling started. */
959 #ifdef CONFIG_MEMORY_HOTPLUG
960 struct mutex kswapd_lock;
962 struct task_struct *kswapd; /* Protected by kswapd_lock */
964 enum zone_type kswapd_highest_zoneidx;
966 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
968 #ifdef CONFIG_COMPACTION
969 int kcompactd_max_order;
970 enum zone_type kcompactd_highest_zoneidx;
971 wait_queue_head_t kcompactd_wait;
972 struct task_struct *kcompactd;
973 bool proactive_compact_trigger;
976 * This is a per-node reserve of pages that are not available
977 * to userspace allocations.
979 unsigned long totalreserve_pages;
983 * node reclaim becomes active if more unmapped pages exist.
985 unsigned long min_unmapped_pages;
986 unsigned long min_slab_pages;
987 #endif /* CONFIG_NUMA */
989 /* Write-intensive fields used by page reclaim */
992 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
994 * If memory initialisation on large machines is deferred then this
995 * is the first PFN that needs to be initialised.
997 unsigned long first_deferred_pfn;
998 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1000 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1001 struct deferred_split deferred_split_queue;
1004 #ifdef CONFIG_NUMA_BALANCING
1005 /* start time in ms of current promote rate limit period */
1006 unsigned int nbp_rl_start;
1007 /* number of promote candidate pages at start time of current rate limit period */
1008 unsigned long nbp_rl_nr_cand;
1009 /* promote threshold in ms */
1010 unsigned int nbp_threshold;
1011 /* start time in ms of current promote threshold adjustment period */
1012 unsigned int nbp_th_start;
1014 * number of promote candidate pages at stat time of current promote
1015 * threshold adjustment period
1017 unsigned long nbp_th_nr_cand;
1019 /* Fields commonly accessed by the page reclaim scanner */
1022 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1024 * Use mem_cgroup_lruvec() to look up lruvecs.
1026 struct lruvec __lruvec;
1028 unsigned long flags;
1030 ZONE_PADDING(_pad2_)
1032 /* Per-node vmstats */
1033 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1034 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1037 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1038 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1040 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1041 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1043 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1045 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1048 static inline bool pgdat_is_empty(pg_data_t *pgdat)
1050 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
1053 #include <linux/memory_hotplug.h>
1055 void build_all_zonelists(pg_data_t *pgdat);
1056 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1057 enum zone_type highest_zoneidx);
1058 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1059 int highest_zoneidx, unsigned int alloc_flags,
1061 bool zone_watermark_ok(struct zone *z, unsigned int order,
1062 unsigned long mark, int highest_zoneidx,
1063 unsigned int alloc_flags);
1064 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1065 unsigned long mark, int highest_zoneidx);
1067 * Memory initialization context, use to differentiate memory added by
1068 * the platform statically or via memory hotplug interface.
1070 enum meminit_context {
1075 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1076 unsigned long size);
1078 extern void lruvec_init(struct lruvec *lruvec);
1080 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1083 return lruvec->pgdat;
1085 return container_of(lruvec, struct pglist_data, __lruvec);
1089 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1090 int local_memory_node(int node_id);
1092 static inline int local_memory_node(int node_id) { return node_id; };
1096 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1098 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1100 #ifdef CONFIG_ZONE_DEVICE
1101 static inline bool zone_is_zone_device(struct zone *zone)
1103 return zone_idx(zone) == ZONE_DEVICE;
1106 static inline bool zone_is_zone_device(struct zone *zone)
1113 * Returns true if a zone has pages managed by the buddy allocator.
1114 * All the reclaim decisions have to use this function rather than
1115 * populated_zone(). If the whole zone is reserved then we can easily
1116 * end up with populated_zone() && !managed_zone().
1118 static inline bool managed_zone(struct zone *zone)
1120 return zone_managed_pages(zone);
1123 /* Returns true if a zone has memory */
1124 static inline bool populated_zone(struct zone *zone)
1126 return zone->present_pages;
1130 static inline int zone_to_nid(struct zone *zone)
1135 static inline void zone_set_nid(struct zone *zone, int nid)
1140 static inline int zone_to_nid(struct zone *zone)
1145 static inline void zone_set_nid(struct zone *zone, int nid) {}
1148 extern int movable_zone;
1150 static inline int is_highmem_idx(enum zone_type idx)
1152 #ifdef CONFIG_HIGHMEM
1153 return (idx == ZONE_HIGHMEM ||
1154 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1161 * is_highmem - helper function to quickly check if a struct zone is a
1162 * highmem zone or not. This is an attempt to keep references
1163 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1164 * @zone: pointer to struct zone variable
1165 * Return: 1 for a highmem zone, 0 otherwise
1167 static inline int is_highmem(struct zone *zone)
1169 return is_highmem_idx(zone_idx(zone));
1172 #ifdef CONFIG_ZONE_DMA
1173 bool has_managed_dma(void);
1175 static inline bool has_managed_dma(void)
1181 /* These two functions are used to setup the per zone pages min values */
1184 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1186 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1187 size_t *, loff_t *);
1188 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1189 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1190 size_t *, loff_t *);
1191 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1192 void *, size_t *, loff_t *);
1193 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1194 void *, size_t *, loff_t *);
1195 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1196 void *, size_t *, loff_t *);
1197 int numa_zonelist_order_handler(struct ctl_table *, int,
1198 void *, size_t *, loff_t *);
1199 extern int percpu_pagelist_high_fraction;
1200 extern char numa_zonelist_order[];
1201 #define NUMA_ZONELIST_ORDER_LEN 16
1205 extern struct pglist_data contig_page_data;
1206 static inline struct pglist_data *NODE_DATA(int nid)
1208 return &contig_page_data;
1211 #else /* CONFIG_NUMA */
1213 #include <asm/mmzone.h>
1215 #endif /* !CONFIG_NUMA */
1217 extern struct pglist_data *first_online_pgdat(void);
1218 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1219 extern struct zone *next_zone(struct zone *zone);
1222 * for_each_online_pgdat - helper macro to iterate over all online nodes
1223 * @pgdat: pointer to a pg_data_t variable
1225 #define for_each_online_pgdat(pgdat) \
1226 for (pgdat = first_online_pgdat(); \
1228 pgdat = next_online_pgdat(pgdat))
1230 * for_each_zone - helper macro to iterate over all memory zones
1231 * @zone: pointer to struct zone variable
1233 * The user only needs to declare the zone variable, for_each_zone
1236 #define for_each_zone(zone) \
1237 for (zone = (first_online_pgdat())->node_zones; \
1239 zone = next_zone(zone))
1241 #define for_each_populated_zone(zone) \
1242 for (zone = (first_online_pgdat())->node_zones; \
1244 zone = next_zone(zone)) \
1245 if (!populated_zone(zone)) \
1246 ; /* do nothing */ \
1249 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1251 return zoneref->zone;
1254 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1256 return zoneref->zone_idx;
1259 static inline int zonelist_node_idx(struct zoneref *zoneref)
1261 return zone_to_nid(zoneref->zone);
1264 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1265 enum zone_type highest_zoneidx,
1269 * 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
1270 * @z: The cursor used as a starting point for the search
1271 * @highest_zoneidx: The zone index of the highest zone to return
1272 * @nodes: An optional nodemask to filter the zonelist with
1274 * This function returns the next zone at or below a given zone index that is
1275 * within the allowed nodemask using a cursor as the starting point for the
1276 * search. The zoneref returned is a cursor that represents the current zone
1277 * being examined. It should be advanced by one before calling
1278 * next_zones_zonelist again.
1280 * Return: the next zone at or below highest_zoneidx within the allowed
1281 * nodemask using a cursor within a zonelist as a starting point
1283 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1284 enum zone_type highest_zoneidx,
1287 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1289 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1293 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1294 * @zonelist: The zonelist to search for a suitable zone
1295 * @highest_zoneidx: The zone index of the highest zone to return
1296 * @nodes: An optional nodemask to filter the zonelist with
1298 * This function returns the first zone at or below a given zone index that is
1299 * within the allowed nodemask. The zoneref returned is a cursor that can be
1300 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1301 * one before calling.
1303 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1304 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1305 * update due to cpuset modification.
1307 * Return: Zoneref pointer for the first suitable zone found
1309 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1310 enum zone_type highest_zoneidx,
1313 return next_zones_zonelist(zonelist->_zonerefs,
1314 highest_zoneidx, nodes);
1318 * 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
1319 * @zone: The current zone in the iterator
1320 * @z: The current pointer within zonelist->_zonerefs being iterated
1321 * @zlist: The zonelist being iterated
1322 * @highidx: The zone index of the highest zone to return
1323 * @nodemask: Nodemask allowed by the allocator
1325 * This iterator iterates though all zones at or below a given zone index and
1326 * within a given nodemask
1328 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1329 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1331 z = next_zones_zonelist(++z, highidx, nodemask), \
1332 zone = zonelist_zone(z))
1334 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1335 for (zone = z->zone; \
1337 z = next_zones_zonelist(++z, highidx, nodemask), \
1338 zone = zonelist_zone(z))
1342 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1343 * @zone: The current zone in the iterator
1344 * @z: The current pointer within zonelist->zones being iterated
1345 * @zlist: The zonelist being iterated
1346 * @highidx: The zone index of the highest zone to return
1348 * This iterator iterates though all zones at or below a given zone index.
1350 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1351 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1353 /* Whether the 'nodes' are all movable nodes */
1354 static inline bool movable_only_nodes(nodemask_t *nodes)
1356 struct zonelist *zonelist;
1360 if (nodes_empty(*nodes))
1364 * We can chose arbitrary node from the nodemask to get a
1365 * zonelist as they are interlinked. We just need to find
1366 * at least one zone that can satisfy kernel allocations.
1368 nid = first_node(*nodes);
1369 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1370 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1371 return (!z->zone) ? true : false;
1375 #ifdef CONFIG_SPARSEMEM
1376 #include <asm/sparsemem.h>
1379 #ifdef CONFIG_FLATMEM
1380 #define pfn_to_nid(pfn) (0)
1383 #ifdef CONFIG_SPARSEMEM
1386 * PA_SECTION_SHIFT physical address to/from section number
1387 * PFN_SECTION_SHIFT pfn to/from section number
1389 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1390 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1392 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1394 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1395 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1397 #define SECTION_BLOCKFLAGS_BITS \
1398 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1400 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1401 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1404 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1406 return pfn >> PFN_SECTION_SHIFT;
1408 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1410 return sec << PFN_SECTION_SHIFT;
1413 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1414 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1416 #define SUBSECTION_SHIFT 21
1417 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1419 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1420 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1421 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1423 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1424 #error Subsection size exceeds section size
1426 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1429 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1430 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1432 struct mem_section_usage {
1433 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1434 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1436 /* See declaration of similar field in struct zone */
1437 unsigned long pageblock_flags[0];
1440 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1444 struct mem_section {
1446 * This is, logically, a pointer to an array of struct
1447 * pages. However, it is stored with some other magic.
1448 * (see sparse.c::sparse_init_one_section())
1450 * Additionally during early boot we encode node id of
1451 * the location of the section here to guide allocation.
1452 * (see sparse.c::memory_present())
1454 * Making it a UL at least makes someone do a cast
1455 * before using it wrong.
1457 unsigned long section_mem_map;
1459 struct mem_section_usage *usage;
1460 #ifdef CONFIG_PAGE_EXTENSION
1462 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1463 * section. (see page_ext.h about this.)
1465 struct page_ext *page_ext;
1469 * WARNING: mem_section must be a power-of-2 in size for the
1470 * calculation and use of SECTION_ROOT_MASK to make sense.
1474 #ifdef CONFIG_SPARSEMEM_EXTREME
1475 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1477 #define SECTIONS_PER_ROOT 1
1480 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1481 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1482 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1484 #ifdef CONFIG_SPARSEMEM_EXTREME
1485 extern struct mem_section **mem_section;
1487 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1490 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1492 return ms->usage->pageblock_flags;
1495 static inline struct mem_section *__nr_to_section(unsigned long nr)
1497 unsigned long root = SECTION_NR_TO_ROOT(nr);
1499 if (unlikely(root >= NR_SECTION_ROOTS))
1502 #ifdef CONFIG_SPARSEMEM_EXTREME
1503 if (!mem_section || !mem_section[root])
1506 return &mem_section[root][nr & SECTION_ROOT_MASK];
1508 extern size_t mem_section_usage_size(void);
1511 * We use the lower bits of the mem_map pointer to store
1512 * a little bit of information. The pointer is calculated
1513 * as mem_map - section_nr_to_pfn(pnum). The result is
1514 * aligned to the minimum alignment of the two values:
1515 * 1. All mem_map arrays are page-aligned.
1516 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1517 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1518 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1519 * worst combination is powerpc with 256k pages,
1520 * which results in PFN_SECTION_SHIFT equal 6.
1521 * To sum it up, at least 6 bits are available on all architectures.
1522 * However, we can exceed 6 bits on some other architectures except
1523 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1524 * with the worst case of 64K pages on arm64) if we make sure the
1525 * exceeded bit is not applicable to powerpc.
1528 SECTION_MARKED_PRESENT_BIT,
1529 SECTION_HAS_MEM_MAP_BIT,
1530 SECTION_IS_ONLINE_BIT,
1531 SECTION_IS_EARLY_BIT,
1532 #ifdef CONFIG_ZONE_DEVICE
1533 SECTION_TAINT_ZONE_DEVICE_BIT,
1535 SECTION_MAP_LAST_BIT,
1538 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1539 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1540 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1541 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1542 #ifdef CONFIG_ZONE_DEVICE
1543 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1545 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1546 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1548 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1550 unsigned long map = section->section_mem_map;
1551 map &= SECTION_MAP_MASK;
1552 return (struct page *)map;
1555 static inline int present_section(struct mem_section *section)
1557 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1560 static inline int present_section_nr(unsigned long nr)
1562 return present_section(__nr_to_section(nr));
1565 static inline int valid_section(struct mem_section *section)
1567 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1570 static inline int early_section(struct mem_section *section)
1572 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1575 static inline int valid_section_nr(unsigned long nr)
1577 return valid_section(__nr_to_section(nr));
1580 static inline int online_section(struct mem_section *section)
1582 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1585 #ifdef CONFIG_ZONE_DEVICE
1586 static inline int online_device_section(struct mem_section *section)
1588 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1590 return section && ((section->section_mem_map & flags) == flags);
1593 static inline int online_device_section(struct mem_section *section)
1599 static inline int online_section_nr(unsigned long nr)
1601 return online_section(__nr_to_section(nr));
1604 #ifdef CONFIG_MEMORY_HOTPLUG
1605 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1606 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1609 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1611 return __nr_to_section(pfn_to_section_nr(pfn));
1614 extern unsigned long __highest_present_section_nr;
1616 static inline int subsection_map_index(unsigned long pfn)
1618 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1621 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1622 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1624 int idx = subsection_map_index(pfn);
1626 return test_bit(idx, ms->usage->subsection_map);
1629 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1635 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1637 * pfn_valid - check if there is a valid memory map entry for a PFN
1638 * @pfn: the page frame number to check
1640 * Check if there is a valid memory map entry aka struct page for the @pfn.
1641 * Note, that availability of the memory map entry does not imply that
1642 * there is actual usable memory at that @pfn. The struct page may
1643 * represent a hole or an unusable page frame.
1645 * Return: 1 for PFNs that have memory map entries and 0 otherwise
1647 static inline int pfn_valid(unsigned long pfn)
1649 struct mem_section *ms;
1652 * Ensure the upper PAGE_SHIFT bits are clear in the
1653 * pfn. Else it might lead to false positives when
1654 * some of the upper bits are set, but the lower bits
1655 * match a valid pfn.
1657 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
1660 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1662 ms = __pfn_to_section(pfn);
1663 if (!valid_section(ms))
1666 * Traditionally early sections always returned pfn_valid() for
1667 * the entire section-sized span.
1669 return early_section(ms) || pfn_section_valid(ms, pfn);
1673 static inline int pfn_in_present_section(unsigned long pfn)
1675 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1677 return present_section(__pfn_to_section(pfn));
1680 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1682 while (++section_nr <= __highest_present_section_nr) {
1683 if (present_section_nr(section_nr))
1691 * These are _only_ used during initialisation, therefore they
1692 * can use __initdata ... They could have names to indicate
1696 #define pfn_to_nid(pfn) \
1698 unsigned long __pfn_to_nid_pfn = (pfn); \
1699 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1702 #define pfn_to_nid(pfn) (0)
1705 void sparse_init(void);
1707 #define sparse_init() do {} while (0)
1708 #define sparse_index_init(_sec, _nid) do {} while (0)
1709 #define pfn_in_present_section pfn_valid
1710 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1711 #endif /* CONFIG_SPARSEMEM */
1713 #endif /* !__GENERATING_BOUNDS.H */
1714 #endif /* !__ASSEMBLY__ */
1715 #endif /* _LINUX_MMZONE_H */