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/list_nulls.h>
11 #include <linux/wait.h>
12 #include <linux/bitops.h>
13 #include <linux/cache.h>
14 #include <linux/threads.h>
15 #include <linux/numa.h>
16 #include <linux/init.h>
17 #include <linux/seqlock.h>
18 #include <linux/nodemask.h>
19 #include <linux/pageblock-flags.h>
20 #include <linux/page-flags-layout.h>
21 #include <linux/atomic.h>
22 #include <linux/mm_types.h>
23 #include <linux/page-flags.h>
24 #include <linux/local_lock.h>
25 #include <linux/zswap.h>
28 /* Free memory management - zoned buddy allocator. */
29 #ifndef CONFIG_ARCH_FORCE_MAX_ORDER
32 #define MAX_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
34 #define MAX_ORDER_NR_PAGES (1 << MAX_ORDER)
36 #define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
39 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
40 * costly to service. That is between allocation orders which should
41 * coalesce naturally under reasonable reclaim pressure and those which
44 #define PAGE_ALLOC_COSTLY_ORDER 3
50 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
51 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
54 * MIGRATE_CMA migration type is designed to mimic the way
55 * ZONE_MOVABLE works. Only movable pages can be allocated
56 * from MIGRATE_CMA pageblocks and page allocator never
57 * implicitly change migration type of MIGRATE_CMA pageblock.
59 * The way to use it is to change migratetype of a range of
60 * pageblocks to MIGRATE_CMA which can be done by
61 * __free_pageblock_cma() function.
65 #ifdef CONFIG_MEMORY_ISOLATION
66 MIGRATE_ISOLATE, /* can't allocate from here */
71 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
72 extern const char * const migratetype_names[MIGRATE_TYPES];
75 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
76 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
78 # define is_migrate_cma(migratetype) false
79 # define is_migrate_cma_page(_page) false
82 static inline bool is_migrate_movable(int mt)
84 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
88 * Check whether a migratetype can be merged with another migratetype.
90 * It is only mergeable when it can fall back to other migratetypes for
91 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
93 static inline bool migratetype_is_mergeable(int mt)
95 return mt < MIGRATE_PCPTYPES;
98 #define for_each_migratetype_order(order, type) \
99 for (order = 0; order <= MAX_ORDER; order++) \
100 for (type = 0; type < MIGRATE_TYPES; type++)
102 extern int page_group_by_mobility_disabled;
104 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
106 #define get_pageblock_migratetype(page) \
107 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
109 #define folio_migratetype(folio) \
110 get_pfnblock_flags_mask(&folio->page, folio_pfn(folio), \
113 struct list_head free_list[MIGRATE_TYPES];
114 unsigned long nr_free;
120 enum numa_stat_item {
121 NUMA_HIT, /* allocated in intended node */
122 NUMA_MISS, /* allocated in non intended node */
123 NUMA_FOREIGN, /* was intended here, hit elsewhere */
124 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
125 NUMA_LOCAL, /* allocation from local node */
126 NUMA_OTHER, /* allocation from other node */
127 NR_VM_NUMA_EVENT_ITEMS
130 #define NR_VM_NUMA_EVENT_ITEMS 0
133 enum zone_stat_item {
134 /* First 128 byte cacheline (assuming 64 bit words) */
136 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
137 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
139 NR_ZONE_INACTIVE_FILE,
142 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
143 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
144 /* Second 128 byte cacheline */
146 #if IS_ENABLED(CONFIG_ZSMALLOC)
147 NR_ZSPAGES, /* allocated in zsmalloc */
150 #ifdef CONFIG_UNACCEPTED_MEMORY
153 NR_VM_ZONE_STAT_ITEMS };
155 enum node_stat_item {
157 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
158 NR_ACTIVE_ANON, /* " " " " " */
159 NR_INACTIVE_FILE, /* " " " " " */
160 NR_ACTIVE_FILE, /* " " " " " */
161 NR_UNEVICTABLE, /* " " " " " */
162 NR_SLAB_RECLAIMABLE_B,
163 NR_SLAB_UNRECLAIMABLE_B,
164 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
165 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
167 WORKINGSET_REFAULT_BASE,
168 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
169 WORKINGSET_REFAULT_FILE,
170 WORKINGSET_ACTIVATE_BASE,
171 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
172 WORKINGSET_ACTIVATE_FILE,
173 WORKINGSET_RESTORE_BASE,
174 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
175 WORKINGSET_RESTORE_FILE,
176 WORKINGSET_NODERECLAIM,
177 NR_ANON_MAPPED, /* Mapped anonymous pages */
178 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
179 only modified from process context */
183 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
184 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
191 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
192 NR_DIRTIED, /* page dirtyings since bootup */
193 NR_WRITTEN, /* page writings since bootup */
194 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
195 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
196 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
197 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
198 NR_KERNEL_STACK_KB, /* measured in KiB */
199 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
200 NR_KERNEL_SCS_KB, /* measured in KiB */
202 NR_PAGETABLE, /* used for pagetables */
203 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
207 #ifdef CONFIG_NUMA_BALANCING
208 PGPROMOTE_SUCCESS, /* promote successfully */
209 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
211 /* PGDEMOTE_*: pages demoted */
215 NR_VM_NODE_STAT_ITEMS
219 * Returns true if the item should be printed in THPs (/proc/vmstat
220 * currently prints number of anon, file and shmem THPs. But the item
221 * is charged in pages).
223 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
225 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
228 return item == NR_ANON_THPS ||
229 item == NR_FILE_THPS ||
230 item == NR_SHMEM_THPS ||
231 item == NR_SHMEM_PMDMAPPED ||
232 item == NR_FILE_PMDMAPPED;
236 * Returns true if the value is measured in bytes (most vmstat values are
237 * measured in pages). This defines the API part, the internal representation
238 * might be different.
240 static __always_inline bool vmstat_item_in_bytes(int idx)
243 * Global and per-node slab counters track slab pages.
244 * It's expected that changes are multiples of PAGE_SIZE.
245 * Internally values are stored in pages.
247 * Per-memcg and per-lruvec counters track memory, consumed
248 * by individual slab objects. These counters are actually
251 return (idx == NR_SLAB_RECLAIMABLE_B ||
252 idx == NR_SLAB_UNRECLAIMABLE_B);
256 * We do arithmetic on the LRU lists in various places in the code,
257 * so it is important to keep the active lists LRU_ACTIVE higher in
258 * the array than the corresponding inactive lists, and to keep
259 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
261 * This has to be kept in sync with the statistics in zone_stat_item
262 * above and the descriptions in vmstat_text in mm/vmstat.c
269 LRU_INACTIVE_ANON = LRU_BASE,
270 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
271 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
272 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
277 enum vmscan_throttle_state {
278 VMSCAN_THROTTLE_WRITEBACK,
279 VMSCAN_THROTTLE_ISOLATED,
280 VMSCAN_THROTTLE_NOPROGRESS,
281 VMSCAN_THROTTLE_CONGESTED,
285 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
287 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
289 static inline bool is_file_lru(enum lru_list lru)
291 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
294 static inline bool is_active_lru(enum lru_list lru)
296 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
299 #define WORKINGSET_ANON 0
300 #define WORKINGSET_FILE 1
301 #define ANON_AND_FILE 2
305 * An lruvec has many dirty pages backed by a congested BDI:
306 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
307 * It can be cleared by cgroup reclaim or kswapd.
308 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
309 * It can only be cleared by kswapd.
311 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
312 * reclaim, but not vice versa. This only applies to the root cgroup.
313 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
314 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
317 LRUVEC_CGROUP_CONGESTED,
318 LRUVEC_NODE_CONGESTED,
321 #endif /* !__GENERATING_BOUNDS_H */
324 * Evictable pages are divided into multiple generations. The youngest and the
325 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
326 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
327 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
328 * corresponding generation. The gen counter in folio->flags stores gen+1 while
329 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
331 * A page is added to the youngest generation on faulting. The aging needs to
332 * check the accessed bit at least twice before handing this page over to the
333 * eviction. The first check takes care of the accessed bit set on the initial
334 * fault; the second check makes sure this page hasn't been used since then.
335 * This process, AKA second chance, requires a minimum of two generations,
336 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
337 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
338 * rest of generations, if they exist, are considered inactive. See
339 * lru_gen_is_active().
341 * PG_active is always cleared while a page is on one of lrugen->folios[] so
342 * that the aging needs not to worry about it. And it's set again when a page
343 * considered active is isolated for non-reclaiming purposes, e.g., migration.
344 * See lru_gen_add_folio() and lru_gen_del_folio().
346 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
347 * number of categories of the active/inactive LRU when keeping track of
348 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
351 #define MIN_NR_GENS 2U
352 #define MAX_NR_GENS 4U
355 * Each generation is divided into multiple tiers. A page accessed N times
356 * through file descriptors is in tier order_base_2(N). A page in the first tier
357 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
358 * tables or read ahead. A page in any other tier (N>1) is marked by
359 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
360 * supported without using additional bits in folio->flags.
362 * In contrast to moving across generations which requires the LRU lock, moving
363 * across tiers only involves atomic operations on folio->flags and therefore
364 * has a negligible cost in the buffered access path. In the eviction path,
365 * comparisons of refaulted/(evicted+protected) from the first tier and the
366 * rest infer whether pages accessed multiple times through file descriptors
367 * are statistically hot and thus worth protecting.
369 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
370 * number of categories of the active/inactive LRU when keeping track of
371 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
374 #define MAX_NR_TIERS 4U
376 #ifndef __GENERATING_BOUNDS_H
379 struct page_vma_mapped_walk;
381 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
382 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
384 #ifdef CONFIG_LRU_GEN
394 LRU_GEN_NONLEAF_YOUNG,
398 #define MIN_LRU_BATCH BITS_PER_LONG
399 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
401 /* whether to keep historical stats from evicted generations */
402 #ifdef CONFIG_LRU_GEN_STATS
403 #define NR_HIST_GENS MAX_NR_GENS
405 #define NR_HIST_GENS 1U
409 * The youngest generation number is stored in max_seq for both anon and file
410 * types as they are aged on an equal footing. The oldest generation numbers are
411 * stored in min_seq[] separately for anon and file types as clean file pages
412 * can be evicted regardless of swap constraints.
414 * Normally anon and file min_seq are in sync. But if swapping is constrained,
415 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
418 * The number of pages in each generation is eventually consistent and therefore
419 * can be transiently negative when reset_batch_size() is pending.
421 struct lru_gen_folio {
422 /* the aging increments the youngest generation number */
423 unsigned long max_seq;
424 /* the eviction increments the oldest generation numbers */
425 unsigned long min_seq[ANON_AND_FILE];
426 /* the birth time of each generation in jiffies */
427 unsigned long timestamps[MAX_NR_GENS];
428 /* the multi-gen LRU lists, lazily sorted on eviction */
429 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
430 /* the multi-gen LRU sizes, eventually consistent */
431 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
432 /* the exponential moving average of refaulted */
433 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
434 /* the exponential moving average of evicted+protected */
435 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
436 /* the first tier doesn't need protection, hence the minus one */
437 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
438 /* can be modified without holding the LRU lock */
439 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
440 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
441 /* whether the multi-gen LRU is enabled */
443 /* the memcg generation this lru_gen_folio belongs to */
445 /* the list segment this lru_gen_folio belongs to */
447 /* per-node lru_gen_folio list for global reclaim */
448 struct hlist_nulls_node list;
452 MM_LEAF_TOTAL, /* total leaf entries */
453 MM_LEAF_OLD, /* old leaf entries */
454 MM_LEAF_YOUNG, /* young leaf entries */
455 MM_NONLEAF_TOTAL, /* total non-leaf entries */
456 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
457 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
461 /* double-buffering Bloom filters */
462 #define NR_BLOOM_FILTERS 2
464 struct lru_gen_mm_state {
465 /* set to max_seq after each iteration */
467 /* where the current iteration continues after */
468 struct list_head *head;
469 /* where the last iteration ended before */
470 struct list_head *tail;
471 /* Bloom filters flip after each iteration */
472 unsigned long *filters[NR_BLOOM_FILTERS];
473 /* the mm stats for debugging */
474 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
477 struct lru_gen_mm_walk {
478 /* the lruvec under reclaim */
479 struct lruvec *lruvec;
480 /* unstable max_seq from lru_gen_folio */
481 unsigned long max_seq;
482 /* the next address within an mm to scan */
483 unsigned long next_addr;
484 /* to batch promoted pages */
485 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
486 /* to batch the mm stats */
487 int mm_stats[NR_MM_STATS];
488 /* total batched items */
495 * For each node, memcgs are divided into two generations: the old and the
496 * young. For each generation, memcgs are randomly sharded into multiple bins
497 * to improve scalability. For each bin, the hlist_nulls is virtually divided
498 * into three segments: the head, the tail and the default.
500 * An onlining memcg is added to the tail of a random bin in the old generation.
501 * The eviction starts at the head of a random bin in the old generation. The
502 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
503 * the old generation, is incremented when all its bins become empty.
505 * There are four operations:
506 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
507 * current generation (old or young) and updates its "seg" to "head";
508 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
509 * current generation (old or young) and updates its "seg" to "tail";
510 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
511 * generation, updates its "gen" to "old" and resets its "seg" to "default";
512 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
513 * young generation, updates its "gen" to "young" and resets its "seg" to
516 * The events that trigger the above operations are:
517 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
518 * 2. The first attempt to reclaim a memcg below low, which triggers
520 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
521 * threshold, which triggers MEMCG_LRU_TAIL;
522 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
523 * threshold, which triggers MEMCG_LRU_YOUNG;
524 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
525 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
526 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
529 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
530 * of their max_seq counters ensures the eventual fairness to all eligible
531 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
532 * 2. There are only two valid generations: old (seq) and young (seq+1).
533 * MEMCG_NR_GENS is set to three so that when reading the generation counter
534 * locklessly, a stale value (seq-1) does not wraparound to young.
536 #define MEMCG_NR_GENS 3
537 #define MEMCG_NR_BINS 8
539 struct lru_gen_memcg {
540 /* the per-node memcg generation counter */
542 /* each memcg has one lru_gen_folio per node */
543 unsigned long nr_memcgs[MEMCG_NR_GENS];
544 /* per-node lru_gen_folio list for global reclaim */
545 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
546 /* protects the above */
550 void lru_gen_init_pgdat(struct pglist_data *pgdat);
551 void lru_gen_init_lruvec(struct lruvec *lruvec);
552 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
554 void lru_gen_init_memcg(struct mem_cgroup *memcg);
555 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
556 void lru_gen_online_memcg(struct mem_cgroup *memcg);
557 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
558 void lru_gen_release_memcg(struct mem_cgroup *memcg);
559 void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
561 #else /* !CONFIG_LRU_GEN */
563 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
567 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
571 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
575 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
579 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
583 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
587 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
591 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
595 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
599 #endif /* CONFIG_LRU_GEN */
602 struct list_head lists[NR_LRU_LISTS];
603 /* per lruvec lru_lock for memcg */
606 * These track the cost of reclaiming one LRU - file or anon -
607 * over the other. As the observed cost of reclaiming one LRU
608 * increases, the reclaim scan balance tips toward the other.
610 unsigned long anon_cost;
611 unsigned long file_cost;
612 /* Non-resident age, driven by LRU movement */
613 atomic_long_t nonresident_age;
614 /* Refaults at the time of last reclaim cycle */
615 unsigned long refaults[ANON_AND_FILE];
616 /* Various lruvec state flags (enum lruvec_flags) */
618 #ifdef CONFIG_LRU_GEN
619 /* evictable pages divided into generations */
620 struct lru_gen_folio lrugen;
621 #ifdef CONFIG_LRU_GEN_WALKS_MMU
622 /* to concurrently iterate lru_gen_mm_list */
623 struct lru_gen_mm_state mm_state;
625 #endif /* CONFIG_LRU_GEN */
627 struct pglist_data *pgdat;
629 struct zswap_lruvec_state zswap_lruvec_state;
632 /* Isolate for asynchronous migration */
633 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
634 /* Isolate unevictable pages */
635 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
637 /* LRU Isolation modes. */
638 typedef unsigned __bitwise isolate_mode_t;
640 enum zone_watermarks {
649 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
650 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
651 * it should not contribute to serious fragmentation causing THP allocation
654 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
659 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
660 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
662 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
663 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
664 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
665 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
668 * Flags used in pcp->flags field.
670 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
671 * previous page freeing. To avoid to drain PCP for an accident
672 * high-order page freeing.
674 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
675 * draining PCP for consecutive high-order pages freeing without
676 * allocation if data cache slice of CPU is large enough. To reduce
677 * zone lock contention and keep cache-hot pages reusing.
679 #define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
680 #define PCPF_FREE_HIGH_BATCH BIT(1)
682 struct per_cpu_pages {
683 spinlock_t lock; /* Protects lists field */
684 int count; /* number of pages in the list */
685 int high; /* high watermark, emptying needed */
686 int high_min; /* min high watermark */
687 int high_max; /* max high watermark */
688 int batch; /* chunk size for buddy add/remove */
689 u8 flags; /* protected by pcp->lock */
690 u8 alloc_factor; /* batch scaling factor during allocate */
692 u8 expire; /* When 0, remote pagesets are drained */
694 short free_count; /* consecutive free count */
696 /* Lists of pages, one per migrate type stored on the pcp-lists */
697 struct list_head lists[NR_PCP_LISTS];
698 } ____cacheline_aligned_in_smp;
700 struct per_cpu_zonestat {
702 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
707 * Low priority inaccurate counters that are only folded
708 * on demand. Use a large type to avoid the overhead of
709 * folding during refresh_cpu_vm_stats.
711 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
715 struct per_cpu_nodestat {
717 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
720 #endif /* !__GENERATING_BOUNDS.H */
724 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
725 * to DMA to all of the addressable memory (ZONE_NORMAL).
726 * On architectures where this area covers the whole 32 bit address
727 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
728 * DMA addressing constraints. This distinction is important as a 32bit
729 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
730 * platforms may need both zones as they support peripherals with
731 * different DMA addressing limitations.
733 #ifdef CONFIG_ZONE_DMA
736 #ifdef CONFIG_ZONE_DMA32
740 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
741 * performed on pages in ZONE_NORMAL if the DMA devices support
742 * transfers to all addressable memory.
745 #ifdef CONFIG_HIGHMEM
747 * A memory area that is only addressable by the kernel through
748 * mapping portions into its own address space. This is for example
749 * used by i386 to allow the kernel to address the memory beyond
750 * 900MB. The kernel will set up special mappings (page
751 * table entries on i386) for each page that the kernel needs to
757 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
758 * movable pages with few exceptional cases described below. Main use
759 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
760 * likely to succeed, and to locally limit unmovable allocations - e.g.,
761 * to increase the number of THP/huge pages. Notable special cases are:
763 * 1. Pinned pages: (long-term) pinning of movable pages might
764 * essentially turn such pages unmovable. Therefore, we do not allow
765 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
766 * faulted, they come from the right zone right away. However, it is
767 * still possible that address space already has pages in
768 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
769 * touches that memory before pinning). In such case we migrate them
770 * to a different zone. When migration fails - pinning fails.
771 * 2. memblock allocations: kernelcore/movablecore setups might create
772 * situations where ZONE_MOVABLE contains unmovable allocations
773 * after boot. Memory offlining and allocations fail early.
774 * 3. Memory holes: kernelcore/movablecore setups might create very rare
775 * situations where ZONE_MOVABLE contains memory holes after boot,
776 * for example, if we have sections that are only partially
777 * populated. Memory offlining and allocations fail early.
778 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
779 * memory offlining, such pages cannot be allocated.
780 * 5. Unmovable PG_offline pages: in paravirtualized environments,
781 * hotplugged memory blocks might only partially be managed by the
782 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
783 * parts not manged by the buddy are unmovable PG_offline pages. In
784 * some cases (virtio-mem), such pages can be skipped during
785 * memory offlining, however, cannot be moved/allocated. These
786 * techniques might use alloc_contig_range() to hide previously
787 * exposed pages from the buddy again (e.g., to implement some sort
788 * of memory unplug in virtio-mem).
789 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
790 * situations where ZERO_PAGE(0) which is allocated differently
791 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
792 * cannot be migrated.
793 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
794 * memory to the MOVABLE zone, the vmemmap pages are also placed in
795 * such zone. Such pages cannot be really moved around as they are
796 * self-stored in the range, but they are treated as movable when
797 * the range they describe is about to be offlined.
799 * In general, no unmovable allocations that degrade memory offlining
800 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
801 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
802 * if has_unmovable_pages() states that there are no unmovable pages,
803 * there can be false negatives).
806 #ifdef CONFIG_ZONE_DEVICE
813 #ifndef __GENERATING_BOUNDS_H
815 #define ASYNC_AND_SYNC 2
818 /* Read-mostly fields */
820 /* zone watermarks, access with *_wmark_pages(zone) macros */
821 unsigned long _watermark[NR_WMARK];
822 unsigned long watermark_boost;
824 unsigned long nr_reserved_highatomic;
827 * We don't know if the memory that we're going to allocate will be
828 * freeable or/and it will be released eventually, so to avoid totally
829 * wasting several GB of ram we must reserve some of the lower zone
830 * memory (otherwise we risk to run OOM on the lower zones despite
831 * there being tons of freeable ram on the higher zones). This array is
832 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
835 long lowmem_reserve[MAX_NR_ZONES];
840 struct pglist_data *zone_pgdat;
841 struct per_cpu_pages __percpu *per_cpu_pageset;
842 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
844 * the high and batch values are copied to individual pagesets for
847 int pageset_high_min;
848 int pageset_high_max;
851 #ifndef CONFIG_SPARSEMEM
853 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
854 * In SPARSEMEM, this map is stored in struct mem_section
856 unsigned long *pageblock_flags;
857 #endif /* CONFIG_SPARSEMEM */
859 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
860 unsigned long zone_start_pfn;
863 * spanned_pages is the total pages spanned by the zone, including
864 * holes, which is calculated as:
865 * spanned_pages = zone_end_pfn - zone_start_pfn;
867 * present_pages is physical pages existing within the zone, which
869 * present_pages = spanned_pages - absent_pages(pages in holes);
871 * present_early_pages is present pages existing within the zone
872 * located on memory available since early boot, excluding hotplugged
875 * managed_pages is present pages managed by the buddy system, which
876 * is calculated as (reserved_pages includes pages allocated by the
877 * bootmem allocator):
878 * managed_pages = present_pages - reserved_pages;
880 * cma pages is present pages that are assigned for CMA use
883 * So present_pages may be used by memory hotplug or memory power
884 * management logic to figure out unmanaged pages by checking
885 * (present_pages - managed_pages). And managed_pages should be used
886 * by page allocator and vm scanner to calculate all kinds of watermarks
891 * zone_start_pfn and spanned_pages are protected by span_seqlock.
892 * It is a seqlock because it has to be read outside of zone->lock,
893 * and it is done in the main allocator path. But, it is written
894 * quite infrequently.
896 * The span_seq lock is declared along with zone->lock because it is
897 * frequently read in proximity to zone->lock. It's good to
898 * give them a chance of being in the same cacheline.
900 * Write access to present_pages at runtime should be protected by
901 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
902 * present_pages should use get_online_mems() to get a stable value.
904 atomic_long_t managed_pages;
905 unsigned long spanned_pages;
906 unsigned long present_pages;
907 #if defined(CONFIG_MEMORY_HOTPLUG)
908 unsigned long present_early_pages;
911 unsigned long cma_pages;
916 #ifdef CONFIG_MEMORY_ISOLATION
918 * Number of isolated pageblock. It is used to solve incorrect
919 * freepage counting problem due to racy retrieving migratetype
920 * of pageblock. Protected by zone->lock.
922 unsigned long nr_isolate_pageblock;
925 #ifdef CONFIG_MEMORY_HOTPLUG
926 /* see spanned/present_pages for more description */
927 seqlock_t span_seqlock;
932 /* Write-intensive fields used from the page allocator */
933 CACHELINE_PADDING(_pad1_);
935 /* free areas of different sizes */
936 struct free_area free_area[MAX_ORDER + 1];
938 #ifdef CONFIG_UNACCEPTED_MEMORY
939 /* Pages to be accepted. All pages on the list are MAX_ORDER */
940 struct list_head unaccepted_pages;
943 /* zone flags, see below */
946 /* Primarily protects free_area */
949 /* Write-intensive fields used by compaction and vmstats. */
950 CACHELINE_PADDING(_pad2_);
953 * When free pages are below this point, additional steps are taken
954 * when reading the number of free pages to avoid per-cpu counter
955 * drift allowing watermarks to be breached
957 unsigned long percpu_drift_mark;
959 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
960 /* pfn where compaction free scanner should start */
961 unsigned long compact_cached_free_pfn;
962 /* pfn where compaction migration scanner should start */
963 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
964 unsigned long compact_init_migrate_pfn;
965 unsigned long compact_init_free_pfn;
968 #ifdef CONFIG_COMPACTION
970 * On compaction failure, 1<<compact_defer_shift compactions
971 * are skipped before trying again. The number attempted since
972 * last failure is tracked with compact_considered.
973 * compact_order_failed is the minimum compaction failed order.
975 unsigned int compact_considered;
976 unsigned int compact_defer_shift;
977 int compact_order_failed;
980 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
981 /* Set to true when the PG_migrate_skip bits should be cleared */
982 bool compact_blockskip_flush;
987 CACHELINE_PADDING(_pad3_);
988 /* Zone statistics */
989 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
990 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
991 } ____cacheline_internodealigned_in_smp;
994 PGDAT_DIRTY, /* reclaim scanning has recently found
995 * many dirty file pages at the tail
998 PGDAT_WRITEBACK, /* reclaim scanning has recently found
999 * many pages under writeback
1001 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1005 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1006 * Cleared when kswapd is woken.
1008 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1009 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1012 static inline unsigned long zone_managed_pages(struct zone *zone)
1014 return (unsigned long)atomic_long_read(&zone->managed_pages);
1017 static inline unsigned long zone_cma_pages(struct zone *zone)
1020 return zone->cma_pages;
1026 static inline unsigned long zone_end_pfn(const struct zone *zone)
1028 return zone->zone_start_pfn + zone->spanned_pages;
1031 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1033 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1036 static inline bool zone_is_initialized(struct zone *zone)
1038 return zone->initialized;
1041 static inline bool zone_is_empty(struct zone *zone)
1043 return zone->spanned_pages == 0;
1046 #ifndef BUILD_VDSO32_64
1048 * The zone field is never updated after free_area_init_core()
1049 * sets it, so none of the operations on it need to be atomic.
1052 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1053 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1054 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1055 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1056 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1057 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1058 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1059 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1062 * Define the bit shifts to access each section. For non-existent
1063 * sections we define the shift as 0; that plus a 0 mask ensures
1064 * the compiler will optimise away reference to them.
1066 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1067 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1068 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1069 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1070 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1072 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1073 #ifdef NODE_NOT_IN_PAGE_FLAGS
1074 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1075 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1076 SECTIONS_PGOFF : ZONES_PGOFF)
1078 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1079 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1080 NODES_PGOFF : ZONES_PGOFF)
1083 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1085 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1086 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1087 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1088 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1089 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1090 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1092 static inline enum zone_type page_zonenum(const struct page *page)
1094 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1095 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1098 static inline enum zone_type folio_zonenum(const struct folio *folio)
1100 return page_zonenum(&folio->page);
1103 #ifdef CONFIG_ZONE_DEVICE
1104 static inline bool is_zone_device_page(const struct page *page)
1106 return page_zonenum(page) == ZONE_DEVICE;
1110 * Consecutive zone device pages should not be merged into the same sgl
1111 * or bvec segment with other types of pages or if they belong to different
1112 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1113 * without scanning the entire segment. This helper returns true either if
1114 * both pages are not zone device pages or both pages are zone device pages
1115 * with the same pgmap.
1117 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1118 const struct page *b)
1120 if (is_zone_device_page(a) != is_zone_device_page(b))
1122 if (!is_zone_device_page(a))
1124 return a->pgmap == b->pgmap;
1127 extern void memmap_init_zone_device(struct zone *, unsigned long,
1128 unsigned long, struct dev_pagemap *);
1130 static inline bool is_zone_device_page(const struct page *page)
1134 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1135 const struct page *b)
1141 static inline bool folio_is_zone_device(const struct folio *folio)
1143 return is_zone_device_page(&folio->page);
1146 static inline bool is_zone_movable_page(const struct page *page)
1148 return page_zonenum(page) == ZONE_MOVABLE;
1151 static inline bool folio_is_zone_movable(const struct folio *folio)
1153 return folio_zonenum(folio) == ZONE_MOVABLE;
1158 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1159 * intersection with the given zone
1161 static inline bool zone_intersects(struct zone *zone,
1162 unsigned long start_pfn, unsigned long nr_pages)
1164 if (zone_is_empty(zone))
1166 if (start_pfn >= zone_end_pfn(zone) ||
1167 start_pfn + nr_pages <= zone->zone_start_pfn)
1174 * The "priority" of VM scanning is how much of the queues we will scan in one
1175 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1176 * queues ("queue_length >> 12") during an aging round.
1178 #define DEF_PRIORITY 12
1180 /* Maximum number of zones on a zonelist */
1181 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1184 ZONELIST_FALLBACK, /* zonelist with fallback */
1187 * The NUMA zonelists are doubled because we need zonelists that
1188 * restrict the allocations to a single node for __GFP_THISNODE.
1190 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1196 * This struct contains information about a zone in a zonelist. It is stored
1197 * here to avoid dereferences into large structures and lookups of tables
1200 struct zone *zone; /* Pointer to actual zone */
1201 int zone_idx; /* zone_idx(zoneref->zone) */
1205 * One allocation request operates on a zonelist. A zonelist
1206 * is a list of zones, the first one is the 'goal' of the
1207 * allocation, the other zones are fallback zones, in decreasing
1210 * To speed the reading of the zonelist, the zonerefs contain the zone index
1211 * of the entry being read. Helper functions to access information given
1212 * a struct zoneref are
1214 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1215 * zonelist_zone_idx() - Return the index of the zone for an entry
1216 * zonelist_node_idx() - Return the index of the node for an entry
1219 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1223 * The array of struct pages for flatmem.
1224 * It must be declared for SPARSEMEM as well because there are configurations
1225 * that rely on that.
1227 extern struct page *mem_map;
1229 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1230 struct deferred_split {
1231 spinlock_t split_queue_lock;
1232 struct list_head split_queue;
1233 unsigned long split_queue_len;
1237 #ifdef CONFIG_MEMORY_FAILURE
1239 * Per NUMA node memory failure handling statistics.
1241 struct memory_failure_stats {
1243 * Number of raw pages poisoned.
1244 * Cases not accounted: memory outside kernel control, offline page,
1245 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1246 * error events, and unpoison actions from hwpoison_unpoison.
1248 unsigned long total;
1250 * Recovery results of poisoned raw pages handled by memory_failure,
1251 * in sync with mf_result.
1252 * total = ignored + failed + delayed + recovered.
1253 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1255 unsigned long ignored;
1256 unsigned long failed;
1257 unsigned long delayed;
1258 unsigned long recovered;
1263 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1264 * it's memory layout. On UMA machines there is a single pglist_data which
1265 * describes the whole memory.
1267 * Memory statistics and page replacement data structures are maintained on a
1270 typedef struct pglist_data {
1272 * node_zones contains just the zones for THIS node. Not all of the
1273 * zones may be populated, but it is the full list. It is referenced by
1274 * this node's node_zonelists as well as other node's node_zonelists.
1276 struct zone node_zones[MAX_NR_ZONES];
1279 * node_zonelists contains references to all zones in all nodes.
1280 * Generally the first zones will be references to this node's
1283 struct zonelist node_zonelists[MAX_ZONELISTS];
1285 int nr_zones; /* number of populated zones in this node */
1286 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1287 struct page *node_mem_map;
1288 #ifdef CONFIG_PAGE_EXTENSION
1289 struct page_ext *node_page_ext;
1292 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1294 * Must be held any time you expect node_start_pfn,
1295 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1296 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1299 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1300 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1301 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1303 * Nests above zone->lock and zone->span_seqlock
1305 spinlock_t node_size_lock;
1307 unsigned long node_start_pfn;
1308 unsigned long node_present_pages; /* total number of physical pages */
1309 unsigned long node_spanned_pages; /* total size of physical page
1310 range, including holes */
1312 wait_queue_head_t kswapd_wait;
1313 wait_queue_head_t pfmemalloc_wait;
1315 /* workqueues for throttling reclaim for different reasons. */
1316 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1318 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1319 unsigned long nr_reclaim_start; /* nr pages written while throttled
1320 * when throttling started. */
1321 #ifdef CONFIG_MEMORY_HOTPLUG
1322 struct mutex kswapd_lock;
1324 struct task_struct *kswapd; /* Protected by kswapd_lock */
1326 enum zone_type kswapd_highest_zoneidx;
1328 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1330 #ifdef CONFIG_COMPACTION
1331 int kcompactd_max_order;
1332 enum zone_type kcompactd_highest_zoneidx;
1333 wait_queue_head_t kcompactd_wait;
1334 struct task_struct *kcompactd;
1335 bool proactive_compact_trigger;
1338 * This is a per-node reserve of pages that are not available
1339 * to userspace allocations.
1341 unsigned long totalreserve_pages;
1345 * node reclaim becomes active if more unmapped pages exist.
1347 unsigned long min_unmapped_pages;
1348 unsigned long min_slab_pages;
1349 #endif /* CONFIG_NUMA */
1351 /* Write-intensive fields used by page reclaim */
1352 CACHELINE_PADDING(_pad1_);
1354 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1356 * If memory initialisation on large machines is deferred then this
1357 * is the first PFN that needs to be initialised.
1359 unsigned long first_deferred_pfn;
1360 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1362 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1363 struct deferred_split deferred_split_queue;
1366 #ifdef CONFIG_NUMA_BALANCING
1367 /* start time in ms of current promote rate limit period */
1368 unsigned int nbp_rl_start;
1369 /* number of promote candidate pages at start time of current rate limit period */
1370 unsigned long nbp_rl_nr_cand;
1371 /* promote threshold in ms */
1372 unsigned int nbp_threshold;
1373 /* start time in ms of current promote threshold adjustment period */
1374 unsigned int nbp_th_start;
1376 * number of promote candidate pages at start time of current promote
1377 * threshold adjustment period
1379 unsigned long nbp_th_nr_cand;
1381 /* Fields commonly accessed by the page reclaim scanner */
1384 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1386 * Use mem_cgroup_lruvec() to look up lruvecs.
1388 struct lruvec __lruvec;
1390 unsigned long flags;
1392 #ifdef CONFIG_LRU_GEN
1393 /* kswap mm walk data */
1394 struct lru_gen_mm_walk mm_walk;
1395 /* lru_gen_folio list */
1396 struct lru_gen_memcg memcg_lru;
1399 CACHELINE_PADDING(_pad2_);
1401 /* Per-node vmstats */
1402 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1403 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1405 struct memory_tier __rcu *memtier;
1407 #ifdef CONFIG_MEMORY_FAILURE
1408 struct memory_failure_stats mf_stats;
1412 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1413 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1415 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1416 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1418 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1420 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1423 #include <linux/memory_hotplug.h>
1425 void build_all_zonelists(pg_data_t *pgdat);
1426 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1427 enum zone_type highest_zoneidx);
1428 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1429 int highest_zoneidx, unsigned int alloc_flags,
1431 bool zone_watermark_ok(struct zone *z, unsigned int order,
1432 unsigned long mark, int highest_zoneidx,
1433 unsigned int alloc_flags);
1434 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1435 unsigned long mark, int highest_zoneidx);
1437 * Memory initialization context, use to differentiate memory added by
1438 * the platform statically or via memory hotplug interface.
1440 enum meminit_context {
1445 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1446 unsigned long size);
1448 extern void lruvec_init(struct lruvec *lruvec);
1450 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1453 return lruvec->pgdat;
1455 return container_of(lruvec, struct pglist_data, __lruvec);
1459 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1460 int local_memory_node(int node_id);
1462 static inline int local_memory_node(int node_id) { return node_id; };
1466 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1468 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1470 #ifdef CONFIG_ZONE_DEVICE
1471 static inline bool zone_is_zone_device(struct zone *zone)
1473 return zone_idx(zone) == ZONE_DEVICE;
1476 static inline bool zone_is_zone_device(struct zone *zone)
1483 * Returns true if a zone has pages managed by the buddy allocator.
1484 * All the reclaim decisions have to use this function rather than
1485 * populated_zone(). If the whole zone is reserved then we can easily
1486 * end up with populated_zone() && !managed_zone().
1488 static inline bool managed_zone(struct zone *zone)
1490 return zone_managed_pages(zone);
1493 /* Returns true if a zone has memory */
1494 static inline bool populated_zone(struct zone *zone)
1496 return zone->present_pages;
1500 static inline int zone_to_nid(struct zone *zone)
1505 static inline void zone_set_nid(struct zone *zone, int nid)
1510 static inline int zone_to_nid(struct zone *zone)
1515 static inline void zone_set_nid(struct zone *zone, int nid) {}
1518 extern int movable_zone;
1520 static inline int is_highmem_idx(enum zone_type idx)
1522 #ifdef CONFIG_HIGHMEM
1523 return (idx == ZONE_HIGHMEM ||
1524 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1531 * is_highmem - helper function to quickly check if a struct zone is a
1532 * highmem zone or not. This is an attempt to keep references
1533 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1534 * @zone: pointer to struct zone variable
1535 * Return: 1 for a highmem zone, 0 otherwise
1537 static inline int is_highmem(struct zone *zone)
1539 return is_highmem_idx(zone_idx(zone));
1542 #ifdef CONFIG_ZONE_DMA
1543 bool has_managed_dma(void);
1545 static inline bool has_managed_dma(void)
1554 extern struct pglist_data contig_page_data;
1555 static inline struct pglist_data *NODE_DATA(int nid)
1557 return &contig_page_data;
1560 #else /* CONFIG_NUMA */
1562 #include <asm/mmzone.h>
1564 #endif /* !CONFIG_NUMA */
1566 extern struct pglist_data *first_online_pgdat(void);
1567 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1568 extern struct zone *next_zone(struct zone *zone);
1571 * for_each_online_pgdat - helper macro to iterate over all online nodes
1572 * @pgdat: pointer to a pg_data_t variable
1574 #define for_each_online_pgdat(pgdat) \
1575 for (pgdat = first_online_pgdat(); \
1577 pgdat = next_online_pgdat(pgdat))
1579 * for_each_zone - helper macro to iterate over all memory zones
1580 * @zone: pointer to struct zone variable
1582 * The user only needs to declare the zone variable, for_each_zone
1585 #define for_each_zone(zone) \
1586 for (zone = (first_online_pgdat())->node_zones; \
1588 zone = next_zone(zone))
1590 #define for_each_populated_zone(zone) \
1591 for (zone = (first_online_pgdat())->node_zones; \
1593 zone = next_zone(zone)) \
1594 if (!populated_zone(zone)) \
1595 ; /* do nothing */ \
1598 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1600 return zoneref->zone;
1603 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1605 return zoneref->zone_idx;
1608 static inline int zonelist_node_idx(struct zoneref *zoneref)
1610 return zone_to_nid(zoneref->zone);
1613 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1614 enum zone_type highest_zoneidx,
1618 * 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
1619 * @z: The cursor used as a starting point for the search
1620 * @highest_zoneidx: The zone index of the highest zone to return
1621 * @nodes: An optional nodemask to filter the zonelist with
1623 * This function returns the next zone at or below a given zone index that is
1624 * within the allowed nodemask using a cursor as the starting point for the
1625 * search. The zoneref returned is a cursor that represents the current zone
1626 * being examined. It should be advanced by one before calling
1627 * next_zones_zonelist again.
1629 * Return: the next zone at or below highest_zoneidx within the allowed
1630 * nodemask using a cursor within a zonelist as a starting point
1632 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1633 enum zone_type highest_zoneidx,
1636 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1638 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1642 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1643 * @zonelist: The zonelist to search for a suitable zone
1644 * @highest_zoneidx: The zone index of the highest zone to return
1645 * @nodes: An optional nodemask to filter the zonelist with
1647 * This function returns the first zone at or below a given zone index that is
1648 * within the allowed nodemask. The zoneref returned is a cursor that can be
1649 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1650 * one before calling.
1652 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1653 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1654 * update due to cpuset modification.
1656 * Return: Zoneref pointer for the first suitable zone found
1658 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1659 enum zone_type highest_zoneidx,
1662 return next_zones_zonelist(zonelist->_zonerefs,
1663 highest_zoneidx, nodes);
1667 * 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
1668 * @zone: The current zone in the iterator
1669 * @z: The current pointer within zonelist->_zonerefs being iterated
1670 * @zlist: The zonelist being iterated
1671 * @highidx: The zone index of the highest zone to return
1672 * @nodemask: Nodemask allowed by the allocator
1674 * This iterator iterates though all zones at or below a given zone index and
1675 * within a given nodemask
1677 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1678 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1680 z = next_zones_zonelist(++z, highidx, nodemask), \
1681 zone = zonelist_zone(z))
1683 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1684 for (zone = z->zone; \
1686 z = next_zones_zonelist(++z, highidx, nodemask), \
1687 zone = zonelist_zone(z))
1691 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1692 * @zone: The current zone in the iterator
1693 * @z: The current pointer within zonelist->zones being iterated
1694 * @zlist: The zonelist being iterated
1695 * @highidx: The zone index of the highest zone to return
1697 * This iterator iterates though all zones at or below a given zone index.
1699 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1700 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1702 /* Whether the 'nodes' are all movable nodes */
1703 static inline bool movable_only_nodes(nodemask_t *nodes)
1705 struct zonelist *zonelist;
1709 if (nodes_empty(*nodes))
1713 * We can chose arbitrary node from the nodemask to get a
1714 * zonelist as they are interlinked. We just need to find
1715 * at least one zone that can satisfy kernel allocations.
1717 nid = first_node(*nodes);
1718 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1719 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1720 return (!z->zone) ? true : false;
1724 #ifdef CONFIG_SPARSEMEM
1725 #include <asm/sparsemem.h>
1728 #ifdef CONFIG_FLATMEM
1729 #define pfn_to_nid(pfn) (0)
1732 #ifdef CONFIG_SPARSEMEM
1735 * PA_SECTION_SHIFT physical address to/from section number
1736 * PFN_SECTION_SHIFT pfn to/from section number
1738 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1739 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1741 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1743 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1744 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1746 #define SECTION_BLOCKFLAGS_BITS \
1747 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1749 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1750 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1753 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1755 return pfn >> PFN_SECTION_SHIFT;
1757 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1759 return sec << PFN_SECTION_SHIFT;
1762 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1763 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1765 #define SUBSECTION_SHIFT 21
1766 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1768 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1769 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1770 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1772 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1773 #error Subsection size exceeds section size
1775 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1778 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1779 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1781 struct mem_section_usage {
1782 struct rcu_head rcu;
1783 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1784 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1786 /* See declaration of similar field in struct zone */
1787 unsigned long pageblock_flags[0];
1790 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1794 struct mem_section {
1796 * This is, logically, a pointer to an array of struct
1797 * pages. However, it is stored with some other magic.
1798 * (see sparse.c::sparse_init_one_section())
1800 * Additionally during early boot we encode node id of
1801 * the location of the section here to guide allocation.
1802 * (see sparse.c::memory_present())
1804 * Making it a UL at least makes someone do a cast
1805 * before using it wrong.
1807 unsigned long section_mem_map;
1809 struct mem_section_usage *usage;
1810 #ifdef CONFIG_PAGE_EXTENSION
1812 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1813 * section. (see page_ext.h about this.)
1815 struct page_ext *page_ext;
1819 * WARNING: mem_section must be a power-of-2 in size for the
1820 * calculation and use of SECTION_ROOT_MASK to make sense.
1824 #ifdef CONFIG_SPARSEMEM_EXTREME
1825 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1827 #define SECTIONS_PER_ROOT 1
1830 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1831 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1832 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1834 #ifdef CONFIG_SPARSEMEM_EXTREME
1835 extern struct mem_section **mem_section;
1837 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1840 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1842 return ms->usage->pageblock_flags;
1845 static inline struct mem_section *__nr_to_section(unsigned long nr)
1847 unsigned long root = SECTION_NR_TO_ROOT(nr);
1849 if (unlikely(root >= NR_SECTION_ROOTS))
1852 #ifdef CONFIG_SPARSEMEM_EXTREME
1853 if (!mem_section || !mem_section[root])
1856 return &mem_section[root][nr & SECTION_ROOT_MASK];
1858 extern size_t mem_section_usage_size(void);
1861 * We use the lower bits of the mem_map pointer to store
1862 * a little bit of information. The pointer is calculated
1863 * as mem_map - section_nr_to_pfn(pnum). The result is
1864 * aligned to the minimum alignment of the two values:
1865 * 1. All mem_map arrays are page-aligned.
1866 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1867 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1868 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1869 * worst combination is powerpc with 256k pages,
1870 * which results in PFN_SECTION_SHIFT equal 6.
1871 * To sum it up, at least 6 bits are available on all architectures.
1872 * However, we can exceed 6 bits on some other architectures except
1873 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1874 * with the worst case of 64K pages on arm64) if we make sure the
1875 * exceeded bit is not applicable to powerpc.
1878 SECTION_MARKED_PRESENT_BIT,
1879 SECTION_HAS_MEM_MAP_BIT,
1880 SECTION_IS_ONLINE_BIT,
1881 SECTION_IS_EARLY_BIT,
1882 #ifdef CONFIG_ZONE_DEVICE
1883 SECTION_TAINT_ZONE_DEVICE_BIT,
1885 SECTION_MAP_LAST_BIT,
1888 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1889 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1890 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1891 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1892 #ifdef CONFIG_ZONE_DEVICE
1893 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1895 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1896 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1898 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1900 unsigned long map = section->section_mem_map;
1901 map &= SECTION_MAP_MASK;
1902 return (struct page *)map;
1905 static inline int present_section(struct mem_section *section)
1907 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1910 static inline int present_section_nr(unsigned long nr)
1912 return present_section(__nr_to_section(nr));
1915 static inline int valid_section(struct mem_section *section)
1917 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1920 static inline int early_section(struct mem_section *section)
1922 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1925 static inline int valid_section_nr(unsigned long nr)
1927 return valid_section(__nr_to_section(nr));
1930 static inline int online_section(struct mem_section *section)
1932 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1935 #ifdef CONFIG_ZONE_DEVICE
1936 static inline int online_device_section(struct mem_section *section)
1938 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1940 return section && ((section->section_mem_map & flags) == flags);
1943 static inline int online_device_section(struct mem_section *section)
1949 static inline int online_section_nr(unsigned long nr)
1951 return online_section(__nr_to_section(nr));
1954 #ifdef CONFIG_MEMORY_HOTPLUG
1955 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1956 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1959 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1961 return __nr_to_section(pfn_to_section_nr(pfn));
1964 extern unsigned long __highest_present_section_nr;
1966 static inline int subsection_map_index(unsigned long pfn)
1968 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1971 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1972 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1974 int idx = subsection_map_index(pfn);
1976 return test_bit(idx, READ_ONCE(ms->usage)->subsection_map);
1979 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1985 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1987 * pfn_valid - check if there is a valid memory map entry for a PFN
1988 * @pfn: the page frame number to check
1990 * Check if there is a valid memory map entry aka struct page for the @pfn.
1991 * Note, that availability of the memory map entry does not imply that
1992 * there is actual usable memory at that @pfn. The struct page may
1993 * represent a hole or an unusable page frame.
1995 * Return: 1 for PFNs that have memory map entries and 0 otherwise
1997 static inline int pfn_valid(unsigned long pfn)
1999 struct mem_section *ms;
2003 * Ensure the upper PAGE_SHIFT bits are clear in the
2004 * pfn. Else it might lead to false positives when
2005 * some of the upper bits are set, but the lower bits
2006 * match a valid pfn.
2008 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2011 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2013 ms = __pfn_to_section(pfn);
2015 if (!valid_section(ms)) {
2020 * Traditionally early sections always returned pfn_valid() for
2021 * the entire section-sized span.
2023 ret = early_section(ms) || pfn_section_valid(ms, pfn);
2030 static inline int pfn_in_present_section(unsigned long pfn)
2032 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2034 return present_section(__pfn_to_section(pfn));
2037 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2039 while (++section_nr <= __highest_present_section_nr) {
2040 if (present_section_nr(section_nr))
2048 * These are _only_ used during initialisation, therefore they
2049 * can use __initdata ... They could have names to indicate
2053 #define pfn_to_nid(pfn) \
2055 unsigned long __pfn_to_nid_pfn = (pfn); \
2056 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2059 #define pfn_to_nid(pfn) (0)
2062 void sparse_init(void);
2064 #define sparse_init() do {} while (0)
2065 #define sparse_index_init(_sec, _nid) do {} while (0)
2066 #define pfn_in_present_section pfn_valid
2067 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2068 #endif /* CONFIG_SPARSEMEM */
2070 #endif /* !__GENERATING_BOUNDS.H */
2071 #endif /* !__ASSEMBLY__ */
2072 #endif /* _LINUX_MMZONE_H */