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)
109 struct list_head free_list[MIGRATE_TYPES];
110 unsigned long nr_free;
116 enum numa_stat_item {
117 NUMA_HIT, /* allocated in intended node */
118 NUMA_MISS, /* allocated in non intended node */
119 NUMA_FOREIGN, /* was intended here, hit elsewhere */
120 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
121 NUMA_LOCAL, /* allocation from local node */
122 NUMA_OTHER, /* allocation from other node */
123 NR_VM_NUMA_EVENT_ITEMS
126 #define NR_VM_NUMA_EVENT_ITEMS 0
129 enum zone_stat_item {
130 /* First 128 byte cacheline (assuming 64 bit words) */
132 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
133 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
135 NR_ZONE_INACTIVE_FILE,
138 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
139 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
140 /* Second 128 byte cacheline */
142 #if IS_ENABLED(CONFIG_ZSMALLOC)
143 NR_ZSPAGES, /* allocated in zsmalloc */
146 NR_VM_ZONE_STAT_ITEMS };
148 enum node_stat_item {
150 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
151 NR_ACTIVE_ANON, /* " " " " " */
152 NR_INACTIVE_FILE, /* " " " " " */
153 NR_ACTIVE_FILE, /* " " " " " */
154 NR_UNEVICTABLE, /* " " " " " */
155 NR_SLAB_RECLAIMABLE_B,
156 NR_SLAB_UNRECLAIMABLE_B,
157 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
158 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
160 WORKINGSET_REFAULT_BASE,
161 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
162 WORKINGSET_REFAULT_FILE,
163 WORKINGSET_ACTIVATE_BASE,
164 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
165 WORKINGSET_ACTIVATE_FILE,
166 WORKINGSET_RESTORE_BASE,
167 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
168 WORKINGSET_RESTORE_FILE,
169 WORKINGSET_NODERECLAIM,
170 NR_ANON_MAPPED, /* Mapped anonymous pages */
171 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
172 only modified from process context */
176 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
177 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
184 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
185 NR_DIRTIED, /* page dirtyings since bootup */
186 NR_WRITTEN, /* page writings since bootup */
187 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
188 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
189 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
190 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
191 NR_KERNEL_STACK_KB, /* measured in KiB */
192 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
193 NR_KERNEL_SCS_KB, /* measured in KiB */
195 NR_PAGETABLE, /* used for pagetables */
196 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
200 #ifdef CONFIG_NUMA_BALANCING
201 PGPROMOTE_SUCCESS, /* promote successfully */
202 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
204 NR_VM_NODE_STAT_ITEMS
208 * Returns true if the item should be printed in THPs (/proc/vmstat
209 * currently prints number of anon, file and shmem THPs. But the item
210 * is charged in pages).
212 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
214 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
217 return item == NR_ANON_THPS ||
218 item == NR_FILE_THPS ||
219 item == NR_SHMEM_THPS ||
220 item == NR_SHMEM_PMDMAPPED ||
221 item == NR_FILE_PMDMAPPED;
225 * Returns true if the value is measured in bytes (most vmstat values are
226 * measured in pages). This defines the API part, the internal representation
227 * might be different.
229 static __always_inline bool vmstat_item_in_bytes(int idx)
232 * Global and per-node slab counters track slab pages.
233 * It's expected that changes are multiples of PAGE_SIZE.
234 * Internally values are stored in pages.
236 * Per-memcg and per-lruvec counters track memory, consumed
237 * by individual slab objects. These counters are actually
240 return (idx == NR_SLAB_RECLAIMABLE_B ||
241 idx == NR_SLAB_UNRECLAIMABLE_B);
245 * We do arithmetic on the LRU lists in various places in the code,
246 * so it is important to keep the active lists LRU_ACTIVE higher in
247 * the array than the corresponding inactive lists, and to keep
248 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
250 * This has to be kept in sync with the statistics in zone_stat_item
251 * above and the descriptions in vmstat_text in mm/vmstat.c
258 LRU_INACTIVE_ANON = LRU_BASE,
259 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
260 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
261 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
266 enum vmscan_throttle_state {
267 VMSCAN_THROTTLE_WRITEBACK,
268 VMSCAN_THROTTLE_ISOLATED,
269 VMSCAN_THROTTLE_NOPROGRESS,
270 VMSCAN_THROTTLE_CONGESTED,
274 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
276 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
278 static inline bool is_file_lru(enum lru_list lru)
280 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
283 static inline bool is_active_lru(enum lru_list lru)
285 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
288 #define WORKINGSET_ANON 0
289 #define WORKINGSET_FILE 1
290 #define ANON_AND_FILE 2
293 LRUVEC_CONGESTED, /* lruvec has many dirty pages
294 * backed by a congested BDI
298 #endif /* !__GENERATING_BOUNDS_H */
301 * Evictable pages are divided into multiple generations. The youngest and the
302 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
303 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
304 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
305 * corresponding generation. The gen counter in folio->flags stores gen+1 while
306 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
308 * A page is added to the youngest generation on faulting. The aging needs to
309 * check the accessed bit at least twice before handing this page over to the
310 * eviction. The first check takes care of the accessed bit set on the initial
311 * fault; the second check makes sure this page hasn't been used since then.
312 * This process, AKA second chance, requires a minimum of two generations,
313 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
314 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
315 * rest of generations, if they exist, are considered inactive. See
316 * lru_gen_is_active().
318 * PG_active is always cleared while a page is on one of lrugen->folios[] so
319 * that the aging needs not to worry about it. And it's set again when a page
320 * considered active is isolated for non-reclaiming purposes, e.g., migration.
321 * See lru_gen_add_folio() and lru_gen_del_folio().
323 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
324 * number of categories of the active/inactive LRU when keeping track of
325 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
328 #define MIN_NR_GENS 2U
329 #define MAX_NR_GENS 4U
332 * Each generation is divided into multiple tiers. A page accessed N times
333 * through file descriptors is in tier order_base_2(N). A page in the first tier
334 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
335 * tables or read ahead. A page in any other tier (N>1) is marked by
336 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
337 * supported without using additional bits in folio->flags.
339 * In contrast to moving across generations which requires the LRU lock, moving
340 * across tiers only involves atomic operations on folio->flags and therefore
341 * has a negligible cost in the buffered access path. In the eviction path,
342 * comparisons of refaulted/(evicted+protected) from the first tier and the
343 * rest infer whether pages accessed multiple times through file descriptors
344 * are statistically hot and thus worth protecting.
346 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
347 * number of categories of the active/inactive LRU when keeping track of
348 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
351 #define MAX_NR_TIERS 4U
353 #ifndef __GENERATING_BOUNDS_H
356 struct page_vma_mapped_walk;
358 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
359 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
361 #ifdef CONFIG_LRU_GEN
371 LRU_GEN_NONLEAF_YOUNG,
375 #define MIN_LRU_BATCH BITS_PER_LONG
376 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
378 /* whether to keep historical stats from evicted generations */
379 #ifdef CONFIG_LRU_GEN_STATS
380 #define NR_HIST_GENS MAX_NR_GENS
382 #define NR_HIST_GENS 1U
386 * The youngest generation number is stored in max_seq for both anon and file
387 * types as they are aged on an equal footing. The oldest generation numbers are
388 * stored in min_seq[] separately for anon and file types as clean file pages
389 * can be evicted regardless of swap constraints.
391 * Normally anon and file min_seq are in sync. But if swapping is constrained,
392 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
395 * The number of pages in each generation is eventually consistent and therefore
396 * can be transiently negative when reset_batch_size() is pending.
398 struct lru_gen_folio {
399 /* the aging increments the youngest generation number */
400 unsigned long max_seq;
401 /* the eviction increments the oldest generation numbers */
402 unsigned long min_seq[ANON_AND_FILE];
403 /* the birth time of each generation in jiffies */
404 unsigned long timestamps[MAX_NR_GENS];
405 /* the multi-gen LRU lists, lazily sorted on eviction */
406 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
407 /* the multi-gen LRU sizes, eventually consistent */
408 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
409 /* the exponential moving average of refaulted */
410 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
411 /* the exponential moving average of evicted+protected */
412 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
413 /* the first tier doesn't need protection, hence the minus one */
414 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
415 /* can be modified without holding the LRU lock */
416 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
417 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
418 /* whether the multi-gen LRU is enabled */
421 /* the memcg generation this lru_gen_folio belongs to */
423 /* the list segment this lru_gen_folio belongs to */
425 /* per-node lru_gen_folio list for global reclaim */
426 struct hlist_nulls_node list;
431 MM_LEAF_TOTAL, /* total leaf entries */
432 MM_LEAF_OLD, /* old leaf entries */
433 MM_LEAF_YOUNG, /* young leaf entries */
434 MM_NONLEAF_TOTAL, /* total non-leaf entries */
435 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
436 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
440 /* double-buffering Bloom filters */
441 #define NR_BLOOM_FILTERS 2
443 struct lru_gen_mm_state {
444 /* set to max_seq after each iteration */
446 /* where the current iteration continues after */
447 struct list_head *head;
448 /* where the last iteration ended before */
449 struct list_head *tail;
450 /* Bloom filters flip after each iteration */
451 unsigned long *filters[NR_BLOOM_FILTERS];
452 /* the mm stats for debugging */
453 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
456 struct lru_gen_mm_walk {
457 /* the lruvec under reclaim */
458 struct lruvec *lruvec;
459 /* unstable max_seq from lru_gen_folio */
460 unsigned long max_seq;
461 /* the next address within an mm to scan */
462 unsigned long next_addr;
463 /* to batch promoted pages */
464 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
465 /* to batch the mm stats */
466 int mm_stats[NR_MM_STATS];
467 /* total batched items */
473 void lru_gen_init_lruvec(struct lruvec *lruvec);
474 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
479 * For each node, memcgs are divided into two generations: the old and the
480 * young. For each generation, memcgs are randomly sharded into multiple bins
481 * to improve scalability. For each bin, the hlist_nulls is virtually divided
482 * into three segments: the head, the tail and the default.
484 * An onlining memcg is added to the tail of a random bin in the old generation.
485 * The eviction starts at the head of a random bin in the old generation. The
486 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
487 * the old generation, is incremented when all its bins become empty.
489 * There are four operations:
490 * 1. MEMCG_LRU_HEAD, which moves an memcg to the head of a random bin in its
491 * current generation (old or young) and updates its "seg" to "head";
492 * 2. MEMCG_LRU_TAIL, which moves an memcg to the tail of a random bin in its
493 * current generation (old or young) and updates its "seg" to "tail";
494 * 3. MEMCG_LRU_OLD, which moves an memcg to the head of a random bin in the old
495 * generation, updates its "gen" to "old" and resets its "seg" to "default";
496 * 4. MEMCG_LRU_YOUNG, which moves an memcg to the tail of a random bin in the
497 * young generation, updates its "gen" to "young" and resets its "seg" to
500 * The events that trigger the above operations are:
501 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
502 * 2. The first attempt to reclaim an memcg below low, which triggers
504 * 3. The first attempt to reclaim an memcg below reclaimable size threshold,
505 * which triggers MEMCG_LRU_TAIL;
506 * 4. The second attempt to reclaim an memcg below reclaimable size threshold,
507 * which triggers MEMCG_LRU_YOUNG;
508 * 5. Attempting to reclaim an memcg below min, which triggers MEMCG_LRU_YOUNG;
509 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
510 * 7. Offlining an memcg, which triggers MEMCG_LRU_OLD.
512 * Note that memcg LRU only applies to global reclaim, and the round-robin
513 * incrementing of their max_seq counters ensures the eventual fairness to all
514 * eligible memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
516 #define MEMCG_NR_GENS 2
517 #define MEMCG_NR_BINS 8
519 struct lru_gen_memcg {
520 /* the per-node memcg generation counter */
522 /* each memcg has one lru_gen_folio per node */
523 unsigned long nr_memcgs[MEMCG_NR_GENS];
524 /* per-node lru_gen_folio list for global reclaim */
525 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
526 /* protects the above */
530 void lru_gen_init_pgdat(struct pglist_data *pgdat);
532 void lru_gen_init_memcg(struct mem_cgroup *memcg);
533 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
534 void lru_gen_online_memcg(struct mem_cgroup *memcg);
535 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
536 void lru_gen_release_memcg(struct mem_cgroup *memcg);
537 void lru_gen_soft_reclaim(struct lruvec *lruvec);
539 #else /* !CONFIG_MEMCG */
541 #define MEMCG_NR_GENS 1
543 struct lru_gen_memcg {
546 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
550 #endif /* CONFIG_MEMCG */
552 #else /* !CONFIG_LRU_GEN */
554 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
558 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
562 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
568 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
572 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
576 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
580 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
584 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
588 static inline void lru_gen_soft_reclaim(struct lruvec *lruvec)
592 #endif /* CONFIG_MEMCG */
594 #endif /* CONFIG_LRU_GEN */
597 struct list_head lists[NR_LRU_LISTS];
598 /* per lruvec lru_lock for memcg */
601 * These track the cost of reclaiming one LRU - file or anon -
602 * over the other. As the observed cost of reclaiming one LRU
603 * increases, the reclaim scan balance tips toward the other.
605 unsigned long anon_cost;
606 unsigned long file_cost;
607 /* Non-resident age, driven by LRU movement */
608 atomic_long_t nonresident_age;
609 /* Refaults at the time of last reclaim cycle */
610 unsigned long refaults[ANON_AND_FILE];
611 /* Various lruvec state flags (enum lruvec_flags) */
613 #ifdef CONFIG_LRU_GEN
614 /* evictable pages divided into generations */
615 struct lru_gen_folio lrugen;
616 /* to concurrently iterate lru_gen_mm_list */
617 struct lru_gen_mm_state mm_state;
620 struct pglist_data *pgdat;
624 /* Isolate unmapped pages */
625 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
626 /* Isolate for asynchronous migration */
627 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
628 /* Isolate unevictable pages */
629 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
631 /* LRU Isolation modes. */
632 typedef unsigned __bitwise isolate_mode_t;
634 enum zone_watermarks {
643 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
644 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
645 * it should not contribute to serious fragmentation causing THP allocation
648 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
653 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
654 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
656 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
657 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
658 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
659 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
661 /* Fields and list protected by pagesets local_lock in page_alloc.c */
662 struct per_cpu_pages {
663 spinlock_t lock; /* Protects lists field */
664 int count; /* number of pages in the list */
665 int high; /* high watermark, emptying needed */
666 int batch; /* chunk size for buddy add/remove */
667 short free_factor; /* batch scaling factor during free */
669 short expire; /* When 0, remote pagesets are drained */
672 /* Lists of pages, one per migrate type stored on the pcp-lists */
673 struct list_head lists[NR_PCP_LISTS];
674 } ____cacheline_aligned_in_smp;
676 struct per_cpu_zonestat {
678 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
683 * Low priority inaccurate counters that are only folded
684 * on demand. Use a large type to avoid the overhead of
685 * folding during refresh_cpu_vm_stats.
687 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
691 struct per_cpu_nodestat {
693 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
696 #endif /* !__GENERATING_BOUNDS.H */
700 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
701 * to DMA to all of the addressable memory (ZONE_NORMAL).
702 * On architectures where this area covers the whole 32 bit address
703 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
704 * DMA addressing constraints. This distinction is important as a 32bit
705 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
706 * platforms may need both zones as they support peripherals with
707 * different DMA addressing limitations.
709 #ifdef CONFIG_ZONE_DMA
712 #ifdef CONFIG_ZONE_DMA32
716 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
717 * performed on pages in ZONE_NORMAL if the DMA devices support
718 * transfers to all addressable memory.
721 #ifdef CONFIG_HIGHMEM
723 * A memory area that is only addressable by the kernel through
724 * mapping portions into its own address space. This is for example
725 * used by i386 to allow the kernel to address the memory beyond
726 * 900MB. The kernel will set up special mappings (page
727 * table entries on i386) for each page that the kernel needs to
733 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
734 * movable pages with few exceptional cases described below. Main use
735 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
736 * likely to succeed, and to locally limit unmovable allocations - e.g.,
737 * to increase the number of THP/huge pages. Notable special cases are:
739 * 1. Pinned pages: (long-term) pinning of movable pages might
740 * essentially turn such pages unmovable. Therefore, we do not allow
741 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
742 * faulted, they come from the right zone right away. However, it is
743 * still possible that address space already has pages in
744 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
745 * touches that memory before pinning). In such case we migrate them
746 * to a different zone. When migration fails - pinning fails.
747 * 2. memblock allocations: kernelcore/movablecore setups might create
748 * situations where ZONE_MOVABLE contains unmovable allocations
749 * after boot. Memory offlining and allocations fail early.
750 * 3. Memory holes: kernelcore/movablecore setups might create very rare
751 * situations where ZONE_MOVABLE contains memory holes after boot,
752 * for example, if we have sections that are only partially
753 * populated. Memory offlining and allocations fail early.
754 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
755 * memory offlining, such pages cannot be allocated.
756 * 5. Unmovable PG_offline pages: in paravirtualized environments,
757 * hotplugged memory blocks might only partially be managed by the
758 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
759 * parts not manged by the buddy are unmovable PG_offline pages. In
760 * some cases (virtio-mem), such pages can be skipped during
761 * memory offlining, however, cannot be moved/allocated. These
762 * techniques might use alloc_contig_range() to hide previously
763 * exposed pages from the buddy again (e.g., to implement some sort
764 * of memory unplug in virtio-mem).
765 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
766 * situations where ZERO_PAGE(0) which is allocated differently
767 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
768 * cannot be migrated.
769 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
770 * memory to the MOVABLE zone, the vmemmap pages are also placed in
771 * such zone. Such pages cannot be really moved around as they are
772 * self-stored in the range, but they are treated as movable when
773 * the range they describe is about to be offlined.
775 * In general, no unmovable allocations that degrade memory offlining
776 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
777 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
778 * if has_unmovable_pages() states that there are no unmovable pages,
779 * there can be false negatives).
782 #ifdef CONFIG_ZONE_DEVICE
789 #ifndef __GENERATING_BOUNDS_H
791 #define ASYNC_AND_SYNC 2
794 /* Read-mostly fields */
796 /* zone watermarks, access with *_wmark_pages(zone) macros */
797 unsigned long _watermark[NR_WMARK];
798 unsigned long watermark_boost;
800 unsigned long nr_reserved_highatomic;
803 * We don't know if the memory that we're going to allocate will be
804 * freeable or/and it will be released eventually, so to avoid totally
805 * wasting several GB of ram we must reserve some of the lower zone
806 * memory (otherwise we risk to run OOM on the lower zones despite
807 * there being tons of freeable ram on the higher zones). This array is
808 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
811 long lowmem_reserve[MAX_NR_ZONES];
816 struct pglist_data *zone_pgdat;
817 struct per_cpu_pages __percpu *per_cpu_pageset;
818 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
820 * the high and batch values are copied to individual pagesets for
826 #ifndef CONFIG_SPARSEMEM
828 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
829 * In SPARSEMEM, this map is stored in struct mem_section
831 unsigned long *pageblock_flags;
832 #endif /* CONFIG_SPARSEMEM */
834 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
835 unsigned long zone_start_pfn;
838 * spanned_pages is the total pages spanned by the zone, including
839 * holes, which is calculated as:
840 * spanned_pages = zone_end_pfn - zone_start_pfn;
842 * present_pages is physical pages existing within the zone, which
844 * present_pages = spanned_pages - absent_pages(pages in holes);
846 * present_early_pages is present pages existing within the zone
847 * located on memory available since early boot, excluding hotplugged
850 * managed_pages is present pages managed by the buddy system, which
851 * is calculated as (reserved_pages includes pages allocated by the
852 * bootmem allocator):
853 * managed_pages = present_pages - reserved_pages;
855 * cma pages is present pages that are assigned for CMA use
858 * So present_pages may be used by memory hotplug or memory power
859 * management logic to figure out unmanaged pages by checking
860 * (present_pages - managed_pages). And managed_pages should be used
861 * by page allocator and vm scanner to calculate all kinds of watermarks
866 * zone_start_pfn and spanned_pages are protected by span_seqlock.
867 * It is a seqlock because it has to be read outside of zone->lock,
868 * and it is done in the main allocator path. But, it is written
869 * quite infrequently.
871 * The span_seq lock is declared along with zone->lock because it is
872 * frequently read in proximity to zone->lock. It's good to
873 * give them a chance of being in the same cacheline.
875 * Write access to present_pages at runtime should be protected by
876 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
877 * present_pages should use get_online_mems() to get a stable value.
879 atomic_long_t managed_pages;
880 unsigned long spanned_pages;
881 unsigned long present_pages;
882 #if defined(CONFIG_MEMORY_HOTPLUG)
883 unsigned long present_early_pages;
886 unsigned long cma_pages;
891 #ifdef CONFIG_MEMORY_ISOLATION
893 * Number of isolated pageblock. It is used to solve incorrect
894 * freepage counting problem due to racy retrieving migratetype
895 * of pageblock. Protected by zone->lock.
897 unsigned long nr_isolate_pageblock;
900 #ifdef CONFIG_MEMORY_HOTPLUG
901 /* see spanned/present_pages for more description */
902 seqlock_t span_seqlock;
907 /* Write-intensive fields used from the page allocator */
908 CACHELINE_PADDING(_pad1_);
910 /* free areas of different sizes */
911 struct free_area free_area[MAX_ORDER + 1];
913 /* zone flags, see below */
916 /* Primarily protects free_area */
919 /* Write-intensive fields used by compaction and vmstats. */
920 CACHELINE_PADDING(_pad2_);
923 * When free pages are below this point, additional steps are taken
924 * when reading the number of free pages to avoid per-cpu counter
925 * drift allowing watermarks to be breached
927 unsigned long percpu_drift_mark;
929 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
930 /* pfn where compaction free scanner should start */
931 unsigned long compact_cached_free_pfn;
932 /* pfn where compaction migration scanner should start */
933 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
934 unsigned long compact_init_migrate_pfn;
935 unsigned long compact_init_free_pfn;
938 #ifdef CONFIG_COMPACTION
940 * On compaction failure, 1<<compact_defer_shift compactions
941 * are skipped before trying again. The number attempted since
942 * last failure is tracked with compact_considered.
943 * compact_order_failed is the minimum compaction failed order.
945 unsigned int compact_considered;
946 unsigned int compact_defer_shift;
947 int compact_order_failed;
950 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
951 /* Set to true when the PG_migrate_skip bits should be cleared */
952 bool compact_blockskip_flush;
957 CACHELINE_PADDING(_pad3_);
958 /* Zone statistics */
959 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
960 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
961 } ____cacheline_internodealigned_in_smp;
964 PGDAT_DIRTY, /* reclaim scanning has recently found
965 * many dirty file pages at the tail
968 PGDAT_WRITEBACK, /* reclaim scanning has recently found
969 * many pages under writeback
971 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
975 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
976 * Cleared when kswapd is woken.
978 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
981 static inline unsigned long zone_managed_pages(struct zone *zone)
983 return (unsigned long)atomic_long_read(&zone->managed_pages);
986 static inline unsigned long zone_cma_pages(struct zone *zone)
989 return zone->cma_pages;
995 static inline unsigned long zone_end_pfn(const struct zone *zone)
997 return zone->zone_start_pfn + zone->spanned_pages;
1000 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1002 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1005 static inline bool zone_is_initialized(struct zone *zone)
1007 return zone->initialized;
1010 static inline bool zone_is_empty(struct zone *zone)
1012 return zone->spanned_pages == 0;
1015 #ifndef BUILD_VDSO32_64
1017 * The zone field is never updated after free_area_init_core()
1018 * sets it, so none of the operations on it need to be atomic.
1021 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1022 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1023 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1024 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1025 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1026 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1027 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1028 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1031 * Define the bit shifts to access each section. For non-existent
1032 * sections we define the shift as 0; that plus a 0 mask ensures
1033 * the compiler will optimise away reference to them.
1035 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1036 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1037 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1038 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1039 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1041 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1042 #ifdef NODE_NOT_IN_PAGE_FLAGS
1043 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1044 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1045 SECTIONS_PGOFF : ZONES_PGOFF)
1047 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1048 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1049 NODES_PGOFF : ZONES_PGOFF)
1052 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1054 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1055 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1056 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1057 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1058 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1059 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1061 static inline enum zone_type page_zonenum(const struct page *page)
1063 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1064 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1067 static inline enum zone_type folio_zonenum(const struct folio *folio)
1069 return page_zonenum(&folio->page);
1072 #ifdef CONFIG_ZONE_DEVICE
1073 static inline bool is_zone_device_page(const struct page *page)
1075 return page_zonenum(page) == ZONE_DEVICE;
1079 * Consecutive zone device pages should not be merged into the same sgl
1080 * or bvec segment with other types of pages or if they belong to different
1081 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1082 * without scanning the entire segment. This helper returns true either if
1083 * both pages are not zone device pages or both pages are zone device pages
1084 * with the same pgmap.
1086 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1087 const struct page *b)
1089 if (is_zone_device_page(a) != is_zone_device_page(b))
1091 if (!is_zone_device_page(a))
1093 return a->pgmap == b->pgmap;
1096 extern void memmap_init_zone_device(struct zone *, unsigned long,
1097 unsigned long, struct dev_pagemap *);
1099 static inline bool is_zone_device_page(const struct page *page)
1103 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1104 const struct page *b)
1110 static inline bool folio_is_zone_device(const struct folio *folio)
1112 return is_zone_device_page(&folio->page);
1115 static inline bool is_zone_movable_page(const struct page *page)
1117 return page_zonenum(page) == ZONE_MOVABLE;
1122 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1123 * intersection with the given zone
1125 static inline bool zone_intersects(struct zone *zone,
1126 unsigned long start_pfn, unsigned long nr_pages)
1128 if (zone_is_empty(zone))
1130 if (start_pfn >= zone_end_pfn(zone) ||
1131 start_pfn + nr_pages <= zone->zone_start_pfn)
1138 * The "priority" of VM scanning is how much of the queues we will scan in one
1139 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1140 * queues ("queue_length >> 12") during an aging round.
1142 #define DEF_PRIORITY 12
1144 /* Maximum number of zones on a zonelist */
1145 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1148 ZONELIST_FALLBACK, /* zonelist with fallback */
1151 * The NUMA zonelists are doubled because we need zonelists that
1152 * restrict the allocations to a single node for __GFP_THISNODE.
1154 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1160 * This struct contains information about a zone in a zonelist. It is stored
1161 * here to avoid dereferences into large structures and lookups of tables
1164 struct zone *zone; /* Pointer to actual zone */
1165 int zone_idx; /* zone_idx(zoneref->zone) */
1169 * One allocation request operates on a zonelist. A zonelist
1170 * is a list of zones, the first one is the 'goal' of the
1171 * allocation, the other zones are fallback zones, in decreasing
1174 * To speed the reading of the zonelist, the zonerefs contain the zone index
1175 * of the entry being read. Helper functions to access information given
1176 * a struct zoneref are
1178 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1179 * zonelist_zone_idx() - Return the index of the zone for an entry
1180 * zonelist_node_idx() - Return the index of the node for an entry
1183 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1187 * The array of struct pages for flatmem.
1188 * It must be declared for SPARSEMEM as well because there are configurations
1189 * that rely on that.
1191 extern struct page *mem_map;
1193 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1194 struct deferred_split {
1195 spinlock_t split_queue_lock;
1196 struct list_head split_queue;
1197 unsigned long split_queue_len;
1201 #ifdef CONFIG_MEMORY_FAILURE
1203 * Per NUMA node memory failure handling statistics.
1205 struct memory_failure_stats {
1207 * Number of raw pages poisoned.
1208 * Cases not accounted: memory outside kernel control, offline page,
1209 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1210 * error events, and unpoison actions from hwpoison_unpoison.
1212 unsigned long total;
1214 * Recovery results of poisoned raw pages handled by memory_failure,
1215 * in sync with mf_result.
1216 * total = ignored + failed + delayed + recovered.
1217 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1219 unsigned long ignored;
1220 unsigned long failed;
1221 unsigned long delayed;
1222 unsigned long recovered;
1227 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1228 * it's memory layout. On UMA machines there is a single pglist_data which
1229 * describes the whole memory.
1231 * Memory statistics and page replacement data structures are maintained on a
1234 typedef struct pglist_data {
1236 * node_zones contains just the zones for THIS node. Not all of the
1237 * zones may be populated, but it is the full list. It is referenced by
1238 * this node's node_zonelists as well as other node's node_zonelists.
1240 struct zone node_zones[MAX_NR_ZONES];
1243 * node_zonelists contains references to all zones in all nodes.
1244 * Generally the first zones will be references to this node's
1247 struct zonelist node_zonelists[MAX_ZONELISTS];
1249 int nr_zones; /* number of populated zones in this node */
1250 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1251 struct page *node_mem_map;
1252 #ifdef CONFIG_PAGE_EXTENSION
1253 struct page_ext *node_page_ext;
1256 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1258 * Must be held any time you expect node_start_pfn,
1259 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1260 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1263 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1264 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1265 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1267 * Nests above zone->lock and zone->span_seqlock
1269 spinlock_t node_size_lock;
1271 unsigned long node_start_pfn;
1272 unsigned long node_present_pages; /* total number of physical pages */
1273 unsigned long node_spanned_pages; /* total size of physical page
1274 range, including holes */
1276 wait_queue_head_t kswapd_wait;
1277 wait_queue_head_t pfmemalloc_wait;
1279 /* workqueues for throttling reclaim for different reasons. */
1280 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1282 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1283 unsigned long nr_reclaim_start; /* nr pages written while throttled
1284 * when throttling started. */
1285 #ifdef CONFIG_MEMORY_HOTPLUG
1286 struct mutex kswapd_lock;
1288 struct task_struct *kswapd; /* Protected by kswapd_lock */
1290 enum zone_type kswapd_highest_zoneidx;
1292 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1294 #ifdef CONFIG_COMPACTION
1295 int kcompactd_max_order;
1296 enum zone_type kcompactd_highest_zoneidx;
1297 wait_queue_head_t kcompactd_wait;
1298 struct task_struct *kcompactd;
1299 bool proactive_compact_trigger;
1302 * This is a per-node reserve of pages that are not available
1303 * to userspace allocations.
1305 unsigned long totalreserve_pages;
1309 * node reclaim becomes active if more unmapped pages exist.
1311 unsigned long min_unmapped_pages;
1312 unsigned long min_slab_pages;
1313 #endif /* CONFIG_NUMA */
1315 /* Write-intensive fields used by page reclaim */
1316 CACHELINE_PADDING(_pad1_);
1318 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1320 * If memory initialisation on large machines is deferred then this
1321 * is the first PFN that needs to be initialised.
1323 unsigned long first_deferred_pfn;
1324 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1326 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1327 struct deferred_split deferred_split_queue;
1330 #ifdef CONFIG_NUMA_BALANCING
1331 /* start time in ms of current promote rate limit period */
1332 unsigned int nbp_rl_start;
1333 /* number of promote candidate pages at start time of current rate limit period */
1334 unsigned long nbp_rl_nr_cand;
1335 /* promote threshold in ms */
1336 unsigned int nbp_threshold;
1337 /* start time in ms of current promote threshold adjustment period */
1338 unsigned int nbp_th_start;
1340 * number of promote candidate pages at start time of current promote
1341 * threshold adjustment period
1343 unsigned long nbp_th_nr_cand;
1345 /* Fields commonly accessed by the page reclaim scanner */
1348 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1350 * Use mem_cgroup_lruvec() to look up lruvecs.
1352 struct lruvec __lruvec;
1354 unsigned long flags;
1356 #ifdef CONFIG_LRU_GEN
1357 /* kswap mm walk data */
1358 struct lru_gen_mm_walk mm_walk;
1359 /* lru_gen_folio list */
1360 struct lru_gen_memcg memcg_lru;
1363 CACHELINE_PADDING(_pad2_);
1365 /* Per-node vmstats */
1366 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1367 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1369 struct memory_tier __rcu *memtier;
1371 #ifdef CONFIG_MEMORY_FAILURE
1372 struct memory_failure_stats mf_stats;
1376 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1377 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1379 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1380 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1382 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1384 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1387 #include <linux/memory_hotplug.h>
1389 void build_all_zonelists(pg_data_t *pgdat);
1390 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1391 enum zone_type highest_zoneidx);
1392 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1393 int highest_zoneidx, unsigned int alloc_flags,
1395 bool zone_watermark_ok(struct zone *z, unsigned int order,
1396 unsigned long mark, int highest_zoneidx,
1397 unsigned int alloc_flags);
1398 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1399 unsigned long mark, int highest_zoneidx);
1401 * Memory initialization context, use to differentiate memory added by
1402 * the platform statically or via memory hotplug interface.
1404 enum meminit_context {
1409 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1410 unsigned long size);
1412 extern void lruvec_init(struct lruvec *lruvec);
1414 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1417 return lruvec->pgdat;
1419 return container_of(lruvec, struct pglist_data, __lruvec);
1423 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1424 int local_memory_node(int node_id);
1426 static inline int local_memory_node(int node_id) { return node_id; };
1430 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1432 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1434 #ifdef CONFIG_ZONE_DEVICE
1435 static inline bool zone_is_zone_device(struct zone *zone)
1437 return zone_idx(zone) == ZONE_DEVICE;
1440 static inline bool zone_is_zone_device(struct zone *zone)
1447 * Returns true if a zone has pages managed by the buddy allocator.
1448 * All the reclaim decisions have to use this function rather than
1449 * populated_zone(). If the whole zone is reserved then we can easily
1450 * end up with populated_zone() && !managed_zone().
1452 static inline bool managed_zone(struct zone *zone)
1454 return zone_managed_pages(zone);
1457 /* Returns true if a zone has memory */
1458 static inline bool populated_zone(struct zone *zone)
1460 return zone->present_pages;
1464 static inline int zone_to_nid(struct zone *zone)
1469 static inline void zone_set_nid(struct zone *zone, int nid)
1474 static inline int zone_to_nid(struct zone *zone)
1479 static inline void zone_set_nid(struct zone *zone, int nid) {}
1482 extern int movable_zone;
1484 static inline int is_highmem_idx(enum zone_type idx)
1486 #ifdef CONFIG_HIGHMEM
1487 return (idx == ZONE_HIGHMEM ||
1488 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1495 * is_highmem - helper function to quickly check if a struct zone is a
1496 * highmem zone or not. This is an attempt to keep references
1497 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1498 * @zone: pointer to struct zone variable
1499 * Return: 1 for a highmem zone, 0 otherwise
1501 static inline int is_highmem(struct zone *zone)
1503 return is_highmem_idx(zone_idx(zone));
1506 #ifdef CONFIG_ZONE_DMA
1507 bool has_managed_dma(void);
1509 static inline bool has_managed_dma(void)
1515 /* These two functions are used to setup the per zone pages min values */
1518 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1520 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1521 size_t *, loff_t *);
1522 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1523 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1524 size_t *, loff_t *);
1525 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1526 void *, size_t *, loff_t *);
1527 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1528 void *, size_t *, loff_t *);
1529 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1530 void *, size_t *, loff_t *);
1531 int numa_zonelist_order_handler(struct ctl_table *, int,
1532 void *, size_t *, loff_t *);
1533 extern int percpu_pagelist_high_fraction;
1534 extern char numa_zonelist_order[];
1535 #define NUMA_ZONELIST_ORDER_LEN 16
1539 extern struct pglist_data contig_page_data;
1540 static inline struct pglist_data *NODE_DATA(int nid)
1542 return &contig_page_data;
1545 #else /* CONFIG_NUMA */
1547 #include <asm/mmzone.h>
1549 #endif /* !CONFIG_NUMA */
1551 extern struct pglist_data *first_online_pgdat(void);
1552 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1553 extern struct zone *next_zone(struct zone *zone);
1556 * for_each_online_pgdat - helper macro to iterate over all online nodes
1557 * @pgdat: pointer to a pg_data_t variable
1559 #define for_each_online_pgdat(pgdat) \
1560 for (pgdat = first_online_pgdat(); \
1562 pgdat = next_online_pgdat(pgdat))
1564 * for_each_zone - helper macro to iterate over all memory zones
1565 * @zone: pointer to struct zone variable
1567 * The user only needs to declare the zone variable, for_each_zone
1570 #define for_each_zone(zone) \
1571 for (zone = (first_online_pgdat())->node_zones; \
1573 zone = next_zone(zone))
1575 #define for_each_populated_zone(zone) \
1576 for (zone = (first_online_pgdat())->node_zones; \
1578 zone = next_zone(zone)) \
1579 if (!populated_zone(zone)) \
1580 ; /* do nothing */ \
1583 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1585 return zoneref->zone;
1588 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1590 return zoneref->zone_idx;
1593 static inline int zonelist_node_idx(struct zoneref *zoneref)
1595 return zone_to_nid(zoneref->zone);
1598 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1599 enum zone_type highest_zoneidx,
1603 * 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
1604 * @z: The cursor used as a starting point for the search
1605 * @highest_zoneidx: The zone index of the highest zone to return
1606 * @nodes: An optional nodemask to filter the zonelist with
1608 * This function returns the next zone at or below a given zone index that is
1609 * within the allowed nodemask using a cursor as the starting point for the
1610 * search. The zoneref returned is a cursor that represents the current zone
1611 * being examined. It should be advanced by one before calling
1612 * next_zones_zonelist again.
1614 * Return: the next zone at or below highest_zoneidx within the allowed
1615 * nodemask using a cursor within a zonelist as a starting point
1617 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1618 enum zone_type highest_zoneidx,
1621 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1623 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1627 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1628 * @zonelist: The zonelist to search for a suitable zone
1629 * @highest_zoneidx: The zone index of the highest zone to return
1630 * @nodes: An optional nodemask to filter the zonelist with
1632 * This function returns the first zone at or below a given zone index that is
1633 * within the allowed nodemask. The zoneref returned is a cursor that can be
1634 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1635 * one before calling.
1637 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1638 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1639 * update due to cpuset modification.
1641 * Return: Zoneref pointer for the first suitable zone found
1643 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1644 enum zone_type highest_zoneidx,
1647 return next_zones_zonelist(zonelist->_zonerefs,
1648 highest_zoneidx, nodes);
1652 * 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
1653 * @zone: The current zone in the iterator
1654 * @z: The current pointer within zonelist->_zonerefs being iterated
1655 * @zlist: The zonelist being iterated
1656 * @highidx: The zone index of the highest zone to return
1657 * @nodemask: Nodemask allowed by the allocator
1659 * This iterator iterates though all zones at or below a given zone index and
1660 * within a given nodemask
1662 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1663 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1665 z = next_zones_zonelist(++z, highidx, nodemask), \
1666 zone = zonelist_zone(z))
1668 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1669 for (zone = z->zone; \
1671 z = next_zones_zonelist(++z, highidx, nodemask), \
1672 zone = zonelist_zone(z))
1676 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1677 * @zone: The current zone in the iterator
1678 * @z: The current pointer within zonelist->zones being iterated
1679 * @zlist: The zonelist being iterated
1680 * @highidx: The zone index of the highest zone to return
1682 * This iterator iterates though all zones at or below a given zone index.
1684 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1685 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1687 /* Whether the 'nodes' are all movable nodes */
1688 static inline bool movable_only_nodes(nodemask_t *nodes)
1690 struct zonelist *zonelist;
1694 if (nodes_empty(*nodes))
1698 * We can chose arbitrary node from the nodemask to get a
1699 * zonelist as they are interlinked. We just need to find
1700 * at least one zone that can satisfy kernel allocations.
1702 nid = first_node(*nodes);
1703 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1704 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1705 return (!z->zone) ? true : false;
1709 #ifdef CONFIG_SPARSEMEM
1710 #include <asm/sparsemem.h>
1713 #ifdef CONFIG_FLATMEM
1714 #define pfn_to_nid(pfn) (0)
1717 #ifdef CONFIG_SPARSEMEM
1720 * PA_SECTION_SHIFT physical address to/from section number
1721 * PFN_SECTION_SHIFT pfn to/from section number
1723 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1724 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1726 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1728 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1729 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1731 #define SECTION_BLOCKFLAGS_BITS \
1732 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1734 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1735 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1738 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1740 return pfn >> PFN_SECTION_SHIFT;
1742 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1744 return sec << PFN_SECTION_SHIFT;
1747 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1748 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1750 #define SUBSECTION_SHIFT 21
1751 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1753 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1754 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1755 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1757 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1758 #error Subsection size exceeds section size
1760 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1763 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1764 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1766 struct mem_section_usage {
1767 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1768 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1770 /* See declaration of similar field in struct zone */
1771 unsigned long pageblock_flags[0];
1774 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1778 struct mem_section {
1780 * This is, logically, a pointer to an array of struct
1781 * pages. However, it is stored with some other magic.
1782 * (see sparse.c::sparse_init_one_section())
1784 * Additionally during early boot we encode node id of
1785 * the location of the section here to guide allocation.
1786 * (see sparse.c::memory_present())
1788 * Making it a UL at least makes someone do a cast
1789 * before using it wrong.
1791 unsigned long section_mem_map;
1793 struct mem_section_usage *usage;
1794 #ifdef CONFIG_PAGE_EXTENSION
1796 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1797 * section. (see page_ext.h about this.)
1799 struct page_ext *page_ext;
1803 * WARNING: mem_section must be a power-of-2 in size for the
1804 * calculation and use of SECTION_ROOT_MASK to make sense.
1808 #ifdef CONFIG_SPARSEMEM_EXTREME
1809 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1811 #define SECTIONS_PER_ROOT 1
1814 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1815 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1816 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1818 #ifdef CONFIG_SPARSEMEM_EXTREME
1819 extern struct mem_section **mem_section;
1821 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1824 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1826 return ms->usage->pageblock_flags;
1829 static inline struct mem_section *__nr_to_section(unsigned long nr)
1831 unsigned long root = SECTION_NR_TO_ROOT(nr);
1833 if (unlikely(root >= NR_SECTION_ROOTS))
1836 #ifdef CONFIG_SPARSEMEM_EXTREME
1837 if (!mem_section || !mem_section[root])
1840 return &mem_section[root][nr & SECTION_ROOT_MASK];
1842 extern size_t mem_section_usage_size(void);
1845 * We use the lower bits of the mem_map pointer to store
1846 * a little bit of information. The pointer is calculated
1847 * as mem_map - section_nr_to_pfn(pnum). The result is
1848 * aligned to the minimum alignment of the two values:
1849 * 1. All mem_map arrays are page-aligned.
1850 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1851 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1852 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1853 * worst combination is powerpc with 256k pages,
1854 * which results in PFN_SECTION_SHIFT equal 6.
1855 * To sum it up, at least 6 bits are available on all architectures.
1856 * However, we can exceed 6 bits on some other architectures except
1857 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1858 * with the worst case of 64K pages on arm64) if we make sure the
1859 * exceeded bit is not applicable to powerpc.
1862 SECTION_MARKED_PRESENT_BIT,
1863 SECTION_HAS_MEM_MAP_BIT,
1864 SECTION_IS_ONLINE_BIT,
1865 SECTION_IS_EARLY_BIT,
1866 #ifdef CONFIG_ZONE_DEVICE
1867 SECTION_TAINT_ZONE_DEVICE_BIT,
1869 SECTION_MAP_LAST_BIT,
1872 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1873 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1874 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1875 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1876 #ifdef CONFIG_ZONE_DEVICE
1877 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1879 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1880 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1882 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1884 unsigned long map = section->section_mem_map;
1885 map &= SECTION_MAP_MASK;
1886 return (struct page *)map;
1889 static inline int present_section(struct mem_section *section)
1891 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1894 static inline int present_section_nr(unsigned long nr)
1896 return present_section(__nr_to_section(nr));
1899 static inline int valid_section(struct mem_section *section)
1901 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1904 static inline int early_section(struct mem_section *section)
1906 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1909 static inline int valid_section_nr(unsigned long nr)
1911 return valid_section(__nr_to_section(nr));
1914 static inline int online_section(struct mem_section *section)
1916 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1919 #ifdef CONFIG_ZONE_DEVICE
1920 static inline int online_device_section(struct mem_section *section)
1922 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1924 return section && ((section->section_mem_map & flags) == flags);
1927 static inline int online_device_section(struct mem_section *section)
1933 static inline int online_section_nr(unsigned long nr)
1935 return online_section(__nr_to_section(nr));
1938 #ifdef CONFIG_MEMORY_HOTPLUG
1939 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1940 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1943 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1945 return __nr_to_section(pfn_to_section_nr(pfn));
1948 extern unsigned long __highest_present_section_nr;
1950 static inline int subsection_map_index(unsigned long pfn)
1952 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1955 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1956 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1958 int idx = subsection_map_index(pfn);
1960 return test_bit(idx, ms->usage->subsection_map);
1963 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1969 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1971 * pfn_valid - check if there is a valid memory map entry for a PFN
1972 * @pfn: the page frame number to check
1974 * Check if there is a valid memory map entry aka struct page for the @pfn.
1975 * Note, that availability of the memory map entry does not imply that
1976 * there is actual usable memory at that @pfn. The struct page may
1977 * represent a hole or an unusable page frame.
1979 * Return: 1 for PFNs that have memory map entries and 0 otherwise
1981 static inline int pfn_valid(unsigned long pfn)
1983 struct mem_section *ms;
1986 * Ensure the upper PAGE_SHIFT bits are clear in the
1987 * pfn. Else it might lead to false positives when
1988 * some of the upper bits are set, but the lower bits
1989 * match a valid pfn.
1991 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
1994 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1996 ms = __pfn_to_section(pfn);
1997 if (!valid_section(ms))
2000 * Traditionally early sections always returned pfn_valid() for
2001 * the entire section-sized span.
2003 return early_section(ms) || pfn_section_valid(ms, pfn);
2007 static inline int pfn_in_present_section(unsigned long pfn)
2009 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2011 return present_section(__pfn_to_section(pfn));
2014 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2016 while (++section_nr <= __highest_present_section_nr) {
2017 if (present_section_nr(section_nr))
2025 * These are _only_ used during initialisation, therefore they
2026 * can use __initdata ... They could have names to indicate
2030 #define pfn_to_nid(pfn) \
2032 unsigned long __pfn_to_nid_pfn = (pfn); \
2033 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2036 #define pfn_to_nid(pfn) (0)
2039 void sparse_init(void);
2041 #define sparse_init() do {} while (0)
2042 #define sparse_index_init(_sec, _nid) do {} while (0)
2043 #define pfn_in_present_section pfn_valid
2044 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2045 #endif /* CONFIG_SPARSEMEM */
2047 #endif /* !__GENERATING_BOUNDS.H */
2048 #endif /* !__ASSEMBLY__ */
2049 #endif /* _LINUX_MMZONE_H */