1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_MMZONE_H
3#define _LINUX_MMZONE_H
4
5#ifndef __ASSEMBLY__
6#ifndef __GENERATING_BOUNDS_H
7
8#include <linux/spinlock.h>
9#include <linux/list.h>
10#include <linux/list_nulls.h>
11#include <linux/wait.h>
12#include <linux/bitops.h>
13#include <linux/cache.h>
14#include <linux/threads.h>
15#include <linux/numa.h>
16#include <linux/init.h>
17#include <linux/seqlock.h>
18#include <linux/nodemask.h>
19#include <linux/pageblock-flags.h>
20#include <linux/page-flags-layout.h>
21#include <linux/atomic.h>
22#include <linux/mm_types.h>
23#include <linux/page-flags.h>
24#include <linux/local_lock.h>
25#include <asm/page.h>
26
27/* Free memory management - zoned buddy allocator. */
28#ifndef CONFIG_ARCH_FORCE_MAX_ORDER
29#define MAX_ORDER 10
30#else
31#define MAX_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
32#endif
33#define MAX_ORDER_NR_PAGES (1 << MAX_ORDER)
34
35#define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
36
37/*
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
41 * will not.
42 */
43#define PAGE_ALLOC_COSTLY_ORDER 3
44
45enum migratetype {
46 MIGRATE_UNMOVABLE,
47 MIGRATE_MOVABLE,
48 MIGRATE_RECLAIMABLE,
49 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
50 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
51#ifdef CONFIG_CMA
52 /*
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.
57 *
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.
61 */
62 MIGRATE_CMA,
63#endif
64#ifdef CONFIG_MEMORY_ISOLATION
65 MIGRATE_ISOLATE, /* can't allocate from here */
66#endif
67 MIGRATE_TYPES
68};
69
70/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
71extern const char * const migratetype_names[MIGRATE_TYPES];
72
73#ifdef CONFIG_CMA
74# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
75# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76#else
77# define is_migrate_cma(migratetype) false
78# define is_migrate_cma_page(_page) false
79#endif
80
81static inline bool is_migrate_movable(int mt)
82{
83 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
84}
85
86/*
87 * Check whether a migratetype can be merged with another migratetype.
88 *
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.
91 */
92static inline bool migratetype_is_mergeable(int mt)
93{
94 return mt < MIGRATE_PCPTYPES;
95}
96
97#define for_each_migratetype_order(order, type) \
98 for (order = 0; order <= MAX_ORDER; order++) \
99 for (type = 0; type < MIGRATE_TYPES; type++)
100
101extern int page_group_by_mobility_disabled;
102
103#define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
104
105#define get_pageblock_migratetype(page) \
106 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
107
108#define folio_migratetype(folio) \
109 get_pfnblock_flags_mask(&folio->page, folio_pfn(folio), \
110 MIGRATETYPE_MASK)
111struct free_area {
112 struct list_head free_list[MIGRATE_TYPES];
113 unsigned long nr_free;
114};
115
116struct pglist_data;
117
118#ifdef CONFIG_NUMA
119enum numa_stat_item {
120 NUMA_HIT, /* allocated in intended node */
121 NUMA_MISS, /* allocated in non intended node */
122 NUMA_FOREIGN, /* was intended here, hit elsewhere */
123 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
124 NUMA_LOCAL, /* allocation from local node */
125 NUMA_OTHER, /* allocation from other node */
126 NR_VM_NUMA_EVENT_ITEMS
127};
128#else
129#define NR_VM_NUMA_EVENT_ITEMS 0
130#endif
131
132enum zone_stat_item {
133 /* First 128 byte cacheline (assuming 64 bit words) */
134 NR_FREE_PAGES,
135 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
136 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
137 NR_ZONE_ACTIVE_ANON,
138 NR_ZONE_INACTIVE_FILE,
139 NR_ZONE_ACTIVE_FILE,
140 NR_ZONE_UNEVICTABLE,
141 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
142 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
143 /* Second 128 byte cacheline */
144 NR_BOUNCE,
145#if IS_ENABLED(CONFIG_ZSMALLOC)
146 NR_ZSPAGES, /* allocated in zsmalloc */
147#endif
148 NR_FREE_CMA_PAGES,
149#ifdef CONFIG_UNACCEPTED_MEMORY
150 NR_UNACCEPTED,
151#endif
152 NR_VM_ZONE_STAT_ITEMS };
153
154enum node_stat_item {
155 NR_LRU_BASE,
156 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
157 NR_ACTIVE_ANON, /* " " " " " */
158 NR_INACTIVE_FILE, /* " " " " " */
159 NR_ACTIVE_FILE, /* " " " " " */
160 NR_UNEVICTABLE, /* " " " " " */
161 NR_SLAB_RECLAIMABLE_B,
162 NR_SLAB_UNRECLAIMABLE_B,
163 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
164 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
165 WORKINGSET_NODES,
166 WORKINGSET_REFAULT_BASE,
167 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
168 WORKINGSET_REFAULT_FILE,
169 WORKINGSET_ACTIVATE_BASE,
170 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
171 WORKINGSET_ACTIVATE_FILE,
172 WORKINGSET_RESTORE_BASE,
173 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
174 WORKINGSET_RESTORE_FILE,
175 WORKINGSET_NODERECLAIM,
176 NR_ANON_MAPPED, /* Mapped anonymous pages */
177 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
178 only modified from process context */
179 NR_FILE_PAGES,
180 NR_FILE_DIRTY,
181 NR_WRITEBACK,
182 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
183 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
184 NR_SHMEM_THPS,
185 NR_SHMEM_PMDMAPPED,
186 NR_FILE_THPS,
187 NR_FILE_PMDMAPPED,
188 NR_ANON_THPS,
189 NR_VMSCAN_WRITE,
190 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
191 NR_DIRTIED, /* page dirtyings since bootup */
192 NR_WRITTEN, /* page writings since bootup */
193 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
194 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
195 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
196 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
197 NR_KERNEL_STACK_KB, /* measured in KiB */
198#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
199 NR_KERNEL_SCS_KB, /* measured in KiB */
200#endif
201 NR_PAGETABLE, /* used for pagetables */
202 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
203#ifdef CONFIG_SWAP
204 NR_SWAPCACHE,
205#endif
206#ifdef CONFIG_NUMA_BALANCING
207 PGPROMOTE_SUCCESS, /* promote successfully */
208 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
209#endif
210 NR_VM_NODE_STAT_ITEMS
211};
212
213/*
214 * Returns true if the item should be printed in THPs (/proc/vmstat
215 * currently prints number of anon, file and shmem THPs. But the item
216 * is charged in pages).
217 */
218static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
219{
220 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
221 return false;
222
223 return item == NR_ANON_THPS ||
224 item == NR_FILE_THPS ||
225 item == NR_SHMEM_THPS ||
226 item == NR_SHMEM_PMDMAPPED ||
227 item == NR_FILE_PMDMAPPED;
228}
229
230/*
231 * Returns true if the value is measured in bytes (most vmstat values are
232 * measured in pages). This defines the API part, the internal representation
233 * might be different.
234 */
235static __always_inline bool vmstat_item_in_bytes(int idx)
236{
237 /*
238 * Global and per-node slab counters track slab pages.
239 * It's expected that changes are multiples of PAGE_SIZE.
240 * Internally values are stored in pages.
241 *
242 * Per-memcg and per-lruvec counters track memory, consumed
243 * by individual slab objects. These counters are actually
244 * byte-precise.
245 */
246 return (idx == NR_SLAB_RECLAIMABLE_B ||
247 idx == NR_SLAB_UNRECLAIMABLE_B);
248}
249
250/*
251 * We do arithmetic on the LRU lists in various places in the code,
252 * so it is important to keep the active lists LRU_ACTIVE higher in
253 * the array than the corresponding inactive lists, and to keep
254 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
255 *
256 * This has to be kept in sync with the statistics in zone_stat_item
257 * above and the descriptions in vmstat_text in mm/vmstat.c
258 */
259#define LRU_BASE 0
260#define LRU_ACTIVE 1
261#define LRU_FILE 2
262
263enum lru_list {
264 LRU_INACTIVE_ANON = LRU_BASE,
265 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
266 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
267 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
268 LRU_UNEVICTABLE,
269 NR_LRU_LISTS
270};
271
272enum vmscan_throttle_state {
273 VMSCAN_THROTTLE_WRITEBACK,
274 VMSCAN_THROTTLE_ISOLATED,
275 VMSCAN_THROTTLE_NOPROGRESS,
276 VMSCAN_THROTTLE_CONGESTED,
277 NR_VMSCAN_THROTTLE,
278};
279
280#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
281
282#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
283
284static inline bool is_file_lru(enum lru_list lru)
285{
286 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
287}
288
289static inline bool is_active_lru(enum lru_list lru)
290{
291 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
292}
293
294#define WORKINGSET_ANON 0
295#define WORKINGSET_FILE 1
296#define ANON_AND_FILE 2
297
298enum lruvec_flags {
299 /*
300 * An lruvec has many dirty pages backed by a congested BDI:
301 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
302 * It can be cleared by cgroup reclaim or kswapd.
303 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
304 * It can only be cleared by kswapd.
305 *
306 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
307 * reclaim, but not vice versa. This only applies to the root cgroup.
308 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
309 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
310 * by kswapd).
311 */
312 LRUVEC_CGROUP_CONGESTED,
313 LRUVEC_NODE_CONGESTED,
314};
315
316#endif /* !__GENERATING_BOUNDS_H */
317
318/*
319 * Evictable pages are divided into multiple generations. The youngest and the
320 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
321 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
322 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
323 * corresponding generation. The gen counter in folio->flags stores gen+1 while
324 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
325 *
326 * A page is added to the youngest generation on faulting. The aging needs to
327 * check the accessed bit at least twice before handing this page over to the
328 * eviction. The first check takes care of the accessed bit set on the initial
329 * fault; the second check makes sure this page hasn't been used since then.
330 * This process, AKA second chance, requires a minimum of two generations,
331 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
332 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
333 * rest of generations, if they exist, are considered inactive. See
334 * lru_gen_is_active().
335 *
336 * PG_active is always cleared while a page is on one of lrugen->folios[] so
337 * that the aging needs not to worry about it. And it's set again when a page
338 * considered active is isolated for non-reclaiming purposes, e.g., migration.
339 * See lru_gen_add_folio() and lru_gen_del_folio().
340 *
341 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
342 * number of categories of the active/inactive LRU when keeping track of
343 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
344 * in folio->flags.
345 */
346#define MIN_NR_GENS 2U
347#define MAX_NR_GENS 4U
348
349/*
350 * Each generation is divided into multiple tiers. A page accessed N times
351 * through file descriptors is in tier order_base_2(N). A page in the first tier
352 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
353 * tables or read ahead. A page in any other tier (N>1) is marked by
354 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
355 * supported without using additional bits in folio->flags.
356 *
357 * In contrast to moving across generations which requires the LRU lock, moving
358 * across tiers only involves atomic operations on folio->flags and therefore
359 * has a negligible cost in the buffered access path. In the eviction path,
360 * comparisons of refaulted/(evicted+protected) from the first tier and the
361 * rest infer whether pages accessed multiple times through file descriptors
362 * are statistically hot and thus worth protecting.
363 *
364 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
365 * number of categories of the active/inactive LRU when keeping track of
366 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
367 * folio->flags.
368 */
369#define MAX_NR_TIERS 4U
370
371#ifndef __GENERATING_BOUNDS_H
372
373struct lruvec;
374struct page_vma_mapped_walk;
375
376#define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
377#define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
378
379#ifdef CONFIG_LRU_GEN
380
381enum {
382 LRU_GEN_ANON,
383 LRU_GEN_FILE,
384};
385
386enum {
387 LRU_GEN_CORE,
388 LRU_GEN_MM_WALK,
389 LRU_GEN_NONLEAF_YOUNG,
390 NR_LRU_GEN_CAPS
391};
392
393#define MIN_LRU_BATCH BITS_PER_LONG
394#define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
395
396/* whether to keep historical stats from evicted generations */
397#ifdef CONFIG_LRU_GEN_STATS
398#define NR_HIST_GENS MAX_NR_GENS
399#else
400#define NR_HIST_GENS 1U
401#endif
402
403/*
404 * The youngest generation number is stored in max_seq for both anon and file
405 * types as they are aged on an equal footing. The oldest generation numbers are
406 * stored in min_seq[] separately for anon and file types as clean file pages
407 * can be evicted regardless of swap constraints.
408 *
409 * Normally anon and file min_seq are in sync. But if swapping is constrained,
410 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
411 * min_seq behind.
412 *
413 * The number of pages in each generation is eventually consistent and therefore
414 * can be transiently negative when reset_batch_size() is pending.
415 */
416struct lru_gen_folio {
417 /* the aging increments the youngest generation number */
418 unsigned long max_seq;
419 /* the eviction increments the oldest generation numbers */
420 unsigned long min_seq[ANON_AND_FILE];
421 /* the birth time of each generation in jiffies */
422 unsigned long timestamps[MAX_NR_GENS];
423 /* the multi-gen LRU lists, lazily sorted on eviction */
424 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
425 /* the multi-gen LRU sizes, eventually consistent */
426 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
427 /* the exponential moving average of refaulted */
428 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
429 /* the exponential moving average of evicted+protected */
430 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
431 /* the first tier doesn't need protection, hence the minus one */
432 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
433 /* can be modified without holding the LRU lock */
434 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
435 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
436 /* whether the multi-gen LRU is enabled */
437 bool enabled;
438#ifdef CONFIG_MEMCG
439 /* the memcg generation this lru_gen_folio belongs to */
440 u8 gen;
441 /* the list segment this lru_gen_folio belongs to */
442 u8 seg;
443 /* per-node lru_gen_folio list for global reclaim */
444 struct hlist_nulls_node list;
445#endif
446};
447
448enum {
449 MM_LEAF_TOTAL, /* total leaf entries */
450 MM_LEAF_OLD, /* old leaf entries */
451 MM_LEAF_YOUNG, /* young leaf entries */
452 MM_NONLEAF_TOTAL, /* total non-leaf entries */
453 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
454 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
455 NR_MM_STATS
456};
457
458/* double-buffering Bloom filters */
459#define NR_BLOOM_FILTERS 2
460
461struct lru_gen_mm_state {
462 /* set to max_seq after each iteration */
463 unsigned long seq;
464 /* where the current iteration continues after */
465 struct list_head *head;
466 /* where the last iteration ended before */
467 struct list_head *tail;
468 /* Bloom filters flip after each iteration */
469 unsigned long *filters[NR_BLOOM_FILTERS];
470 /* the mm stats for debugging */
471 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
472};
473
474struct lru_gen_mm_walk {
475 /* the lruvec under reclaim */
476 struct lruvec *lruvec;
477 /* unstable max_seq from lru_gen_folio */
478 unsigned long max_seq;
479 /* the next address within an mm to scan */
480 unsigned long next_addr;
481 /* to batch promoted pages */
482 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
483 /* to batch the mm stats */
484 int mm_stats[NR_MM_STATS];
485 /* total batched items */
486 int batched;
487 bool can_swap;
488 bool force_scan;
489};
490
491void lru_gen_init_lruvec(struct lruvec *lruvec);
492void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
493
494#ifdef CONFIG_MEMCG
495
496/*
497 * For each node, memcgs are divided into two generations: the old and the
498 * young. For each generation, memcgs are randomly sharded into multiple bins
499 * to improve scalability. For each bin, the hlist_nulls is virtually divided
500 * into three segments: the head, the tail and the default.
501 *
502 * An onlining memcg is added to the tail of a random bin in the old generation.
503 * The eviction starts at the head of a random bin in the old generation. The
504 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
505 * the old generation, is incremented when all its bins become empty.
506 *
507 * There are four operations:
508 * 1. MEMCG_LRU_HEAD, which moves an memcg to the head of a random bin in its
509 * current generation (old or young) and updates its "seg" to "head";
510 * 2. MEMCG_LRU_TAIL, which moves an memcg to the tail of a random bin in its
511 * current generation (old or young) and updates its "seg" to "tail";
512 * 3. MEMCG_LRU_OLD, which moves an memcg to the head of a random bin in the old
513 * generation, updates its "gen" to "old" and resets its "seg" to "default";
514 * 4. MEMCG_LRU_YOUNG, which moves an memcg to the tail of a random bin in the
515 * young generation, updates its "gen" to "young" and resets its "seg" to
516 * "default".
517 *
518 * The events that trigger the above operations are:
519 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
520 * 2. The first attempt to reclaim an memcg below low, which triggers
521 * MEMCG_LRU_TAIL;
522 * 3. The first attempt to reclaim an memcg below reclaimable size threshold,
523 * which triggers MEMCG_LRU_TAIL;
524 * 4. The second attempt to reclaim an memcg below reclaimable size threshold,
525 * which triggers MEMCG_LRU_YOUNG;
526 * 5. Attempting to reclaim an memcg below min, which triggers MEMCG_LRU_YOUNG;
527 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
528 * 7. Offlining an memcg, which triggers MEMCG_LRU_OLD.
529 *
530 * Note that memcg LRU only applies to global reclaim, and the round-robin
531 * incrementing of their max_seq counters ensures the eventual fairness to all
532 * eligible memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
533 */
534#define MEMCG_NR_GENS 2
535#define MEMCG_NR_BINS 8
536
537struct lru_gen_memcg {
538 /* the per-node memcg generation counter */
539 unsigned long seq;
540 /* each memcg has one lru_gen_folio per node */
541 unsigned long nr_memcgs[MEMCG_NR_GENS];
542 /* per-node lru_gen_folio list for global reclaim */
543 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
544 /* protects the above */
545 spinlock_t lock;
546};
547
548void lru_gen_init_pgdat(struct pglist_data *pgdat);
549
550void lru_gen_init_memcg(struct mem_cgroup *memcg);
551void lru_gen_exit_memcg(struct mem_cgroup *memcg);
552void lru_gen_online_memcg(struct mem_cgroup *memcg);
553void lru_gen_offline_memcg(struct mem_cgroup *memcg);
554void lru_gen_release_memcg(struct mem_cgroup *memcg);
555void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
556
557#else /* !CONFIG_MEMCG */
558
559#define MEMCG_NR_GENS 1
560
561struct lru_gen_memcg {
562};
563
564static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
565{
566}
567
568#endif /* CONFIG_MEMCG */
569
570#else /* !CONFIG_LRU_GEN */
571
572static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
573{
574}
575
576static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
577{
578}
579
580static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
581{
582}
583
584#ifdef CONFIG_MEMCG
585
586static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
587{
588}
589
590static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
591{
592}
593
594static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
595{
596}
597
598static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
599{
600}
601
602static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
603{
604}
605
606static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
607{
608}
609
610#endif /* CONFIG_MEMCG */
611
612#endif /* CONFIG_LRU_GEN */
613
614struct lruvec {
615 struct list_head lists[NR_LRU_LISTS];
616 /* per lruvec lru_lock for memcg */
617 spinlock_t lru_lock;
618 /*
619 * These track the cost of reclaiming one LRU - file or anon -
620 * over the other. As the observed cost of reclaiming one LRU
621 * increases, the reclaim scan balance tips toward the other.
622 */
623 unsigned long anon_cost;
624 unsigned long file_cost;
625 /* Non-resident age, driven by LRU movement */
626 atomic_long_t nonresident_age;
627 /* Refaults at the time of last reclaim cycle */
628 unsigned long refaults[ANON_AND_FILE];
629 /* Various lruvec state flags (enum lruvec_flags) */
630 unsigned long flags;
631#ifdef CONFIG_LRU_GEN
632 /* evictable pages divided into generations */
633 struct lru_gen_folio lrugen;
634 /* to concurrently iterate lru_gen_mm_list */
635 struct lru_gen_mm_state mm_state;
636#endif
637#ifdef CONFIG_MEMCG
638 struct pglist_data *pgdat;
639#endif
640};
641
642/* Isolate for asynchronous migration */
643#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
644/* Isolate unevictable pages */
645#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
646
647/* LRU Isolation modes. */
648typedef unsigned __bitwise isolate_mode_t;
649
650enum zone_watermarks {
651 WMARK_MIN,
652 WMARK_LOW,
653 WMARK_HIGH,
654 WMARK_PROMO,
655 NR_WMARK
656};
657
658/*
659 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
660 * for THP which will usually be GFP_MOVABLE. Even if it is another type,
661 * it should not contribute to serious fragmentation causing THP allocation
662 * failures.
663 */
664#ifdef CONFIG_TRANSPARENT_HUGEPAGE
665#define NR_PCP_THP 1
666#else
667#define NR_PCP_THP 0
668#endif
669#define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
670#define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
671
672#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
673#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
674#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
675#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
676
677/*
678 * Flags used in pcp->flags field.
679 *
680 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
681 * previous page freeing. To avoid to drain PCP for an accident
682 * high-order page freeing.
683 *
684 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
685 * draining PCP for consecutive high-order pages freeing without
686 * allocation if data cache slice of CPU is large enough. To reduce
687 * zone lock contention and keep cache-hot pages reusing.
688 */
689#define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
690#define PCPF_FREE_HIGH_BATCH BIT(1)
691
692struct per_cpu_pages {
693 spinlock_t lock; /* Protects lists field */
694 int count; /* number of pages in the list */
695 int high; /* high watermark, emptying needed */
696 int high_min; /* min high watermark */
697 int high_max; /* max high watermark */
698 int batch; /* chunk size for buddy add/remove */
699 u8 flags; /* protected by pcp->lock */
700 u8 alloc_factor; /* batch scaling factor during allocate */
701#ifdef CONFIG_NUMA
702 u8 expire; /* When 0, remote pagesets are drained */
703#endif
704 short free_count; /* consecutive free count */
705
706 /* Lists of pages, one per migrate type stored on the pcp-lists */
707 struct list_head lists[NR_PCP_LISTS];
708} ____cacheline_aligned_in_smp;
709
710struct per_cpu_zonestat {
711#ifdef CONFIG_SMP
712 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
713 s8 stat_threshold;
714#endif
715#ifdef CONFIG_NUMA
716 /*
717 * Low priority inaccurate counters that are only folded
718 * on demand. Use a large type to avoid the overhead of
719 * folding during refresh_cpu_vm_stats.
720 */
721 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
722#endif
723};
724
725struct per_cpu_nodestat {
726 s8 stat_threshold;
727 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
728};
729
730#endif /* !__GENERATING_BOUNDS.H */
731
732enum zone_type {
733 /*
734 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
735 * to DMA to all of the addressable memory (ZONE_NORMAL).
736 * On architectures where this area covers the whole 32 bit address
737 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
738 * DMA addressing constraints. This distinction is important as a 32bit
739 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
740 * platforms may need both zones as they support peripherals with
741 * different DMA addressing limitations.
742 */
743#ifdef CONFIG_ZONE_DMA
744 ZONE_DMA,
745#endif
746#ifdef CONFIG_ZONE_DMA32
747 ZONE_DMA32,
748#endif
749 /*
750 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
751 * performed on pages in ZONE_NORMAL if the DMA devices support
752 * transfers to all addressable memory.
753 */
754 ZONE_NORMAL,
755#ifdef CONFIG_HIGHMEM
756 /*
757 * A memory area that is only addressable by the kernel through
758 * mapping portions into its own address space. This is for example
759 * used by i386 to allow the kernel to address the memory beyond
760 * 900MB. The kernel will set up special mappings (page
761 * table entries on i386) for each page that the kernel needs to
762 * access.
763 */
764 ZONE_HIGHMEM,
765#endif
766 /*
767 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
768 * movable pages with few exceptional cases described below. Main use
769 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
770 * likely to succeed, and to locally limit unmovable allocations - e.g.,
771 * to increase the number of THP/huge pages. Notable special cases are:
772 *
773 * 1. Pinned pages: (long-term) pinning of movable pages might
774 * essentially turn such pages unmovable. Therefore, we do not allow
775 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
776 * faulted, they come from the right zone right away. However, it is
777 * still possible that address space already has pages in
778 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
779 * touches that memory before pinning). In such case we migrate them
780 * to a different zone. When migration fails - pinning fails.
781 * 2. memblock allocations: kernelcore/movablecore setups might create
782 * situations where ZONE_MOVABLE contains unmovable allocations
783 * after boot. Memory offlining and allocations fail early.
784 * 3. Memory holes: kernelcore/movablecore setups might create very rare
785 * situations where ZONE_MOVABLE contains memory holes after boot,
786 * for example, if we have sections that are only partially
787 * populated. Memory offlining and allocations fail early.
788 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
789 * memory offlining, such pages cannot be allocated.
790 * 5. Unmovable PG_offline pages: in paravirtualized environments,
791 * hotplugged memory blocks might only partially be managed by the
792 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
793 * parts not manged by the buddy are unmovable PG_offline pages. In
794 * some cases (virtio-mem), such pages can be skipped during
795 * memory offlining, however, cannot be moved/allocated. These
796 * techniques might use alloc_contig_range() to hide previously
797 * exposed pages from the buddy again (e.g., to implement some sort
798 * of memory unplug in virtio-mem).
799 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
800 * situations where ZERO_PAGE(0) which is allocated differently
801 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
802 * cannot be migrated.
803 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
804 * memory to the MOVABLE zone, the vmemmap pages are also placed in
805 * such zone. Such pages cannot be really moved around as they are
806 * self-stored in the range, but they are treated as movable when
807 * the range they describe is about to be offlined.
808 *
809 * In general, no unmovable allocations that degrade memory offlining
810 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
811 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
812 * if has_unmovable_pages() states that there are no unmovable pages,
813 * there can be false negatives).
814 */
815 ZONE_MOVABLE,
816#ifdef CONFIG_ZONE_DEVICE
817 ZONE_DEVICE,
818#endif
819 __MAX_NR_ZONES
820
821};
822
823#ifndef __GENERATING_BOUNDS_H
824
825#define ASYNC_AND_SYNC 2
826
827struct zone {
828 /* Read-mostly fields */
829
830 /* zone watermarks, access with *_wmark_pages(zone) macros */
831 unsigned long _watermark[NR_WMARK];
832 unsigned long watermark_boost;
833
834 unsigned long nr_reserved_highatomic;
835
836 /*
837 * We don't know if the memory that we're going to allocate will be
838 * freeable or/and it will be released eventually, so to avoid totally
839 * wasting several GB of ram we must reserve some of the lower zone
840 * memory (otherwise we risk to run OOM on the lower zones despite
841 * there being tons of freeable ram on the higher zones). This array is
842 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
843 * changes.
844 */
845 long lowmem_reserve[MAX_NR_ZONES];
846
847#ifdef CONFIG_NUMA
848 int node;
849#endif
850 struct pglist_data *zone_pgdat;
851 struct per_cpu_pages __percpu *per_cpu_pageset;
852 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
853 /*
854 * the high and batch values are copied to individual pagesets for
855 * faster access
856 */
857 int pageset_high_min;
858 int pageset_high_max;
859 int pageset_batch;
860
861#ifndef CONFIG_SPARSEMEM
862 /*
863 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
864 * In SPARSEMEM, this map is stored in struct mem_section
865 */
866 unsigned long *pageblock_flags;
867#endif /* CONFIG_SPARSEMEM */
868
869 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
870 unsigned long zone_start_pfn;
871
872 /*
873 * spanned_pages is the total pages spanned by the zone, including
874 * holes, which is calculated as:
875 * spanned_pages = zone_end_pfn - zone_start_pfn;
876 *
877 * present_pages is physical pages existing within the zone, which
878 * is calculated as:
879 * present_pages = spanned_pages - absent_pages(pages in holes);
880 *
881 * present_early_pages is present pages existing within the zone
882 * located on memory available since early boot, excluding hotplugged
883 * memory.
884 *
885 * managed_pages is present pages managed by the buddy system, which
886 * is calculated as (reserved_pages includes pages allocated by the
887 * bootmem allocator):
888 * managed_pages = present_pages - reserved_pages;
889 *
890 * cma pages is present pages that are assigned for CMA use
891 * (MIGRATE_CMA).
892 *
893 * So present_pages may be used by memory hotplug or memory power
894 * management logic to figure out unmanaged pages by checking
895 * (present_pages - managed_pages). And managed_pages should be used
896 * by page allocator and vm scanner to calculate all kinds of watermarks
897 * and thresholds.
898 *
899 * Locking rules:
900 *
901 * zone_start_pfn and spanned_pages are protected by span_seqlock.
902 * It is a seqlock because it has to be read outside of zone->lock,
903 * and it is done in the main allocator path. But, it is written
904 * quite infrequently.
905 *
906 * The span_seq lock is declared along with zone->lock because it is
907 * frequently read in proximity to zone->lock. It's good to
908 * give them a chance of being in the same cacheline.
909 *
910 * Write access to present_pages at runtime should be protected by
911 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
912 * present_pages should use get_online_mems() to get a stable value.
913 */
914 atomic_long_t managed_pages;
915 unsigned long spanned_pages;
916 unsigned long present_pages;
917#if defined(CONFIG_MEMORY_HOTPLUG)
918 unsigned long present_early_pages;
919#endif
920#ifdef CONFIG_CMA
921 unsigned long cma_pages;
922#endif
923
924 const char *name;
925
926#ifdef CONFIG_MEMORY_ISOLATION
927 /*
928 * Number of isolated pageblock. It is used to solve incorrect
929 * freepage counting problem due to racy retrieving migratetype
930 * of pageblock. Protected by zone->lock.
931 */
932 unsigned long nr_isolate_pageblock;
933#endif
934
935#ifdef CONFIG_MEMORY_HOTPLUG
936 /* see spanned/present_pages for more description */
937 seqlock_t span_seqlock;
938#endif
939
940 int initialized;
941
942 /* Write-intensive fields used from the page allocator */
943 CACHELINE_PADDING(_pad1_);
944
945 /* free areas of different sizes */
946 struct free_area free_area[MAX_ORDER + 1];
947
948#ifdef CONFIG_UNACCEPTED_MEMORY
949 /* Pages to be accepted. All pages on the list are MAX_ORDER */
950 struct list_head unaccepted_pages;
951#endif
952
953 /* zone flags, see below */
954 unsigned long flags;
955
956 /* Primarily protects free_area */
957 spinlock_t lock;
958
959 /* Write-intensive fields used by compaction and vmstats. */
960 CACHELINE_PADDING(_pad2_);
961
962 /*
963 * When free pages are below this point, additional steps are taken
964 * when reading the number of free pages to avoid per-cpu counter
965 * drift allowing watermarks to be breached
966 */
967 unsigned long percpu_drift_mark;
968
969#if defined CONFIG_COMPACTION || defined CONFIG_CMA
970 /* pfn where compaction free scanner should start */
971 unsigned long compact_cached_free_pfn;
972 /* pfn where compaction migration scanner should start */
973 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
974 unsigned long compact_init_migrate_pfn;
975 unsigned long compact_init_free_pfn;
976#endif
977
978#ifdef CONFIG_COMPACTION
979 /*
980 * On compaction failure, 1<<compact_defer_shift compactions
981 * are skipped before trying again. The number attempted since
982 * last failure is tracked with compact_considered.
983 * compact_order_failed is the minimum compaction failed order.
984 */
985 unsigned int compact_considered;
986 unsigned int compact_defer_shift;
987 int compact_order_failed;
988#endif
989
990#if defined CONFIG_COMPACTION || defined CONFIG_CMA
991 /* Set to true when the PG_migrate_skip bits should be cleared */
992 bool compact_blockskip_flush;
993#endif
994
995 bool contiguous;
996
997 CACHELINE_PADDING(_pad3_);
998 /* Zone statistics */
999 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
1000 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
1001} ____cacheline_internodealigned_in_smp;
1002
1003enum pgdat_flags {
1004 PGDAT_DIRTY, /* reclaim scanning has recently found
1005 * many dirty file pages at the tail
1006 * of the LRU.
1007 */
1008 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1009 * many pages under writeback
1010 */
1011 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1012};
1013
1014enum zone_flags {
1015 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1016 * Cleared when kswapd is woken.
1017 */
1018 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1019 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1020};
1021
1022static inline unsigned long zone_managed_pages(struct zone *zone)
1023{
1024 return (unsigned long)atomic_long_read(v: &zone->managed_pages);
1025}
1026
1027static inline unsigned long zone_cma_pages(struct zone *zone)
1028{
1029#ifdef CONFIG_CMA
1030 return zone->cma_pages;
1031#else
1032 return 0;
1033#endif
1034}
1035
1036static inline unsigned long zone_end_pfn(const struct zone *zone)
1037{
1038 return zone->zone_start_pfn + zone->spanned_pages;
1039}
1040
1041static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1042{
1043 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1044}
1045
1046static inline bool zone_is_initialized(struct zone *zone)
1047{
1048 return zone->initialized;
1049}
1050
1051static inline bool zone_is_empty(struct zone *zone)
1052{
1053 return zone->spanned_pages == 0;
1054}
1055
1056#ifndef BUILD_VDSO32_64
1057/*
1058 * The zone field is never updated after free_area_init_core()
1059 * sets it, so none of the operations on it need to be atomic.
1060 */
1061
1062/* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1063#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1064#define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1065#define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1066#define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1067#define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1068#define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1069#define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1070
1071/*
1072 * Define the bit shifts to access each section. For non-existent
1073 * sections we define the shift as 0; that plus a 0 mask ensures
1074 * the compiler will optimise away reference to them.
1075 */
1076#define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1077#define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1078#define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1079#define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1080#define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1081
1082/* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1083#ifdef NODE_NOT_IN_PAGE_FLAGS
1084#define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1085#define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1086 SECTIONS_PGOFF : ZONES_PGOFF)
1087#else
1088#define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1089#define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1090 NODES_PGOFF : ZONES_PGOFF)
1091#endif
1092
1093#define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1094
1095#define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1096#define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1097#define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1098#define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1099#define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1100#define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1101
1102static inline enum zone_type page_zonenum(const struct page *page)
1103{
1104 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1105 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1106}
1107
1108static inline enum zone_type folio_zonenum(const struct folio *folio)
1109{
1110 return page_zonenum(page: &folio->page);
1111}
1112
1113#ifdef CONFIG_ZONE_DEVICE
1114static inline bool is_zone_device_page(const struct page *page)
1115{
1116 return page_zonenum(page) == ZONE_DEVICE;
1117}
1118
1119/*
1120 * Consecutive zone device pages should not be merged into the same sgl
1121 * or bvec segment with other types of pages or if they belong to different
1122 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1123 * without scanning the entire segment. This helper returns true either if
1124 * both pages are not zone device pages or both pages are zone device pages
1125 * with the same pgmap.
1126 */
1127static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1128 const struct page *b)
1129{
1130 if (is_zone_device_page(page: a) != is_zone_device_page(page: b))
1131 return false;
1132 if (!is_zone_device_page(page: a))
1133 return true;
1134 return a->pgmap == b->pgmap;
1135}
1136
1137extern void memmap_init_zone_device(struct zone *, unsigned long,
1138 unsigned long, struct dev_pagemap *);
1139#else
1140static inline bool is_zone_device_page(const struct page *page)
1141{
1142 return false;
1143}
1144static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1145 const struct page *b)
1146{
1147 return true;
1148}
1149#endif
1150
1151static inline bool folio_is_zone_device(const struct folio *folio)
1152{
1153 return is_zone_device_page(page: &folio->page);
1154}
1155
1156static inline bool is_zone_movable_page(const struct page *page)
1157{
1158 return page_zonenum(page) == ZONE_MOVABLE;
1159}
1160
1161static inline bool folio_is_zone_movable(const struct folio *folio)
1162{
1163 return folio_zonenum(folio) == ZONE_MOVABLE;
1164}
1165#endif
1166
1167/*
1168 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1169 * intersection with the given zone
1170 */
1171static inline bool zone_intersects(struct zone *zone,
1172 unsigned long start_pfn, unsigned long nr_pages)
1173{
1174 if (zone_is_empty(zone))
1175 return false;
1176 if (start_pfn >= zone_end_pfn(zone) ||
1177 start_pfn + nr_pages <= zone->zone_start_pfn)
1178 return false;
1179
1180 return true;
1181}
1182
1183/*
1184 * The "priority" of VM scanning is how much of the queues we will scan in one
1185 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1186 * queues ("queue_length >> 12") during an aging round.
1187 */
1188#define DEF_PRIORITY 12
1189
1190/* Maximum number of zones on a zonelist */
1191#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1192
1193enum {
1194 ZONELIST_FALLBACK, /* zonelist with fallback */
1195#ifdef CONFIG_NUMA
1196 /*
1197 * The NUMA zonelists are doubled because we need zonelists that
1198 * restrict the allocations to a single node for __GFP_THISNODE.
1199 */
1200 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1201#endif
1202 MAX_ZONELISTS
1203};
1204
1205/*
1206 * This struct contains information about a zone in a zonelist. It is stored
1207 * here to avoid dereferences into large structures and lookups of tables
1208 */
1209struct zoneref {
1210 struct zone *zone; /* Pointer to actual zone */
1211 int zone_idx; /* zone_idx(zoneref->zone) */
1212};
1213
1214/*
1215 * One allocation request operates on a zonelist. A zonelist
1216 * is a list of zones, the first one is the 'goal' of the
1217 * allocation, the other zones are fallback zones, in decreasing
1218 * priority.
1219 *
1220 * To speed the reading of the zonelist, the zonerefs contain the zone index
1221 * of the entry being read. Helper functions to access information given
1222 * a struct zoneref are
1223 *
1224 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1225 * zonelist_zone_idx() - Return the index of the zone for an entry
1226 * zonelist_node_idx() - Return the index of the node for an entry
1227 */
1228struct zonelist {
1229 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1230};
1231
1232/*
1233 * The array of struct pages for flatmem.
1234 * It must be declared for SPARSEMEM as well because there are configurations
1235 * that rely on that.
1236 */
1237extern struct page *mem_map;
1238
1239#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1240struct deferred_split {
1241 spinlock_t split_queue_lock;
1242 struct list_head split_queue;
1243 unsigned long split_queue_len;
1244};
1245#endif
1246
1247#ifdef CONFIG_MEMORY_FAILURE
1248/*
1249 * Per NUMA node memory failure handling statistics.
1250 */
1251struct memory_failure_stats {
1252 /*
1253 * Number of raw pages poisoned.
1254 * Cases not accounted: memory outside kernel control, offline page,
1255 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1256 * error events, and unpoison actions from hwpoison_unpoison.
1257 */
1258 unsigned long total;
1259 /*
1260 * Recovery results of poisoned raw pages handled by memory_failure,
1261 * in sync with mf_result.
1262 * total = ignored + failed + delayed + recovered.
1263 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1264 */
1265 unsigned long ignored;
1266 unsigned long failed;
1267 unsigned long delayed;
1268 unsigned long recovered;
1269};
1270#endif
1271
1272/*
1273 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1274 * it's memory layout. On UMA machines there is a single pglist_data which
1275 * describes the whole memory.
1276 *
1277 * Memory statistics and page replacement data structures are maintained on a
1278 * per-zone basis.
1279 */
1280typedef struct pglist_data {
1281 /*
1282 * node_zones contains just the zones for THIS node. Not all of the
1283 * zones may be populated, but it is the full list. It is referenced by
1284 * this node's node_zonelists as well as other node's node_zonelists.
1285 */
1286 struct zone node_zones[MAX_NR_ZONES];
1287
1288 /*
1289 * node_zonelists contains references to all zones in all nodes.
1290 * Generally the first zones will be references to this node's
1291 * node_zones.
1292 */
1293 struct zonelist node_zonelists[MAX_ZONELISTS];
1294
1295 int nr_zones; /* number of populated zones in this node */
1296#ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1297 struct page *node_mem_map;
1298#ifdef CONFIG_PAGE_EXTENSION
1299 struct page_ext *node_page_ext;
1300#endif
1301#endif
1302#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1303 /*
1304 * Must be held any time you expect node_start_pfn,
1305 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1306 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1307 * init.
1308 *
1309 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1310 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1311 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1312 *
1313 * Nests above zone->lock and zone->span_seqlock
1314 */
1315 spinlock_t node_size_lock;
1316#endif
1317 unsigned long node_start_pfn;
1318 unsigned long node_present_pages; /* total number of physical pages */
1319 unsigned long node_spanned_pages; /* total size of physical page
1320 range, including holes */
1321 int node_id;
1322 wait_queue_head_t kswapd_wait;
1323 wait_queue_head_t pfmemalloc_wait;
1324
1325 /* workqueues for throttling reclaim for different reasons. */
1326 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1327
1328 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1329 unsigned long nr_reclaim_start; /* nr pages written while throttled
1330 * when throttling started. */
1331#ifdef CONFIG_MEMORY_HOTPLUG
1332 struct mutex kswapd_lock;
1333#endif
1334 struct task_struct *kswapd; /* Protected by kswapd_lock */
1335 int kswapd_order;
1336 enum zone_type kswapd_highest_zoneidx;
1337
1338 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1339
1340#ifdef CONFIG_COMPACTION
1341 int kcompactd_max_order;
1342 enum zone_type kcompactd_highest_zoneidx;
1343 wait_queue_head_t kcompactd_wait;
1344 struct task_struct *kcompactd;
1345 bool proactive_compact_trigger;
1346#endif
1347 /*
1348 * This is a per-node reserve of pages that are not available
1349 * to userspace allocations.
1350 */
1351 unsigned long totalreserve_pages;
1352
1353#ifdef CONFIG_NUMA
1354 /*
1355 * node reclaim becomes active if more unmapped pages exist.
1356 */
1357 unsigned long min_unmapped_pages;
1358 unsigned long min_slab_pages;
1359#endif /* CONFIG_NUMA */
1360
1361 /* Write-intensive fields used by page reclaim */
1362 CACHELINE_PADDING(_pad1_);
1363
1364#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1365 /*
1366 * If memory initialisation on large machines is deferred then this
1367 * is the first PFN that needs to be initialised.
1368 */
1369 unsigned long first_deferred_pfn;
1370#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1371
1372#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1373 struct deferred_split deferred_split_queue;
1374#endif
1375
1376#ifdef CONFIG_NUMA_BALANCING
1377 /* start time in ms of current promote rate limit period */
1378 unsigned int nbp_rl_start;
1379 /* number of promote candidate pages at start time of current rate limit period */
1380 unsigned long nbp_rl_nr_cand;
1381 /* promote threshold in ms */
1382 unsigned int nbp_threshold;
1383 /* start time in ms of current promote threshold adjustment period */
1384 unsigned int nbp_th_start;
1385 /*
1386 * number of promote candidate pages at start time of current promote
1387 * threshold adjustment period
1388 */
1389 unsigned long nbp_th_nr_cand;
1390#endif
1391 /* Fields commonly accessed by the page reclaim scanner */
1392
1393 /*
1394 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1395 *
1396 * Use mem_cgroup_lruvec() to look up lruvecs.
1397 */
1398 struct lruvec __lruvec;
1399
1400 unsigned long flags;
1401
1402#ifdef CONFIG_LRU_GEN
1403 /* kswap mm walk data */
1404 struct lru_gen_mm_walk mm_walk;
1405 /* lru_gen_folio list */
1406 struct lru_gen_memcg memcg_lru;
1407#endif
1408
1409 CACHELINE_PADDING(_pad2_);
1410
1411 /* Per-node vmstats */
1412 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1413 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1414#ifdef CONFIG_NUMA
1415 struct memory_tier __rcu *memtier;
1416#endif
1417#ifdef CONFIG_MEMORY_FAILURE
1418 struct memory_failure_stats mf_stats;
1419#endif
1420} pg_data_t;
1421
1422#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1423#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1424
1425#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1426#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1427
1428static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1429{
1430 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1431}
1432
1433#include <linux/memory_hotplug.h>
1434
1435void build_all_zonelists(pg_data_t *pgdat);
1436void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1437 enum zone_type highest_zoneidx);
1438bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1439 int highest_zoneidx, unsigned int alloc_flags,
1440 long free_pages);
1441bool zone_watermark_ok(struct zone *z, unsigned int order,
1442 unsigned long mark, int highest_zoneidx,
1443 unsigned int alloc_flags);
1444bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1445 unsigned long mark, int highest_zoneidx);
1446/*
1447 * Memory initialization context, use to differentiate memory added by
1448 * the platform statically or via memory hotplug interface.
1449 */
1450enum meminit_context {
1451 MEMINIT_EARLY,
1452 MEMINIT_HOTPLUG,
1453};
1454
1455extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1456 unsigned long size);
1457
1458extern void lruvec_init(struct lruvec *lruvec);
1459
1460static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1461{
1462#ifdef CONFIG_MEMCG
1463 return lruvec->pgdat;
1464#else
1465 return container_of(lruvec, struct pglist_data, __lruvec);
1466#endif
1467}
1468
1469#ifdef CONFIG_HAVE_MEMORYLESS_NODES
1470int local_memory_node(int node_id);
1471#else
1472static inline int local_memory_node(int node_id) { return node_id; };
1473#endif
1474
1475/*
1476 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1477 */
1478#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1479
1480#ifdef CONFIG_ZONE_DEVICE
1481static inline bool zone_is_zone_device(struct zone *zone)
1482{
1483 return zone_idx(zone) == ZONE_DEVICE;
1484}
1485#else
1486static inline bool zone_is_zone_device(struct zone *zone)
1487{
1488 return false;
1489}
1490#endif
1491
1492/*
1493 * Returns true if a zone has pages managed by the buddy allocator.
1494 * All the reclaim decisions have to use this function rather than
1495 * populated_zone(). If the whole zone is reserved then we can easily
1496 * end up with populated_zone() && !managed_zone().
1497 */
1498static inline bool managed_zone(struct zone *zone)
1499{
1500 return zone_managed_pages(zone);
1501}
1502
1503/* Returns true if a zone has memory */
1504static inline bool populated_zone(struct zone *zone)
1505{
1506 return zone->present_pages;
1507}
1508
1509#ifdef CONFIG_NUMA
1510static inline int zone_to_nid(struct zone *zone)
1511{
1512 return zone->node;
1513}
1514
1515static inline void zone_set_nid(struct zone *zone, int nid)
1516{
1517 zone->node = nid;
1518}
1519#else
1520static inline int zone_to_nid(struct zone *zone)
1521{
1522 return 0;
1523}
1524
1525static inline void zone_set_nid(struct zone *zone, int nid) {}
1526#endif
1527
1528extern int movable_zone;
1529
1530static inline int is_highmem_idx(enum zone_type idx)
1531{
1532#ifdef CONFIG_HIGHMEM
1533 return (idx == ZONE_HIGHMEM ||
1534 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1535#else
1536 return 0;
1537#endif
1538}
1539
1540/**
1541 * is_highmem - helper function to quickly check if a struct zone is a
1542 * highmem zone or not. This is an attempt to keep references
1543 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1544 * @zone: pointer to struct zone variable
1545 * Return: 1 for a highmem zone, 0 otherwise
1546 */
1547static inline int is_highmem(struct zone *zone)
1548{
1549 return is_highmem_idx(zone_idx(zone));
1550}
1551
1552#ifdef CONFIG_ZONE_DMA
1553bool has_managed_dma(void);
1554#else
1555static inline bool has_managed_dma(void)
1556{
1557 return false;
1558}
1559#endif
1560
1561
1562#ifndef CONFIG_NUMA
1563
1564extern struct pglist_data contig_page_data;
1565static inline struct pglist_data *NODE_DATA(int nid)
1566{
1567 return &contig_page_data;
1568}
1569
1570#else /* CONFIG_NUMA */
1571
1572#include <asm/mmzone.h>
1573
1574#endif /* !CONFIG_NUMA */
1575
1576extern struct pglist_data *first_online_pgdat(void);
1577extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1578extern struct zone *next_zone(struct zone *zone);
1579
1580/**
1581 * for_each_online_pgdat - helper macro to iterate over all online nodes
1582 * @pgdat: pointer to a pg_data_t variable
1583 */
1584#define for_each_online_pgdat(pgdat) \
1585 for (pgdat = first_online_pgdat(); \
1586 pgdat; \
1587 pgdat = next_online_pgdat(pgdat))
1588/**
1589 * for_each_zone - helper macro to iterate over all memory zones
1590 * @zone: pointer to struct zone variable
1591 *
1592 * The user only needs to declare the zone variable, for_each_zone
1593 * fills it in.
1594 */
1595#define for_each_zone(zone) \
1596 for (zone = (first_online_pgdat())->node_zones; \
1597 zone; \
1598 zone = next_zone(zone))
1599
1600#define for_each_populated_zone(zone) \
1601 for (zone = (first_online_pgdat())->node_zones; \
1602 zone; \
1603 zone = next_zone(zone)) \
1604 if (!populated_zone(zone)) \
1605 ; /* do nothing */ \
1606 else
1607
1608static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1609{
1610 return zoneref->zone;
1611}
1612
1613static inline int zonelist_zone_idx(struct zoneref *zoneref)
1614{
1615 return zoneref->zone_idx;
1616}
1617
1618static inline int zonelist_node_idx(struct zoneref *zoneref)
1619{
1620 return zone_to_nid(zone: zoneref->zone);
1621}
1622
1623struct zoneref *__next_zones_zonelist(struct zoneref *z,
1624 enum zone_type highest_zoneidx,
1625 nodemask_t *nodes);
1626
1627/**
1628 * 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
1629 * @z: The cursor used as a starting point for the search
1630 * @highest_zoneidx: The zone index of the highest zone to return
1631 * @nodes: An optional nodemask to filter the zonelist with
1632 *
1633 * This function returns the next zone at or below a given zone index that is
1634 * within the allowed nodemask using a cursor as the starting point for the
1635 * search. The zoneref returned is a cursor that represents the current zone
1636 * being examined. It should be advanced by one before calling
1637 * next_zones_zonelist again.
1638 *
1639 * Return: the next zone at or below highest_zoneidx within the allowed
1640 * nodemask using a cursor within a zonelist as a starting point
1641 */
1642static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1643 enum zone_type highest_zoneidx,
1644 nodemask_t *nodes)
1645{
1646 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1647 return z;
1648 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1649}
1650
1651/**
1652 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1653 * @zonelist: The zonelist to search for a suitable zone
1654 * @highest_zoneidx: The zone index of the highest zone to return
1655 * @nodes: An optional nodemask to filter the zonelist with
1656 *
1657 * This function returns the first zone at or below a given zone index that is
1658 * within the allowed nodemask. The zoneref returned is a cursor that can be
1659 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1660 * one before calling.
1661 *
1662 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1663 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1664 * update due to cpuset modification.
1665 *
1666 * Return: Zoneref pointer for the first suitable zone found
1667 */
1668static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1669 enum zone_type highest_zoneidx,
1670 nodemask_t *nodes)
1671{
1672 return next_zones_zonelist(z: zonelist->_zonerefs,
1673 highest_zoneidx, nodes);
1674}
1675
1676/**
1677 * 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
1678 * @zone: The current zone in the iterator
1679 * @z: The current pointer within zonelist->_zonerefs being iterated
1680 * @zlist: The zonelist being iterated
1681 * @highidx: The zone index of the highest zone to return
1682 * @nodemask: Nodemask allowed by the allocator
1683 *
1684 * This iterator iterates though all zones at or below a given zone index and
1685 * within a given nodemask
1686 */
1687#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1688 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1689 zone; \
1690 z = next_zones_zonelist(++z, highidx, nodemask), \
1691 zone = zonelist_zone(z))
1692
1693#define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1694 for (zone = z->zone; \
1695 zone; \
1696 z = next_zones_zonelist(++z, highidx, nodemask), \
1697 zone = zonelist_zone(z))
1698
1699
1700/**
1701 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1702 * @zone: The current zone in the iterator
1703 * @z: The current pointer within zonelist->zones being iterated
1704 * @zlist: The zonelist being iterated
1705 * @highidx: The zone index of the highest zone to return
1706 *
1707 * This iterator iterates though all zones at or below a given zone index.
1708 */
1709#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1710 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1711
1712/* Whether the 'nodes' are all movable nodes */
1713static inline bool movable_only_nodes(nodemask_t *nodes)
1714{
1715 struct zonelist *zonelist;
1716 struct zoneref *z;
1717 int nid;
1718
1719 if (nodes_empty(*nodes))
1720 return false;
1721
1722 /*
1723 * We can chose arbitrary node from the nodemask to get a
1724 * zonelist as they are interlinked. We just need to find
1725 * at least one zone that can satisfy kernel allocations.
1726 */
1727 nid = first_node(*nodes);
1728 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1729 z = first_zones_zonelist(zonelist, highest_zoneidx: ZONE_NORMAL, nodes);
1730 return (!z->zone) ? true : false;
1731}
1732
1733
1734#ifdef CONFIG_SPARSEMEM
1735#include <asm/sparsemem.h>
1736#endif
1737
1738#ifdef CONFIG_FLATMEM
1739#define pfn_to_nid(pfn) (0)
1740#endif
1741
1742#ifdef CONFIG_SPARSEMEM
1743
1744/*
1745 * PA_SECTION_SHIFT physical address to/from section number
1746 * PFN_SECTION_SHIFT pfn to/from section number
1747 */
1748#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1749#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1750
1751#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1752
1753#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1754#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1755
1756#define SECTION_BLOCKFLAGS_BITS \
1757 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1758
1759#if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1760#error Allocator MAX_ORDER exceeds SECTION_SIZE
1761#endif
1762
1763static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1764{
1765 return pfn >> PFN_SECTION_SHIFT;
1766}
1767static inline unsigned long section_nr_to_pfn(unsigned long sec)
1768{
1769 return sec << PFN_SECTION_SHIFT;
1770}
1771
1772#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1773#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1774
1775#define SUBSECTION_SHIFT 21
1776#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1777
1778#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1779#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1780#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1781
1782#if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1783#error Subsection size exceeds section size
1784#else
1785#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1786#endif
1787
1788#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1789#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1790
1791struct mem_section_usage {
1792#ifdef CONFIG_SPARSEMEM_VMEMMAP
1793 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1794#endif
1795 /* See declaration of similar field in struct zone */
1796 unsigned long pageblock_flags[0];
1797};
1798
1799void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1800
1801struct page;
1802struct page_ext;
1803struct mem_section {
1804 /*
1805 * This is, logically, a pointer to an array of struct
1806 * pages. However, it is stored with some other magic.
1807 * (see sparse.c::sparse_init_one_section())
1808 *
1809 * Additionally during early boot we encode node id of
1810 * the location of the section here to guide allocation.
1811 * (see sparse.c::memory_present())
1812 *
1813 * Making it a UL at least makes someone do a cast
1814 * before using it wrong.
1815 */
1816 unsigned long section_mem_map;
1817
1818 struct mem_section_usage *usage;
1819#ifdef CONFIG_PAGE_EXTENSION
1820 /*
1821 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1822 * section. (see page_ext.h about this.)
1823 */
1824 struct page_ext *page_ext;
1825 unsigned long pad;
1826#endif
1827 /*
1828 * WARNING: mem_section must be a power-of-2 in size for the
1829 * calculation and use of SECTION_ROOT_MASK to make sense.
1830 */
1831};
1832
1833#ifdef CONFIG_SPARSEMEM_EXTREME
1834#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1835#else
1836#define SECTIONS_PER_ROOT 1
1837#endif
1838
1839#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1840#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1841#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1842
1843#ifdef CONFIG_SPARSEMEM_EXTREME
1844extern struct mem_section **mem_section;
1845#else
1846extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1847#endif
1848
1849static inline unsigned long *section_to_usemap(struct mem_section *ms)
1850{
1851 return ms->usage->pageblock_flags;
1852}
1853
1854static inline struct mem_section *__nr_to_section(unsigned long nr)
1855{
1856 unsigned long root = SECTION_NR_TO_ROOT(nr);
1857
1858 if (unlikely(root >= NR_SECTION_ROOTS))
1859 return NULL;
1860
1861#ifdef CONFIG_SPARSEMEM_EXTREME
1862 if (!mem_section || !mem_section[root])
1863 return NULL;
1864#endif
1865 return &mem_section[root][nr & SECTION_ROOT_MASK];
1866}
1867extern size_t mem_section_usage_size(void);
1868
1869/*
1870 * We use the lower bits of the mem_map pointer to store
1871 * a little bit of information. The pointer is calculated
1872 * as mem_map - section_nr_to_pfn(pnum). The result is
1873 * aligned to the minimum alignment of the two values:
1874 * 1. All mem_map arrays are page-aligned.
1875 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1876 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1877 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1878 * worst combination is powerpc with 256k pages,
1879 * which results in PFN_SECTION_SHIFT equal 6.
1880 * To sum it up, at least 6 bits are available on all architectures.
1881 * However, we can exceed 6 bits on some other architectures except
1882 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1883 * with the worst case of 64K pages on arm64) if we make sure the
1884 * exceeded bit is not applicable to powerpc.
1885 */
1886enum {
1887 SECTION_MARKED_PRESENT_BIT,
1888 SECTION_HAS_MEM_MAP_BIT,
1889 SECTION_IS_ONLINE_BIT,
1890 SECTION_IS_EARLY_BIT,
1891#ifdef CONFIG_ZONE_DEVICE
1892 SECTION_TAINT_ZONE_DEVICE_BIT,
1893#endif
1894 SECTION_MAP_LAST_BIT,
1895};
1896
1897#define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1898#define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1899#define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1900#define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1901#ifdef CONFIG_ZONE_DEVICE
1902#define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1903#endif
1904#define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1905#define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1906
1907static inline struct page *__section_mem_map_addr(struct mem_section *section)
1908{
1909 unsigned long map = section->section_mem_map;
1910 map &= SECTION_MAP_MASK;
1911 return (struct page *)map;
1912}
1913
1914static inline int present_section(struct mem_section *section)
1915{
1916 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1917}
1918
1919static inline int present_section_nr(unsigned long nr)
1920{
1921 return present_section(section: __nr_to_section(nr));
1922}
1923
1924static inline int valid_section(struct mem_section *section)
1925{
1926 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1927}
1928
1929static inline int early_section(struct mem_section *section)
1930{
1931 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1932}
1933
1934static inline int valid_section_nr(unsigned long nr)
1935{
1936 return valid_section(section: __nr_to_section(nr));
1937}
1938
1939static inline int online_section(struct mem_section *section)
1940{
1941 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1942}
1943
1944#ifdef CONFIG_ZONE_DEVICE
1945static inline int online_device_section(struct mem_section *section)
1946{
1947 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1948
1949 return section && ((section->section_mem_map & flags) == flags);
1950}
1951#else
1952static inline int online_device_section(struct mem_section *section)
1953{
1954 return 0;
1955}
1956#endif
1957
1958static inline int online_section_nr(unsigned long nr)
1959{
1960 return online_section(section: __nr_to_section(nr));
1961}
1962
1963#ifdef CONFIG_MEMORY_HOTPLUG
1964void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1965void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1966#endif
1967
1968static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1969{
1970 return __nr_to_section(nr: pfn_to_section_nr(pfn));
1971}
1972
1973extern unsigned long __highest_present_section_nr;
1974
1975static inline int subsection_map_index(unsigned long pfn)
1976{
1977 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1978}
1979
1980#ifdef CONFIG_SPARSEMEM_VMEMMAP
1981static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1982{
1983 int idx = subsection_map_index(pfn);
1984
1985 return test_bit(idx, ms->usage->subsection_map);
1986}
1987#else
1988static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1989{
1990 return 1;
1991}
1992#endif
1993
1994#ifndef CONFIG_HAVE_ARCH_PFN_VALID
1995/**
1996 * pfn_valid - check if there is a valid memory map entry for a PFN
1997 * @pfn: the page frame number to check
1998 *
1999 * Check if there is a valid memory map entry aka struct page for the @pfn.
2000 * Note, that availability of the memory map entry does not imply that
2001 * there is actual usable memory at that @pfn. The struct page may
2002 * represent a hole or an unusable page frame.
2003 *
2004 * Return: 1 for PFNs that have memory map entries and 0 otherwise
2005 */
2006static inline int pfn_valid(unsigned long pfn)
2007{
2008 struct mem_section *ms;
2009
2010 /*
2011 * Ensure the upper PAGE_SHIFT bits are clear in the
2012 * pfn. Else it might lead to false positives when
2013 * some of the upper bits are set, but the lower bits
2014 * match a valid pfn.
2015 */
2016 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2017 return 0;
2018
2019 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2020 return 0;
2021 ms = __pfn_to_section(pfn);
2022 if (!valid_section(section: ms))
2023 return 0;
2024 /*
2025 * Traditionally early sections always returned pfn_valid() for
2026 * the entire section-sized span.
2027 */
2028 return early_section(section: ms) || pfn_section_valid(ms, pfn);
2029}
2030#endif
2031
2032static inline int pfn_in_present_section(unsigned long pfn)
2033{
2034 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2035 return 0;
2036 return present_section(section: __pfn_to_section(pfn));
2037}
2038
2039static inline unsigned long next_present_section_nr(unsigned long section_nr)
2040{
2041 while (++section_nr <= __highest_present_section_nr) {
2042 if (present_section_nr(nr: section_nr))
2043 return section_nr;
2044 }
2045
2046 return -1;
2047}
2048
2049/*
2050 * These are _only_ used during initialisation, therefore they
2051 * can use __initdata ... They could have names to indicate
2052 * this restriction.
2053 */
2054#ifdef CONFIG_NUMA
2055#define pfn_to_nid(pfn) \
2056({ \
2057 unsigned long __pfn_to_nid_pfn = (pfn); \
2058 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2059})
2060#else
2061#define pfn_to_nid(pfn) (0)
2062#endif
2063
2064void sparse_init(void);
2065#else
2066#define sparse_init() do {} while (0)
2067#define sparse_index_init(_sec, _nid) do {} while (0)
2068#define pfn_in_present_section pfn_valid
2069#define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2070#endif /* CONFIG_SPARSEMEM */
2071
2072#endif /* !__GENERATING_BOUNDS.H */
2073#endif /* !__ASSEMBLY__ */
2074#endif /* _LINUX_MMZONE_H */
2075

source code of linux/include/linux/mmzone.h