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
30 #define MAX_PAGE_ORDER 10
32 #define MAX_PAGE_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
34 #define MAX_ORDER_NR_PAGES (1 << MAX_PAGE_ORDER)
36 #define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
38 #define NR_PAGE_ORDERS (MAX_PAGE_ORDER + 1)
41 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
42 * costly to service. That is between allocation orders which should
43 * coalesce naturally under reasonable reclaim pressure and those which
46 #define PAGE_ALLOC_COSTLY_ORDER 3
52 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
53 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
56 * MIGRATE_CMA migration type is designed to mimic the way
57 * ZONE_MOVABLE works. Only movable pages can be allocated
58 * from MIGRATE_CMA pageblocks and page allocator never
59 * implicitly change migration type of MIGRATE_CMA pageblock.
61 * The way to use it is to change migratetype of a range of
62 * pageblocks to MIGRATE_CMA which can be done by
63 * __free_pageblock_cma() function.
67 #ifdef CONFIG_MEMORY_ISOLATION
68 MIGRATE_ISOLATE, /* can't allocate from here */
73 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
74 extern const char * const migratetype_names[MIGRATE_TYPES];
77 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
78 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
80 # define is_migrate_cma(migratetype) false
81 # define is_migrate_cma_page(_page) false
84 static inline bool is_migrate_movable(int mt)
86 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
90 * Check whether a migratetype can be merged with another migratetype.
92 * It is only mergeable when it can fall back to other migratetypes for
93 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
95 static inline bool migratetype_is_mergeable(int mt)
97 return mt < MIGRATE_PCPTYPES;
100 #define for_each_migratetype_order(order, type) \
101 for (order = 0; order < NR_PAGE_ORDERS; order++) \
102 for (type = 0; type < MIGRATE_TYPES; type++)
104 extern int page_group_by_mobility_disabled;
106 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
108 #define get_pageblock_migratetype(page) \
109 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
111 #define folio_migratetype(folio) \
112 get_pfnblock_flags_mask(&folio->page, folio_pfn(folio), \
115 struct list_head free_list[MIGRATE_TYPES];
116 unsigned long nr_free;
122 enum numa_stat_item {
123 NUMA_HIT, /* allocated in intended node */
124 NUMA_MISS, /* allocated in non intended node */
125 NUMA_FOREIGN, /* was intended here, hit elsewhere */
126 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
127 NUMA_LOCAL, /* allocation from local node */
128 NUMA_OTHER, /* allocation from other node */
129 NR_VM_NUMA_EVENT_ITEMS
132 #define NR_VM_NUMA_EVENT_ITEMS 0
135 enum zone_stat_item {
136 /* First 128 byte cacheline (assuming 64 bit words) */
138 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
139 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
141 NR_ZONE_INACTIVE_FILE,
144 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
145 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
146 /* Second 128 byte cacheline */
148 #if IS_ENABLED(CONFIG_ZSMALLOC)
149 NR_ZSPAGES, /* allocated in zsmalloc */
152 #ifdef CONFIG_UNACCEPTED_MEMORY
155 NR_VM_ZONE_STAT_ITEMS };
157 enum node_stat_item {
159 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
160 NR_ACTIVE_ANON, /* " " " " " */
161 NR_INACTIVE_FILE, /* " " " " " */
162 NR_ACTIVE_FILE, /* " " " " " */
163 NR_UNEVICTABLE, /* " " " " " */
164 NR_SLAB_RECLAIMABLE_B,
165 NR_SLAB_UNRECLAIMABLE_B,
166 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
167 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
169 WORKINGSET_REFAULT_BASE,
170 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
171 WORKINGSET_REFAULT_FILE,
172 WORKINGSET_ACTIVATE_BASE,
173 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
174 WORKINGSET_ACTIVATE_FILE,
175 WORKINGSET_RESTORE_BASE,
176 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
177 WORKINGSET_RESTORE_FILE,
178 WORKINGSET_NODERECLAIM,
179 NR_ANON_MAPPED, /* Mapped anonymous pages */
180 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
181 only modified from process context */
185 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
186 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
193 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
194 NR_DIRTIED, /* page dirtyings since bootup */
195 NR_WRITTEN, /* page writings since bootup */
196 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
197 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
198 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
199 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
200 NR_KERNEL_STACK_KB, /* measured in KiB */
201 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
202 NR_KERNEL_SCS_KB, /* measured in KiB */
204 NR_PAGETABLE, /* used for pagetables */
205 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
209 #ifdef CONFIG_NUMA_BALANCING
210 PGPROMOTE_SUCCESS, /* promote successfully */
211 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
213 /* PGDEMOTE_*: pages demoted */
217 NR_VM_NODE_STAT_ITEMS
221 * Returns true if the item should be printed in THPs (/proc/vmstat
222 * currently prints number of anon, file and shmem THPs. But the item
223 * is charged in pages).
225 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
227 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
230 return item == NR_ANON_THPS ||
231 item == NR_FILE_THPS ||
232 item == NR_SHMEM_THPS ||
233 item == NR_SHMEM_PMDMAPPED ||
234 item == NR_FILE_PMDMAPPED;
238 * Returns true if the value is measured in bytes (most vmstat values are
239 * measured in pages). This defines the API part, the internal representation
240 * might be different.
242 static __always_inline bool vmstat_item_in_bytes(int idx)
245 * Global and per-node slab counters track slab pages.
246 * It's expected that changes are multiples of PAGE_SIZE.
247 * Internally values are stored in pages.
249 * Per-memcg and per-lruvec counters track memory, consumed
250 * by individual slab objects. These counters are actually
253 return (idx == NR_SLAB_RECLAIMABLE_B ||
254 idx == NR_SLAB_UNRECLAIMABLE_B);
258 * We do arithmetic on the LRU lists in various places in the code,
259 * so it is important to keep the active lists LRU_ACTIVE higher in
260 * the array than the corresponding inactive lists, and to keep
261 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
263 * This has to be kept in sync with the statistics in zone_stat_item
264 * above and the descriptions in vmstat_text in mm/vmstat.c
271 LRU_INACTIVE_ANON = LRU_BASE,
272 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
273 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
274 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
279 enum vmscan_throttle_state {
280 VMSCAN_THROTTLE_WRITEBACK,
281 VMSCAN_THROTTLE_ISOLATED,
282 VMSCAN_THROTTLE_NOPROGRESS,
283 VMSCAN_THROTTLE_CONGESTED,
287 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
289 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
291 static inline bool is_file_lru(enum lru_list lru)
293 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
296 static inline bool is_active_lru(enum lru_list lru)
298 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
301 #define WORKINGSET_ANON 0
302 #define WORKINGSET_FILE 1
303 #define ANON_AND_FILE 2
307 * An lruvec has many dirty pages backed by a congested BDI:
308 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
309 * It can be cleared by cgroup reclaim or kswapd.
310 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
311 * It can only be cleared by kswapd.
313 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
314 * reclaim, but not vice versa. This only applies to the root cgroup.
315 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
316 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
319 LRUVEC_CGROUP_CONGESTED,
320 LRUVEC_NODE_CONGESTED,
323 #endif /* !__GENERATING_BOUNDS_H */
326 * Evictable pages are divided into multiple generations. The youngest and the
327 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
328 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
329 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
330 * corresponding generation. The gen counter in folio->flags stores gen+1 while
331 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
333 * A page is added to the youngest generation on faulting. The aging needs to
334 * check the accessed bit at least twice before handing this page over to the
335 * eviction. The first check takes care of the accessed bit set on the initial
336 * fault; the second check makes sure this page hasn't been used since then.
337 * This process, AKA second chance, requires a minimum of two generations,
338 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
339 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
340 * rest of generations, if they exist, are considered inactive. See
341 * lru_gen_is_active().
343 * PG_active is always cleared while a page is on one of lrugen->folios[] so
344 * that the aging needs not to worry about it. And it's set again when a page
345 * considered active is isolated for non-reclaiming purposes, e.g., migration.
346 * See lru_gen_add_folio() and lru_gen_del_folio().
348 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
349 * number of categories of the active/inactive LRU when keeping track of
350 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
353 #define MIN_NR_GENS 2U
354 #define MAX_NR_GENS 4U
357 * Each generation is divided into multiple tiers. A page accessed N times
358 * through file descriptors is in tier order_base_2(N). A page in the first tier
359 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
360 * tables or read ahead. A page in any other tier (N>1) is marked by
361 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
362 * supported without using additional bits in folio->flags.
364 * In contrast to moving across generations which requires the LRU lock, moving
365 * across tiers only involves atomic operations on folio->flags and therefore
366 * has a negligible cost in the buffered access path. In the eviction path,
367 * comparisons of refaulted/(evicted+protected) from the first tier and the
368 * rest infer whether pages accessed multiple times through file descriptors
369 * are statistically hot and thus worth protecting.
371 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
372 * number of categories of the active/inactive LRU when keeping track of
373 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
376 #define MAX_NR_TIERS 4U
378 #ifndef __GENERATING_BOUNDS_H
381 struct page_vma_mapped_walk;
383 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
384 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
386 #ifdef CONFIG_LRU_GEN
396 LRU_GEN_NONLEAF_YOUNG,
400 #define MIN_LRU_BATCH BITS_PER_LONG
401 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
403 /* whether to keep historical stats from evicted generations */
404 #ifdef CONFIG_LRU_GEN_STATS
405 #define NR_HIST_GENS MAX_NR_GENS
407 #define NR_HIST_GENS 1U
411 * The youngest generation number is stored in max_seq for both anon and file
412 * types as they are aged on an equal footing. The oldest generation numbers are
413 * stored in min_seq[] separately for anon and file types as clean file pages
414 * can be evicted regardless of swap constraints.
416 * Normally anon and file min_seq are in sync. But if swapping is constrained,
417 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
420 * The number of pages in each generation is eventually consistent and therefore
421 * can be transiently negative when reset_batch_size() is pending.
423 struct lru_gen_folio {
424 /* the aging increments the youngest generation number */
425 unsigned long max_seq;
426 /* the eviction increments the oldest generation numbers */
427 unsigned long min_seq[ANON_AND_FILE];
428 /* the birth time of each generation in jiffies */
429 unsigned long timestamps[MAX_NR_GENS];
430 /* the multi-gen LRU lists, lazily sorted on eviction */
431 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
432 /* the multi-gen LRU sizes, eventually consistent */
433 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
434 /* the exponential moving average of refaulted */
435 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
436 /* the exponential moving average of evicted+protected */
437 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
438 /* the first tier doesn't need protection, hence the minus one */
439 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
440 /* can be modified without holding the LRU lock */
441 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
442 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
443 /* whether the multi-gen LRU is enabled */
445 /* the memcg generation this lru_gen_folio belongs to */
447 /* the list segment this lru_gen_folio belongs to */
449 /* per-node lru_gen_folio list for global reclaim */
450 struct hlist_nulls_node list;
454 MM_LEAF_TOTAL, /* total leaf entries */
455 MM_LEAF_OLD, /* old leaf entries */
456 MM_LEAF_YOUNG, /* young leaf entries */
457 MM_NONLEAF_TOTAL, /* total non-leaf entries */
458 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
459 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
463 /* double-buffering Bloom filters */
464 #define NR_BLOOM_FILTERS 2
466 struct lru_gen_mm_state {
467 /* set to max_seq after each iteration */
469 /* where the current iteration continues after */
470 struct list_head *head;
471 /* where the last iteration ended before */
472 struct list_head *tail;
473 /* Bloom filters flip after each iteration */
474 unsigned long *filters[NR_BLOOM_FILTERS];
475 /* the mm stats for debugging */
476 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
479 struct lru_gen_mm_walk {
480 /* the lruvec under reclaim */
481 struct lruvec *lruvec;
482 /* unstable max_seq from lru_gen_folio */
483 unsigned long max_seq;
484 /* the next address within an mm to scan */
485 unsigned long next_addr;
486 /* to batch promoted pages */
487 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
488 /* to batch the mm stats */
489 int mm_stats[NR_MM_STATS];
490 /* total batched items */
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);
553 void lru_gen_init_lruvec(struct lruvec *lruvec);
554 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
556 void lru_gen_init_memcg(struct mem_cgroup *memcg);
557 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
558 void lru_gen_online_memcg(struct mem_cgroup *memcg);
559 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
560 void lru_gen_release_memcg(struct mem_cgroup *memcg);
561 void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
563 #else /* !CONFIG_LRU_GEN */
565 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
569 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
573 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
577 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
581 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
585 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
589 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
593 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
597 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
601 #endif /* CONFIG_LRU_GEN */
604 struct list_head lists[NR_LRU_LISTS];
605 /* per lruvec lru_lock for memcg */
608 * These track the cost of reclaiming one LRU - file or anon -
609 * over the other. As the observed cost of reclaiming one LRU
610 * increases, the reclaim scan balance tips toward the other.
612 unsigned long anon_cost;
613 unsigned long file_cost;
614 /* Non-resident age, driven by LRU movement */
615 atomic_long_t nonresident_age;
616 /* Refaults at the time of last reclaim cycle */
617 unsigned long refaults[ANON_AND_FILE];
618 /* Various lruvec state flags (enum lruvec_flags) */
620 #ifdef CONFIG_LRU_GEN
621 /* evictable pages divided into generations */
622 struct lru_gen_folio lrugen;
623 #ifdef CONFIG_LRU_GEN_WALKS_MMU
624 /* to concurrently iterate lru_gen_mm_list */
625 struct lru_gen_mm_state mm_state;
627 #endif /* CONFIG_LRU_GEN */
629 struct pglist_data *pgdat;
631 struct zswap_lruvec_state zswap_lruvec_state;
634 /* Isolate for asynchronous migration */
635 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
636 /* Isolate unevictable pages */
637 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
639 /* LRU Isolation modes. */
640 typedef unsigned __bitwise isolate_mode_t;
642 enum zone_watermarks {
651 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
652 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
653 * it should not contribute to serious fragmentation causing THP allocation
656 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
661 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
662 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
664 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
665 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
666 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
667 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
670 * Flags used in pcp->flags field.
672 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
673 * previous page freeing. To avoid to drain PCP for an accident
674 * high-order page freeing.
676 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
677 * draining PCP for consecutive high-order pages freeing without
678 * allocation if data cache slice of CPU is large enough. To reduce
679 * zone lock contention and keep cache-hot pages reusing.
681 #define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
682 #define PCPF_FREE_HIGH_BATCH BIT(1)
684 struct per_cpu_pages {
685 spinlock_t lock; /* Protects lists field */
686 int count; /* number of pages in the list */
687 int high; /* high watermark, emptying needed */
688 int high_min; /* min high watermark */
689 int high_max; /* max high watermark */
690 int batch; /* chunk size for buddy add/remove */
691 u8 flags; /* protected by pcp->lock */
692 u8 alloc_factor; /* batch scaling factor during allocate */
694 u8 expire; /* When 0, remote pagesets are drained */
696 short free_count; /* consecutive free count */
698 /* Lists of pages, one per migrate type stored on the pcp-lists */
699 struct list_head lists[NR_PCP_LISTS];
700 } ____cacheline_aligned_in_smp;
702 struct per_cpu_zonestat {
704 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
709 * Low priority inaccurate counters that are only folded
710 * on demand. Use a large type to avoid the overhead of
711 * folding during refresh_cpu_vm_stats.
713 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
717 struct per_cpu_nodestat {
719 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
722 #endif /* !__GENERATING_BOUNDS.H */
726 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
727 * to DMA to all of the addressable memory (ZONE_NORMAL).
728 * On architectures where this area covers the whole 32 bit address
729 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
730 * DMA addressing constraints. This distinction is important as a 32bit
731 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
732 * platforms may need both zones as they support peripherals with
733 * different DMA addressing limitations.
735 #ifdef CONFIG_ZONE_DMA
738 #ifdef CONFIG_ZONE_DMA32
742 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
743 * performed on pages in ZONE_NORMAL if the DMA devices support
744 * transfers to all addressable memory.
747 #ifdef CONFIG_HIGHMEM
749 * A memory area that is only addressable by the kernel through
750 * mapping portions into its own address space. This is for example
751 * used by i386 to allow the kernel to address the memory beyond
752 * 900MB. The kernel will set up special mappings (page
753 * table entries on i386) for each page that the kernel needs to
759 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
760 * movable pages with few exceptional cases described below. Main use
761 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
762 * likely to succeed, and to locally limit unmovable allocations - e.g.,
763 * to increase the number of THP/huge pages. Notable special cases are:
765 * 1. Pinned pages: (long-term) pinning of movable pages might
766 * essentially turn such pages unmovable. Therefore, we do not allow
767 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
768 * faulted, they come from the right zone right away. However, it is
769 * still possible that address space already has pages in
770 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
771 * touches that memory before pinning). In such case we migrate them
772 * to a different zone. When migration fails - pinning fails.
773 * 2. memblock allocations: kernelcore/movablecore setups might create
774 * situations where ZONE_MOVABLE contains unmovable allocations
775 * after boot. Memory offlining and allocations fail early.
776 * 3. Memory holes: kernelcore/movablecore setups might create very rare
777 * situations where ZONE_MOVABLE contains memory holes after boot,
778 * for example, if we have sections that are only partially
779 * populated. Memory offlining and allocations fail early.
780 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
781 * memory offlining, such pages cannot be allocated.
782 * 5. Unmovable PG_offline pages: in paravirtualized environments,
783 * hotplugged memory blocks might only partially be managed by the
784 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
785 * parts not manged by the buddy are unmovable PG_offline pages. In
786 * some cases (virtio-mem), such pages can be skipped during
787 * memory offlining, however, cannot be moved/allocated. These
788 * techniques might use alloc_contig_range() to hide previously
789 * exposed pages from the buddy again (e.g., to implement some sort
790 * of memory unplug in virtio-mem).
791 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
792 * situations where ZERO_PAGE(0) which is allocated differently
793 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
794 * cannot be migrated.
795 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
796 * memory to the MOVABLE zone, the vmemmap pages are also placed in
797 * such zone. Such pages cannot be really moved around as they are
798 * self-stored in the range, but they are treated as movable when
799 * the range they describe is about to be offlined.
801 * In general, no unmovable allocations that degrade memory offlining
802 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
803 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
804 * if has_unmovable_pages() states that there are no unmovable pages,
805 * there can be false negatives).
808 #ifdef CONFIG_ZONE_DEVICE
815 #ifndef __GENERATING_BOUNDS_H
817 #define ASYNC_AND_SYNC 2
820 /* Read-mostly fields */
822 /* zone watermarks, access with *_wmark_pages(zone) macros */
823 unsigned long _watermark[NR_WMARK];
824 unsigned long watermark_boost;
826 unsigned long nr_reserved_highatomic;
829 * We don't know if the memory that we're going to allocate will be
830 * freeable or/and it will be released eventually, so to avoid totally
831 * wasting several GB of ram we must reserve some of the lower zone
832 * memory (otherwise we risk to run OOM on the lower zones despite
833 * there being tons of freeable ram on the higher zones). This array is
834 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
837 long lowmem_reserve[MAX_NR_ZONES];
842 struct pglist_data *zone_pgdat;
843 struct per_cpu_pages __percpu *per_cpu_pageset;
844 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
846 * the high and batch values are copied to individual pagesets for
849 int pageset_high_min;
850 int pageset_high_max;
853 #ifndef CONFIG_SPARSEMEM
855 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
856 * In SPARSEMEM, this map is stored in struct mem_section
858 unsigned long *pageblock_flags;
859 #endif /* CONFIG_SPARSEMEM */
861 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
862 unsigned long zone_start_pfn;
865 * spanned_pages is the total pages spanned by the zone, including
866 * holes, which is calculated as:
867 * spanned_pages = zone_end_pfn - zone_start_pfn;
869 * present_pages is physical pages existing within the zone, which
871 * present_pages = spanned_pages - absent_pages(pages in holes);
873 * present_early_pages is present pages existing within the zone
874 * located on memory available since early boot, excluding hotplugged
877 * managed_pages is present pages managed by the buddy system, which
878 * is calculated as (reserved_pages includes pages allocated by the
879 * bootmem allocator):
880 * managed_pages = present_pages - reserved_pages;
882 * cma pages is present pages that are assigned for CMA use
885 * So present_pages may be used by memory hotplug or memory power
886 * management logic to figure out unmanaged pages by checking
887 * (present_pages - managed_pages). And managed_pages should be used
888 * by page allocator and vm scanner to calculate all kinds of watermarks
893 * zone_start_pfn and spanned_pages are protected by span_seqlock.
894 * It is a seqlock because it has to be read outside of zone->lock,
895 * and it is done in the main allocator path. But, it is written
896 * quite infrequently.
898 * The span_seq lock is declared along with zone->lock because it is
899 * frequently read in proximity to zone->lock. It's good to
900 * give them a chance of being in the same cacheline.
902 * Write access to present_pages at runtime should be protected by
903 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
904 * present_pages should use get_online_mems() to get a stable value.
906 atomic_long_t managed_pages;
907 unsigned long spanned_pages;
908 unsigned long present_pages;
909 #if defined(CONFIG_MEMORY_HOTPLUG)
910 unsigned long present_early_pages;
913 unsigned long cma_pages;
918 #ifdef CONFIG_MEMORY_ISOLATION
920 * Number of isolated pageblock. It is used to solve incorrect
921 * freepage counting problem due to racy retrieving migratetype
922 * of pageblock. Protected by zone->lock.
924 unsigned long nr_isolate_pageblock;
927 #ifdef CONFIG_MEMORY_HOTPLUG
928 /* see spanned/present_pages for more description */
929 seqlock_t span_seqlock;
934 /* Write-intensive fields used from the page allocator */
935 CACHELINE_PADDING(_pad1_);
937 /* free areas of different sizes */
938 struct free_area free_area[NR_PAGE_ORDERS];
940 #ifdef CONFIG_UNACCEPTED_MEMORY
941 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
942 struct list_head unaccepted_pages;
945 /* zone flags, see below */
948 /* Primarily protects free_area */
951 /* Write-intensive fields used by compaction and vmstats. */
952 CACHELINE_PADDING(_pad2_);
955 * When free pages are below this point, additional steps are taken
956 * when reading the number of free pages to avoid per-cpu counter
957 * drift allowing watermarks to be breached
959 unsigned long percpu_drift_mark;
961 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
962 /* pfn where compaction free scanner should start */
963 unsigned long compact_cached_free_pfn;
964 /* pfn where compaction migration scanner should start */
965 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
966 unsigned long compact_init_migrate_pfn;
967 unsigned long compact_init_free_pfn;
970 #ifdef CONFIG_COMPACTION
972 * On compaction failure, 1<<compact_defer_shift compactions
973 * are skipped before trying again. The number attempted since
974 * last failure is tracked with compact_considered.
975 * compact_order_failed is the minimum compaction failed order.
977 unsigned int compact_considered;
978 unsigned int compact_defer_shift;
979 int compact_order_failed;
982 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
983 /* Set to true when the PG_migrate_skip bits should be cleared */
984 bool compact_blockskip_flush;
989 CACHELINE_PADDING(_pad3_);
990 /* Zone statistics */
991 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
992 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
993 } ____cacheline_internodealigned_in_smp;
996 PGDAT_DIRTY, /* reclaim scanning has recently found
997 * many dirty file pages at the tail
1000 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1001 * many pages under writeback
1003 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1007 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1008 * Cleared when kswapd is woken.
1010 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1011 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1014 static inline unsigned long zone_managed_pages(struct zone *zone)
1016 return (unsigned long)atomic_long_read(&zone->managed_pages);
1019 static inline unsigned long zone_cma_pages(struct zone *zone)
1022 return zone->cma_pages;
1028 static inline unsigned long zone_end_pfn(const struct zone *zone)
1030 return zone->zone_start_pfn + zone->spanned_pages;
1033 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1035 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1038 static inline bool zone_is_initialized(struct zone *zone)
1040 return zone->initialized;
1043 static inline bool zone_is_empty(struct zone *zone)
1045 return zone->spanned_pages == 0;
1048 #ifndef BUILD_VDSO32_64
1050 * The zone field is never updated after free_area_init_core()
1051 * sets it, so none of the operations on it need to be atomic.
1054 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1055 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1056 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1057 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1058 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1059 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1060 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1061 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1064 * Define the bit shifts to access each section. For non-existent
1065 * sections we define the shift as 0; that plus a 0 mask ensures
1066 * the compiler will optimise away reference to them.
1068 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1069 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1070 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1071 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1072 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1074 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1075 #ifdef NODE_NOT_IN_PAGE_FLAGS
1076 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1077 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1078 SECTIONS_PGOFF : ZONES_PGOFF)
1080 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1081 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1082 NODES_PGOFF : ZONES_PGOFF)
1085 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1087 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1088 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1089 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1090 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1091 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1092 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1094 static inline enum zone_type page_zonenum(const struct page *page)
1096 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1097 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1100 static inline enum zone_type folio_zonenum(const struct folio *folio)
1102 return page_zonenum(&folio->page);
1105 #ifdef CONFIG_ZONE_DEVICE
1106 static inline bool is_zone_device_page(const struct page *page)
1108 return page_zonenum(page) == ZONE_DEVICE;
1112 * Consecutive zone device pages should not be merged into the same sgl
1113 * or bvec segment with other types of pages or if they belong to different
1114 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1115 * without scanning the entire segment. This helper returns true either if
1116 * both pages are not zone device pages or both pages are zone device pages
1117 * with the same pgmap.
1119 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1120 const struct page *b)
1122 if (is_zone_device_page(a) != is_zone_device_page(b))
1124 if (!is_zone_device_page(a))
1126 return a->pgmap == b->pgmap;
1129 extern void memmap_init_zone_device(struct zone *, unsigned long,
1130 unsigned long, struct dev_pagemap *);
1132 static inline bool is_zone_device_page(const struct page *page)
1136 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1137 const struct page *b)
1143 static inline bool folio_is_zone_device(const struct folio *folio)
1145 return is_zone_device_page(&folio->page);
1148 static inline bool is_zone_movable_page(const struct page *page)
1150 return page_zonenum(page) == ZONE_MOVABLE;
1153 static inline bool folio_is_zone_movable(const struct folio *folio)
1155 return folio_zonenum(folio) == ZONE_MOVABLE;
1160 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1161 * intersection with the given zone
1163 static inline bool zone_intersects(struct zone *zone,
1164 unsigned long start_pfn, unsigned long nr_pages)
1166 if (zone_is_empty(zone))
1168 if (start_pfn >= zone_end_pfn(zone) ||
1169 start_pfn + nr_pages <= zone->zone_start_pfn)
1176 * The "priority" of VM scanning is how much of the queues we will scan in one
1177 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1178 * queues ("queue_length >> 12") during an aging round.
1180 #define DEF_PRIORITY 12
1182 /* Maximum number of zones on a zonelist */
1183 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1186 ZONELIST_FALLBACK, /* zonelist with fallback */
1189 * The NUMA zonelists are doubled because we need zonelists that
1190 * restrict the allocations to a single node for __GFP_THISNODE.
1192 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1198 * This struct contains information about a zone in a zonelist. It is stored
1199 * here to avoid dereferences into large structures and lookups of tables
1202 struct zone *zone; /* Pointer to actual zone */
1203 int zone_idx; /* zone_idx(zoneref->zone) */
1207 * One allocation request operates on a zonelist. A zonelist
1208 * is a list of zones, the first one is the 'goal' of the
1209 * allocation, the other zones are fallback zones, in decreasing
1212 * To speed the reading of the zonelist, the zonerefs contain the zone index
1213 * of the entry being read. Helper functions to access information given
1214 * a struct zoneref are
1216 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1217 * zonelist_zone_idx() - Return the index of the zone for an entry
1218 * zonelist_node_idx() - Return the index of the node for an entry
1221 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1225 * The array of struct pages for flatmem.
1226 * It must be declared for SPARSEMEM as well because there are configurations
1227 * that rely on that.
1229 extern struct page *mem_map;
1231 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1232 struct deferred_split {
1233 spinlock_t split_queue_lock;
1234 struct list_head split_queue;
1235 unsigned long split_queue_len;
1239 #ifdef CONFIG_MEMORY_FAILURE
1241 * Per NUMA node memory failure handling statistics.
1243 struct memory_failure_stats {
1245 * Number of raw pages poisoned.
1246 * Cases not accounted: memory outside kernel control, offline page,
1247 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1248 * error events, and unpoison actions from hwpoison_unpoison.
1250 unsigned long total;
1252 * Recovery results of poisoned raw pages handled by memory_failure,
1253 * in sync with mf_result.
1254 * total = ignored + failed + delayed + recovered.
1255 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1257 unsigned long ignored;
1258 unsigned long failed;
1259 unsigned long delayed;
1260 unsigned long recovered;
1265 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1266 * it's memory layout. On UMA machines there is a single pglist_data which
1267 * describes the whole memory.
1269 * Memory statistics and page replacement data structures are maintained on a
1272 typedef struct pglist_data {
1274 * node_zones contains just the zones for THIS node. Not all of the
1275 * zones may be populated, but it is the full list. It is referenced by
1276 * this node's node_zonelists as well as other node's node_zonelists.
1278 struct zone node_zones[MAX_NR_ZONES];
1281 * node_zonelists contains references to all zones in all nodes.
1282 * Generally the first zones will be references to this node's
1285 struct zonelist node_zonelists[MAX_ZONELISTS];
1287 int nr_zones; /* number of populated zones in this node */
1288 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1289 struct page *node_mem_map;
1290 #ifdef CONFIG_PAGE_EXTENSION
1291 struct page_ext *node_page_ext;
1294 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1296 * Must be held any time you expect node_start_pfn,
1297 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1298 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1301 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1302 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1303 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1305 * Nests above zone->lock and zone->span_seqlock
1307 spinlock_t node_size_lock;
1309 unsigned long node_start_pfn;
1310 unsigned long node_present_pages; /* total number of physical pages */
1311 unsigned long node_spanned_pages; /* total size of physical page
1312 range, including holes */
1314 wait_queue_head_t kswapd_wait;
1315 wait_queue_head_t pfmemalloc_wait;
1317 /* workqueues for throttling reclaim for different reasons. */
1318 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1320 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1321 unsigned long nr_reclaim_start; /* nr pages written while throttled
1322 * when throttling started. */
1323 #ifdef CONFIG_MEMORY_HOTPLUG
1324 struct mutex kswapd_lock;
1326 struct task_struct *kswapd; /* Protected by kswapd_lock */
1328 enum zone_type kswapd_highest_zoneidx;
1330 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1332 #ifdef CONFIG_COMPACTION
1333 int kcompactd_max_order;
1334 enum zone_type kcompactd_highest_zoneidx;
1335 wait_queue_head_t kcompactd_wait;
1336 struct task_struct *kcompactd;
1337 bool proactive_compact_trigger;
1340 * This is a per-node reserve of pages that are not available
1341 * to userspace allocations.
1343 unsigned long totalreserve_pages;
1347 * node reclaim becomes active if more unmapped pages exist.
1349 unsigned long min_unmapped_pages;
1350 unsigned long min_slab_pages;
1351 #endif /* CONFIG_NUMA */
1353 /* Write-intensive fields used by page reclaim */
1354 CACHELINE_PADDING(_pad1_);
1356 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1358 * If memory initialisation on large machines is deferred then this
1359 * is the first PFN that needs to be initialised.
1361 unsigned long first_deferred_pfn;
1362 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1364 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1365 struct deferred_split deferred_split_queue;
1368 #ifdef CONFIG_NUMA_BALANCING
1369 /* start time in ms of current promote rate limit period */
1370 unsigned int nbp_rl_start;
1371 /* number of promote candidate pages at start time of current rate limit period */
1372 unsigned long nbp_rl_nr_cand;
1373 /* promote threshold in ms */
1374 unsigned int nbp_threshold;
1375 /* start time in ms of current promote threshold adjustment period */
1376 unsigned int nbp_th_start;
1378 * number of promote candidate pages at start time of current promote
1379 * threshold adjustment period
1381 unsigned long nbp_th_nr_cand;
1383 /* Fields commonly accessed by the page reclaim scanner */
1386 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1388 * Use mem_cgroup_lruvec() to look up lruvecs.
1390 struct lruvec __lruvec;
1392 unsigned long flags;
1394 #ifdef CONFIG_LRU_GEN
1395 /* kswap mm walk data */
1396 struct lru_gen_mm_walk mm_walk;
1397 /* lru_gen_folio list */
1398 struct lru_gen_memcg memcg_lru;
1401 CACHELINE_PADDING(_pad2_);
1403 /* Per-node vmstats */
1404 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1405 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1407 struct memory_tier __rcu *memtier;
1409 #ifdef CONFIG_MEMORY_FAILURE
1410 struct memory_failure_stats mf_stats;
1414 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1415 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1417 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1418 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1420 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1422 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1425 #include <linux/memory_hotplug.h>
1427 void build_all_zonelists(pg_data_t *pgdat);
1428 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1429 enum zone_type highest_zoneidx);
1430 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1431 int highest_zoneidx, unsigned int alloc_flags,
1433 bool zone_watermark_ok(struct zone *z, unsigned int order,
1434 unsigned long mark, int highest_zoneidx,
1435 unsigned int alloc_flags);
1436 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1437 unsigned long mark, int highest_zoneidx);
1439 * Memory initialization context, use to differentiate memory added by
1440 * the platform statically or via memory hotplug interface.
1442 enum meminit_context {
1447 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1448 unsigned long size);
1450 extern void lruvec_init(struct lruvec *lruvec);
1452 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1455 return lruvec->pgdat;
1457 return container_of(lruvec, struct pglist_data, __lruvec);
1461 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1462 int local_memory_node(int node_id);
1464 static inline int local_memory_node(int node_id) { return node_id; };
1468 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1470 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1472 #ifdef CONFIG_ZONE_DEVICE
1473 static inline bool zone_is_zone_device(struct zone *zone)
1475 return zone_idx(zone) == ZONE_DEVICE;
1478 static inline bool zone_is_zone_device(struct zone *zone)
1485 * Returns true if a zone has pages managed by the buddy allocator.
1486 * All the reclaim decisions have to use this function rather than
1487 * populated_zone(). If the whole zone is reserved then we can easily
1488 * end up with populated_zone() && !managed_zone().
1490 static inline bool managed_zone(struct zone *zone)
1492 return zone_managed_pages(zone);
1495 /* Returns true if a zone has memory */
1496 static inline bool populated_zone(struct zone *zone)
1498 return zone->present_pages;
1502 static inline int zone_to_nid(struct zone *zone)
1507 static inline void zone_set_nid(struct zone *zone, int nid)
1512 static inline int zone_to_nid(struct zone *zone)
1517 static inline void zone_set_nid(struct zone *zone, int nid) {}
1520 extern int movable_zone;
1522 static inline int is_highmem_idx(enum zone_type idx)
1524 #ifdef CONFIG_HIGHMEM
1525 return (idx == ZONE_HIGHMEM ||
1526 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1533 * is_highmem - helper function to quickly check if a struct zone is a
1534 * highmem zone or not. This is an attempt to keep references
1535 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1536 * @zone: pointer to struct zone variable
1537 * Return: 1 for a highmem zone, 0 otherwise
1539 static inline int is_highmem(struct zone *zone)
1541 return is_highmem_idx(zone_idx(zone));
1544 #ifdef CONFIG_ZONE_DMA
1545 bool has_managed_dma(void);
1547 static inline bool has_managed_dma(void)
1556 extern struct pglist_data contig_page_data;
1557 static inline struct pglist_data *NODE_DATA(int nid)
1559 return &contig_page_data;
1562 #else /* CONFIG_NUMA */
1564 #include <asm/mmzone.h>
1566 #endif /* !CONFIG_NUMA */
1568 extern struct pglist_data *first_online_pgdat(void);
1569 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1570 extern struct zone *next_zone(struct zone *zone);
1573 * for_each_online_pgdat - helper macro to iterate over all online nodes
1574 * @pgdat: pointer to a pg_data_t variable
1576 #define for_each_online_pgdat(pgdat) \
1577 for (pgdat = first_online_pgdat(); \
1579 pgdat = next_online_pgdat(pgdat))
1581 * for_each_zone - helper macro to iterate over all memory zones
1582 * @zone: pointer to struct zone variable
1584 * The user only needs to declare the zone variable, for_each_zone
1587 #define for_each_zone(zone) \
1588 for (zone = (first_online_pgdat())->node_zones; \
1590 zone = next_zone(zone))
1592 #define for_each_populated_zone(zone) \
1593 for (zone = (first_online_pgdat())->node_zones; \
1595 zone = next_zone(zone)) \
1596 if (!populated_zone(zone)) \
1597 ; /* do nothing */ \
1600 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1602 return zoneref->zone;
1605 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1607 return zoneref->zone_idx;
1610 static inline int zonelist_node_idx(struct zoneref *zoneref)
1612 return zone_to_nid(zoneref->zone);
1615 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1616 enum zone_type highest_zoneidx,
1620 * 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
1621 * @z: The cursor used as a starting point for the search
1622 * @highest_zoneidx: The zone index of the highest zone to return
1623 * @nodes: An optional nodemask to filter the zonelist with
1625 * This function returns the next zone at or below a given zone index that is
1626 * within the allowed nodemask using a cursor as the starting point for the
1627 * search. The zoneref returned is a cursor that represents the current zone
1628 * being examined. It should be advanced by one before calling
1629 * next_zones_zonelist again.
1631 * Return: the next zone at or below highest_zoneidx within the allowed
1632 * nodemask using a cursor within a zonelist as a starting point
1634 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1635 enum zone_type highest_zoneidx,
1638 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1640 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1644 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1645 * @zonelist: The zonelist to search for a suitable zone
1646 * @highest_zoneidx: The zone index of the highest zone to return
1647 * @nodes: An optional nodemask to filter the zonelist with
1649 * This function returns the first zone at or below a given zone index that is
1650 * within the allowed nodemask. The zoneref returned is a cursor that can be
1651 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1652 * one before calling.
1654 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1655 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1656 * update due to cpuset modification.
1658 * Return: Zoneref pointer for the first suitable zone found
1660 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1661 enum zone_type highest_zoneidx,
1664 return next_zones_zonelist(zonelist->_zonerefs,
1665 highest_zoneidx, nodes);
1669 * 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
1670 * @zone: The current zone in the iterator
1671 * @z: The current pointer within zonelist->_zonerefs being iterated
1672 * @zlist: The zonelist being iterated
1673 * @highidx: The zone index of the highest zone to return
1674 * @nodemask: Nodemask allowed by the allocator
1676 * This iterator iterates though all zones at or below a given zone index and
1677 * within a given nodemask
1679 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1680 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1682 z = next_zones_zonelist(++z, highidx, nodemask), \
1683 zone = zonelist_zone(z))
1685 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1686 for (zone = z->zone; \
1688 z = next_zones_zonelist(++z, highidx, nodemask), \
1689 zone = zonelist_zone(z))
1693 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1694 * @zone: The current zone in the iterator
1695 * @z: The current pointer within zonelist->zones being iterated
1696 * @zlist: The zonelist being iterated
1697 * @highidx: The zone index of the highest zone to return
1699 * This iterator iterates though all zones at or below a given zone index.
1701 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1702 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1704 /* Whether the 'nodes' are all movable nodes */
1705 static inline bool movable_only_nodes(nodemask_t *nodes)
1707 struct zonelist *zonelist;
1711 if (nodes_empty(*nodes))
1715 * We can chose arbitrary node from the nodemask to get a
1716 * zonelist as they are interlinked. We just need to find
1717 * at least one zone that can satisfy kernel allocations.
1719 nid = first_node(*nodes);
1720 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1721 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1722 return (!z->zone) ? true : false;
1726 #ifdef CONFIG_SPARSEMEM
1727 #include <asm/sparsemem.h>
1730 #ifdef CONFIG_FLATMEM
1731 #define pfn_to_nid(pfn) (0)
1734 #ifdef CONFIG_SPARSEMEM
1737 * PA_SECTION_SHIFT physical address to/from section number
1738 * PFN_SECTION_SHIFT pfn to/from section number
1740 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1741 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1743 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1745 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1746 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1748 #define SECTION_BLOCKFLAGS_BITS \
1749 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1751 #if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1752 #error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1755 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1757 return pfn >> PFN_SECTION_SHIFT;
1759 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1761 return sec << PFN_SECTION_SHIFT;
1764 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1765 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1767 #define SUBSECTION_SHIFT 21
1768 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1770 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1771 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1772 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1774 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1775 #error Subsection size exceeds section size
1777 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1780 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1781 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1783 struct mem_section_usage {
1784 struct rcu_head rcu;
1785 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1786 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1788 /* See declaration of similar field in struct zone */
1789 unsigned long pageblock_flags[0];
1792 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1796 struct mem_section {
1798 * This is, logically, a pointer to an array of struct
1799 * pages. However, it is stored with some other magic.
1800 * (see sparse.c::sparse_init_one_section())
1802 * Additionally during early boot we encode node id of
1803 * the location of the section here to guide allocation.
1804 * (see sparse.c::memory_present())
1806 * Making it a UL at least makes someone do a cast
1807 * before using it wrong.
1809 unsigned long section_mem_map;
1811 struct mem_section_usage *usage;
1812 #ifdef CONFIG_PAGE_EXTENSION
1814 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1815 * section. (see page_ext.h about this.)
1817 struct page_ext *page_ext;
1821 * WARNING: mem_section must be a power-of-2 in size for the
1822 * calculation and use of SECTION_ROOT_MASK to make sense.
1826 #ifdef CONFIG_SPARSEMEM_EXTREME
1827 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1829 #define SECTIONS_PER_ROOT 1
1832 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1833 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1834 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1836 #ifdef CONFIG_SPARSEMEM_EXTREME
1837 extern struct mem_section **mem_section;
1839 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1842 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1844 return ms->usage->pageblock_flags;
1847 static inline struct mem_section *__nr_to_section(unsigned long nr)
1849 unsigned long root = SECTION_NR_TO_ROOT(nr);
1851 if (unlikely(root >= NR_SECTION_ROOTS))
1854 #ifdef CONFIG_SPARSEMEM_EXTREME
1855 if (!mem_section || !mem_section[root])
1858 return &mem_section[root][nr & SECTION_ROOT_MASK];
1860 extern size_t mem_section_usage_size(void);
1863 * We use the lower bits of the mem_map pointer to store
1864 * a little bit of information. The pointer is calculated
1865 * as mem_map - section_nr_to_pfn(pnum). The result is
1866 * aligned to the minimum alignment of the two values:
1867 * 1. All mem_map arrays are page-aligned.
1868 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1869 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1870 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1871 * worst combination is powerpc with 256k pages,
1872 * which results in PFN_SECTION_SHIFT equal 6.
1873 * To sum it up, at least 6 bits are available on all architectures.
1874 * However, we can exceed 6 bits on some other architectures except
1875 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1876 * with the worst case of 64K pages on arm64) if we make sure the
1877 * exceeded bit is not applicable to powerpc.
1880 SECTION_MARKED_PRESENT_BIT,
1881 SECTION_HAS_MEM_MAP_BIT,
1882 SECTION_IS_ONLINE_BIT,
1883 SECTION_IS_EARLY_BIT,
1884 #ifdef CONFIG_ZONE_DEVICE
1885 SECTION_TAINT_ZONE_DEVICE_BIT,
1887 SECTION_MAP_LAST_BIT,
1890 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1891 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1892 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1893 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1894 #ifdef CONFIG_ZONE_DEVICE
1895 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1897 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1898 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1900 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1902 unsigned long map = section->section_mem_map;
1903 map &= SECTION_MAP_MASK;
1904 return (struct page *)map;
1907 static inline int present_section(struct mem_section *section)
1909 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1912 static inline int present_section_nr(unsigned long nr)
1914 return present_section(__nr_to_section(nr));
1917 static inline int valid_section(struct mem_section *section)
1919 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1922 static inline int early_section(struct mem_section *section)
1924 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1927 static inline int valid_section_nr(unsigned long nr)
1929 return valid_section(__nr_to_section(nr));
1932 static inline int online_section(struct mem_section *section)
1934 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1937 #ifdef CONFIG_ZONE_DEVICE
1938 static inline int online_device_section(struct mem_section *section)
1940 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1942 return section && ((section->section_mem_map & flags) == flags);
1945 static inline int online_device_section(struct mem_section *section)
1951 static inline int online_section_nr(unsigned long nr)
1953 return online_section(__nr_to_section(nr));
1956 #ifdef CONFIG_MEMORY_HOTPLUG
1957 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1958 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1961 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1963 return __nr_to_section(pfn_to_section_nr(pfn));
1966 extern unsigned long __highest_present_section_nr;
1968 static inline int subsection_map_index(unsigned long pfn)
1970 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1973 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1974 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1976 int idx = subsection_map_index(pfn);
1978 return test_bit(idx, READ_ONCE(ms->usage)->subsection_map);
1981 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1987 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1989 * pfn_valid - check if there is a valid memory map entry for a PFN
1990 * @pfn: the page frame number to check
1992 * Check if there is a valid memory map entry aka struct page for the @pfn.
1993 * Note, that availability of the memory map entry does not imply that
1994 * there is actual usable memory at that @pfn. The struct page may
1995 * represent a hole or an unusable page frame.
1997 * Return: 1 for PFNs that have memory map entries and 0 otherwise
1999 static inline int pfn_valid(unsigned long pfn)
2001 struct mem_section *ms;
2005 * Ensure the upper PAGE_SHIFT bits are clear in the
2006 * pfn. Else it might lead to false positives when
2007 * some of the upper bits are set, but the lower bits
2008 * match a valid pfn.
2010 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2013 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2015 ms = __pfn_to_section(pfn);
2017 if (!valid_section(ms)) {
2022 * Traditionally early sections always returned pfn_valid() for
2023 * the entire section-sized span.
2025 ret = early_section(ms) || pfn_section_valid(ms, pfn);
2032 static inline int pfn_in_present_section(unsigned long pfn)
2034 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2036 return present_section(__pfn_to_section(pfn));
2039 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2041 while (++section_nr <= __highest_present_section_nr) {
2042 if (present_section_nr(section_nr))
2050 * These are _only_ used during initialisation, therefore they
2051 * can use __initdata ... They could have names to indicate
2055 #define pfn_to_nid(pfn) \
2057 unsigned long __pfn_to_nid_pfn = (pfn); \
2058 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2061 #define pfn_to_nid(pfn) (0)
2064 void sparse_init(void);
2066 #define sparse_init() do {} while (0)
2067 #define sparse_index_init(_sec, _nid) do {} while (0)
2068 #define pfn_in_present_section pfn_valid
2069 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2070 #endif /* CONFIG_SPARSEMEM */
2072 #endif /* !__GENERATING_BOUNDS.H */
2073 #endif /* !__ASSEMBLY__ */
2074 #endif /* _LINUX_MMZONE_H */