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>
27 /* Free memory management - zoned buddy allocator. */
28 #ifndef CONFIG_ARCH_FORCE_MAX_ORDER
31 #define MAX_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
33 #define MAX_ORDER_NR_PAGES (1 << MAX_ORDER)
35 #define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
38 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
39 * costly to service. That is between allocation orders which should
40 * coalesce naturally under reasonable reclaim pressure and those which
43 #define PAGE_ALLOC_COSTLY_ORDER 3
49 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
50 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
53 * MIGRATE_CMA migration type is designed to mimic the way
54 * ZONE_MOVABLE works. Only movable pages can be allocated
55 * from MIGRATE_CMA pageblocks and page allocator never
56 * implicitly change migration type of MIGRATE_CMA pageblock.
58 * The way to use it is to change migratetype of a range of
59 * pageblocks to MIGRATE_CMA which can be done by
60 * __free_pageblock_cma() function.
64 #ifdef CONFIG_MEMORY_ISOLATION
65 MIGRATE_ISOLATE, /* can't allocate from here */
70 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
71 extern const char * const migratetype_names[MIGRATE_TYPES];
74 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
75 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
77 # define is_migrate_cma(migratetype) false
78 # define is_migrate_cma_page(_page) false
81 static inline bool is_migrate_movable(int mt)
83 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
87 * Check whether a migratetype can be merged with another migratetype.
89 * It is only mergeable when it can fall back to other migratetypes for
90 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
92 static inline bool migratetype_is_mergeable(int mt)
94 return mt < MIGRATE_PCPTYPES;
97 #define for_each_migratetype_order(order, type) \
98 for (order = 0; order <= MAX_ORDER; order++) \
99 for (type = 0; type < MIGRATE_TYPES; type++)
101 extern int page_group_by_mobility_disabled;
103 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
105 #define get_pageblock_migratetype(page) \
106 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
108 #define folio_migratetype(folio) \
109 get_pfnblock_flags_mask(&folio->page, folio_pfn(folio), \
112 struct list_head free_list[MIGRATE_TYPES];
113 unsigned long nr_free;
119 enum numa_stat_item {
120 NUMA_HIT, /* allocated in intended node */
121 NUMA_MISS, /* allocated in non intended node */
122 NUMA_FOREIGN, /* was intended here, hit elsewhere */
123 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
124 NUMA_LOCAL, /* allocation from local node */
125 NUMA_OTHER, /* allocation from other node */
126 NR_VM_NUMA_EVENT_ITEMS
129 #define NR_VM_NUMA_EVENT_ITEMS 0
132 enum zone_stat_item {
133 /* First 128 byte cacheline (assuming 64 bit words) */
135 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
136 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
138 NR_ZONE_INACTIVE_FILE,
141 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
142 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
143 /* Second 128 byte cacheline */
145 #if IS_ENABLED(CONFIG_ZSMALLOC)
146 NR_ZSPAGES, /* allocated in zsmalloc */
149 #ifdef CONFIG_UNACCEPTED_MEMORY
152 NR_VM_ZONE_STAT_ITEMS };
154 enum node_stat_item {
156 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
157 NR_ACTIVE_ANON, /* " " " " " */
158 NR_INACTIVE_FILE, /* " " " " " */
159 NR_ACTIVE_FILE, /* " " " " " */
160 NR_UNEVICTABLE, /* " " " " " */
161 NR_SLAB_RECLAIMABLE_B,
162 NR_SLAB_UNRECLAIMABLE_B,
163 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
164 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
166 WORKINGSET_REFAULT_BASE,
167 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
168 WORKINGSET_REFAULT_FILE,
169 WORKINGSET_ACTIVATE_BASE,
170 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
171 WORKINGSET_ACTIVATE_FILE,
172 WORKINGSET_RESTORE_BASE,
173 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
174 WORKINGSET_RESTORE_FILE,
175 WORKINGSET_NODERECLAIM,
176 NR_ANON_MAPPED, /* Mapped anonymous pages */
177 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
178 only modified from process context */
182 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
183 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
190 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
191 NR_DIRTIED, /* page dirtyings since bootup */
192 NR_WRITTEN, /* page writings since bootup */
193 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
194 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
195 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
196 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
197 NR_KERNEL_STACK_KB, /* measured in KiB */
198 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
199 NR_KERNEL_SCS_KB, /* measured in KiB */
201 NR_PAGETABLE, /* used for pagetables */
202 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
206 #ifdef CONFIG_NUMA_BALANCING
207 PGPROMOTE_SUCCESS, /* promote successfully */
208 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
210 NR_VM_NODE_STAT_ITEMS
214 * Returns true if the item should be printed in THPs (/proc/vmstat
215 * currently prints number of anon, file and shmem THPs. But the item
216 * is charged in pages).
218 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
220 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
223 return item == NR_ANON_THPS ||
224 item == NR_FILE_THPS ||
225 item == NR_SHMEM_THPS ||
226 item == NR_SHMEM_PMDMAPPED ||
227 item == NR_FILE_PMDMAPPED;
231 * Returns true if the value is measured in bytes (most vmstat values are
232 * measured in pages). This defines the API part, the internal representation
233 * might be different.
235 static __always_inline bool vmstat_item_in_bytes(int idx)
238 * Global and per-node slab counters track slab pages.
239 * It's expected that changes are multiples of PAGE_SIZE.
240 * Internally values are stored in pages.
242 * Per-memcg and per-lruvec counters track memory, consumed
243 * by individual slab objects. These counters are actually
246 return (idx == NR_SLAB_RECLAIMABLE_B ||
247 idx == NR_SLAB_UNRECLAIMABLE_B);
251 * We do arithmetic on the LRU lists in various places in the code,
252 * so it is important to keep the active lists LRU_ACTIVE higher in
253 * the array than the corresponding inactive lists, and to keep
254 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
256 * This has to be kept in sync with the statistics in zone_stat_item
257 * above and the descriptions in vmstat_text in mm/vmstat.c
264 LRU_INACTIVE_ANON = LRU_BASE,
265 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
266 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
267 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
272 enum vmscan_throttle_state {
273 VMSCAN_THROTTLE_WRITEBACK,
274 VMSCAN_THROTTLE_ISOLATED,
275 VMSCAN_THROTTLE_NOPROGRESS,
276 VMSCAN_THROTTLE_CONGESTED,
280 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
282 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
284 static inline bool is_file_lru(enum lru_list lru)
286 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
289 static inline bool is_active_lru(enum lru_list lru)
291 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
294 #define WORKINGSET_ANON 0
295 #define WORKINGSET_FILE 1
296 #define ANON_AND_FILE 2
300 * An lruvec has many dirty pages backed by a congested BDI:
301 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
302 * It can be cleared by cgroup reclaim or kswapd.
303 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
304 * It can only be cleared by kswapd.
306 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
307 * reclaim, but not vice versa. This only applies to the root cgroup.
308 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
309 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
312 LRUVEC_CGROUP_CONGESTED,
313 LRUVEC_NODE_CONGESTED,
316 #endif /* !__GENERATING_BOUNDS_H */
319 * Evictable pages are divided into multiple generations. The youngest and the
320 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
321 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
322 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
323 * corresponding generation. The gen counter in folio->flags stores gen+1 while
324 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
326 * A page is added to the youngest generation on faulting. The aging needs to
327 * check the accessed bit at least twice before handing this page over to the
328 * eviction. The first check takes care of the accessed bit set on the initial
329 * fault; the second check makes sure this page hasn't been used since then.
330 * This process, AKA second chance, requires a minimum of two generations,
331 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
332 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
333 * rest of generations, if they exist, are considered inactive. See
334 * lru_gen_is_active().
336 * PG_active is always cleared while a page is on one of lrugen->folios[] so
337 * that the aging needs not to worry about it. And it's set again when a page
338 * considered active is isolated for non-reclaiming purposes, e.g., migration.
339 * See lru_gen_add_folio() and lru_gen_del_folio().
341 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
342 * number of categories of the active/inactive LRU when keeping track of
343 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
346 #define MIN_NR_GENS 2U
347 #define MAX_NR_GENS 4U
350 * Each generation is divided into multiple tiers. A page accessed N times
351 * through file descriptors is in tier order_base_2(N). A page in the first tier
352 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
353 * tables or read ahead. A page in any other tier (N>1) is marked by
354 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
355 * supported without using additional bits in folio->flags.
357 * In contrast to moving across generations which requires the LRU lock, moving
358 * across tiers only involves atomic operations on folio->flags and therefore
359 * has a negligible cost in the buffered access path. In the eviction path,
360 * comparisons of refaulted/(evicted+protected) from the first tier and the
361 * rest infer whether pages accessed multiple times through file descriptors
362 * are statistically hot and thus worth protecting.
364 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
365 * number of categories of the active/inactive LRU when keeping track of
366 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
369 #define MAX_NR_TIERS 4U
371 #ifndef __GENERATING_BOUNDS_H
374 struct page_vma_mapped_walk;
376 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
377 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
379 #ifdef CONFIG_LRU_GEN
389 LRU_GEN_NONLEAF_YOUNG,
393 #define MIN_LRU_BATCH BITS_PER_LONG
394 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
396 /* whether to keep historical stats from evicted generations */
397 #ifdef CONFIG_LRU_GEN_STATS
398 #define NR_HIST_GENS MAX_NR_GENS
400 #define NR_HIST_GENS 1U
404 * The youngest generation number is stored in max_seq for both anon and file
405 * types as they are aged on an equal footing. The oldest generation numbers are
406 * stored in min_seq[] separately for anon and file types as clean file pages
407 * can be evicted regardless of swap constraints.
409 * Normally anon and file min_seq are in sync. But if swapping is constrained,
410 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
413 * The number of pages in each generation is eventually consistent and therefore
414 * can be transiently negative when reset_batch_size() is pending.
416 struct lru_gen_folio {
417 /* the aging increments the youngest generation number */
418 unsigned long max_seq;
419 /* the eviction increments the oldest generation numbers */
420 unsigned long min_seq[ANON_AND_FILE];
421 /* the birth time of each generation in jiffies */
422 unsigned long timestamps[MAX_NR_GENS];
423 /* the multi-gen LRU lists, lazily sorted on eviction */
424 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
425 /* the multi-gen LRU sizes, eventually consistent */
426 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
427 /* the exponential moving average of refaulted */
428 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
429 /* the exponential moving average of evicted+protected */
430 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
431 /* the first tier doesn't need protection, hence the minus one */
432 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
433 /* can be modified without holding the LRU lock */
434 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
435 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
436 /* whether the multi-gen LRU is enabled */
439 /* the memcg generation this lru_gen_folio belongs to */
441 /* the list segment this lru_gen_folio belongs to */
443 /* per-node lru_gen_folio list for global reclaim */
444 struct hlist_nulls_node list;
449 MM_LEAF_TOTAL, /* total leaf entries */
450 MM_LEAF_OLD, /* old leaf entries */
451 MM_LEAF_YOUNG, /* young leaf entries */
452 MM_NONLEAF_TOTAL, /* total non-leaf entries */
453 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
454 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
458 /* double-buffering Bloom filters */
459 #define NR_BLOOM_FILTERS 2
461 struct lru_gen_mm_state {
462 /* set to max_seq after each iteration */
464 /* where the current iteration continues after */
465 struct list_head *head;
466 /* where the last iteration ended before */
467 struct list_head *tail;
468 /* Bloom filters flip after each iteration */
469 unsigned long *filters[NR_BLOOM_FILTERS];
470 /* the mm stats for debugging */
471 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
474 struct lru_gen_mm_walk {
475 /* the lruvec under reclaim */
476 struct lruvec *lruvec;
477 /* unstable max_seq from lru_gen_folio */
478 unsigned long max_seq;
479 /* the next address within an mm to scan */
480 unsigned long next_addr;
481 /* to batch promoted pages */
482 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
483 /* to batch the mm stats */
484 int mm_stats[NR_MM_STATS];
485 /* total batched items */
491 void lru_gen_init_lruvec(struct lruvec *lruvec);
492 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
497 * For each node, memcgs are divided into two generations: the old and the
498 * young. For each generation, memcgs are randomly sharded into multiple bins
499 * to improve scalability. For each bin, the hlist_nulls is virtually divided
500 * into three segments: the head, the tail and the default.
502 * An onlining memcg is added to the tail of a random bin in the old generation.
503 * The eviction starts at the head of a random bin in the old generation. The
504 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
505 * the old generation, is incremented when all its bins become empty.
507 * There are four operations:
508 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
509 * current generation (old or young) and updates its "seg" to "head";
510 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
511 * current generation (old or young) and updates its "seg" to "tail";
512 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
513 * generation, updates its "gen" to "old" and resets its "seg" to "default";
514 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
515 * young generation, updates its "gen" to "young" and resets its "seg" to
518 * The events that trigger the above operations are:
519 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
520 * 2. The first attempt to reclaim a memcg below low, which triggers
522 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
523 * threshold, which triggers MEMCG_LRU_TAIL;
524 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
525 * threshold, which triggers MEMCG_LRU_YOUNG;
526 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
527 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
528 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
531 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
532 * of their max_seq counters ensures the eventual fairness to all eligible
533 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
534 * 2. There are only two valid generations: old (seq) and young (seq+1).
535 * MEMCG_NR_GENS is set to three so that when reading the generation counter
536 * locklessly, a stale value (seq-1) does not wraparound to young.
538 #define MEMCG_NR_GENS 3
539 #define MEMCG_NR_BINS 8
541 struct lru_gen_memcg {
542 /* the per-node memcg generation counter */
544 /* each memcg has one lru_gen_folio per node */
545 unsigned long nr_memcgs[MEMCG_NR_GENS];
546 /* per-node lru_gen_folio list for global reclaim */
547 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
548 /* protects the above */
552 void lru_gen_init_pgdat(struct pglist_data *pgdat);
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_MEMCG */
563 #define MEMCG_NR_GENS 1
565 struct lru_gen_memcg {
568 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
572 #endif /* CONFIG_MEMCG */
574 #else /* !CONFIG_LRU_GEN */
576 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
580 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
584 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
590 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
594 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
598 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
602 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
606 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
610 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
614 #endif /* CONFIG_MEMCG */
616 #endif /* CONFIG_LRU_GEN */
619 struct list_head lists[NR_LRU_LISTS];
620 /* per lruvec lru_lock for memcg */
623 * These track the cost of reclaiming one LRU - file or anon -
624 * over the other. As the observed cost of reclaiming one LRU
625 * increases, the reclaim scan balance tips toward the other.
627 unsigned long anon_cost;
628 unsigned long file_cost;
629 /* Non-resident age, driven by LRU movement */
630 atomic_long_t nonresident_age;
631 /* Refaults at the time of last reclaim cycle */
632 unsigned long refaults[ANON_AND_FILE];
633 /* Various lruvec state flags (enum lruvec_flags) */
635 #ifdef CONFIG_LRU_GEN
636 /* evictable pages divided into generations */
637 struct lru_gen_folio lrugen;
638 /* to concurrently iterate lru_gen_mm_list */
639 struct lru_gen_mm_state mm_state;
642 struct pglist_data *pgdat;
646 /* Isolate for asynchronous migration */
647 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
648 /* Isolate unevictable pages */
649 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
651 /* LRU Isolation modes. */
652 typedef unsigned __bitwise isolate_mode_t;
654 enum zone_watermarks {
663 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
664 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
665 * it should not contribute to serious fragmentation causing THP allocation
668 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
673 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
674 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
676 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
677 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
678 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
679 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
682 * Flags used in pcp->flags field.
684 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
685 * previous page freeing. To avoid to drain PCP for an accident
686 * high-order page freeing.
688 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
689 * draining PCP for consecutive high-order pages freeing without
690 * allocation if data cache slice of CPU is large enough. To reduce
691 * zone lock contention and keep cache-hot pages reusing.
693 #define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
694 #define PCPF_FREE_HIGH_BATCH BIT(1)
696 struct per_cpu_pages {
697 spinlock_t lock; /* Protects lists field */
698 int count; /* number of pages in the list */
699 int high; /* high watermark, emptying needed */
700 int high_min; /* min high watermark */
701 int high_max; /* max high watermark */
702 int batch; /* chunk size for buddy add/remove */
703 u8 flags; /* protected by pcp->lock */
704 u8 alloc_factor; /* batch scaling factor during allocate */
706 u8 expire; /* When 0, remote pagesets are drained */
708 short free_count; /* consecutive free count */
710 /* Lists of pages, one per migrate type stored on the pcp-lists */
711 struct list_head lists[NR_PCP_LISTS];
712 } ____cacheline_aligned_in_smp;
714 struct per_cpu_zonestat {
716 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
721 * Low priority inaccurate counters that are only folded
722 * on demand. Use a large type to avoid the overhead of
723 * folding during refresh_cpu_vm_stats.
725 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
729 struct per_cpu_nodestat {
731 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
734 #endif /* !__GENERATING_BOUNDS.H */
738 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
739 * to DMA to all of the addressable memory (ZONE_NORMAL).
740 * On architectures where this area covers the whole 32 bit address
741 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
742 * DMA addressing constraints. This distinction is important as a 32bit
743 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
744 * platforms may need both zones as they support peripherals with
745 * different DMA addressing limitations.
747 #ifdef CONFIG_ZONE_DMA
750 #ifdef CONFIG_ZONE_DMA32
754 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
755 * performed on pages in ZONE_NORMAL if the DMA devices support
756 * transfers to all addressable memory.
759 #ifdef CONFIG_HIGHMEM
761 * A memory area that is only addressable by the kernel through
762 * mapping portions into its own address space. This is for example
763 * used by i386 to allow the kernel to address the memory beyond
764 * 900MB. The kernel will set up special mappings (page
765 * table entries on i386) for each page that the kernel needs to
771 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
772 * movable pages with few exceptional cases described below. Main use
773 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
774 * likely to succeed, and to locally limit unmovable allocations - e.g.,
775 * to increase the number of THP/huge pages. Notable special cases are:
777 * 1. Pinned pages: (long-term) pinning of movable pages might
778 * essentially turn such pages unmovable. Therefore, we do not allow
779 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
780 * faulted, they come from the right zone right away. However, it is
781 * still possible that address space already has pages in
782 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
783 * touches that memory before pinning). In such case we migrate them
784 * to a different zone. When migration fails - pinning fails.
785 * 2. memblock allocations: kernelcore/movablecore setups might create
786 * situations where ZONE_MOVABLE contains unmovable allocations
787 * after boot. Memory offlining and allocations fail early.
788 * 3. Memory holes: kernelcore/movablecore setups might create very rare
789 * situations where ZONE_MOVABLE contains memory holes after boot,
790 * for example, if we have sections that are only partially
791 * populated. Memory offlining and allocations fail early.
792 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
793 * memory offlining, such pages cannot be allocated.
794 * 5. Unmovable PG_offline pages: in paravirtualized environments,
795 * hotplugged memory blocks might only partially be managed by the
796 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
797 * parts not manged by the buddy are unmovable PG_offline pages. In
798 * some cases (virtio-mem), such pages can be skipped during
799 * memory offlining, however, cannot be moved/allocated. These
800 * techniques might use alloc_contig_range() to hide previously
801 * exposed pages from the buddy again (e.g., to implement some sort
802 * of memory unplug in virtio-mem).
803 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
804 * situations where ZERO_PAGE(0) which is allocated differently
805 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
806 * cannot be migrated.
807 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
808 * memory to the MOVABLE zone, the vmemmap pages are also placed in
809 * such zone. Such pages cannot be really moved around as they are
810 * self-stored in the range, but they are treated as movable when
811 * the range they describe is about to be offlined.
813 * In general, no unmovable allocations that degrade memory offlining
814 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
815 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
816 * if has_unmovable_pages() states that there are no unmovable pages,
817 * there can be false negatives).
820 #ifdef CONFIG_ZONE_DEVICE
827 #ifndef __GENERATING_BOUNDS_H
829 #define ASYNC_AND_SYNC 2
832 /* Read-mostly fields */
834 /* zone watermarks, access with *_wmark_pages(zone) macros */
835 unsigned long _watermark[NR_WMARK];
836 unsigned long watermark_boost;
838 unsigned long nr_reserved_highatomic;
841 * We don't know if the memory that we're going to allocate will be
842 * freeable or/and it will be released eventually, so to avoid totally
843 * wasting several GB of ram we must reserve some of the lower zone
844 * memory (otherwise we risk to run OOM on the lower zones despite
845 * there being tons of freeable ram on the higher zones). This array is
846 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
849 long lowmem_reserve[MAX_NR_ZONES];
854 struct pglist_data *zone_pgdat;
855 struct per_cpu_pages __percpu *per_cpu_pageset;
856 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
858 * the high and batch values are copied to individual pagesets for
861 int pageset_high_min;
862 int pageset_high_max;
865 #ifndef CONFIG_SPARSEMEM
867 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
868 * In SPARSEMEM, this map is stored in struct mem_section
870 unsigned long *pageblock_flags;
871 #endif /* CONFIG_SPARSEMEM */
873 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
874 unsigned long zone_start_pfn;
877 * spanned_pages is the total pages spanned by the zone, including
878 * holes, which is calculated as:
879 * spanned_pages = zone_end_pfn - zone_start_pfn;
881 * present_pages is physical pages existing within the zone, which
883 * present_pages = spanned_pages - absent_pages(pages in holes);
885 * present_early_pages is present pages existing within the zone
886 * located on memory available since early boot, excluding hotplugged
889 * managed_pages is present pages managed by the buddy system, which
890 * is calculated as (reserved_pages includes pages allocated by the
891 * bootmem allocator):
892 * managed_pages = present_pages - reserved_pages;
894 * cma pages is present pages that are assigned for CMA use
897 * So present_pages may be used by memory hotplug or memory power
898 * management logic to figure out unmanaged pages by checking
899 * (present_pages - managed_pages). And managed_pages should be used
900 * by page allocator and vm scanner to calculate all kinds of watermarks
905 * zone_start_pfn and spanned_pages are protected by span_seqlock.
906 * It is a seqlock because it has to be read outside of zone->lock,
907 * and it is done in the main allocator path. But, it is written
908 * quite infrequently.
910 * The span_seq lock is declared along with zone->lock because it is
911 * frequently read in proximity to zone->lock. It's good to
912 * give them a chance of being in the same cacheline.
914 * Write access to present_pages at runtime should be protected by
915 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
916 * present_pages should use get_online_mems() to get a stable value.
918 atomic_long_t managed_pages;
919 unsigned long spanned_pages;
920 unsigned long present_pages;
921 #if defined(CONFIG_MEMORY_HOTPLUG)
922 unsigned long present_early_pages;
925 unsigned long cma_pages;
930 #ifdef CONFIG_MEMORY_ISOLATION
932 * Number of isolated pageblock. It is used to solve incorrect
933 * freepage counting problem due to racy retrieving migratetype
934 * of pageblock. Protected by zone->lock.
936 unsigned long nr_isolate_pageblock;
939 #ifdef CONFIG_MEMORY_HOTPLUG
940 /* see spanned/present_pages for more description */
941 seqlock_t span_seqlock;
946 /* Write-intensive fields used from the page allocator */
947 CACHELINE_PADDING(_pad1_);
949 /* free areas of different sizes */
950 struct free_area free_area[MAX_ORDER + 1];
952 #ifdef CONFIG_UNACCEPTED_MEMORY
953 /* Pages to be accepted. All pages on the list are MAX_ORDER */
954 struct list_head unaccepted_pages;
957 /* zone flags, see below */
960 /* Primarily protects free_area */
963 /* Write-intensive fields used by compaction and vmstats. */
964 CACHELINE_PADDING(_pad2_);
967 * When free pages are below this point, additional steps are taken
968 * when reading the number of free pages to avoid per-cpu counter
969 * drift allowing watermarks to be breached
971 unsigned long percpu_drift_mark;
973 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
974 /* pfn where compaction free scanner should start */
975 unsigned long compact_cached_free_pfn;
976 /* pfn where compaction migration scanner should start */
977 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
978 unsigned long compact_init_migrate_pfn;
979 unsigned long compact_init_free_pfn;
982 #ifdef CONFIG_COMPACTION
984 * On compaction failure, 1<<compact_defer_shift compactions
985 * are skipped before trying again. The number attempted since
986 * last failure is tracked with compact_considered.
987 * compact_order_failed is the minimum compaction failed order.
989 unsigned int compact_considered;
990 unsigned int compact_defer_shift;
991 int compact_order_failed;
994 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
995 /* Set to true when the PG_migrate_skip bits should be cleared */
996 bool compact_blockskip_flush;
1001 CACHELINE_PADDING(_pad3_);
1002 /* Zone statistics */
1003 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
1004 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
1005 } ____cacheline_internodealigned_in_smp;
1008 PGDAT_DIRTY, /* reclaim scanning has recently found
1009 * many dirty file pages at the tail
1012 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1013 * many pages under writeback
1015 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1019 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1020 * Cleared when kswapd is woken.
1022 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1023 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1026 static inline unsigned long zone_managed_pages(struct zone *zone)
1028 return (unsigned long)atomic_long_read(&zone->managed_pages);
1031 static inline unsigned long zone_cma_pages(struct zone *zone)
1034 return zone->cma_pages;
1040 static inline unsigned long zone_end_pfn(const struct zone *zone)
1042 return zone->zone_start_pfn + zone->spanned_pages;
1045 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1047 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1050 static inline bool zone_is_initialized(struct zone *zone)
1052 return zone->initialized;
1055 static inline bool zone_is_empty(struct zone *zone)
1057 return zone->spanned_pages == 0;
1060 #ifndef BUILD_VDSO32_64
1062 * The zone field is never updated after free_area_init_core()
1063 * sets it, so none of the operations on it need to be atomic.
1066 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1067 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1068 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1069 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1070 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1071 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1072 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1073 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1076 * Define the bit shifts to access each section. For non-existent
1077 * sections we define the shift as 0; that plus a 0 mask ensures
1078 * the compiler will optimise away reference to them.
1080 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1081 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1082 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1083 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1084 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1086 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1087 #ifdef NODE_NOT_IN_PAGE_FLAGS
1088 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1089 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1090 SECTIONS_PGOFF : ZONES_PGOFF)
1092 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1093 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1094 NODES_PGOFF : ZONES_PGOFF)
1097 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1099 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1100 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1101 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1102 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1103 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1104 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1106 static inline enum zone_type page_zonenum(const struct page *page)
1108 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1109 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1112 static inline enum zone_type folio_zonenum(const struct folio *folio)
1114 return page_zonenum(&folio->page);
1117 #ifdef CONFIG_ZONE_DEVICE
1118 static inline bool is_zone_device_page(const struct page *page)
1120 return page_zonenum(page) == ZONE_DEVICE;
1124 * Consecutive zone device pages should not be merged into the same sgl
1125 * or bvec segment with other types of pages or if they belong to different
1126 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1127 * without scanning the entire segment. This helper returns true either if
1128 * both pages are not zone device pages or both pages are zone device pages
1129 * with the same pgmap.
1131 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1132 const struct page *b)
1134 if (is_zone_device_page(a) != is_zone_device_page(b))
1136 if (!is_zone_device_page(a))
1138 return a->pgmap == b->pgmap;
1141 extern void memmap_init_zone_device(struct zone *, unsigned long,
1142 unsigned long, struct dev_pagemap *);
1144 static inline bool is_zone_device_page(const struct page *page)
1148 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1149 const struct page *b)
1155 static inline bool folio_is_zone_device(const struct folio *folio)
1157 return is_zone_device_page(&folio->page);
1160 static inline bool is_zone_movable_page(const struct page *page)
1162 return page_zonenum(page) == ZONE_MOVABLE;
1165 static inline bool folio_is_zone_movable(const struct folio *folio)
1167 return folio_zonenum(folio) == ZONE_MOVABLE;
1172 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1173 * intersection with the given zone
1175 static inline bool zone_intersects(struct zone *zone,
1176 unsigned long start_pfn, unsigned long nr_pages)
1178 if (zone_is_empty(zone))
1180 if (start_pfn >= zone_end_pfn(zone) ||
1181 start_pfn + nr_pages <= zone->zone_start_pfn)
1188 * The "priority" of VM scanning is how much of the queues we will scan in one
1189 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1190 * queues ("queue_length >> 12") during an aging round.
1192 #define DEF_PRIORITY 12
1194 /* Maximum number of zones on a zonelist */
1195 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1198 ZONELIST_FALLBACK, /* zonelist with fallback */
1201 * The NUMA zonelists are doubled because we need zonelists that
1202 * restrict the allocations to a single node for __GFP_THISNODE.
1204 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1210 * This struct contains information about a zone in a zonelist. It is stored
1211 * here to avoid dereferences into large structures and lookups of tables
1214 struct zone *zone; /* Pointer to actual zone */
1215 int zone_idx; /* zone_idx(zoneref->zone) */
1219 * One allocation request operates on a zonelist. A zonelist
1220 * is a list of zones, the first one is the 'goal' of the
1221 * allocation, the other zones are fallback zones, in decreasing
1224 * To speed the reading of the zonelist, the zonerefs contain the zone index
1225 * of the entry being read. Helper functions to access information given
1226 * a struct zoneref are
1228 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1229 * zonelist_zone_idx() - Return the index of the zone for an entry
1230 * zonelist_node_idx() - Return the index of the node for an entry
1233 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1237 * The array of struct pages for flatmem.
1238 * It must be declared for SPARSEMEM as well because there are configurations
1239 * that rely on that.
1241 extern struct page *mem_map;
1243 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1244 struct deferred_split {
1245 spinlock_t split_queue_lock;
1246 struct list_head split_queue;
1247 unsigned long split_queue_len;
1251 #ifdef CONFIG_MEMORY_FAILURE
1253 * Per NUMA node memory failure handling statistics.
1255 struct memory_failure_stats {
1257 * Number of raw pages poisoned.
1258 * Cases not accounted: memory outside kernel control, offline page,
1259 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1260 * error events, and unpoison actions from hwpoison_unpoison.
1262 unsigned long total;
1264 * Recovery results of poisoned raw pages handled by memory_failure,
1265 * in sync with mf_result.
1266 * total = ignored + failed + delayed + recovered.
1267 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1269 unsigned long ignored;
1270 unsigned long failed;
1271 unsigned long delayed;
1272 unsigned long recovered;
1277 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1278 * it's memory layout. On UMA machines there is a single pglist_data which
1279 * describes the whole memory.
1281 * Memory statistics and page replacement data structures are maintained on a
1284 typedef struct pglist_data {
1286 * node_zones contains just the zones for THIS node. Not all of the
1287 * zones may be populated, but it is the full list. It is referenced by
1288 * this node's node_zonelists as well as other node's node_zonelists.
1290 struct zone node_zones[MAX_NR_ZONES];
1293 * node_zonelists contains references to all zones in all nodes.
1294 * Generally the first zones will be references to this node's
1297 struct zonelist node_zonelists[MAX_ZONELISTS];
1299 int nr_zones; /* number of populated zones in this node */
1300 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1301 struct page *node_mem_map;
1302 #ifdef CONFIG_PAGE_EXTENSION
1303 struct page_ext *node_page_ext;
1306 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1308 * Must be held any time you expect node_start_pfn,
1309 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1310 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1313 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1314 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1315 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1317 * Nests above zone->lock and zone->span_seqlock
1319 spinlock_t node_size_lock;
1321 unsigned long node_start_pfn;
1322 unsigned long node_present_pages; /* total number of physical pages */
1323 unsigned long node_spanned_pages; /* total size of physical page
1324 range, including holes */
1326 wait_queue_head_t kswapd_wait;
1327 wait_queue_head_t pfmemalloc_wait;
1329 /* workqueues for throttling reclaim for different reasons. */
1330 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1332 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1333 unsigned long nr_reclaim_start; /* nr pages written while throttled
1334 * when throttling started. */
1335 #ifdef CONFIG_MEMORY_HOTPLUG
1336 struct mutex kswapd_lock;
1338 struct task_struct *kswapd; /* Protected by kswapd_lock */
1340 enum zone_type kswapd_highest_zoneidx;
1342 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1344 #ifdef CONFIG_COMPACTION
1345 int kcompactd_max_order;
1346 enum zone_type kcompactd_highest_zoneidx;
1347 wait_queue_head_t kcompactd_wait;
1348 struct task_struct *kcompactd;
1349 bool proactive_compact_trigger;
1352 * This is a per-node reserve of pages that are not available
1353 * to userspace allocations.
1355 unsigned long totalreserve_pages;
1359 * node reclaim becomes active if more unmapped pages exist.
1361 unsigned long min_unmapped_pages;
1362 unsigned long min_slab_pages;
1363 #endif /* CONFIG_NUMA */
1365 /* Write-intensive fields used by page reclaim */
1366 CACHELINE_PADDING(_pad1_);
1368 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1370 * If memory initialisation on large machines is deferred then this
1371 * is the first PFN that needs to be initialised.
1373 unsigned long first_deferred_pfn;
1374 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1376 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1377 struct deferred_split deferred_split_queue;
1380 #ifdef CONFIG_NUMA_BALANCING
1381 /* start time in ms of current promote rate limit period */
1382 unsigned int nbp_rl_start;
1383 /* number of promote candidate pages at start time of current rate limit period */
1384 unsigned long nbp_rl_nr_cand;
1385 /* promote threshold in ms */
1386 unsigned int nbp_threshold;
1387 /* start time in ms of current promote threshold adjustment period */
1388 unsigned int nbp_th_start;
1390 * number of promote candidate pages at start time of current promote
1391 * threshold adjustment period
1393 unsigned long nbp_th_nr_cand;
1395 /* Fields commonly accessed by the page reclaim scanner */
1398 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1400 * Use mem_cgroup_lruvec() to look up lruvecs.
1402 struct lruvec __lruvec;
1404 unsigned long flags;
1406 #ifdef CONFIG_LRU_GEN
1407 /* kswap mm walk data */
1408 struct lru_gen_mm_walk mm_walk;
1409 /* lru_gen_folio list */
1410 struct lru_gen_memcg memcg_lru;
1413 CACHELINE_PADDING(_pad2_);
1415 /* Per-node vmstats */
1416 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1417 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1419 struct memory_tier __rcu *memtier;
1421 #ifdef CONFIG_MEMORY_FAILURE
1422 struct memory_failure_stats mf_stats;
1426 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1427 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1429 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1430 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1432 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1434 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1437 #include <linux/memory_hotplug.h>
1439 void build_all_zonelists(pg_data_t *pgdat);
1440 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1441 enum zone_type highest_zoneidx);
1442 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1443 int highest_zoneidx, unsigned int alloc_flags,
1445 bool zone_watermark_ok(struct zone *z, unsigned int order,
1446 unsigned long mark, int highest_zoneidx,
1447 unsigned int alloc_flags);
1448 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1449 unsigned long mark, int highest_zoneidx);
1451 * Memory initialization context, use to differentiate memory added by
1452 * the platform statically or via memory hotplug interface.
1454 enum meminit_context {
1459 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1460 unsigned long size);
1462 extern void lruvec_init(struct lruvec *lruvec);
1464 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1467 return lruvec->pgdat;
1469 return container_of(lruvec, struct pglist_data, __lruvec);
1473 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1474 int local_memory_node(int node_id);
1476 static inline int local_memory_node(int node_id) { return node_id; };
1480 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1482 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1484 #ifdef CONFIG_ZONE_DEVICE
1485 static inline bool zone_is_zone_device(struct zone *zone)
1487 return zone_idx(zone) == ZONE_DEVICE;
1490 static inline bool zone_is_zone_device(struct zone *zone)
1497 * Returns true if a zone has pages managed by the buddy allocator.
1498 * All the reclaim decisions have to use this function rather than
1499 * populated_zone(). If the whole zone is reserved then we can easily
1500 * end up with populated_zone() && !managed_zone().
1502 static inline bool managed_zone(struct zone *zone)
1504 return zone_managed_pages(zone);
1507 /* Returns true if a zone has memory */
1508 static inline bool populated_zone(struct zone *zone)
1510 return zone->present_pages;
1514 static inline int zone_to_nid(struct zone *zone)
1519 static inline void zone_set_nid(struct zone *zone, int nid)
1524 static inline int zone_to_nid(struct zone *zone)
1529 static inline void zone_set_nid(struct zone *zone, int nid) {}
1532 extern int movable_zone;
1534 static inline int is_highmem_idx(enum zone_type idx)
1536 #ifdef CONFIG_HIGHMEM
1537 return (idx == ZONE_HIGHMEM ||
1538 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1545 * is_highmem - helper function to quickly check if a struct zone is a
1546 * highmem zone or not. This is an attempt to keep references
1547 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1548 * @zone: pointer to struct zone variable
1549 * Return: 1 for a highmem zone, 0 otherwise
1551 static inline int is_highmem(struct zone *zone)
1553 return is_highmem_idx(zone_idx(zone));
1556 #ifdef CONFIG_ZONE_DMA
1557 bool has_managed_dma(void);
1559 static inline bool has_managed_dma(void)
1568 extern struct pglist_data contig_page_data;
1569 static inline struct pglist_data *NODE_DATA(int nid)
1571 return &contig_page_data;
1574 #else /* CONFIG_NUMA */
1576 #include <asm/mmzone.h>
1578 #endif /* !CONFIG_NUMA */
1580 extern struct pglist_data *first_online_pgdat(void);
1581 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1582 extern struct zone *next_zone(struct zone *zone);
1585 * for_each_online_pgdat - helper macro to iterate over all online nodes
1586 * @pgdat: pointer to a pg_data_t variable
1588 #define for_each_online_pgdat(pgdat) \
1589 for (pgdat = first_online_pgdat(); \
1591 pgdat = next_online_pgdat(pgdat))
1593 * for_each_zone - helper macro to iterate over all memory zones
1594 * @zone: pointer to struct zone variable
1596 * The user only needs to declare the zone variable, for_each_zone
1599 #define for_each_zone(zone) \
1600 for (zone = (first_online_pgdat())->node_zones; \
1602 zone = next_zone(zone))
1604 #define for_each_populated_zone(zone) \
1605 for (zone = (first_online_pgdat())->node_zones; \
1607 zone = next_zone(zone)) \
1608 if (!populated_zone(zone)) \
1609 ; /* do nothing */ \
1612 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1614 return zoneref->zone;
1617 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1619 return zoneref->zone_idx;
1622 static inline int zonelist_node_idx(struct zoneref *zoneref)
1624 return zone_to_nid(zoneref->zone);
1627 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1628 enum zone_type highest_zoneidx,
1632 * 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
1633 * @z: The cursor used as a starting point for the search
1634 * @highest_zoneidx: The zone index of the highest zone to return
1635 * @nodes: An optional nodemask to filter the zonelist with
1637 * This function returns the next zone at or below a given zone index that is
1638 * within the allowed nodemask using a cursor as the starting point for the
1639 * search. The zoneref returned is a cursor that represents the current zone
1640 * being examined. It should be advanced by one before calling
1641 * next_zones_zonelist again.
1643 * Return: the next zone at or below highest_zoneidx within the allowed
1644 * nodemask using a cursor within a zonelist as a starting point
1646 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1647 enum zone_type highest_zoneidx,
1650 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1652 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1656 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1657 * @zonelist: The zonelist to search for a suitable zone
1658 * @highest_zoneidx: The zone index of the highest zone to return
1659 * @nodes: An optional nodemask to filter the zonelist with
1661 * This function returns the first zone at or below a given zone index that is
1662 * within the allowed nodemask. The zoneref returned is a cursor that can be
1663 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1664 * one before calling.
1666 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1667 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1668 * update due to cpuset modification.
1670 * Return: Zoneref pointer for the first suitable zone found
1672 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1673 enum zone_type highest_zoneidx,
1676 return next_zones_zonelist(zonelist->_zonerefs,
1677 highest_zoneidx, nodes);
1681 * 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
1682 * @zone: The current zone in the iterator
1683 * @z: The current pointer within zonelist->_zonerefs being iterated
1684 * @zlist: The zonelist being iterated
1685 * @highidx: The zone index of the highest zone to return
1686 * @nodemask: Nodemask allowed by the allocator
1688 * This iterator iterates though all zones at or below a given zone index and
1689 * within a given nodemask
1691 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1692 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1694 z = next_zones_zonelist(++z, highidx, nodemask), \
1695 zone = zonelist_zone(z))
1697 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1698 for (zone = z->zone; \
1700 z = next_zones_zonelist(++z, highidx, nodemask), \
1701 zone = zonelist_zone(z))
1705 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1706 * @zone: The current zone in the iterator
1707 * @z: The current pointer within zonelist->zones being iterated
1708 * @zlist: The zonelist being iterated
1709 * @highidx: The zone index of the highest zone to return
1711 * This iterator iterates though all zones at or below a given zone index.
1713 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1714 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1716 /* Whether the 'nodes' are all movable nodes */
1717 static inline bool movable_only_nodes(nodemask_t *nodes)
1719 struct zonelist *zonelist;
1723 if (nodes_empty(*nodes))
1727 * We can chose arbitrary node from the nodemask to get a
1728 * zonelist as they are interlinked. We just need to find
1729 * at least one zone that can satisfy kernel allocations.
1731 nid = first_node(*nodes);
1732 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1733 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1734 return (!z->zone) ? true : false;
1738 #ifdef CONFIG_SPARSEMEM
1739 #include <asm/sparsemem.h>
1742 #ifdef CONFIG_FLATMEM
1743 #define pfn_to_nid(pfn) (0)
1746 #ifdef CONFIG_SPARSEMEM
1749 * PA_SECTION_SHIFT physical address to/from section number
1750 * PFN_SECTION_SHIFT pfn to/from section number
1752 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1753 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1755 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1757 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1758 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1760 #define SECTION_BLOCKFLAGS_BITS \
1761 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1763 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1764 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1767 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1769 return pfn >> PFN_SECTION_SHIFT;
1771 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1773 return sec << PFN_SECTION_SHIFT;
1776 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1777 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1779 #define SUBSECTION_SHIFT 21
1780 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1782 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1783 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1784 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1786 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1787 #error Subsection size exceeds section size
1789 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1792 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1793 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1795 struct mem_section_usage {
1796 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1797 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1799 /* See declaration of similar field in struct zone */
1800 unsigned long pageblock_flags[0];
1803 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1807 struct mem_section {
1809 * This is, logically, a pointer to an array of struct
1810 * pages. However, it is stored with some other magic.
1811 * (see sparse.c::sparse_init_one_section())
1813 * Additionally during early boot we encode node id of
1814 * the location of the section here to guide allocation.
1815 * (see sparse.c::memory_present())
1817 * Making it a UL at least makes someone do a cast
1818 * before using it wrong.
1820 unsigned long section_mem_map;
1822 struct mem_section_usage *usage;
1823 #ifdef CONFIG_PAGE_EXTENSION
1825 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1826 * section. (see page_ext.h about this.)
1828 struct page_ext *page_ext;
1832 * WARNING: mem_section must be a power-of-2 in size for the
1833 * calculation and use of SECTION_ROOT_MASK to make sense.
1837 #ifdef CONFIG_SPARSEMEM_EXTREME
1838 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1840 #define SECTIONS_PER_ROOT 1
1843 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1844 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1845 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1847 #ifdef CONFIG_SPARSEMEM_EXTREME
1848 extern struct mem_section **mem_section;
1850 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1853 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1855 return ms->usage->pageblock_flags;
1858 static inline struct mem_section *__nr_to_section(unsigned long nr)
1860 unsigned long root = SECTION_NR_TO_ROOT(nr);
1862 if (unlikely(root >= NR_SECTION_ROOTS))
1865 #ifdef CONFIG_SPARSEMEM_EXTREME
1866 if (!mem_section || !mem_section[root])
1869 return &mem_section[root][nr & SECTION_ROOT_MASK];
1871 extern size_t mem_section_usage_size(void);
1874 * We use the lower bits of the mem_map pointer to store
1875 * a little bit of information. The pointer is calculated
1876 * as mem_map - section_nr_to_pfn(pnum). The result is
1877 * aligned to the minimum alignment of the two values:
1878 * 1. All mem_map arrays are page-aligned.
1879 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1880 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1881 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1882 * worst combination is powerpc with 256k pages,
1883 * which results in PFN_SECTION_SHIFT equal 6.
1884 * To sum it up, at least 6 bits are available on all architectures.
1885 * However, we can exceed 6 bits on some other architectures except
1886 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1887 * with the worst case of 64K pages on arm64) if we make sure the
1888 * exceeded bit is not applicable to powerpc.
1891 SECTION_MARKED_PRESENT_BIT,
1892 SECTION_HAS_MEM_MAP_BIT,
1893 SECTION_IS_ONLINE_BIT,
1894 SECTION_IS_EARLY_BIT,
1895 #ifdef CONFIG_ZONE_DEVICE
1896 SECTION_TAINT_ZONE_DEVICE_BIT,
1898 SECTION_MAP_LAST_BIT,
1901 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1902 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1903 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1904 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1905 #ifdef CONFIG_ZONE_DEVICE
1906 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1908 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1909 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1911 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1913 unsigned long map = section->section_mem_map;
1914 map &= SECTION_MAP_MASK;
1915 return (struct page *)map;
1918 static inline int present_section(struct mem_section *section)
1920 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1923 static inline int present_section_nr(unsigned long nr)
1925 return present_section(__nr_to_section(nr));
1928 static inline int valid_section(struct mem_section *section)
1930 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1933 static inline int early_section(struct mem_section *section)
1935 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1938 static inline int valid_section_nr(unsigned long nr)
1940 return valid_section(__nr_to_section(nr));
1943 static inline int online_section(struct mem_section *section)
1945 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1948 #ifdef CONFIG_ZONE_DEVICE
1949 static inline int online_device_section(struct mem_section *section)
1951 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1953 return section && ((section->section_mem_map & flags) == flags);
1956 static inline int online_device_section(struct mem_section *section)
1962 static inline int online_section_nr(unsigned long nr)
1964 return online_section(__nr_to_section(nr));
1967 #ifdef CONFIG_MEMORY_HOTPLUG
1968 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1969 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1972 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1974 return __nr_to_section(pfn_to_section_nr(pfn));
1977 extern unsigned long __highest_present_section_nr;
1979 static inline int subsection_map_index(unsigned long pfn)
1981 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1984 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1985 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1987 int idx = subsection_map_index(pfn);
1989 return test_bit(idx, ms->usage->subsection_map);
1992 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1998 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
2000 * pfn_valid - check if there is a valid memory map entry for a PFN
2001 * @pfn: the page frame number to check
2003 * Check if there is a valid memory map entry aka struct page for the @pfn.
2004 * Note, that availability of the memory map entry does not imply that
2005 * there is actual usable memory at that @pfn. The struct page may
2006 * represent a hole or an unusable page frame.
2008 * Return: 1 for PFNs that have memory map entries and 0 otherwise
2010 static inline int pfn_valid(unsigned long pfn)
2012 struct mem_section *ms;
2015 * Ensure the upper PAGE_SHIFT bits are clear in the
2016 * pfn. Else it might lead to false positives when
2017 * some of the upper bits are set, but the lower bits
2018 * match a valid pfn.
2020 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2023 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2025 ms = __pfn_to_section(pfn);
2026 if (!valid_section(ms))
2029 * Traditionally early sections always returned pfn_valid() for
2030 * the entire section-sized span.
2032 return early_section(ms) || pfn_section_valid(ms, pfn);
2036 static inline int pfn_in_present_section(unsigned long pfn)
2038 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2040 return present_section(__pfn_to_section(pfn));
2043 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2045 while (++section_nr <= __highest_present_section_nr) {
2046 if (present_section_nr(section_nr))
2054 * These are _only_ used during initialisation, therefore they
2055 * can use __initdata ... They could have names to indicate
2059 #define pfn_to_nid(pfn) \
2061 unsigned long __pfn_to_nid_pfn = (pfn); \
2062 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2065 #define pfn_to_nid(pfn) (0)
2068 void sparse_init(void);
2070 #define sparse_init() do {} while (0)
2071 #define sparse_index_init(_sec, _nid) do {} while (0)
2072 #define pfn_in_present_section pfn_valid
2073 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2074 #endif /* CONFIG_SPARSEMEM */
2076 #endif /* !__GENERATING_BOUNDS.H */
2077 #endif /* !__ASSEMBLY__ */
2078 #endif /* _LINUX_MMZONE_H */