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>
25 /* Free memory management - zoned buddy allocator. */
26 #ifndef CONFIG_FORCE_MAX_ZONEORDER
29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
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
39 #define PAGE_ALLOC_COSTLY_ORDER 3
45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
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.
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
63 #ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE, /* can't allocate from here */
69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70 extern const char * const migratetype_names[MIGRATE_TYPES];
73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76 # define is_migrate_cma(migratetype) false
77 # define is_migrate_cma_page(_page) false
80 static inline bool is_migrate_movable(int mt)
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
85 #define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
89 extern int page_group_by_mobility_disabled;
91 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
92 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
94 #define get_pageblock_migratetype(page) \
95 get_pfnblock_flags_mask(page, page_to_pfn(page), \
96 PB_migrate_end, MIGRATETYPE_MASK)
99 struct list_head free_list[MIGRATE_TYPES];
100 unsigned long nr_free;
103 /* Used for pages not on another list */
104 static inline void add_to_free_area(struct page *page, struct free_area *area,
107 list_add(&page->lru, &area->free_list[migratetype]);
111 /* Used for pages not on another list */
112 static inline void add_to_free_area_tail(struct page *page, struct free_area *area,
115 list_add_tail(&page->lru, &area->free_list[migratetype]);
119 #ifdef CONFIG_SHUFFLE_PAGE_ALLOCATOR
120 /* Used to preserve page allocation order entropy */
121 void add_to_free_area_random(struct page *page, struct free_area *area,
124 static inline void add_to_free_area_random(struct page *page,
125 struct free_area *area, int migratetype)
127 add_to_free_area(page, area, migratetype);
131 /* Used for pages which are on another list */
132 static inline void move_to_free_area(struct page *page, struct free_area *area,
135 list_move(&page->lru, &area->free_list[migratetype]);
138 static inline struct page *get_page_from_free_area(struct free_area *area,
141 return list_first_entry_or_null(&area->free_list[migratetype],
145 static inline void del_page_from_free_area(struct page *page,
146 struct free_area *area)
148 list_del(&page->lru);
149 __ClearPageBuddy(page);
150 set_page_private(page, 0);
154 static inline bool free_area_empty(struct free_area *area, int migratetype)
156 return list_empty(&area->free_list[migratetype]);
162 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
163 * So add a wild amount of padding here to ensure that they fall into separate
164 * cachelines. There are very few zone structures in the machine, so space
165 * consumption is not a concern here.
167 #if defined(CONFIG_SMP)
168 struct zone_padding {
170 } ____cacheline_internodealigned_in_smp;
171 #define ZONE_PADDING(name) struct zone_padding name;
173 #define ZONE_PADDING(name)
177 enum numa_stat_item {
178 NUMA_HIT, /* allocated in intended node */
179 NUMA_MISS, /* allocated in non intended node */
180 NUMA_FOREIGN, /* was intended here, hit elsewhere */
181 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
182 NUMA_LOCAL, /* allocation from local node */
183 NUMA_OTHER, /* allocation from other node */
184 NR_VM_NUMA_STAT_ITEMS
187 #define NR_VM_NUMA_STAT_ITEMS 0
190 enum zone_stat_item {
191 /* First 128 byte cacheline (assuming 64 bit words) */
193 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
194 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
196 NR_ZONE_INACTIVE_FILE,
199 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
200 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
201 NR_PAGETABLE, /* used for pagetables */
202 NR_KERNEL_STACK_KB, /* measured in KiB */
203 /* Second 128 byte cacheline */
205 #if IS_ENABLED(CONFIG_ZSMALLOC)
206 NR_ZSPAGES, /* allocated in zsmalloc */
209 NR_VM_ZONE_STAT_ITEMS };
211 enum node_stat_item {
213 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
214 NR_ACTIVE_ANON, /* " " " " " */
215 NR_INACTIVE_FILE, /* " " " " " */
216 NR_ACTIVE_FILE, /* " " " " " */
217 NR_UNEVICTABLE, /* " " " " " */
219 NR_SLAB_UNRECLAIMABLE,
220 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
221 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
226 WORKINGSET_NODERECLAIM,
227 NR_ANON_MAPPED, /* Mapped anonymous pages */
228 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
229 only modified from process context */
233 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
234 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
240 NR_UNSTABLE_NFS, /* NFS unstable pages */
242 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
243 NR_DIRTIED, /* page dirtyings since bootup */
244 NR_WRITTEN, /* page writings since bootup */
245 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
246 NR_VM_NODE_STAT_ITEMS
250 * We do arithmetic on the LRU lists in various places in the code,
251 * so it is important to keep the active lists LRU_ACTIVE higher in
252 * the array than the corresponding inactive lists, and to keep
253 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
255 * This has to be kept in sync with the statistics in zone_stat_item
256 * above and the descriptions in vmstat_text in mm/vmstat.c
263 LRU_INACTIVE_ANON = LRU_BASE,
264 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
265 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
266 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
271 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
273 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
275 static inline bool is_file_lru(enum lru_list lru)
277 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
280 static inline bool is_active_lru(enum lru_list lru)
282 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
285 #define ANON_AND_FILE 2
287 struct zone_reclaim_stat {
289 * The pageout code in vmscan.c keeps track of how many of the
290 * mem/swap backed and file backed pages are referenced.
291 * The higher the rotated/scanned ratio, the more valuable
294 * The anon LRU stats live in [0], file LRU stats in [1]
296 unsigned long recent_rotated[2];
297 unsigned long recent_scanned[2];
301 LRUVEC_CONGESTED, /* lruvec has many dirty pages
302 * backed by a congested BDI
306 #endif /* !__GENERATING_BOUNDS_H */
309 * Evictable pages are divided into multiple generations. The youngest and the
310 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
311 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
312 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
313 * corresponding generation. The gen counter in page->flags stores gen+1 while
314 * a page is on one of lrugen->lists[]. Otherwise it stores 0.
316 * A page is added to the youngest generation on faulting. The aging needs to
317 * check the accessed bit at least twice before handing this page over to the
318 * eviction. The first check takes care of the accessed bit set on the initial
319 * fault; the second check makes sure this page hasn't been used since then.
320 * This process, AKA second chance, requires a minimum of two generations,
321 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
322 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
323 * rest of generations, if they exist, are considered inactive. See
324 * lru_gen_is_active().
326 * PG_active is always cleared while a page is on one of lrugen->lists[] so that
327 * the aging needs not to worry about it. And it's set again when a page
328 * considered active is isolated for non-reclaiming purposes, e.g., migration.
329 * See lru_gen_add_page() and lru_gen_del_page().
331 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
332 * number of categories of the active/inactive LRU when keeping track of
333 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
336 #define MIN_NR_GENS 2U
337 #define MAX_NR_GENS 4U
340 * Each generation is divided into multiple tiers. A page accessed N times
341 * through file descriptors is in tier order_base_2(N). A page in the first tier
342 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
343 * tables or read ahead. A page in any other tier (N>1) is marked by
344 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
345 * supported without using additional bits in page->flags.
347 * In contrast to moving across generations which requires the LRU lock, moving
348 * across tiers only involves atomic operations on page->flags and therefore
349 * has a negligible cost in the buffered access path. In the eviction path,
350 * comparisons of refaulted/(evicted+protected) from the first tier and the
351 * rest infer whether pages accessed multiple times through file descriptors
352 * are statistically hot and thus worth protecting.
354 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
355 * number of categories of the active/inactive LRU when keeping track of
356 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
359 #define MAX_NR_TIERS 4U
361 #ifndef __GENERATING_BOUNDS_H
364 struct page_vma_mapped_walk;
366 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
367 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
369 #ifdef CONFIG_LRU_GEN
379 LRU_GEN_NONLEAF_YOUNG,
383 #define MIN_LRU_BATCH BITS_PER_LONG
384 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
386 /* whether to keep historical stats from evicted generations */
387 #ifdef CONFIG_LRU_GEN_STATS
388 #define NR_HIST_GENS MAX_NR_GENS
390 #define NR_HIST_GENS 1U
394 * The youngest generation number is stored in max_seq for both anon and file
395 * types as they are aged on an equal footing. The oldest generation numbers are
396 * stored in min_seq[] separately for anon and file types as clean file pages
397 * can be evicted regardless of swap constraints.
399 * Normally anon and file min_seq are in sync. But if swapping is constrained,
400 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
403 * The number of pages in each generation is eventually consistent and therefore
404 * can be transiently negative when reset_batch_size() is pending.
406 struct lru_gen_struct {
407 /* the aging increments the youngest generation number */
408 unsigned long max_seq;
409 /* the eviction increments the oldest generation numbers */
410 unsigned long min_seq[ANON_AND_FILE];
411 /* the birth time of each generation in jiffies */
412 unsigned long timestamps[MAX_NR_GENS];
413 /* the multi-gen LRU lists, lazily sorted on eviction */
414 struct list_head lists[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
415 /* the multi-gen LRU sizes, eventually consistent */
416 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
417 /* the exponential moving average of refaulted */
418 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
419 /* the exponential moving average of evicted+protected */
420 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
421 /* the first tier doesn't need protection, hence the minus one */
422 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
423 /* can be modified without holding the LRU lock */
424 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
425 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
426 /* whether the multi-gen LRU is enabled */
431 MM_LEAF_TOTAL, /* total leaf entries */
432 MM_LEAF_OLD, /* old leaf entries */
433 MM_LEAF_YOUNG, /* young leaf entries */
434 MM_NONLEAF_TOTAL, /* total non-leaf entries */
435 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
436 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
440 /* double-buffering Bloom filters */
441 #define NR_BLOOM_FILTERS 2
443 struct lru_gen_mm_state {
444 /* set to max_seq after each iteration */
446 /* where the current iteration continues (inclusive) */
447 struct list_head *head;
448 /* where the last iteration ended (exclusive) */
449 struct list_head *tail;
450 /* to wait for the last page table walker to finish */
451 struct wait_queue_head wait;
452 /* Bloom filters flip after each iteration */
453 unsigned long *filters[NR_BLOOM_FILTERS];
454 /* the mm stats for debugging */
455 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
456 /* the number of concurrent page table walkers */
460 struct lru_gen_mm_walk {
461 /* the lruvec under reclaim */
462 struct lruvec *lruvec;
463 /* unstable max_seq from lru_gen_struct */
464 unsigned long max_seq;
465 /* the next address within an mm to scan */
466 unsigned long next_addr;
467 /* to batch promoted pages */
468 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
469 /* to batch the mm stats */
470 int mm_stats[NR_MM_STATS];
471 /* total batched items */
477 void lru_gen_init_lruvec(struct lruvec *lruvec);
478 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
481 void lru_gen_init_memcg(struct mem_cgroup *memcg);
482 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
485 #else /* !CONFIG_LRU_GEN */
487 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
491 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
496 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
500 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
505 #endif /* CONFIG_LRU_GEN */
508 struct list_head lists[NR_LRU_LISTS];
509 struct zone_reclaim_stat reclaim_stat;
510 /* Evictions & activations on the inactive file list */
511 atomic_long_t inactive_age;
512 /* Refaults at the time of last reclaim cycle */
513 unsigned long refaults;
514 /* Various lruvec state flags (enum lruvec_flags) */
516 #ifdef CONFIG_LRU_GEN
517 /* evictable pages divided into generations */
518 struct lru_gen_struct lrugen;
519 /* to concurrently iterate lru_gen_mm_list */
520 struct lru_gen_mm_state mm_state;
523 struct pglist_data *pgdat;
527 /* Isolate unmapped file */
528 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
529 /* Isolate for asynchronous migration */
530 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
531 /* Isolate unevictable pages */
532 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
534 /* LRU Isolation modes. */
535 typedef unsigned __bitwise isolate_mode_t;
537 enum zone_watermarks {
544 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
545 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
546 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
547 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
549 struct per_cpu_pages {
550 int count; /* number of pages in the list */
551 int high; /* high watermark, emptying needed */
552 int batch; /* chunk size for buddy add/remove */
554 /* Lists of pages, one per migrate type stored on the pcp-lists */
555 struct list_head lists[MIGRATE_PCPTYPES];
558 struct per_cpu_pageset {
559 struct per_cpu_pages pcp;
562 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
566 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
570 struct per_cpu_nodestat {
572 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
575 #endif /* !__GENERATING_BOUNDS.H */
579 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
580 * to DMA to all of the addressable memory (ZONE_NORMAL).
581 * On architectures where this area covers the whole 32 bit address
582 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
583 * DMA addressing constraints. This distinction is important as a 32bit
584 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
585 * platforms may need both zones as they support peripherals with
586 * different DMA addressing limitations.
590 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
591 * rest of the lower 4G.
593 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
594 * the specific device.
596 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
599 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
600 * depending on the specific device.
602 * - s390 uses ZONE_DMA fixed to the lower 2G.
604 * - ia64 and riscv only use ZONE_DMA32.
606 * - parisc uses neither.
608 #ifdef CONFIG_ZONE_DMA
611 #ifdef CONFIG_ZONE_DMA32
615 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
616 * performed on pages in ZONE_NORMAL if the DMA devices support
617 * transfers to all addressable memory.
620 #ifdef CONFIG_HIGHMEM
622 * A memory area that is only addressable by the kernel through
623 * mapping portions into its own address space. This is for example
624 * used by i386 to allow the kernel to address the memory beyond
625 * 900MB. The kernel will set up special mappings (page
626 * table entries on i386) for each page that the kernel needs to
632 #ifdef CONFIG_ZONE_DEVICE
639 #ifndef __GENERATING_BOUNDS_H
641 #define ASYNC_AND_SYNC 2
644 /* Read-mostly fields */
646 /* zone watermarks, access with *_wmark_pages(zone) macros */
647 unsigned long _watermark[NR_WMARK];
648 unsigned long watermark_boost;
650 unsigned long nr_reserved_highatomic;
653 * We don't know if the memory that we're going to allocate will be
654 * freeable or/and it will be released eventually, so to avoid totally
655 * wasting several GB of ram we must reserve some of the lower zone
656 * memory (otherwise we risk to run OOM on the lower zones despite
657 * there being tons of freeable ram on the higher zones). This array is
658 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
661 long lowmem_reserve[MAX_NR_ZONES];
666 struct pglist_data *zone_pgdat;
667 struct per_cpu_pageset __percpu *pageset;
669 #ifndef CONFIG_SPARSEMEM
671 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
672 * In SPARSEMEM, this map is stored in struct mem_section
674 unsigned long *pageblock_flags;
675 #endif /* CONFIG_SPARSEMEM */
677 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
678 unsigned long zone_start_pfn;
681 * spanned_pages is the total pages spanned by the zone, including
682 * holes, which is calculated as:
683 * spanned_pages = zone_end_pfn - zone_start_pfn;
685 * present_pages is physical pages existing within the zone, which
687 * present_pages = spanned_pages - absent_pages(pages in holes);
689 * managed_pages is present pages managed by the buddy system, which
690 * is calculated as (reserved_pages includes pages allocated by the
691 * bootmem allocator):
692 * managed_pages = present_pages - reserved_pages;
694 * So present_pages may be used by memory hotplug or memory power
695 * management logic to figure out unmanaged pages by checking
696 * (present_pages - managed_pages). And managed_pages should be used
697 * by page allocator and vm scanner to calculate all kinds of watermarks
702 * zone_start_pfn and spanned_pages are protected by span_seqlock.
703 * It is a seqlock because it has to be read outside of zone->lock,
704 * and it is done in the main allocator path. But, it is written
705 * quite infrequently.
707 * The span_seq lock is declared along with zone->lock because it is
708 * frequently read in proximity to zone->lock. It's good to
709 * give them a chance of being in the same cacheline.
711 * Write access to present_pages at runtime should be protected by
712 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
713 * present_pages should get_online_mems() to get a stable value.
715 atomic_long_t managed_pages;
716 unsigned long spanned_pages;
717 unsigned long present_pages;
721 #ifdef CONFIG_MEMORY_ISOLATION
723 * Number of isolated pageblock. It is used to solve incorrect
724 * freepage counting problem due to racy retrieving migratetype
725 * of pageblock. Protected by zone->lock.
727 unsigned long nr_isolate_pageblock;
730 #ifdef CONFIG_MEMORY_HOTPLUG
731 /* see spanned/present_pages for more description */
732 seqlock_t span_seqlock;
737 /* Write-intensive fields used from the page allocator */
740 /* free areas of different sizes */
741 struct free_area free_area[MAX_ORDER];
743 /* zone flags, see below */
746 /* Primarily protects free_area */
749 /* Write-intensive fields used by compaction and vmstats. */
753 * When free pages are below this point, additional steps are taken
754 * when reading the number of free pages to avoid per-cpu counter
755 * drift allowing watermarks to be breached
757 unsigned long percpu_drift_mark;
759 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
760 /* pfn where compaction free scanner should start */
761 unsigned long compact_cached_free_pfn;
762 /* pfn where compaction migration scanner should start */
763 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
764 unsigned long compact_init_migrate_pfn;
765 unsigned long compact_init_free_pfn;
768 #ifdef CONFIG_COMPACTION
770 * On compaction failure, 1<<compact_defer_shift compactions
771 * are skipped before trying again. The number attempted since
772 * last failure is tracked with compact_considered.
774 unsigned int compact_considered;
775 unsigned int compact_defer_shift;
776 int compact_order_failed;
779 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
780 /* Set to true when the PG_migrate_skip bits should be cleared */
781 bool compact_blockskip_flush;
787 /* Zone statistics */
788 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
789 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
790 } ____cacheline_internodealigned_in_smp;
793 PGDAT_DIRTY, /* reclaim scanning has recently found
794 * many dirty file pages at the tail
797 PGDAT_WRITEBACK, /* reclaim scanning has recently found
798 * many pages under writeback
800 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
804 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
805 * Cleared when kswapd is woken.
809 static inline unsigned long zone_managed_pages(struct zone *zone)
811 return (unsigned long)atomic_long_read(&zone->managed_pages);
814 static inline unsigned long zone_end_pfn(const struct zone *zone)
816 return zone->zone_start_pfn + zone->spanned_pages;
819 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
821 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
824 static inline bool zone_is_initialized(struct zone *zone)
826 return zone->initialized;
829 static inline bool zone_is_empty(struct zone *zone)
831 return zone->spanned_pages == 0;
835 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
836 * intersection with the given zone
838 static inline bool zone_intersects(struct zone *zone,
839 unsigned long start_pfn, unsigned long nr_pages)
841 if (zone_is_empty(zone))
843 if (start_pfn >= zone_end_pfn(zone) ||
844 start_pfn + nr_pages <= zone->zone_start_pfn)
851 * The "priority" of VM scanning is how much of the queues we will scan in one
852 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
853 * queues ("queue_length >> 12") during an aging round.
855 #define DEF_PRIORITY 12
857 /* Maximum number of zones on a zonelist */
858 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
861 ZONELIST_FALLBACK, /* zonelist with fallback */
864 * The NUMA zonelists are doubled because we need zonelists that
865 * restrict the allocations to a single node for __GFP_THISNODE.
867 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
873 * This struct contains information about a zone in a zonelist. It is stored
874 * here to avoid dereferences into large structures and lookups of tables
877 struct zone *zone; /* Pointer to actual zone */
878 int zone_idx; /* zone_idx(zoneref->zone) */
882 * One allocation request operates on a zonelist. A zonelist
883 * is a list of zones, the first one is the 'goal' of the
884 * allocation, the other zones are fallback zones, in decreasing
887 * To speed the reading of the zonelist, the zonerefs contain the zone index
888 * of the entry being read. Helper functions to access information given
889 * a struct zoneref are
891 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
892 * zonelist_zone_idx() - Return the index of the zone for an entry
893 * zonelist_node_idx() - Return the index of the node for an entry
896 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
899 #ifndef CONFIG_DISCONTIGMEM
900 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
901 extern struct page *mem_map;
904 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
905 struct deferred_split {
906 spinlock_t split_queue_lock;
907 struct list_head split_queue;
908 unsigned long split_queue_len;
913 * On NUMA machines, each NUMA node would have a pg_data_t to describe
914 * it's memory layout. On UMA machines there is a single pglist_data which
915 * describes the whole memory.
917 * Memory statistics and page replacement data structures are maintained on a
921 typedef struct pglist_data {
922 struct zone node_zones[MAX_NR_ZONES];
923 struct zonelist node_zonelists[MAX_ZONELISTS];
925 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
926 struct page *node_mem_map;
927 #ifdef CONFIG_PAGE_EXTENSION
928 struct page_ext *node_page_ext;
931 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
933 * Must be held any time you expect node_start_pfn,
934 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
935 * Also synchronizes pgdat->first_deferred_pfn during deferred page
938 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
939 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
940 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
942 * Nests above zone->lock and zone->span_seqlock
944 spinlock_t node_size_lock;
946 unsigned long node_start_pfn;
947 unsigned long node_present_pages; /* total number of physical pages */
948 unsigned long node_spanned_pages; /* total size of physical page
949 range, including holes */
951 wait_queue_head_t kswapd_wait;
952 wait_queue_head_t pfmemalloc_wait;
953 struct task_struct *kswapd; /* Protected by
954 mem_hotplug_begin/end() */
956 enum zone_type kswapd_classzone_idx;
958 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
960 #ifdef CONFIG_COMPACTION
961 int kcompactd_max_order;
962 enum zone_type kcompactd_classzone_idx;
963 wait_queue_head_t kcompactd_wait;
964 struct task_struct *kcompactd;
967 * This is a per-node reserve of pages that are not available
968 * to userspace allocations.
970 unsigned long totalreserve_pages;
974 * zone reclaim becomes active if more unmapped pages exist.
976 unsigned long min_unmapped_pages;
977 unsigned long min_slab_pages;
978 #endif /* CONFIG_NUMA */
980 /* Write-intensive fields used by page reclaim */
984 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
986 * If memory initialisation on large machines is deferred then this
987 * is the first PFN that needs to be initialised.
989 unsigned long first_deferred_pfn;
990 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
992 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
993 struct deferred_split deferred_split_queue;
996 /* Fields commonly accessed by the page reclaim scanner */
999 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1001 * Use mem_cgroup_lruvec() to look up lruvecs.
1003 struct lruvec __lruvec;
1005 unsigned long flags;
1007 #ifdef CONFIG_LRU_GEN
1008 /* kswap mm walk data */
1009 struct lru_gen_mm_walk mm_walk;
1012 ZONE_PADDING(_pad2_)
1014 /* Per-node vmstats */
1015 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1016 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1019 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1020 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1021 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1022 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
1024 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
1026 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
1028 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1029 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1031 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1033 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1036 static inline bool pgdat_is_empty(pg_data_t *pgdat)
1038 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
1041 #include <linux/memory_hotplug.h>
1043 void build_all_zonelists(pg_data_t *pgdat);
1044 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1045 enum zone_type classzone_idx);
1046 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1047 int classzone_idx, unsigned int alloc_flags,
1049 bool zone_watermark_ok(struct zone *z, unsigned int order,
1050 unsigned long mark, int classzone_idx,
1051 unsigned int alloc_flags);
1052 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1053 unsigned long mark, int classzone_idx);
1054 enum memmap_context {
1058 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1059 unsigned long size);
1061 extern void lruvec_init(struct lruvec *lruvec);
1063 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1066 return lruvec->pgdat;
1068 return container_of(lruvec, struct pglist_data, __lruvec);
1072 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
1074 #ifdef CONFIG_HAVE_MEMORY_PRESENT
1075 void memory_present(int nid, unsigned long start, unsigned long end);
1077 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
1080 #if defined(CONFIG_SPARSEMEM)
1081 void memblocks_present(void);
1083 static inline void memblocks_present(void) {}
1086 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1087 int local_memory_node(int node_id);
1089 static inline int local_memory_node(int node_id) { return node_id; };
1093 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1095 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1098 * Returns true if a zone has pages managed by the buddy allocator.
1099 * All the reclaim decisions have to use this function rather than
1100 * populated_zone(). If the whole zone is reserved then we can easily
1101 * end up with populated_zone() && !managed_zone().
1103 static inline bool managed_zone(struct zone *zone)
1105 return zone_managed_pages(zone);
1108 /* Returns true if a zone has memory */
1109 static inline bool populated_zone(struct zone *zone)
1111 return zone->present_pages;
1115 static inline int zone_to_nid(struct zone *zone)
1120 static inline void zone_set_nid(struct zone *zone, int nid)
1125 static inline int zone_to_nid(struct zone *zone)
1130 static inline void zone_set_nid(struct zone *zone, int nid) {}
1133 extern int movable_zone;
1135 #ifdef CONFIG_HIGHMEM
1136 static inline int zone_movable_is_highmem(void)
1138 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1139 return movable_zone == ZONE_HIGHMEM;
1141 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
1146 static inline int is_highmem_idx(enum zone_type idx)
1148 #ifdef CONFIG_HIGHMEM
1149 return (idx == ZONE_HIGHMEM ||
1150 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
1157 * is_highmem - helper function to quickly check if a struct zone is a
1158 * highmem zone or not. This is an attempt to keep references
1159 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1160 * @zone - pointer to struct zone variable
1162 static inline int is_highmem(struct zone *zone)
1164 #ifdef CONFIG_HIGHMEM
1165 return is_highmem_idx(zone_idx(zone));
1171 /* These two functions are used to setup the per zone pages min values */
1173 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
1174 void __user *, size_t *, loff_t *);
1175 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
1176 void __user *, size_t *, loff_t *);
1177 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
1178 void __user *, size_t *, loff_t *);
1179 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1180 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
1181 void __user *, size_t *, loff_t *);
1182 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
1183 void __user *, size_t *, loff_t *);
1184 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1185 void __user *, size_t *, loff_t *);
1186 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1187 void __user *, size_t *, loff_t *);
1189 extern int numa_zonelist_order_handler(struct ctl_table *, int,
1190 void __user *, size_t *, loff_t *);
1191 extern char numa_zonelist_order[];
1192 #define NUMA_ZONELIST_ORDER_LEN 16
1194 #ifndef CONFIG_NEED_MULTIPLE_NODES
1196 extern struct pglist_data contig_page_data;
1197 #define NODE_DATA(nid) (&contig_page_data)
1198 #define NODE_MEM_MAP(nid) mem_map
1200 #else /* CONFIG_NEED_MULTIPLE_NODES */
1202 #include <asm/mmzone.h>
1204 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1206 extern struct pglist_data *first_online_pgdat(void);
1207 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1208 extern struct zone *next_zone(struct zone *zone);
1211 * for_each_online_pgdat - helper macro to iterate over all online nodes
1212 * @pgdat - pointer to a pg_data_t variable
1214 #define for_each_online_pgdat(pgdat) \
1215 for (pgdat = first_online_pgdat(); \
1217 pgdat = next_online_pgdat(pgdat))
1219 * for_each_zone - helper macro to iterate over all memory zones
1220 * @zone - pointer to struct zone variable
1222 * The user only needs to declare the zone variable, for_each_zone
1225 #define for_each_zone(zone) \
1226 for (zone = (first_online_pgdat())->node_zones; \
1228 zone = next_zone(zone))
1230 #define for_each_populated_zone(zone) \
1231 for (zone = (first_online_pgdat())->node_zones; \
1233 zone = next_zone(zone)) \
1234 if (!populated_zone(zone)) \
1235 ; /* do nothing */ \
1238 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1240 return zoneref->zone;
1243 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1245 return zoneref->zone_idx;
1248 static inline int zonelist_node_idx(struct zoneref *zoneref)
1250 return zone_to_nid(zoneref->zone);
1253 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1254 enum zone_type highest_zoneidx,
1258 * 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
1259 * @z - The cursor used as a starting point for the search
1260 * @highest_zoneidx - The zone index of the highest zone to return
1261 * @nodes - An optional nodemask to filter the zonelist with
1263 * This function returns the next zone at or below a given zone index that is
1264 * within the allowed nodemask using a cursor as the starting point for the
1265 * search. The zoneref returned is a cursor that represents the current zone
1266 * being examined. It should be advanced by one before calling
1267 * next_zones_zonelist again.
1269 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1270 enum zone_type highest_zoneidx,
1273 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1275 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1279 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1280 * @zonelist - The zonelist to search for a suitable zone
1281 * @highest_zoneidx - The zone index of the highest zone to return
1282 * @nodes - An optional nodemask to filter the zonelist with
1283 * @return - Zoneref pointer for the first suitable zone found (see below)
1285 * This function returns the first zone at or below a given zone index that is
1286 * within the allowed nodemask. The zoneref returned is a cursor that can be
1287 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1288 * one before calling.
1290 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1291 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1292 * update due to cpuset modification.
1294 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1295 enum zone_type highest_zoneidx,
1298 return next_zones_zonelist(zonelist->_zonerefs,
1299 highest_zoneidx, nodes);
1303 * 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
1304 * @zone - The current zone in the iterator
1305 * @z - The current pointer within zonelist->zones being iterated
1306 * @zlist - The zonelist being iterated
1307 * @highidx - The zone index of the highest zone to return
1308 * @nodemask - Nodemask allowed by the allocator
1310 * This iterator iterates though all zones at or below a given zone index and
1311 * within a given nodemask
1313 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1314 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1316 z = next_zones_zonelist(++z, highidx, nodemask), \
1317 zone = zonelist_zone(z))
1319 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1320 for (zone = z->zone; \
1322 z = next_zones_zonelist(++z, highidx, nodemask), \
1323 zone = zonelist_zone(z))
1327 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1328 * @zone - The current zone in the iterator
1329 * @z - The current pointer within zonelist->zones being iterated
1330 * @zlist - The zonelist being iterated
1331 * @highidx - The zone index of the highest zone to return
1333 * This iterator iterates though all zones at or below a given zone index.
1335 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1336 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1338 #ifdef CONFIG_SPARSEMEM
1339 #include <asm/sparsemem.h>
1342 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1343 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1344 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1346 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1351 #ifdef CONFIG_FLATMEM
1352 #define pfn_to_nid(pfn) (0)
1355 #ifdef CONFIG_SPARSEMEM
1358 * SECTION_SHIFT #bits space required to store a section #
1360 * PA_SECTION_SHIFT physical address to/from section number
1361 * PFN_SECTION_SHIFT pfn to/from section number
1363 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1364 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1366 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1368 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1369 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1371 #define SECTION_BLOCKFLAGS_BITS \
1372 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1374 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1375 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1378 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1380 return pfn >> PFN_SECTION_SHIFT;
1382 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1384 return sec << PFN_SECTION_SHIFT;
1387 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1388 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1390 #define SUBSECTION_SHIFT 21
1392 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1393 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1394 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1396 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1397 #error Subsection size exceeds section size
1399 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1402 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1403 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1405 struct mem_section_usage {
1406 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1407 /* See declaration of similar field in struct zone */
1408 unsigned long pageblock_flags[0];
1411 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1415 struct mem_section {
1417 * This is, logically, a pointer to an array of struct
1418 * pages. However, it is stored with some other magic.
1419 * (see sparse.c::sparse_init_one_section())
1421 * Additionally during early boot we encode node id of
1422 * the location of the section here to guide allocation.
1423 * (see sparse.c::memory_present())
1425 * Making it a UL at least makes someone do a cast
1426 * before using it wrong.
1428 unsigned long section_mem_map;
1430 struct mem_section_usage *usage;
1431 #ifdef CONFIG_PAGE_EXTENSION
1433 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1434 * section. (see page_ext.h about this.)
1436 struct page_ext *page_ext;
1440 * WARNING: mem_section must be a power-of-2 in size for the
1441 * calculation and use of SECTION_ROOT_MASK to make sense.
1445 #ifdef CONFIG_SPARSEMEM_EXTREME
1446 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1448 #define SECTIONS_PER_ROOT 1
1451 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1452 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1453 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1455 #ifdef CONFIG_SPARSEMEM_EXTREME
1456 extern struct mem_section **mem_section;
1458 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1461 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1463 return ms->usage->pageblock_flags;
1466 static inline struct mem_section *__nr_to_section(unsigned long nr)
1468 #ifdef CONFIG_SPARSEMEM_EXTREME
1472 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1474 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1476 extern unsigned long __section_nr(struct mem_section *ms);
1477 extern size_t mem_section_usage_size(void);
1480 * We use the lower bits of the mem_map pointer to store
1481 * a little bit of information. The pointer is calculated
1482 * as mem_map - section_nr_to_pfn(pnum). The result is
1483 * aligned to the minimum alignment of the two values:
1484 * 1. All mem_map arrays are page-aligned.
1485 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1486 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1487 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1488 * worst combination is powerpc with 256k pages,
1489 * which results in PFN_SECTION_SHIFT equal 6.
1490 * To sum it up, at least 6 bits are available.
1492 #define SECTION_MARKED_PRESENT (1UL<<0)
1493 #define SECTION_HAS_MEM_MAP (1UL<<1)
1494 #define SECTION_IS_ONLINE (1UL<<2)
1495 #define SECTION_IS_EARLY (1UL<<3)
1496 #define SECTION_MAP_LAST_BIT (1UL<<4)
1497 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1498 #define SECTION_NID_SHIFT 3
1500 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1502 unsigned long map = section->section_mem_map;
1503 map &= SECTION_MAP_MASK;
1504 return (struct page *)map;
1507 static inline int present_section(struct mem_section *section)
1509 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1512 static inline int present_section_nr(unsigned long nr)
1514 return present_section(__nr_to_section(nr));
1517 static inline int valid_section(struct mem_section *section)
1519 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1522 static inline int early_section(struct mem_section *section)
1524 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1527 static inline int valid_section_nr(unsigned long nr)
1529 return valid_section(__nr_to_section(nr));
1532 static inline int online_section(struct mem_section *section)
1534 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1537 static inline int online_section_nr(unsigned long nr)
1539 return online_section(__nr_to_section(nr));
1542 #ifdef CONFIG_MEMORY_HOTPLUG
1543 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1544 #ifdef CONFIG_MEMORY_HOTREMOVE
1545 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1549 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1551 return __nr_to_section(pfn_to_section_nr(pfn));
1554 extern unsigned long __highest_present_section_nr;
1556 static inline int subsection_map_index(unsigned long pfn)
1558 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1561 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1562 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1564 int idx = subsection_map_index(pfn);
1566 return test_bit(idx, ms->usage->subsection_map);
1569 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1575 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1576 static inline int pfn_valid(unsigned long pfn)
1578 struct mem_section *ms;
1580 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1582 ms = __nr_to_section(pfn_to_section_nr(pfn));
1583 if (!valid_section(ms))
1586 * Traditionally early sections always returned pfn_valid() for
1587 * the entire section-sized span.
1589 return early_section(ms) || pfn_section_valid(ms, pfn);
1593 static inline int pfn_present(unsigned long pfn)
1595 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1597 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1601 * These are _only_ used during initialisation, therefore they
1602 * can use __initdata ... They could have names to indicate
1606 #define pfn_to_nid(pfn) \
1608 unsigned long __pfn_to_nid_pfn = (pfn); \
1609 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1612 #define pfn_to_nid(pfn) (0)
1615 #define early_pfn_valid(pfn) pfn_valid(pfn)
1616 void sparse_init(void);
1618 #define sparse_init() do {} while (0)
1619 #define sparse_index_init(_sec, _nid) do {} while (0)
1620 #define pfn_present pfn_valid
1621 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1622 #endif /* CONFIG_SPARSEMEM */
1625 * During memory init memblocks map pfns to nids. The search is expensive and
1626 * this caches recent lookups. The implementation of __early_pfn_to_nid
1627 * may treat start/end as pfns or sections.
1629 struct mminit_pfnnid_cache {
1630 unsigned long last_start;
1631 unsigned long last_end;
1635 #ifndef early_pfn_valid
1636 #define early_pfn_valid(pfn) (1)
1639 void memory_present(int nid, unsigned long start, unsigned long end);
1642 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1643 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1644 * pfn_valid_within() should be used in this case; we optimise this away
1645 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1647 #ifdef CONFIG_HOLES_IN_ZONE
1648 #define pfn_valid_within(pfn) pfn_valid(pfn)
1650 #define pfn_valid_within(pfn) (1)
1653 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1655 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1656 * associated with it or not. This means that a struct page exists for this
1657 * pfn. The caller cannot assume the page is fully initialized in general.
1658 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1659 * will ensure the struct page is fully online and initialized. Special pages
1660 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1662 * In FLATMEM, it is expected that holes always have valid memmap as long as
1663 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1664 * that a valid section has a memmap for the entire section.
1666 * However, an ARM, and maybe other embedded architectures in the future
1667 * free memmap backing holes to save memory on the assumption the memmap is
1668 * never used. The page_zone linkages are then broken even though pfn_valid()
1669 * returns true. A walker of the full memmap must then do this additional
1670 * check to ensure the memmap they are looking at is sane by making sure
1671 * the zone and PFN linkages are still valid. This is expensive, but walkers
1672 * of the full memmap are extremely rare.
1674 bool memmap_valid_within(unsigned long pfn,
1675 struct page *page, struct zone *zone);
1677 static inline bool memmap_valid_within(unsigned long pfn,
1678 struct page *page, struct zone *zone)
1682 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1684 #endif /* !__GENERATING_BOUNDS.H */
1685 #endif /* !__ASSEMBLY__ */
1686 #endif /* _LINUX_MMZONE_H */