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 enum numa_stat_item {
126 NUMA_HIT, /* allocated in intended node */
127 NUMA_MISS, /* allocated in non intended node */
128 NUMA_FOREIGN, /* was intended here, hit elsewhere */
129 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
130 NUMA_LOCAL, /* allocation from local node */
131 NUMA_OTHER, /* allocation from other node */
132 NR_VM_NUMA_EVENT_ITEMS
135 #define NR_VM_NUMA_EVENT_ITEMS 0
138 enum zone_stat_item {
139 /* First 128 byte cacheline (assuming 64 bit words) */
141 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
142 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
144 NR_ZONE_INACTIVE_FILE,
147 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
148 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
149 /* Second 128 byte cacheline */
151 #if IS_ENABLED(CONFIG_ZSMALLOC)
152 NR_ZSPAGES, /* allocated in zsmalloc */
155 NR_VM_ZONE_STAT_ITEMS };
157 enum node_stat_item {
159 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
160 NR_ACTIVE_ANON, /* " " " " " */
161 NR_INACTIVE_FILE, /* " " " " " */
162 NR_ACTIVE_FILE, /* " " " " " */
163 NR_UNEVICTABLE, /* " " " " " */
164 NR_SLAB_RECLAIMABLE_B,
165 NR_SLAB_UNRECLAIMABLE_B,
166 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
167 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
169 WORKINGSET_REFAULT_BASE,
170 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
171 WORKINGSET_REFAULT_FILE,
172 WORKINGSET_ACTIVATE_BASE,
173 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
174 WORKINGSET_ACTIVATE_FILE,
175 WORKINGSET_RESTORE_BASE,
176 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
177 WORKINGSET_RESTORE_FILE,
178 WORKINGSET_NODERECLAIM,
179 NR_ANON_MAPPED, /* Mapped anonymous pages */
180 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
181 only modified from process context */
185 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
186 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
193 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
194 NR_DIRTIED, /* page dirtyings since bootup */
195 NR_WRITTEN, /* page writings since bootup */
196 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
197 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
198 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
199 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
200 NR_KERNEL_STACK_KB, /* measured in KiB */
201 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
202 NR_KERNEL_SCS_KB, /* measured in KiB */
204 NR_PAGETABLE, /* used for pagetables */
205 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
209 #ifdef CONFIG_NUMA_BALANCING
210 PGPROMOTE_SUCCESS, /* promote successfully */
211 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
213 NR_VM_NODE_STAT_ITEMS
217 * Returns true if the item should be printed in THPs (/proc/vmstat
218 * currently prints number of anon, file and shmem THPs. But the item
219 * is charged in pages).
221 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
223 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
226 return item == NR_ANON_THPS ||
227 item == NR_FILE_THPS ||
228 item == NR_SHMEM_THPS ||
229 item == NR_SHMEM_PMDMAPPED ||
230 item == NR_FILE_PMDMAPPED;
234 * Returns true if the value is measured in bytes (most vmstat values are
235 * measured in pages). This defines the API part, the internal representation
236 * might be different.
238 static __always_inline bool vmstat_item_in_bytes(int idx)
241 * Global and per-node slab counters track slab pages.
242 * It's expected that changes are multiples of PAGE_SIZE.
243 * Internally values are stored in pages.
245 * Per-memcg and per-lruvec counters track memory, consumed
246 * by individual slab objects. These counters are actually
249 return (idx == NR_SLAB_RECLAIMABLE_B ||
250 idx == NR_SLAB_UNRECLAIMABLE_B);
254 * We do arithmetic on the LRU lists in various places in the code,
255 * so it is important to keep the active lists LRU_ACTIVE higher in
256 * the array than the corresponding inactive lists, and to keep
257 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
259 * This has to be kept in sync with the statistics in zone_stat_item
260 * above and the descriptions in vmstat_text in mm/vmstat.c
267 LRU_INACTIVE_ANON = LRU_BASE,
268 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
269 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
270 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
275 enum vmscan_throttle_state {
276 VMSCAN_THROTTLE_WRITEBACK,
277 VMSCAN_THROTTLE_ISOLATED,
278 VMSCAN_THROTTLE_NOPROGRESS,
279 VMSCAN_THROTTLE_CONGESTED,
283 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
285 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
287 static inline bool is_file_lru(enum lru_list lru)
289 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
292 static inline bool is_active_lru(enum lru_list lru)
294 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
297 #define WORKINGSET_ANON 0
298 #define WORKINGSET_FILE 1
299 #define ANON_AND_FILE 2
302 LRUVEC_CONGESTED, /* lruvec has many dirty pages
303 * backed by a congested BDI
307 #endif /* !__GENERATING_BOUNDS_H */
310 * Evictable pages are divided into multiple generations. The youngest and the
311 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
312 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
313 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
314 * corresponding generation. The gen counter in folio->flags stores gen+1 while
315 * a page is on one of lrugen->lists[]. Otherwise it stores 0.
317 * A page is added to the youngest generation on faulting. The aging needs to
318 * check the accessed bit at least twice before handing this page over to the
319 * eviction. The first check takes care of the accessed bit set on the initial
320 * fault; the second check makes sure this page hasn't been used since then.
321 * This process, AKA second chance, requires a minimum of two generations,
322 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
323 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
324 * rest of generations, if they exist, are considered inactive. See
325 * lru_gen_is_active().
327 * PG_active is always cleared while a page is on one of lrugen->lists[] so that
328 * the aging needs not to worry about it. And it's set again when a page
329 * considered active is isolated for non-reclaiming purposes, e.g., migration.
330 * See lru_gen_add_folio() and lru_gen_del_folio().
332 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
333 * number of categories of the active/inactive LRU when keeping track of
334 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
337 #define MIN_NR_GENS 2U
338 #define MAX_NR_GENS 4U
341 * Each generation is divided into multiple tiers. A page accessed N times
342 * through file descriptors is in tier order_base_2(N). A page in the first tier
343 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
344 * tables or read ahead. A page in any other tier (N>1) is marked by
345 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
346 * supported without using additional bits in folio->flags.
348 * In contrast to moving across generations which requires the LRU lock, moving
349 * across tiers only involves atomic operations on folio->flags and therefore
350 * has a negligible cost in the buffered access path. In the eviction path,
351 * comparisons of refaulted/(evicted+protected) from the first tier and the
352 * rest infer whether pages accessed multiple times through file descriptors
353 * are statistically hot and thus worth protecting.
355 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
356 * number of categories of the active/inactive LRU when keeping track of
357 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
360 #define MAX_NR_TIERS 4U
362 #ifndef __GENERATING_BOUNDS_H
365 struct page_vma_mapped_walk;
367 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
368 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
370 #ifdef CONFIG_LRU_GEN
380 LRU_GEN_NONLEAF_YOUNG,
384 #define MIN_LRU_BATCH BITS_PER_LONG
385 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
387 /* whether to keep historical stats from evicted generations */
388 #ifdef CONFIG_LRU_GEN_STATS
389 #define NR_HIST_GENS MAX_NR_GENS
391 #define NR_HIST_GENS 1U
395 * The youngest generation number is stored in max_seq for both anon and file
396 * types as they are aged on an equal footing. The oldest generation numbers are
397 * stored in min_seq[] separately for anon and file types as clean file pages
398 * can be evicted regardless of swap constraints.
400 * Normally anon and file min_seq are in sync. But if swapping is constrained,
401 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
404 * The number of pages in each generation is eventually consistent and therefore
405 * can be transiently negative when reset_batch_size() is pending.
407 struct lru_gen_struct {
408 /* the aging increments the youngest generation number */
409 unsigned long max_seq;
410 /* the eviction increments the oldest generation numbers */
411 unsigned long min_seq[ANON_AND_FILE];
412 /* the birth time of each generation in jiffies */
413 unsigned long timestamps[MAX_NR_GENS];
414 /* the multi-gen LRU lists, lazily sorted on eviction */
415 struct list_head lists[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
416 /* the multi-gen LRU sizes, eventually consistent */
417 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
418 /* the exponential moving average of refaulted */
419 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
420 /* the exponential moving average of evicted+protected */
421 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
422 /* the first tier doesn't need protection, hence the minus one */
423 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
424 /* can be modified without holding the LRU lock */
425 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
426 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
427 /* whether the multi-gen LRU is enabled */
432 MM_LEAF_TOTAL, /* total leaf entries */
433 MM_LEAF_OLD, /* old leaf entries */
434 MM_LEAF_YOUNG, /* young leaf entries */
435 MM_NONLEAF_TOTAL, /* total non-leaf entries */
436 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
437 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
441 /* double-buffering Bloom filters */
442 #define NR_BLOOM_FILTERS 2
444 struct lru_gen_mm_state {
445 /* set to max_seq after each iteration */
447 /* where the current iteration continues (inclusive) */
448 struct list_head *head;
449 /* where the last iteration ended (exclusive) */
450 struct list_head *tail;
451 /* to wait for the last page table walker to finish */
452 struct wait_queue_head wait;
453 /* Bloom filters flip after each iteration */
454 unsigned long *filters[NR_BLOOM_FILTERS];
455 /* the mm stats for debugging */
456 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
457 /* the number of concurrent page table walkers */
461 struct lru_gen_mm_walk {
462 /* the lruvec under reclaim */
463 struct lruvec *lruvec;
464 /* unstable max_seq from lru_gen_struct */
465 unsigned long max_seq;
466 /* the next address within an mm to scan */
467 unsigned long next_addr;
468 /* to batch promoted pages */
469 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
470 /* to batch the mm stats */
471 int mm_stats[NR_MM_STATS];
472 /* total batched items */
478 void lru_gen_init_lruvec(struct lruvec *lruvec);
479 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
482 void lru_gen_init_memcg(struct mem_cgroup *memcg);
483 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
486 #else /* !CONFIG_LRU_GEN */
488 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
492 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
497 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
501 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
506 #endif /* CONFIG_LRU_GEN */
509 struct list_head lists[NR_LRU_LISTS];
510 /* per lruvec lru_lock for memcg */
513 * These track the cost of reclaiming one LRU - file or anon -
514 * over the other. As the observed cost of reclaiming one LRU
515 * increases, the reclaim scan balance tips toward the other.
517 unsigned long anon_cost;
518 unsigned long file_cost;
519 /* Non-resident age, driven by LRU movement */
520 atomic_long_t nonresident_age;
521 /* Refaults at the time of last reclaim cycle */
522 unsigned long refaults[ANON_AND_FILE];
523 /* Various lruvec state flags (enum lruvec_flags) */
525 #ifdef CONFIG_LRU_GEN
526 /* evictable pages divided into generations */
527 struct lru_gen_struct lrugen;
528 /* to concurrently iterate lru_gen_mm_list */
529 struct lru_gen_mm_state mm_state;
532 struct pglist_data *pgdat;
536 /* Isolate unmapped pages */
537 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
538 /* Isolate for asynchronous migration */
539 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
540 /* Isolate unevictable pages */
541 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
543 /* LRU Isolation modes. */
544 typedef unsigned __bitwise isolate_mode_t;
546 enum zone_watermarks {
555 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
556 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
557 * it should not contribute to serious fragmentation causing THP allocation
560 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
565 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
566 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
568 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
569 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
570 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
571 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
573 /* Fields and list protected by pagesets local_lock in page_alloc.c */
574 struct per_cpu_pages {
575 spinlock_t lock; /* Protects lists field */
576 int count; /* number of pages in the list */
577 int high; /* high watermark, emptying needed */
578 int batch; /* chunk size for buddy add/remove */
579 short free_factor; /* batch scaling factor during free */
581 short expire; /* When 0, remote pagesets are drained */
584 /* Lists of pages, one per migrate type stored on the pcp-lists */
585 struct list_head lists[NR_PCP_LISTS];
586 } ____cacheline_aligned_in_smp;
588 struct per_cpu_zonestat {
590 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
595 * Low priority inaccurate counters that are only folded
596 * on demand. Use a large type to avoid the overhead of
597 * folding during refresh_cpu_vm_stats.
599 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
603 struct per_cpu_nodestat {
605 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
608 #endif /* !__GENERATING_BOUNDS.H */
612 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
613 * to DMA to all of the addressable memory (ZONE_NORMAL).
614 * On architectures where this area covers the whole 32 bit address
615 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
616 * DMA addressing constraints. This distinction is important as a 32bit
617 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
618 * platforms may need both zones as they support peripherals with
619 * different DMA addressing limitations.
621 #ifdef CONFIG_ZONE_DMA
624 #ifdef CONFIG_ZONE_DMA32
628 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
629 * performed on pages in ZONE_NORMAL if the DMA devices support
630 * transfers to all addressable memory.
633 #ifdef CONFIG_HIGHMEM
635 * A memory area that is only addressable by the kernel through
636 * mapping portions into its own address space. This is for example
637 * used by i386 to allow the kernel to address the memory beyond
638 * 900MB. The kernel will set up special mappings (page
639 * table entries on i386) for each page that the kernel needs to
645 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
646 * movable pages with few exceptional cases described below. Main use
647 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
648 * likely to succeed, and to locally limit unmovable allocations - e.g.,
649 * to increase the number of THP/huge pages. Notable special cases are:
651 * 1. Pinned pages: (long-term) pinning of movable pages might
652 * essentially turn such pages unmovable. Therefore, we do not allow
653 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
654 * faulted, they come from the right zone right away. However, it is
655 * still possible that address space already has pages in
656 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
657 * touches that memory before pinning). In such case we migrate them
658 * to a different zone. When migration fails - pinning fails.
659 * 2. memblock allocations: kernelcore/movablecore setups might create
660 * situations where ZONE_MOVABLE contains unmovable allocations
661 * after boot. Memory offlining and allocations fail early.
662 * 3. Memory holes: kernelcore/movablecore setups might create very rare
663 * situations where ZONE_MOVABLE contains memory holes after boot,
664 * for example, if we have sections that are only partially
665 * populated. Memory offlining and allocations fail early.
666 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
667 * memory offlining, such pages cannot be allocated.
668 * 5. Unmovable PG_offline pages: in paravirtualized environments,
669 * hotplugged memory blocks might only partially be managed by the
670 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
671 * parts not manged by the buddy are unmovable PG_offline pages. In
672 * some cases (virtio-mem), such pages can be skipped during
673 * memory offlining, however, cannot be moved/allocated. These
674 * techniques might use alloc_contig_range() to hide previously
675 * exposed pages from the buddy again (e.g., to implement some sort
676 * of memory unplug in virtio-mem).
677 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
678 * situations where ZERO_PAGE(0) which is allocated differently
679 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
680 * cannot be migrated.
681 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
682 * memory to the MOVABLE zone, the vmemmap pages are also placed in
683 * such zone. Such pages cannot be really moved around as they are
684 * self-stored in the range, but they are treated as movable when
685 * the range they describe is about to be offlined.
687 * In general, no unmovable allocations that degrade memory offlining
688 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
689 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
690 * if has_unmovable_pages() states that there are no unmovable pages,
691 * there can be false negatives).
694 #ifdef CONFIG_ZONE_DEVICE
701 #ifndef __GENERATING_BOUNDS_H
703 #define ASYNC_AND_SYNC 2
706 /* Read-mostly fields */
708 /* zone watermarks, access with *_wmark_pages(zone) macros */
709 unsigned long _watermark[NR_WMARK];
710 unsigned long watermark_boost;
712 unsigned long nr_reserved_highatomic;
715 * We don't know if the memory that we're going to allocate will be
716 * freeable or/and it will be released eventually, so to avoid totally
717 * wasting several GB of ram we must reserve some of the lower zone
718 * memory (otherwise we risk to run OOM on the lower zones despite
719 * there being tons of freeable ram on the higher zones). This array is
720 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
723 long lowmem_reserve[MAX_NR_ZONES];
728 struct pglist_data *zone_pgdat;
729 struct per_cpu_pages __percpu *per_cpu_pageset;
730 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
732 * the high and batch values are copied to individual pagesets for
738 #ifndef CONFIG_SPARSEMEM
740 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
741 * In SPARSEMEM, this map is stored in struct mem_section
743 unsigned long *pageblock_flags;
744 #endif /* CONFIG_SPARSEMEM */
746 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
747 unsigned long zone_start_pfn;
750 * spanned_pages is the total pages spanned by the zone, including
751 * holes, which is calculated as:
752 * spanned_pages = zone_end_pfn - zone_start_pfn;
754 * present_pages is physical pages existing within the zone, which
756 * present_pages = spanned_pages - absent_pages(pages in holes);
758 * present_early_pages is present pages existing within the zone
759 * located on memory available since early boot, excluding hotplugged
762 * managed_pages is present pages managed by the buddy system, which
763 * is calculated as (reserved_pages includes pages allocated by the
764 * bootmem allocator):
765 * managed_pages = present_pages - reserved_pages;
767 * cma pages is present pages that are assigned for CMA use
770 * So present_pages may be used by memory hotplug or memory power
771 * management logic to figure out unmanaged pages by checking
772 * (present_pages - managed_pages). And managed_pages should be used
773 * by page allocator and vm scanner to calculate all kinds of watermarks
778 * zone_start_pfn and spanned_pages are protected by span_seqlock.
779 * It is a seqlock because it has to be read outside of zone->lock,
780 * and it is done in the main allocator path. But, it is written
781 * quite infrequently.
783 * The span_seq lock is declared along with zone->lock because it is
784 * frequently read in proximity to zone->lock. It's good to
785 * give them a chance of being in the same cacheline.
787 * Write access to present_pages at runtime should be protected by
788 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
789 * present_pages should use get_online_mems() to get a stable value.
791 atomic_long_t managed_pages;
792 unsigned long spanned_pages;
793 unsigned long present_pages;
794 #if defined(CONFIG_MEMORY_HOTPLUG)
795 unsigned long present_early_pages;
798 unsigned long cma_pages;
803 #ifdef CONFIG_MEMORY_ISOLATION
805 * Number of isolated pageblock. It is used to solve incorrect
806 * freepage counting problem due to racy retrieving migratetype
807 * of pageblock. Protected by zone->lock.
809 unsigned long nr_isolate_pageblock;
812 #ifdef CONFIG_MEMORY_HOTPLUG
813 /* see spanned/present_pages for more description */
814 seqlock_t span_seqlock;
819 /* Write-intensive fields used from the page allocator */
820 CACHELINE_PADDING(_pad1_);
822 /* free areas of different sizes */
823 struct free_area free_area[MAX_ORDER];
825 /* zone flags, see below */
828 /* Primarily protects free_area */
831 /* Write-intensive fields used by compaction and vmstats. */
832 CACHELINE_PADDING(_pad2_);
835 * When free pages are below this point, additional steps are taken
836 * when reading the number of free pages to avoid per-cpu counter
837 * drift allowing watermarks to be breached
839 unsigned long percpu_drift_mark;
841 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
842 /* pfn where compaction free scanner should start */
843 unsigned long compact_cached_free_pfn;
844 /* pfn where compaction migration scanner should start */
845 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
846 unsigned long compact_init_migrate_pfn;
847 unsigned long compact_init_free_pfn;
850 #ifdef CONFIG_COMPACTION
852 * On compaction failure, 1<<compact_defer_shift compactions
853 * are skipped before trying again. The number attempted since
854 * last failure is tracked with compact_considered.
855 * compact_order_failed is the minimum compaction failed order.
857 unsigned int compact_considered;
858 unsigned int compact_defer_shift;
859 int compact_order_failed;
862 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
863 /* Set to true when the PG_migrate_skip bits should be cleared */
864 bool compact_blockskip_flush;
869 CACHELINE_PADDING(_pad3_);
870 /* Zone statistics */
871 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
872 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
873 } ____cacheline_internodealigned_in_smp;
876 PGDAT_DIRTY, /* reclaim scanning has recently found
877 * many dirty file pages at the tail
880 PGDAT_WRITEBACK, /* reclaim scanning has recently found
881 * many pages under writeback
883 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
887 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
888 * Cleared when kswapd is woken.
890 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
893 static inline unsigned long zone_managed_pages(struct zone *zone)
895 return (unsigned long)atomic_long_read(&zone->managed_pages);
898 static inline unsigned long zone_cma_pages(struct zone *zone)
901 return zone->cma_pages;
907 static inline unsigned long zone_end_pfn(const struct zone *zone)
909 return zone->zone_start_pfn + zone->spanned_pages;
912 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
914 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
917 static inline bool zone_is_initialized(struct zone *zone)
919 return zone->initialized;
922 static inline bool zone_is_empty(struct zone *zone)
924 return zone->spanned_pages == 0;
927 #ifndef BUILD_VDSO32_64
929 * The zone field is never updated after free_area_init_core()
930 * sets it, so none of the operations on it need to be atomic.
933 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
934 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
935 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
936 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
937 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
938 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
939 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
940 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
943 * Define the bit shifts to access each section. For non-existent
944 * sections we define the shift as 0; that plus a 0 mask ensures
945 * the compiler will optimise away reference to them.
947 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
948 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
949 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
950 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
951 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
953 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
954 #ifdef NODE_NOT_IN_PAGE_FLAGS
955 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
956 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
957 SECTIONS_PGOFF : ZONES_PGOFF)
959 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
960 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
961 NODES_PGOFF : ZONES_PGOFF)
964 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
966 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
967 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
968 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
969 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
970 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
971 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
973 static inline enum zone_type page_zonenum(const struct page *page)
975 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
976 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
979 static inline enum zone_type folio_zonenum(const struct folio *folio)
981 return page_zonenum(&folio->page);
984 #ifdef CONFIG_ZONE_DEVICE
985 static inline bool is_zone_device_page(const struct page *page)
987 return page_zonenum(page) == ZONE_DEVICE;
989 extern void memmap_init_zone_device(struct zone *, unsigned long,
990 unsigned long, struct dev_pagemap *);
992 static inline bool is_zone_device_page(const struct page *page)
998 static inline bool folio_is_zone_device(const struct folio *folio)
1000 return is_zone_device_page(&folio->page);
1003 static inline bool is_zone_movable_page(const struct page *page)
1005 return page_zonenum(page) == ZONE_MOVABLE;
1010 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1011 * intersection with the given zone
1013 static inline bool zone_intersects(struct zone *zone,
1014 unsigned long start_pfn, unsigned long nr_pages)
1016 if (zone_is_empty(zone))
1018 if (start_pfn >= zone_end_pfn(zone) ||
1019 start_pfn + nr_pages <= zone->zone_start_pfn)
1026 * The "priority" of VM scanning is how much of the queues we will scan in one
1027 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1028 * queues ("queue_length >> 12") during an aging round.
1030 #define DEF_PRIORITY 12
1032 /* Maximum number of zones on a zonelist */
1033 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1036 ZONELIST_FALLBACK, /* zonelist with fallback */
1039 * The NUMA zonelists are doubled because we need zonelists that
1040 * restrict the allocations to a single node for __GFP_THISNODE.
1042 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1048 * This struct contains information about a zone in a zonelist. It is stored
1049 * here to avoid dereferences into large structures and lookups of tables
1052 struct zone *zone; /* Pointer to actual zone */
1053 int zone_idx; /* zone_idx(zoneref->zone) */
1057 * One allocation request operates on a zonelist. A zonelist
1058 * is a list of zones, the first one is the 'goal' of the
1059 * allocation, the other zones are fallback zones, in decreasing
1062 * To speed the reading of the zonelist, the zonerefs contain the zone index
1063 * of the entry being read. Helper functions to access information given
1064 * a struct zoneref are
1066 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1067 * zonelist_zone_idx() - Return the index of the zone for an entry
1068 * zonelist_node_idx() - Return the index of the node for an entry
1071 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1075 * The array of struct pages for flatmem.
1076 * It must be declared for SPARSEMEM as well because there are configurations
1077 * that rely on that.
1079 extern struct page *mem_map;
1081 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1082 struct deferred_split {
1083 spinlock_t split_queue_lock;
1084 struct list_head split_queue;
1085 unsigned long split_queue_len;
1090 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1091 * it's memory layout. On UMA machines there is a single pglist_data which
1092 * describes the whole memory.
1094 * Memory statistics and page replacement data structures are maintained on a
1097 typedef struct pglist_data {
1099 * node_zones contains just the zones for THIS node. Not all of the
1100 * zones may be populated, but it is the full list. It is referenced by
1101 * this node's node_zonelists as well as other node's node_zonelists.
1103 struct zone node_zones[MAX_NR_ZONES];
1106 * node_zonelists contains references to all zones in all nodes.
1107 * Generally the first zones will be references to this node's
1110 struct zonelist node_zonelists[MAX_ZONELISTS];
1112 int nr_zones; /* number of populated zones in this node */
1113 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1114 struct page *node_mem_map;
1115 #ifdef CONFIG_PAGE_EXTENSION
1116 struct page_ext *node_page_ext;
1119 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1121 * Must be held any time you expect node_start_pfn,
1122 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1123 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1126 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1127 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1128 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1130 * Nests above zone->lock and zone->span_seqlock
1132 spinlock_t node_size_lock;
1134 unsigned long node_start_pfn;
1135 unsigned long node_present_pages; /* total number of physical pages */
1136 unsigned long node_spanned_pages; /* total size of physical page
1137 range, including holes */
1139 wait_queue_head_t kswapd_wait;
1140 wait_queue_head_t pfmemalloc_wait;
1142 /* workqueues for throttling reclaim for different reasons. */
1143 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1145 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1146 unsigned long nr_reclaim_start; /* nr pages written while throttled
1147 * when throttling started. */
1148 #ifdef CONFIG_MEMORY_HOTPLUG
1149 struct mutex kswapd_lock;
1151 struct task_struct *kswapd; /* Protected by kswapd_lock */
1153 enum zone_type kswapd_highest_zoneidx;
1155 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1157 #ifdef CONFIG_COMPACTION
1158 int kcompactd_max_order;
1159 enum zone_type kcompactd_highest_zoneidx;
1160 wait_queue_head_t kcompactd_wait;
1161 struct task_struct *kcompactd;
1162 bool proactive_compact_trigger;
1165 * This is a per-node reserve of pages that are not available
1166 * to userspace allocations.
1168 unsigned long totalreserve_pages;
1172 * node reclaim becomes active if more unmapped pages exist.
1174 unsigned long min_unmapped_pages;
1175 unsigned long min_slab_pages;
1176 #endif /* CONFIG_NUMA */
1178 /* Write-intensive fields used by page reclaim */
1179 CACHELINE_PADDING(_pad1_);
1181 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1183 * If memory initialisation on large machines is deferred then this
1184 * is the first PFN that needs to be initialised.
1186 unsigned long first_deferred_pfn;
1187 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1189 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1190 struct deferred_split deferred_split_queue;
1193 #ifdef CONFIG_NUMA_BALANCING
1194 /* start time in ms of current promote rate limit period */
1195 unsigned int nbp_rl_start;
1196 /* number of promote candidate pages at start time of current rate limit period */
1197 unsigned long nbp_rl_nr_cand;
1198 /* promote threshold in ms */
1199 unsigned int nbp_threshold;
1200 /* start time in ms of current promote threshold adjustment period */
1201 unsigned int nbp_th_start;
1203 * number of promote candidate pages at stat time of current promote
1204 * threshold adjustment period
1206 unsigned long nbp_th_nr_cand;
1208 /* Fields commonly accessed by the page reclaim scanner */
1211 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1213 * Use mem_cgroup_lruvec() to look up lruvecs.
1215 struct lruvec __lruvec;
1217 unsigned long flags;
1219 #ifdef CONFIG_LRU_GEN
1220 /* kswap mm walk data */
1221 struct lru_gen_mm_walk mm_walk;
1224 CACHELINE_PADDING(_pad2_);
1226 /* Per-node vmstats */
1227 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1228 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1230 struct memory_tier __rcu *memtier;
1234 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1235 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1237 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1238 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1240 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1242 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1245 #include <linux/memory_hotplug.h>
1247 void build_all_zonelists(pg_data_t *pgdat);
1248 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1249 enum zone_type highest_zoneidx);
1250 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1251 int highest_zoneidx, unsigned int alloc_flags,
1253 bool zone_watermark_ok(struct zone *z, unsigned int order,
1254 unsigned long mark, int highest_zoneidx,
1255 unsigned int alloc_flags);
1256 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1257 unsigned long mark, int highest_zoneidx);
1259 * Memory initialization context, use to differentiate memory added by
1260 * the platform statically or via memory hotplug interface.
1262 enum meminit_context {
1267 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1268 unsigned long size);
1270 extern void lruvec_init(struct lruvec *lruvec);
1272 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1275 return lruvec->pgdat;
1277 return container_of(lruvec, struct pglist_data, __lruvec);
1281 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1282 int local_memory_node(int node_id);
1284 static inline int local_memory_node(int node_id) { return node_id; };
1288 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1290 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1292 #ifdef CONFIG_ZONE_DEVICE
1293 static inline bool zone_is_zone_device(struct zone *zone)
1295 return zone_idx(zone) == ZONE_DEVICE;
1298 static inline bool zone_is_zone_device(struct zone *zone)
1305 * Returns true if a zone has pages managed by the buddy allocator.
1306 * All the reclaim decisions have to use this function rather than
1307 * populated_zone(). If the whole zone is reserved then we can easily
1308 * end up with populated_zone() && !managed_zone().
1310 static inline bool managed_zone(struct zone *zone)
1312 return zone_managed_pages(zone);
1315 /* Returns true if a zone has memory */
1316 static inline bool populated_zone(struct zone *zone)
1318 return zone->present_pages;
1322 static inline int zone_to_nid(struct zone *zone)
1327 static inline void zone_set_nid(struct zone *zone, int nid)
1332 static inline int zone_to_nid(struct zone *zone)
1337 static inline void zone_set_nid(struct zone *zone, int nid) {}
1340 extern int movable_zone;
1342 static inline int is_highmem_idx(enum zone_type idx)
1344 #ifdef CONFIG_HIGHMEM
1345 return (idx == ZONE_HIGHMEM ||
1346 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1353 * is_highmem - helper function to quickly check if a struct zone is a
1354 * highmem zone or not. This is an attempt to keep references
1355 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1356 * @zone: pointer to struct zone variable
1357 * Return: 1 for a highmem zone, 0 otherwise
1359 static inline int is_highmem(struct zone *zone)
1361 return is_highmem_idx(zone_idx(zone));
1364 #ifdef CONFIG_ZONE_DMA
1365 bool has_managed_dma(void);
1367 static inline bool has_managed_dma(void)
1373 /* These two functions are used to setup the per zone pages min values */
1376 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1378 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1379 size_t *, loff_t *);
1380 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1381 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1382 size_t *, loff_t *);
1383 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1384 void *, size_t *, loff_t *);
1385 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1386 void *, size_t *, loff_t *);
1387 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1388 void *, size_t *, loff_t *);
1389 int numa_zonelist_order_handler(struct ctl_table *, int,
1390 void *, size_t *, loff_t *);
1391 extern int percpu_pagelist_high_fraction;
1392 extern char numa_zonelist_order[];
1393 #define NUMA_ZONELIST_ORDER_LEN 16
1397 extern struct pglist_data contig_page_data;
1398 static inline struct pglist_data *NODE_DATA(int nid)
1400 return &contig_page_data;
1403 #else /* CONFIG_NUMA */
1405 #include <asm/mmzone.h>
1407 #endif /* !CONFIG_NUMA */
1409 extern struct pglist_data *first_online_pgdat(void);
1410 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1411 extern struct zone *next_zone(struct zone *zone);
1414 * for_each_online_pgdat - helper macro to iterate over all online nodes
1415 * @pgdat: pointer to a pg_data_t variable
1417 #define for_each_online_pgdat(pgdat) \
1418 for (pgdat = first_online_pgdat(); \
1420 pgdat = next_online_pgdat(pgdat))
1422 * for_each_zone - helper macro to iterate over all memory zones
1423 * @zone: pointer to struct zone variable
1425 * The user only needs to declare the zone variable, for_each_zone
1428 #define for_each_zone(zone) \
1429 for (zone = (first_online_pgdat())->node_zones; \
1431 zone = next_zone(zone))
1433 #define for_each_populated_zone(zone) \
1434 for (zone = (first_online_pgdat())->node_zones; \
1436 zone = next_zone(zone)) \
1437 if (!populated_zone(zone)) \
1438 ; /* do nothing */ \
1441 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1443 return zoneref->zone;
1446 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1448 return zoneref->zone_idx;
1451 static inline int zonelist_node_idx(struct zoneref *zoneref)
1453 return zone_to_nid(zoneref->zone);
1456 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1457 enum zone_type highest_zoneidx,
1461 * 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
1462 * @z: The cursor used as a starting point for the search
1463 * @highest_zoneidx: The zone index of the highest zone to return
1464 * @nodes: An optional nodemask to filter the zonelist with
1466 * This function returns the next zone at or below a given zone index that is
1467 * within the allowed nodemask using a cursor as the starting point for the
1468 * search. The zoneref returned is a cursor that represents the current zone
1469 * being examined. It should be advanced by one before calling
1470 * next_zones_zonelist again.
1472 * Return: the next zone at or below highest_zoneidx within the allowed
1473 * nodemask using a cursor within a zonelist as a starting point
1475 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1476 enum zone_type highest_zoneidx,
1479 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1481 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1485 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1486 * @zonelist: The zonelist to search for a suitable zone
1487 * @highest_zoneidx: The zone index of the highest zone to return
1488 * @nodes: An optional nodemask to filter the zonelist with
1490 * This function returns the first zone at or below a given zone index that is
1491 * within the allowed nodemask. The zoneref returned is a cursor that can be
1492 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1493 * one before calling.
1495 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1496 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1497 * update due to cpuset modification.
1499 * Return: Zoneref pointer for the first suitable zone found
1501 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1502 enum zone_type highest_zoneidx,
1505 return next_zones_zonelist(zonelist->_zonerefs,
1506 highest_zoneidx, nodes);
1510 * 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
1511 * @zone: The current zone in the iterator
1512 * @z: The current pointer within zonelist->_zonerefs being iterated
1513 * @zlist: The zonelist being iterated
1514 * @highidx: The zone index of the highest zone to return
1515 * @nodemask: Nodemask allowed by the allocator
1517 * This iterator iterates though all zones at or below a given zone index and
1518 * within a given nodemask
1520 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1521 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1523 z = next_zones_zonelist(++z, highidx, nodemask), \
1524 zone = zonelist_zone(z))
1526 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1527 for (zone = z->zone; \
1529 z = next_zones_zonelist(++z, highidx, nodemask), \
1530 zone = zonelist_zone(z))
1534 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1535 * @zone: The current zone in the iterator
1536 * @z: The current pointer within zonelist->zones being iterated
1537 * @zlist: The zonelist being iterated
1538 * @highidx: The zone index of the highest zone to return
1540 * This iterator iterates though all zones at or below a given zone index.
1542 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1543 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1545 /* Whether the 'nodes' are all movable nodes */
1546 static inline bool movable_only_nodes(nodemask_t *nodes)
1548 struct zonelist *zonelist;
1552 if (nodes_empty(*nodes))
1556 * We can chose arbitrary node from the nodemask to get a
1557 * zonelist as they are interlinked. We just need to find
1558 * at least one zone that can satisfy kernel allocations.
1560 nid = first_node(*nodes);
1561 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1562 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1563 return (!z->zone) ? true : false;
1567 #ifdef CONFIG_SPARSEMEM
1568 #include <asm/sparsemem.h>
1571 #ifdef CONFIG_FLATMEM
1572 #define pfn_to_nid(pfn) (0)
1575 #ifdef CONFIG_SPARSEMEM
1578 * PA_SECTION_SHIFT physical address to/from section number
1579 * PFN_SECTION_SHIFT pfn to/from section number
1581 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1582 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1584 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1586 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1587 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1589 #define SECTION_BLOCKFLAGS_BITS \
1590 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1592 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1593 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1596 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1598 return pfn >> PFN_SECTION_SHIFT;
1600 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1602 return sec << PFN_SECTION_SHIFT;
1605 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1606 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1608 #define SUBSECTION_SHIFT 21
1609 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1611 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1612 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1613 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1615 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1616 #error Subsection size exceeds section size
1618 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1621 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1622 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1624 struct mem_section_usage {
1625 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1626 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1628 /* See declaration of similar field in struct zone */
1629 unsigned long pageblock_flags[0];
1632 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1636 struct mem_section {
1638 * This is, logically, a pointer to an array of struct
1639 * pages. However, it is stored with some other magic.
1640 * (see sparse.c::sparse_init_one_section())
1642 * Additionally during early boot we encode node id of
1643 * the location of the section here to guide allocation.
1644 * (see sparse.c::memory_present())
1646 * Making it a UL at least makes someone do a cast
1647 * before using it wrong.
1649 unsigned long section_mem_map;
1651 struct mem_section_usage *usage;
1652 #ifdef CONFIG_PAGE_EXTENSION
1654 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1655 * section. (see page_ext.h about this.)
1657 struct page_ext *page_ext;
1661 * WARNING: mem_section must be a power-of-2 in size for the
1662 * calculation and use of SECTION_ROOT_MASK to make sense.
1666 #ifdef CONFIG_SPARSEMEM_EXTREME
1667 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1669 #define SECTIONS_PER_ROOT 1
1672 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1673 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1674 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1676 #ifdef CONFIG_SPARSEMEM_EXTREME
1677 extern struct mem_section **mem_section;
1679 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1682 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1684 return ms->usage->pageblock_flags;
1687 static inline struct mem_section *__nr_to_section(unsigned long nr)
1689 unsigned long root = SECTION_NR_TO_ROOT(nr);
1691 if (unlikely(root >= NR_SECTION_ROOTS))
1694 #ifdef CONFIG_SPARSEMEM_EXTREME
1695 if (!mem_section || !mem_section[root])
1698 return &mem_section[root][nr & SECTION_ROOT_MASK];
1700 extern size_t mem_section_usage_size(void);
1703 * We use the lower bits of the mem_map pointer to store
1704 * a little bit of information. The pointer is calculated
1705 * as mem_map - section_nr_to_pfn(pnum). The result is
1706 * aligned to the minimum alignment of the two values:
1707 * 1. All mem_map arrays are page-aligned.
1708 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1709 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1710 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1711 * worst combination is powerpc with 256k pages,
1712 * which results in PFN_SECTION_SHIFT equal 6.
1713 * To sum it up, at least 6 bits are available on all architectures.
1714 * However, we can exceed 6 bits on some other architectures except
1715 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1716 * with the worst case of 64K pages on arm64) if we make sure the
1717 * exceeded bit is not applicable to powerpc.
1720 SECTION_MARKED_PRESENT_BIT,
1721 SECTION_HAS_MEM_MAP_BIT,
1722 SECTION_IS_ONLINE_BIT,
1723 SECTION_IS_EARLY_BIT,
1724 #ifdef CONFIG_ZONE_DEVICE
1725 SECTION_TAINT_ZONE_DEVICE_BIT,
1727 SECTION_MAP_LAST_BIT,
1730 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1731 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1732 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1733 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1734 #ifdef CONFIG_ZONE_DEVICE
1735 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1737 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1738 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1740 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1742 unsigned long map = section->section_mem_map;
1743 map &= SECTION_MAP_MASK;
1744 return (struct page *)map;
1747 static inline int present_section(struct mem_section *section)
1749 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1752 static inline int present_section_nr(unsigned long nr)
1754 return present_section(__nr_to_section(nr));
1757 static inline int valid_section(struct mem_section *section)
1759 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1762 static inline int early_section(struct mem_section *section)
1764 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1767 static inline int valid_section_nr(unsigned long nr)
1769 return valid_section(__nr_to_section(nr));
1772 static inline int online_section(struct mem_section *section)
1774 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1777 #ifdef CONFIG_ZONE_DEVICE
1778 static inline int online_device_section(struct mem_section *section)
1780 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1782 return section && ((section->section_mem_map & flags) == flags);
1785 static inline int online_device_section(struct mem_section *section)
1791 static inline int online_section_nr(unsigned long nr)
1793 return online_section(__nr_to_section(nr));
1796 #ifdef CONFIG_MEMORY_HOTPLUG
1797 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1798 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1801 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1803 return __nr_to_section(pfn_to_section_nr(pfn));
1806 extern unsigned long __highest_present_section_nr;
1808 static inline int subsection_map_index(unsigned long pfn)
1810 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1813 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1814 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1816 int idx = subsection_map_index(pfn);
1818 return test_bit(idx, ms->usage->subsection_map);
1821 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1827 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1829 * pfn_valid - check if there is a valid memory map entry for a PFN
1830 * @pfn: the page frame number to check
1832 * Check if there is a valid memory map entry aka struct page for the @pfn.
1833 * Note, that availability of the memory map entry does not imply that
1834 * there is actual usable memory at that @pfn. The struct page may
1835 * represent a hole or an unusable page frame.
1837 * Return: 1 for PFNs that have memory map entries and 0 otherwise
1839 static inline int pfn_valid(unsigned long pfn)
1841 struct mem_section *ms;
1844 * Ensure the upper PAGE_SHIFT bits are clear in the
1845 * pfn. Else it might lead to false positives when
1846 * some of the upper bits are set, but the lower bits
1847 * match a valid pfn.
1849 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
1852 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1854 ms = __pfn_to_section(pfn);
1855 if (!valid_section(ms))
1858 * Traditionally early sections always returned pfn_valid() for
1859 * the entire section-sized span.
1861 return early_section(ms) || pfn_section_valid(ms, pfn);
1865 static inline int pfn_in_present_section(unsigned long pfn)
1867 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1869 return present_section(__pfn_to_section(pfn));
1872 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1874 while (++section_nr <= __highest_present_section_nr) {
1875 if (present_section_nr(section_nr))
1883 * These are _only_ used during initialisation, therefore they
1884 * can use __initdata ... They could have names to indicate
1888 #define pfn_to_nid(pfn) \
1890 unsigned long __pfn_to_nid_pfn = (pfn); \
1891 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1894 #define pfn_to_nid(pfn) (0)
1897 void sparse_init(void);
1899 #define sparse_init() do {} while (0)
1900 #define sparse_index_init(_sec, _nid) do {} while (0)
1901 #define pfn_in_present_section pfn_valid
1902 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1903 #endif /* CONFIG_SPARSEMEM */
1905 #endif /* !__GENERATING_BOUNDS.H */
1906 #endif /* !__ASSEMBLY__ */
1907 #endif /* _LINUX_MMZONE_H */