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;
113 static inline bool free_area_empty(struct free_area *area, int migratetype)
115 return list_empty(&area->free_list[migratetype]);
121 enum numa_stat_item {
122 NUMA_HIT, /* allocated in intended node */
123 NUMA_MISS, /* allocated in non intended node */
124 NUMA_FOREIGN, /* was intended here, hit elsewhere */
125 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
126 NUMA_LOCAL, /* allocation from local node */
127 NUMA_OTHER, /* allocation from other node */
128 NR_VM_NUMA_EVENT_ITEMS
131 #define NR_VM_NUMA_EVENT_ITEMS 0
134 enum zone_stat_item {
135 /* First 128 byte cacheline (assuming 64 bit words) */
137 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
138 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
140 NR_ZONE_INACTIVE_FILE,
143 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
144 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
145 /* Second 128 byte cacheline */
147 #if IS_ENABLED(CONFIG_ZSMALLOC)
148 NR_ZSPAGES, /* allocated in zsmalloc */
151 NR_VM_ZONE_STAT_ITEMS };
153 enum node_stat_item {
155 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
156 NR_ACTIVE_ANON, /* " " " " " */
157 NR_INACTIVE_FILE, /* " " " " " */
158 NR_ACTIVE_FILE, /* " " " " " */
159 NR_UNEVICTABLE, /* " " " " " */
160 NR_SLAB_RECLAIMABLE_B,
161 NR_SLAB_UNRECLAIMABLE_B,
162 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
163 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
165 WORKINGSET_REFAULT_BASE,
166 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
167 WORKINGSET_REFAULT_FILE,
168 WORKINGSET_ACTIVATE_BASE,
169 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
170 WORKINGSET_ACTIVATE_FILE,
171 WORKINGSET_RESTORE_BASE,
172 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
173 WORKINGSET_RESTORE_FILE,
174 WORKINGSET_NODERECLAIM,
175 NR_ANON_MAPPED, /* Mapped anonymous pages */
176 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
177 only modified from process context */
181 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
182 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
189 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
190 NR_DIRTIED, /* page dirtyings since bootup */
191 NR_WRITTEN, /* page writings since bootup */
192 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
193 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
194 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
195 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
196 NR_KERNEL_STACK_KB, /* measured in KiB */
197 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
198 NR_KERNEL_SCS_KB, /* measured in KiB */
200 NR_PAGETABLE, /* used for pagetables */
201 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
205 #ifdef CONFIG_NUMA_BALANCING
206 PGPROMOTE_SUCCESS, /* promote successfully */
207 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
209 NR_VM_NODE_STAT_ITEMS
213 * Returns true if the item should be printed in THPs (/proc/vmstat
214 * currently prints number of anon, file and shmem THPs. But the item
215 * is charged in pages).
217 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
219 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
222 return item == NR_ANON_THPS ||
223 item == NR_FILE_THPS ||
224 item == NR_SHMEM_THPS ||
225 item == NR_SHMEM_PMDMAPPED ||
226 item == NR_FILE_PMDMAPPED;
230 * Returns true if the value is measured in bytes (most vmstat values are
231 * measured in pages). This defines the API part, the internal representation
232 * might be different.
234 static __always_inline bool vmstat_item_in_bytes(int idx)
237 * Global and per-node slab counters track slab pages.
238 * It's expected that changes are multiples of PAGE_SIZE.
239 * Internally values are stored in pages.
241 * Per-memcg and per-lruvec counters track memory, consumed
242 * by individual slab objects. These counters are actually
245 return (idx == NR_SLAB_RECLAIMABLE_B ||
246 idx == NR_SLAB_UNRECLAIMABLE_B);
250 * We do arithmetic on the LRU lists in various places in the code,
251 * so it is important to keep the active lists LRU_ACTIVE higher in
252 * the array than the corresponding inactive lists, and to keep
253 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
255 * This has to be kept in sync with the statistics in zone_stat_item
256 * above and the descriptions in vmstat_text in mm/vmstat.c
263 LRU_INACTIVE_ANON = LRU_BASE,
264 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
265 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
266 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
271 enum vmscan_throttle_state {
272 VMSCAN_THROTTLE_WRITEBACK,
273 VMSCAN_THROTTLE_ISOLATED,
274 VMSCAN_THROTTLE_NOPROGRESS,
275 VMSCAN_THROTTLE_CONGESTED,
279 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
281 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
283 static inline bool is_file_lru(enum lru_list lru)
285 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
288 static inline bool is_active_lru(enum lru_list lru)
290 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
293 #define WORKINGSET_ANON 0
294 #define WORKINGSET_FILE 1
295 #define ANON_AND_FILE 2
298 LRUVEC_CONGESTED, /* lruvec has many dirty pages
299 * backed by a congested BDI
303 #endif /* !__GENERATING_BOUNDS_H */
306 * Evictable pages are divided into multiple generations. The youngest and the
307 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
308 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
309 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
310 * corresponding generation. The gen counter in folio->flags stores gen+1 while
311 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
313 * A page is added to the youngest generation on faulting. The aging needs to
314 * check the accessed bit at least twice before handing this page over to the
315 * eviction. The first check takes care of the accessed bit set on the initial
316 * fault; the second check makes sure this page hasn't been used since then.
317 * This process, AKA second chance, requires a minimum of two generations,
318 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
319 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
320 * rest of generations, if they exist, are considered inactive. See
321 * lru_gen_is_active().
323 * PG_active is always cleared while a page is on one of lrugen->folios[] so
324 * that the aging needs not to worry about it. And it's set again when a page
325 * considered active is isolated for non-reclaiming purposes, e.g., migration.
326 * See lru_gen_add_folio() and lru_gen_del_folio().
328 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
329 * number of categories of the active/inactive LRU when keeping track of
330 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
333 #define MIN_NR_GENS 2U
334 #define MAX_NR_GENS 4U
337 * Each generation is divided into multiple tiers. A page accessed N times
338 * through file descriptors is in tier order_base_2(N). A page in the first tier
339 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
340 * tables or read ahead. A page in any other tier (N>1) is marked by
341 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
342 * supported without using additional bits in folio->flags.
344 * In contrast to moving across generations which requires the LRU lock, moving
345 * across tiers only involves atomic operations on folio->flags and therefore
346 * has a negligible cost in the buffered access path. In the eviction path,
347 * comparisons of refaulted/(evicted+protected) from the first tier and the
348 * rest infer whether pages accessed multiple times through file descriptors
349 * are statistically hot and thus worth protecting.
351 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
352 * number of categories of the active/inactive LRU when keeping track of
353 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
356 #define MAX_NR_TIERS 4U
358 #ifndef __GENERATING_BOUNDS_H
361 struct page_vma_mapped_walk;
363 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
364 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
366 #ifdef CONFIG_LRU_GEN
376 LRU_GEN_NONLEAF_YOUNG,
380 #define MIN_LRU_BATCH BITS_PER_LONG
381 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
383 /* whether to keep historical stats from evicted generations */
384 #ifdef CONFIG_LRU_GEN_STATS
385 #define NR_HIST_GENS MAX_NR_GENS
387 #define NR_HIST_GENS 1U
391 * The youngest generation number is stored in max_seq for both anon and file
392 * types as they are aged on an equal footing. The oldest generation numbers are
393 * stored in min_seq[] separately for anon and file types as clean file pages
394 * can be evicted regardless of swap constraints.
396 * Normally anon and file min_seq are in sync. But if swapping is constrained,
397 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
400 * The number of pages in each generation is eventually consistent and therefore
401 * can be transiently negative when reset_batch_size() is pending.
403 struct lru_gen_folio {
404 /* the aging increments the youngest generation number */
405 unsigned long max_seq;
406 /* the eviction increments the oldest generation numbers */
407 unsigned long min_seq[ANON_AND_FILE];
408 /* the birth time of each generation in jiffies */
409 unsigned long timestamps[MAX_NR_GENS];
410 /* the multi-gen LRU lists, lazily sorted on eviction */
411 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
412 /* the multi-gen LRU sizes, eventually consistent */
413 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
414 /* the exponential moving average of refaulted */
415 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
416 /* the exponential moving average of evicted+protected */
417 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
418 /* the first tier doesn't need protection, hence the minus one */
419 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
420 /* can be modified without holding the LRU lock */
421 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
422 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
423 /* whether the multi-gen LRU is enabled */
426 /* the memcg generation this lru_gen_folio belongs to */
428 /* the list segment this lru_gen_folio belongs to */
430 /* per-node lru_gen_folio list for global reclaim */
431 struct hlist_nulls_node list;
436 MM_LEAF_TOTAL, /* total leaf entries */
437 MM_LEAF_OLD, /* old leaf entries */
438 MM_LEAF_YOUNG, /* young leaf entries */
439 MM_NONLEAF_TOTAL, /* total non-leaf entries */
440 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
441 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
445 /* double-buffering Bloom filters */
446 #define NR_BLOOM_FILTERS 2
448 struct lru_gen_mm_state {
449 /* set to max_seq after each iteration */
451 /* where the current iteration continues (inclusive) */
452 struct list_head *head;
453 /* where the last iteration ended (exclusive) */
454 struct list_head *tail;
455 /* to wait for the last page table walker to finish */
456 struct wait_queue_head wait;
457 /* Bloom filters flip after each iteration */
458 unsigned long *filters[NR_BLOOM_FILTERS];
459 /* the mm stats for debugging */
460 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
461 /* the number of concurrent page table walkers */
465 struct lru_gen_mm_walk {
466 /* the lruvec under reclaim */
467 struct lruvec *lruvec;
468 /* unstable max_seq from lru_gen_folio */
469 unsigned long max_seq;
470 /* the next address within an mm to scan */
471 unsigned long next_addr;
472 /* to batch promoted pages */
473 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
474 /* to batch the mm stats */
475 int mm_stats[NR_MM_STATS];
476 /* total batched items */
482 void lru_gen_init_lruvec(struct lruvec *lruvec);
483 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
488 * For each node, memcgs are divided into two generations: the old and the
489 * young. For each generation, memcgs are randomly sharded into multiple bins
490 * to improve scalability. For each bin, the hlist_nulls is virtually divided
491 * into three segments: the head, the tail and the default.
493 * An onlining memcg is added to the tail of a random bin in the old generation.
494 * The eviction starts at the head of a random bin in the old generation. The
495 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
496 * the old generation, is incremented when all its bins become empty.
498 * There are four operations:
499 * 1. MEMCG_LRU_HEAD, which moves an memcg to the head of a random bin in its
500 * current generation (old or young) and updates its "seg" to "head";
501 * 2. MEMCG_LRU_TAIL, which moves an memcg to the tail of a random bin in its
502 * current generation (old or young) and updates its "seg" to "tail";
503 * 3. MEMCG_LRU_OLD, which moves an memcg to the head of a random bin in the old
504 * generation, updates its "gen" to "old" and resets its "seg" to "default";
505 * 4. MEMCG_LRU_YOUNG, which moves an memcg to the tail of a random bin in the
506 * young generation, updates its "gen" to "young" and resets its "seg" to
509 * The events that trigger the above operations are:
510 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
511 * 2. The first attempt to reclaim an memcg below low, which triggers
513 * 3. The first attempt to reclaim an memcg below reclaimable size threshold,
514 * which triggers MEMCG_LRU_TAIL;
515 * 4. The second attempt to reclaim an memcg below reclaimable size threshold,
516 * which triggers MEMCG_LRU_YOUNG;
517 * 5. Attempting to reclaim an memcg below min, which triggers MEMCG_LRU_YOUNG;
518 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
519 * 7. Offlining an memcg, which triggers MEMCG_LRU_OLD.
521 * Note that memcg LRU only applies to global reclaim, and the round-robin
522 * incrementing of their max_seq counters ensures the eventual fairness to all
523 * eligible memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
525 #define MEMCG_NR_GENS 2
526 #define MEMCG_NR_BINS 8
528 struct lru_gen_memcg {
529 /* the per-node memcg generation counter */
531 /* each memcg has one lru_gen_folio per node */
532 unsigned long nr_memcgs[MEMCG_NR_GENS];
533 /* per-node lru_gen_folio list for global reclaim */
534 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
535 /* protects the above */
539 void lru_gen_init_pgdat(struct pglist_data *pgdat);
541 void lru_gen_init_memcg(struct mem_cgroup *memcg);
542 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
543 void lru_gen_online_memcg(struct mem_cgroup *memcg);
544 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
545 void lru_gen_release_memcg(struct mem_cgroup *memcg);
546 void lru_gen_soft_reclaim(struct lruvec *lruvec);
548 #else /* !CONFIG_MEMCG */
550 #define MEMCG_NR_GENS 1
552 struct lru_gen_memcg {
555 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
559 #endif /* CONFIG_MEMCG */
561 #else /* !CONFIG_LRU_GEN */
563 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
567 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
571 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
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 lruvec *lruvec)
601 #endif /* CONFIG_MEMCG */
603 #endif /* CONFIG_LRU_GEN */
606 struct list_head lists[NR_LRU_LISTS];
607 /* per lruvec lru_lock for memcg */
610 * These track the cost of reclaiming one LRU - file or anon -
611 * over the other. As the observed cost of reclaiming one LRU
612 * increases, the reclaim scan balance tips toward the other.
614 unsigned long anon_cost;
615 unsigned long file_cost;
616 /* Non-resident age, driven by LRU movement */
617 atomic_long_t nonresident_age;
618 /* Refaults at the time of last reclaim cycle */
619 unsigned long refaults[ANON_AND_FILE];
620 /* Various lruvec state flags (enum lruvec_flags) */
622 #ifdef CONFIG_LRU_GEN
623 /* evictable pages divided into generations */
624 struct lru_gen_folio lrugen;
625 /* to concurrently iterate lru_gen_mm_list */
626 struct lru_gen_mm_state mm_state;
629 struct pglist_data *pgdat;
633 /* Isolate unmapped pages */
634 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
635 /* Isolate for asynchronous migration */
636 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
637 /* Isolate unevictable pages */
638 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
640 /* LRU Isolation modes. */
641 typedef unsigned __bitwise isolate_mode_t;
643 enum zone_watermarks {
652 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
653 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
654 * it should not contribute to serious fragmentation causing THP allocation
657 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
662 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
663 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
665 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
666 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
667 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
668 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
670 /* Fields and list protected by pagesets local_lock in page_alloc.c */
671 struct per_cpu_pages {
672 spinlock_t lock; /* Protects lists field */
673 int count; /* number of pages in the list */
674 int high; /* high watermark, emptying needed */
675 int batch; /* chunk size for buddy add/remove */
676 short free_factor; /* batch scaling factor during free */
678 short expire; /* When 0, remote pagesets are drained */
681 /* Lists of pages, one per migrate type stored on the pcp-lists */
682 struct list_head lists[NR_PCP_LISTS];
683 } ____cacheline_aligned_in_smp;
685 struct per_cpu_zonestat {
687 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
692 * Low priority inaccurate counters that are only folded
693 * on demand. Use a large type to avoid the overhead of
694 * folding during refresh_cpu_vm_stats.
696 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
700 struct per_cpu_nodestat {
702 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
705 #endif /* !__GENERATING_BOUNDS.H */
709 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
710 * to DMA to all of the addressable memory (ZONE_NORMAL).
711 * On architectures where this area covers the whole 32 bit address
712 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
713 * DMA addressing constraints. This distinction is important as a 32bit
714 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
715 * platforms may need both zones as they support peripherals with
716 * different DMA addressing limitations.
718 #ifdef CONFIG_ZONE_DMA
721 #ifdef CONFIG_ZONE_DMA32
725 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
726 * performed on pages in ZONE_NORMAL if the DMA devices support
727 * transfers to all addressable memory.
730 #ifdef CONFIG_HIGHMEM
732 * A memory area that is only addressable by the kernel through
733 * mapping portions into its own address space. This is for example
734 * used by i386 to allow the kernel to address the memory beyond
735 * 900MB. The kernel will set up special mappings (page
736 * table entries on i386) for each page that the kernel needs to
742 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
743 * movable pages with few exceptional cases described below. Main use
744 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
745 * likely to succeed, and to locally limit unmovable allocations - e.g.,
746 * to increase the number of THP/huge pages. Notable special cases are:
748 * 1. Pinned pages: (long-term) pinning of movable pages might
749 * essentially turn such pages unmovable. Therefore, we do not allow
750 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
751 * faulted, they come from the right zone right away. However, it is
752 * still possible that address space already has pages in
753 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
754 * touches that memory before pinning). In such case we migrate them
755 * to a different zone. When migration fails - pinning fails.
756 * 2. memblock allocations: kernelcore/movablecore setups might create
757 * situations where ZONE_MOVABLE contains unmovable allocations
758 * after boot. Memory offlining and allocations fail early.
759 * 3. Memory holes: kernelcore/movablecore setups might create very rare
760 * situations where ZONE_MOVABLE contains memory holes after boot,
761 * for example, if we have sections that are only partially
762 * populated. Memory offlining and allocations fail early.
763 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
764 * memory offlining, such pages cannot be allocated.
765 * 5. Unmovable PG_offline pages: in paravirtualized environments,
766 * hotplugged memory blocks might only partially be managed by the
767 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
768 * parts not manged by the buddy are unmovable PG_offline pages. In
769 * some cases (virtio-mem), such pages can be skipped during
770 * memory offlining, however, cannot be moved/allocated. These
771 * techniques might use alloc_contig_range() to hide previously
772 * exposed pages from the buddy again (e.g., to implement some sort
773 * of memory unplug in virtio-mem).
774 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
775 * situations where ZERO_PAGE(0) which is allocated differently
776 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
777 * cannot be migrated.
778 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
779 * memory to the MOVABLE zone, the vmemmap pages are also placed in
780 * such zone. Such pages cannot be really moved around as they are
781 * self-stored in the range, but they are treated as movable when
782 * the range they describe is about to be offlined.
784 * In general, no unmovable allocations that degrade memory offlining
785 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
786 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
787 * if has_unmovable_pages() states that there are no unmovable pages,
788 * there can be false negatives).
791 #ifdef CONFIG_ZONE_DEVICE
798 #ifndef __GENERATING_BOUNDS_H
800 #define ASYNC_AND_SYNC 2
803 /* Read-mostly fields */
805 /* zone watermarks, access with *_wmark_pages(zone) macros */
806 unsigned long _watermark[NR_WMARK];
807 unsigned long watermark_boost;
809 unsigned long nr_reserved_highatomic;
812 * We don't know if the memory that we're going to allocate will be
813 * freeable or/and it will be released eventually, so to avoid totally
814 * wasting several GB of ram we must reserve some of the lower zone
815 * memory (otherwise we risk to run OOM on the lower zones despite
816 * there being tons of freeable ram on the higher zones). This array is
817 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
820 long lowmem_reserve[MAX_NR_ZONES];
825 struct pglist_data *zone_pgdat;
826 struct per_cpu_pages __percpu *per_cpu_pageset;
827 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
829 * the high and batch values are copied to individual pagesets for
835 #ifndef CONFIG_SPARSEMEM
837 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
838 * In SPARSEMEM, this map is stored in struct mem_section
840 unsigned long *pageblock_flags;
841 #endif /* CONFIG_SPARSEMEM */
843 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
844 unsigned long zone_start_pfn;
847 * spanned_pages is the total pages spanned by the zone, including
848 * holes, which is calculated as:
849 * spanned_pages = zone_end_pfn - zone_start_pfn;
851 * present_pages is physical pages existing within the zone, which
853 * present_pages = spanned_pages - absent_pages(pages in holes);
855 * present_early_pages is present pages existing within the zone
856 * located on memory available since early boot, excluding hotplugged
859 * managed_pages is present pages managed by the buddy system, which
860 * is calculated as (reserved_pages includes pages allocated by the
861 * bootmem allocator):
862 * managed_pages = present_pages - reserved_pages;
864 * cma pages is present pages that are assigned for CMA use
867 * So present_pages may be used by memory hotplug or memory power
868 * management logic to figure out unmanaged pages by checking
869 * (present_pages - managed_pages). And managed_pages should be used
870 * by page allocator and vm scanner to calculate all kinds of watermarks
875 * zone_start_pfn and spanned_pages are protected by span_seqlock.
876 * It is a seqlock because it has to be read outside of zone->lock,
877 * and it is done in the main allocator path. But, it is written
878 * quite infrequently.
880 * The span_seq lock is declared along with zone->lock because it is
881 * frequently read in proximity to zone->lock. It's good to
882 * give them a chance of being in the same cacheline.
884 * Write access to present_pages at runtime should be protected by
885 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
886 * present_pages should use get_online_mems() to get a stable value.
888 atomic_long_t managed_pages;
889 unsigned long spanned_pages;
890 unsigned long present_pages;
891 #if defined(CONFIG_MEMORY_HOTPLUG)
892 unsigned long present_early_pages;
895 unsigned long cma_pages;
900 #ifdef CONFIG_MEMORY_ISOLATION
902 * Number of isolated pageblock. It is used to solve incorrect
903 * freepage counting problem due to racy retrieving migratetype
904 * of pageblock. Protected by zone->lock.
906 unsigned long nr_isolate_pageblock;
909 #ifdef CONFIG_MEMORY_HOTPLUG
910 /* see spanned/present_pages for more description */
911 seqlock_t span_seqlock;
916 /* Write-intensive fields used from the page allocator */
917 CACHELINE_PADDING(_pad1_);
919 /* free areas of different sizes */
920 struct free_area free_area[MAX_ORDER + 1];
922 /* zone flags, see below */
925 /* Primarily protects free_area */
928 /* Write-intensive fields used by compaction and vmstats. */
929 CACHELINE_PADDING(_pad2_);
932 * When free pages are below this point, additional steps are taken
933 * when reading the number of free pages to avoid per-cpu counter
934 * drift allowing watermarks to be breached
936 unsigned long percpu_drift_mark;
938 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
939 /* pfn where compaction free scanner should start */
940 unsigned long compact_cached_free_pfn;
941 /* pfn where compaction migration scanner should start */
942 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
943 unsigned long compact_init_migrate_pfn;
944 unsigned long compact_init_free_pfn;
947 #ifdef CONFIG_COMPACTION
949 * On compaction failure, 1<<compact_defer_shift compactions
950 * are skipped before trying again. The number attempted since
951 * last failure is tracked with compact_considered.
952 * compact_order_failed is the minimum compaction failed order.
954 unsigned int compact_considered;
955 unsigned int compact_defer_shift;
956 int compact_order_failed;
959 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
960 /* Set to true when the PG_migrate_skip bits should be cleared */
961 bool compact_blockskip_flush;
966 CACHELINE_PADDING(_pad3_);
967 /* Zone statistics */
968 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
969 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
970 } ____cacheline_internodealigned_in_smp;
973 PGDAT_DIRTY, /* reclaim scanning has recently found
974 * many dirty file pages at the tail
977 PGDAT_WRITEBACK, /* reclaim scanning has recently found
978 * many pages under writeback
980 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
984 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
985 * Cleared when kswapd is woken.
987 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
990 static inline unsigned long zone_managed_pages(struct zone *zone)
992 return (unsigned long)atomic_long_read(&zone->managed_pages);
995 static inline unsigned long zone_cma_pages(struct zone *zone)
998 return zone->cma_pages;
1004 static inline unsigned long zone_end_pfn(const struct zone *zone)
1006 return zone->zone_start_pfn + zone->spanned_pages;
1009 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1011 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1014 static inline bool zone_is_initialized(struct zone *zone)
1016 return zone->initialized;
1019 static inline bool zone_is_empty(struct zone *zone)
1021 return zone->spanned_pages == 0;
1024 #ifndef BUILD_VDSO32_64
1026 * The zone field is never updated after free_area_init_core()
1027 * sets it, so none of the operations on it need to be atomic.
1030 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1031 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1032 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1033 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1034 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1035 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1036 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1037 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1040 * Define the bit shifts to access each section. For non-existent
1041 * sections we define the shift as 0; that plus a 0 mask ensures
1042 * the compiler will optimise away reference to them.
1044 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1045 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1046 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1047 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1048 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1050 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1051 #ifdef NODE_NOT_IN_PAGE_FLAGS
1052 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1053 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1054 SECTIONS_PGOFF : ZONES_PGOFF)
1056 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1057 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1058 NODES_PGOFF : ZONES_PGOFF)
1061 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1063 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1064 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1065 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1066 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1067 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1068 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1070 static inline enum zone_type page_zonenum(const struct page *page)
1072 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1073 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1076 static inline enum zone_type folio_zonenum(const struct folio *folio)
1078 return page_zonenum(&folio->page);
1081 #ifdef CONFIG_ZONE_DEVICE
1082 static inline bool is_zone_device_page(const struct page *page)
1084 return page_zonenum(page) == ZONE_DEVICE;
1088 * Consecutive zone device pages should not be merged into the same sgl
1089 * or bvec segment with other types of pages or if they belong to different
1090 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1091 * without scanning the entire segment. This helper returns true either if
1092 * both pages are not zone device pages or both pages are zone device pages
1093 * with the same pgmap.
1095 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1096 const struct page *b)
1098 if (is_zone_device_page(a) != is_zone_device_page(b))
1100 if (!is_zone_device_page(a))
1102 return a->pgmap == b->pgmap;
1105 extern void memmap_init_zone_device(struct zone *, unsigned long,
1106 unsigned long, struct dev_pagemap *);
1108 static inline bool is_zone_device_page(const struct page *page)
1112 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1113 const struct page *b)
1119 static inline bool folio_is_zone_device(const struct folio *folio)
1121 return is_zone_device_page(&folio->page);
1124 static inline bool is_zone_movable_page(const struct page *page)
1126 return page_zonenum(page) == ZONE_MOVABLE;
1131 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1132 * intersection with the given zone
1134 static inline bool zone_intersects(struct zone *zone,
1135 unsigned long start_pfn, unsigned long nr_pages)
1137 if (zone_is_empty(zone))
1139 if (start_pfn >= zone_end_pfn(zone) ||
1140 start_pfn + nr_pages <= zone->zone_start_pfn)
1147 * The "priority" of VM scanning is how much of the queues we will scan in one
1148 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1149 * queues ("queue_length >> 12") during an aging round.
1151 #define DEF_PRIORITY 12
1153 /* Maximum number of zones on a zonelist */
1154 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1157 ZONELIST_FALLBACK, /* zonelist with fallback */
1160 * The NUMA zonelists are doubled because we need zonelists that
1161 * restrict the allocations to a single node for __GFP_THISNODE.
1163 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1169 * This struct contains information about a zone in a zonelist. It is stored
1170 * here to avoid dereferences into large structures and lookups of tables
1173 struct zone *zone; /* Pointer to actual zone */
1174 int zone_idx; /* zone_idx(zoneref->zone) */
1178 * One allocation request operates on a zonelist. A zonelist
1179 * is a list of zones, the first one is the 'goal' of the
1180 * allocation, the other zones are fallback zones, in decreasing
1183 * To speed the reading of the zonelist, the zonerefs contain the zone index
1184 * of the entry being read. Helper functions to access information given
1185 * a struct zoneref are
1187 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1188 * zonelist_zone_idx() - Return the index of the zone for an entry
1189 * zonelist_node_idx() - Return the index of the node for an entry
1192 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1196 * The array of struct pages for flatmem.
1197 * It must be declared for SPARSEMEM as well because there are configurations
1198 * that rely on that.
1200 extern struct page *mem_map;
1202 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1203 struct deferred_split {
1204 spinlock_t split_queue_lock;
1205 struct list_head split_queue;
1206 unsigned long split_queue_len;
1210 #ifdef CONFIG_MEMORY_FAILURE
1212 * Per NUMA node memory failure handling statistics.
1214 struct memory_failure_stats {
1216 * Number of raw pages poisoned.
1217 * Cases not accounted: memory outside kernel control, offline page,
1218 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1219 * error events, and unpoison actions from hwpoison_unpoison.
1221 unsigned long total;
1223 * Recovery results of poisoned raw pages handled by memory_failure,
1224 * in sync with mf_result.
1225 * total = ignored + failed + delayed + recovered.
1226 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1228 unsigned long ignored;
1229 unsigned long failed;
1230 unsigned long delayed;
1231 unsigned long recovered;
1236 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1237 * it's memory layout. On UMA machines there is a single pglist_data which
1238 * describes the whole memory.
1240 * Memory statistics and page replacement data structures are maintained on a
1243 typedef struct pglist_data {
1245 * node_zones contains just the zones for THIS node. Not all of the
1246 * zones may be populated, but it is the full list. It is referenced by
1247 * this node's node_zonelists as well as other node's node_zonelists.
1249 struct zone node_zones[MAX_NR_ZONES];
1252 * node_zonelists contains references to all zones in all nodes.
1253 * Generally the first zones will be references to this node's
1256 struct zonelist node_zonelists[MAX_ZONELISTS];
1258 int nr_zones; /* number of populated zones in this node */
1259 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1260 struct page *node_mem_map;
1261 #ifdef CONFIG_PAGE_EXTENSION
1262 struct page_ext *node_page_ext;
1265 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1267 * Must be held any time you expect node_start_pfn,
1268 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1269 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1272 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1273 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1274 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1276 * Nests above zone->lock and zone->span_seqlock
1278 spinlock_t node_size_lock;
1280 unsigned long node_start_pfn;
1281 unsigned long node_present_pages; /* total number of physical pages */
1282 unsigned long node_spanned_pages; /* total size of physical page
1283 range, including holes */
1285 wait_queue_head_t kswapd_wait;
1286 wait_queue_head_t pfmemalloc_wait;
1288 /* workqueues for throttling reclaim for different reasons. */
1289 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1291 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1292 unsigned long nr_reclaim_start; /* nr pages written while throttled
1293 * when throttling started. */
1294 #ifdef CONFIG_MEMORY_HOTPLUG
1295 struct mutex kswapd_lock;
1297 struct task_struct *kswapd; /* Protected by kswapd_lock */
1299 enum zone_type kswapd_highest_zoneidx;
1301 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1303 #ifdef CONFIG_COMPACTION
1304 int kcompactd_max_order;
1305 enum zone_type kcompactd_highest_zoneidx;
1306 wait_queue_head_t kcompactd_wait;
1307 struct task_struct *kcompactd;
1308 bool proactive_compact_trigger;
1311 * This is a per-node reserve of pages that are not available
1312 * to userspace allocations.
1314 unsigned long totalreserve_pages;
1318 * node reclaim becomes active if more unmapped pages exist.
1320 unsigned long min_unmapped_pages;
1321 unsigned long min_slab_pages;
1322 #endif /* CONFIG_NUMA */
1324 /* Write-intensive fields used by page reclaim */
1325 CACHELINE_PADDING(_pad1_);
1327 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1329 * If memory initialisation on large machines is deferred then this
1330 * is the first PFN that needs to be initialised.
1332 unsigned long first_deferred_pfn;
1333 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1335 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1336 struct deferred_split deferred_split_queue;
1339 #ifdef CONFIG_NUMA_BALANCING
1340 /* start time in ms of current promote rate limit period */
1341 unsigned int nbp_rl_start;
1342 /* number of promote candidate pages at start time of current rate limit period */
1343 unsigned long nbp_rl_nr_cand;
1344 /* promote threshold in ms */
1345 unsigned int nbp_threshold;
1346 /* start time in ms of current promote threshold adjustment period */
1347 unsigned int nbp_th_start;
1349 * number of promote candidate pages at start time of current promote
1350 * threshold adjustment period
1352 unsigned long nbp_th_nr_cand;
1354 /* Fields commonly accessed by the page reclaim scanner */
1357 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1359 * Use mem_cgroup_lruvec() to look up lruvecs.
1361 struct lruvec __lruvec;
1363 unsigned long flags;
1365 #ifdef CONFIG_LRU_GEN
1366 /* kswap mm walk data */
1367 struct lru_gen_mm_walk mm_walk;
1368 /* lru_gen_folio list */
1369 struct lru_gen_memcg memcg_lru;
1372 CACHELINE_PADDING(_pad2_);
1374 /* Per-node vmstats */
1375 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1376 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1378 struct memory_tier __rcu *memtier;
1380 #ifdef CONFIG_MEMORY_FAILURE
1381 struct memory_failure_stats mf_stats;
1385 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1386 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1388 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1389 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1391 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1393 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1396 #include <linux/memory_hotplug.h>
1398 void build_all_zonelists(pg_data_t *pgdat);
1399 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1400 enum zone_type highest_zoneidx);
1401 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1402 int highest_zoneidx, unsigned int alloc_flags,
1404 bool zone_watermark_ok(struct zone *z, unsigned int order,
1405 unsigned long mark, int highest_zoneidx,
1406 unsigned int alloc_flags);
1407 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1408 unsigned long mark, int highest_zoneidx);
1410 * Memory initialization context, use to differentiate memory added by
1411 * the platform statically or via memory hotplug interface.
1413 enum meminit_context {
1418 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1419 unsigned long size);
1421 extern void lruvec_init(struct lruvec *lruvec);
1423 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1426 return lruvec->pgdat;
1428 return container_of(lruvec, struct pglist_data, __lruvec);
1432 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1433 int local_memory_node(int node_id);
1435 static inline int local_memory_node(int node_id) { return node_id; };
1439 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1441 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1443 #ifdef CONFIG_ZONE_DEVICE
1444 static inline bool zone_is_zone_device(struct zone *zone)
1446 return zone_idx(zone) == ZONE_DEVICE;
1449 static inline bool zone_is_zone_device(struct zone *zone)
1456 * Returns true if a zone has pages managed by the buddy allocator.
1457 * All the reclaim decisions have to use this function rather than
1458 * populated_zone(). If the whole zone is reserved then we can easily
1459 * end up with populated_zone() && !managed_zone().
1461 static inline bool managed_zone(struct zone *zone)
1463 return zone_managed_pages(zone);
1466 /* Returns true if a zone has memory */
1467 static inline bool populated_zone(struct zone *zone)
1469 return zone->present_pages;
1473 static inline int zone_to_nid(struct zone *zone)
1478 static inline void zone_set_nid(struct zone *zone, int nid)
1483 static inline int zone_to_nid(struct zone *zone)
1488 static inline void zone_set_nid(struct zone *zone, int nid) {}
1491 extern int movable_zone;
1493 static inline int is_highmem_idx(enum zone_type idx)
1495 #ifdef CONFIG_HIGHMEM
1496 return (idx == ZONE_HIGHMEM ||
1497 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1504 * is_highmem - helper function to quickly check if a struct zone is a
1505 * highmem zone or not. This is an attempt to keep references
1506 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1507 * @zone: pointer to struct zone variable
1508 * Return: 1 for a highmem zone, 0 otherwise
1510 static inline int is_highmem(struct zone *zone)
1512 return is_highmem_idx(zone_idx(zone));
1515 #ifdef CONFIG_ZONE_DMA
1516 bool has_managed_dma(void);
1518 static inline bool has_managed_dma(void)
1524 /* These two functions are used to setup the per zone pages min values */
1527 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1529 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1530 size_t *, loff_t *);
1531 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1532 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1533 size_t *, loff_t *);
1534 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1535 void *, size_t *, loff_t *);
1536 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1537 void *, size_t *, loff_t *);
1538 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1539 void *, size_t *, loff_t *);
1540 int numa_zonelist_order_handler(struct ctl_table *, int,
1541 void *, size_t *, loff_t *);
1542 extern int percpu_pagelist_high_fraction;
1543 extern char numa_zonelist_order[];
1544 #define NUMA_ZONELIST_ORDER_LEN 16
1548 extern struct pglist_data contig_page_data;
1549 static inline struct pglist_data *NODE_DATA(int nid)
1551 return &contig_page_data;
1554 #else /* CONFIG_NUMA */
1556 #include <asm/mmzone.h>
1558 #endif /* !CONFIG_NUMA */
1560 extern struct pglist_data *first_online_pgdat(void);
1561 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1562 extern struct zone *next_zone(struct zone *zone);
1565 * for_each_online_pgdat - helper macro to iterate over all online nodes
1566 * @pgdat: pointer to a pg_data_t variable
1568 #define for_each_online_pgdat(pgdat) \
1569 for (pgdat = first_online_pgdat(); \
1571 pgdat = next_online_pgdat(pgdat))
1573 * for_each_zone - helper macro to iterate over all memory zones
1574 * @zone: pointer to struct zone variable
1576 * The user only needs to declare the zone variable, for_each_zone
1579 #define for_each_zone(zone) \
1580 for (zone = (first_online_pgdat())->node_zones; \
1582 zone = next_zone(zone))
1584 #define for_each_populated_zone(zone) \
1585 for (zone = (first_online_pgdat())->node_zones; \
1587 zone = next_zone(zone)) \
1588 if (!populated_zone(zone)) \
1589 ; /* do nothing */ \
1592 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1594 return zoneref->zone;
1597 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1599 return zoneref->zone_idx;
1602 static inline int zonelist_node_idx(struct zoneref *zoneref)
1604 return zone_to_nid(zoneref->zone);
1607 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1608 enum zone_type highest_zoneidx,
1612 * 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
1613 * @z: The cursor used as a starting point for the search
1614 * @highest_zoneidx: The zone index of the highest zone to return
1615 * @nodes: An optional nodemask to filter the zonelist with
1617 * This function returns the next zone at or below a given zone index that is
1618 * within the allowed nodemask using a cursor as the starting point for the
1619 * search. The zoneref returned is a cursor that represents the current zone
1620 * being examined. It should be advanced by one before calling
1621 * next_zones_zonelist again.
1623 * Return: the next zone at or below highest_zoneidx within the allowed
1624 * nodemask using a cursor within a zonelist as a starting point
1626 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1627 enum zone_type highest_zoneidx,
1630 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1632 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1636 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1637 * @zonelist: The zonelist to search for a suitable zone
1638 * @highest_zoneidx: The zone index of the highest zone to return
1639 * @nodes: An optional nodemask to filter the zonelist with
1641 * This function returns the first zone at or below a given zone index that is
1642 * within the allowed nodemask. The zoneref returned is a cursor that can be
1643 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1644 * one before calling.
1646 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1647 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1648 * update due to cpuset modification.
1650 * Return: Zoneref pointer for the first suitable zone found
1652 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1653 enum zone_type highest_zoneidx,
1656 return next_zones_zonelist(zonelist->_zonerefs,
1657 highest_zoneidx, nodes);
1661 * 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
1662 * @zone: The current zone in the iterator
1663 * @z: The current pointer within zonelist->_zonerefs being iterated
1664 * @zlist: The zonelist being iterated
1665 * @highidx: The zone index of the highest zone to return
1666 * @nodemask: Nodemask allowed by the allocator
1668 * This iterator iterates though all zones at or below a given zone index and
1669 * within a given nodemask
1671 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1672 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1674 z = next_zones_zonelist(++z, highidx, nodemask), \
1675 zone = zonelist_zone(z))
1677 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1678 for (zone = z->zone; \
1680 z = next_zones_zonelist(++z, highidx, nodemask), \
1681 zone = zonelist_zone(z))
1685 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1686 * @zone: The current zone in the iterator
1687 * @z: The current pointer within zonelist->zones being iterated
1688 * @zlist: The zonelist being iterated
1689 * @highidx: The zone index of the highest zone to return
1691 * This iterator iterates though all zones at or below a given zone index.
1693 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1694 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1696 /* Whether the 'nodes' are all movable nodes */
1697 static inline bool movable_only_nodes(nodemask_t *nodes)
1699 struct zonelist *zonelist;
1703 if (nodes_empty(*nodes))
1707 * We can chose arbitrary node from the nodemask to get a
1708 * zonelist as they are interlinked. We just need to find
1709 * at least one zone that can satisfy kernel allocations.
1711 nid = first_node(*nodes);
1712 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1713 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1714 return (!z->zone) ? true : false;
1718 #ifdef CONFIG_SPARSEMEM
1719 #include <asm/sparsemem.h>
1722 #ifdef CONFIG_FLATMEM
1723 #define pfn_to_nid(pfn) (0)
1726 #ifdef CONFIG_SPARSEMEM
1729 * PA_SECTION_SHIFT physical address to/from section number
1730 * PFN_SECTION_SHIFT pfn to/from section number
1732 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1733 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1735 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1737 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1738 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1740 #define SECTION_BLOCKFLAGS_BITS \
1741 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1743 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1744 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1747 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1749 return pfn >> PFN_SECTION_SHIFT;
1751 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1753 return sec << PFN_SECTION_SHIFT;
1756 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1757 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1759 #define SUBSECTION_SHIFT 21
1760 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1762 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1763 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1764 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1766 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1767 #error Subsection size exceeds section size
1769 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1772 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1773 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1775 struct mem_section_usage {
1776 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1777 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1779 /* See declaration of similar field in struct zone */
1780 unsigned long pageblock_flags[0];
1783 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1787 struct mem_section {
1789 * This is, logically, a pointer to an array of struct
1790 * pages. However, it is stored with some other magic.
1791 * (see sparse.c::sparse_init_one_section())
1793 * Additionally during early boot we encode node id of
1794 * the location of the section here to guide allocation.
1795 * (see sparse.c::memory_present())
1797 * Making it a UL at least makes someone do a cast
1798 * before using it wrong.
1800 unsigned long section_mem_map;
1802 struct mem_section_usage *usage;
1803 #ifdef CONFIG_PAGE_EXTENSION
1805 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1806 * section. (see page_ext.h about this.)
1808 struct page_ext *page_ext;
1812 * WARNING: mem_section must be a power-of-2 in size for the
1813 * calculation and use of SECTION_ROOT_MASK to make sense.
1817 #ifdef CONFIG_SPARSEMEM_EXTREME
1818 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1820 #define SECTIONS_PER_ROOT 1
1823 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1824 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1825 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1827 #ifdef CONFIG_SPARSEMEM_EXTREME
1828 extern struct mem_section **mem_section;
1830 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1833 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1835 return ms->usage->pageblock_flags;
1838 static inline struct mem_section *__nr_to_section(unsigned long nr)
1840 unsigned long root = SECTION_NR_TO_ROOT(nr);
1842 if (unlikely(root >= NR_SECTION_ROOTS))
1845 #ifdef CONFIG_SPARSEMEM_EXTREME
1846 if (!mem_section || !mem_section[root])
1849 return &mem_section[root][nr & SECTION_ROOT_MASK];
1851 extern size_t mem_section_usage_size(void);
1854 * We use the lower bits of the mem_map pointer to store
1855 * a little bit of information. The pointer is calculated
1856 * as mem_map - section_nr_to_pfn(pnum). The result is
1857 * aligned to the minimum alignment of the two values:
1858 * 1. All mem_map arrays are page-aligned.
1859 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1860 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1861 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1862 * worst combination is powerpc with 256k pages,
1863 * which results in PFN_SECTION_SHIFT equal 6.
1864 * To sum it up, at least 6 bits are available on all architectures.
1865 * However, we can exceed 6 bits on some other architectures except
1866 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1867 * with the worst case of 64K pages on arm64) if we make sure the
1868 * exceeded bit is not applicable to powerpc.
1871 SECTION_MARKED_PRESENT_BIT,
1872 SECTION_HAS_MEM_MAP_BIT,
1873 SECTION_IS_ONLINE_BIT,
1874 SECTION_IS_EARLY_BIT,
1875 #ifdef CONFIG_ZONE_DEVICE
1876 SECTION_TAINT_ZONE_DEVICE_BIT,
1878 SECTION_MAP_LAST_BIT,
1881 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1882 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1883 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1884 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1885 #ifdef CONFIG_ZONE_DEVICE
1886 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1888 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1889 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1891 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1893 unsigned long map = section->section_mem_map;
1894 map &= SECTION_MAP_MASK;
1895 return (struct page *)map;
1898 static inline int present_section(struct mem_section *section)
1900 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1903 static inline int present_section_nr(unsigned long nr)
1905 return present_section(__nr_to_section(nr));
1908 static inline int valid_section(struct mem_section *section)
1910 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1913 static inline int early_section(struct mem_section *section)
1915 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1918 static inline int valid_section_nr(unsigned long nr)
1920 return valid_section(__nr_to_section(nr));
1923 static inline int online_section(struct mem_section *section)
1925 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1928 #ifdef CONFIG_ZONE_DEVICE
1929 static inline int online_device_section(struct mem_section *section)
1931 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1933 return section && ((section->section_mem_map & flags) == flags);
1936 static inline int online_device_section(struct mem_section *section)
1942 static inline int online_section_nr(unsigned long nr)
1944 return online_section(__nr_to_section(nr));
1947 #ifdef CONFIG_MEMORY_HOTPLUG
1948 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1949 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1952 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1954 return __nr_to_section(pfn_to_section_nr(pfn));
1957 extern unsigned long __highest_present_section_nr;
1959 static inline int subsection_map_index(unsigned long pfn)
1961 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1964 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1965 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1967 int idx = subsection_map_index(pfn);
1969 return test_bit(idx, ms->usage->subsection_map);
1972 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1978 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1980 * pfn_valid - check if there is a valid memory map entry for a PFN
1981 * @pfn: the page frame number to check
1983 * Check if there is a valid memory map entry aka struct page for the @pfn.
1984 * Note, that availability of the memory map entry does not imply that
1985 * there is actual usable memory at that @pfn. The struct page may
1986 * represent a hole or an unusable page frame.
1988 * Return: 1 for PFNs that have memory map entries and 0 otherwise
1990 static inline int pfn_valid(unsigned long pfn)
1992 struct mem_section *ms;
1995 * Ensure the upper PAGE_SHIFT bits are clear in the
1996 * pfn. Else it might lead to false positives when
1997 * some of the upper bits are set, but the lower bits
1998 * match a valid pfn.
2000 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2003 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2005 ms = __pfn_to_section(pfn);
2006 if (!valid_section(ms))
2009 * Traditionally early sections always returned pfn_valid() for
2010 * the entire section-sized span.
2012 return early_section(ms) || pfn_section_valid(ms, pfn);
2016 static inline int pfn_in_present_section(unsigned long pfn)
2018 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2020 return present_section(__pfn_to_section(pfn));
2023 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2025 while (++section_nr <= __highest_present_section_nr) {
2026 if (present_section_nr(section_nr))
2034 * These are _only_ used during initialisation, therefore they
2035 * can use __initdata ... They could have names to indicate
2039 #define pfn_to_nid(pfn) \
2041 unsigned long __pfn_to_nid_pfn = (pfn); \
2042 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2045 #define pfn_to_nid(pfn) (0)
2048 void sparse_init(void);
2050 #define sparse_init() do {} while (0)
2051 #define sparse_index_init(_sec, _nid) do {} while (0)
2052 #define pfn_in_present_section pfn_valid
2053 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2054 #endif /* CONFIG_SPARSEMEM */
2056 #endif /* !__GENERATING_BOUNDS.H */
2057 #endif /* !__ASSEMBLY__ */
2058 #endif /* _LINUX_MMZONE_H */