xref: /linux-6.15/include/linux/mmzone.h (revision 27d9a0fd)
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 
45 enum 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 */
71 extern 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 
81 static 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  */
92 static 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 
101 extern 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 struct free_area {
109 	struct list_head	free_list[MIGRATE_TYPES];
110 	unsigned long		nr_free;
111 };
112 
113 struct pglist_data;
114 
115 #ifdef CONFIG_NUMA
116 enum numa_stat_item {
117 	NUMA_HIT,		/* allocated in intended node */
118 	NUMA_MISS,		/* allocated in non intended node */
119 	NUMA_FOREIGN,		/* was intended here, hit elsewhere */
120 	NUMA_INTERLEAVE_HIT,	/* interleaver preferred this zone */
121 	NUMA_LOCAL,		/* allocation from local node */
122 	NUMA_OTHER,		/* allocation from other node */
123 	NR_VM_NUMA_EVENT_ITEMS
124 };
125 #else
126 #define NR_VM_NUMA_EVENT_ITEMS 0
127 #endif
128 
129 enum zone_stat_item {
130 	/* First 128 byte cacheline (assuming 64 bit words) */
131 	NR_FREE_PAGES,
132 	NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
133 	NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
134 	NR_ZONE_ACTIVE_ANON,
135 	NR_ZONE_INACTIVE_FILE,
136 	NR_ZONE_ACTIVE_FILE,
137 	NR_ZONE_UNEVICTABLE,
138 	NR_ZONE_WRITE_PENDING,	/* Count of dirty, writeback and unstable pages */
139 	NR_MLOCK,		/* mlock()ed pages found and moved off LRU */
140 	/* Second 128 byte cacheline */
141 	NR_BOUNCE,
142 #if IS_ENABLED(CONFIG_ZSMALLOC)
143 	NR_ZSPAGES,		/* allocated in zsmalloc */
144 #endif
145 	NR_FREE_CMA_PAGES,
146 	NR_VM_ZONE_STAT_ITEMS };
147 
148 enum node_stat_item {
149 	NR_LRU_BASE,
150 	NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
151 	NR_ACTIVE_ANON,		/*  "     "     "   "       "         */
152 	NR_INACTIVE_FILE,	/*  "     "     "   "       "         */
153 	NR_ACTIVE_FILE,		/*  "     "     "   "       "         */
154 	NR_UNEVICTABLE,		/*  "     "     "   "       "         */
155 	NR_SLAB_RECLAIMABLE_B,
156 	NR_SLAB_UNRECLAIMABLE_B,
157 	NR_ISOLATED_ANON,	/* Temporary isolated pages from anon lru */
158 	NR_ISOLATED_FILE,	/* Temporary isolated pages from file lru */
159 	WORKINGSET_NODES,
160 	WORKINGSET_REFAULT_BASE,
161 	WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
162 	WORKINGSET_REFAULT_FILE,
163 	WORKINGSET_ACTIVATE_BASE,
164 	WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
165 	WORKINGSET_ACTIVATE_FILE,
166 	WORKINGSET_RESTORE_BASE,
167 	WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
168 	WORKINGSET_RESTORE_FILE,
169 	WORKINGSET_NODERECLAIM,
170 	NR_ANON_MAPPED,	/* Mapped anonymous pages */
171 	NR_FILE_MAPPED,	/* pagecache pages mapped into pagetables.
172 			   only modified from process context */
173 	NR_FILE_PAGES,
174 	NR_FILE_DIRTY,
175 	NR_WRITEBACK,
176 	NR_WRITEBACK_TEMP,	/* Writeback using temporary buffers */
177 	NR_SHMEM,		/* shmem pages (included tmpfs/GEM pages) */
178 	NR_SHMEM_THPS,
179 	NR_SHMEM_PMDMAPPED,
180 	NR_FILE_THPS,
181 	NR_FILE_PMDMAPPED,
182 	NR_ANON_THPS,
183 	NR_VMSCAN_WRITE,
184 	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
185 	NR_DIRTIED,		/* page dirtyings since bootup */
186 	NR_WRITTEN,		/* page writings since bootup */
187 	NR_THROTTLED_WRITTEN,	/* NR_WRITTEN while reclaim throttled */
188 	NR_KERNEL_MISC_RECLAIMABLE,	/* reclaimable non-slab kernel pages */
189 	NR_FOLL_PIN_ACQUIRED,	/* via: pin_user_page(), gup flag: FOLL_PIN */
190 	NR_FOLL_PIN_RELEASED,	/* pages returned via unpin_user_page() */
191 	NR_KERNEL_STACK_KB,	/* measured in KiB */
192 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
193 	NR_KERNEL_SCS_KB,	/* measured in KiB */
194 #endif
195 	NR_PAGETABLE,		/* used for pagetables */
196 	NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
197 #ifdef CONFIG_SWAP
198 	NR_SWAPCACHE,
199 #endif
200 #ifdef CONFIG_NUMA_BALANCING
201 	PGPROMOTE_SUCCESS,	/* promote successfully */
202 	PGPROMOTE_CANDIDATE,	/* candidate pages to promote */
203 #endif
204 	NR_VM_NODE_STAT_ITEMS
205 };
206 
207 /*
208  * Returns true if the item should be printed in THPs (/proc/vmstat
209  * currently prints number of anon, file and shmem THPs. But the item
210  * is charged in pages).
211  */
212 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
213 {
214 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
215 		return false;
216 
217 	return item == NR_ANON_THPS ||
218 	       item == NR_FILE_THPS ||
219 	       item == NR_SHMEM_THPS ||
220 	       item == NR_SHMEM_PMDMAPPED ||
221 	       item == NR_FILE_PMDMAPPED;
222 }
223 
224 /*
225  * Returns true if the value is measured in bytes (most vmstat values are
226  * measured in pages). This defines the API part, the internal representation
227  * might be different.
228  */
229 static __always_inline bool vmstat_item_in_bytes(int idx)
230 {
231 	/*
232 	 * Global and per-node slab counters track slab pages.
233 	 * It's expected that changes are multiples of PAGE_SIZE.
234 	 * Internally values are stored in pages.
235 	 *
236 	 * Per-memcg and per-lruvec counters track memory, consumed
237 	 * by individual slab objects. These counters are actually
238 	 * byte-precise.
239 	 */
240 	return (idx == NR_SLAB_RECLAIMABLE_B ||
241 		idx == NR_SLAB_UNRECLAIMABLE_B);
242 }
243 
244 /*
245  * We do arithmetic on the LRU lists in various places in the code,
246  * so it is important to keep the active lists LRU_ACTIVE higher in
247  * the array than the corresponding inactive lists, and to keep
248  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
249  *
250  * This has to be kept in sync with the statistics in zone_stat_item
251  * above and the descriptions in vmstat_text in mm/vmstat.c
252  */
253 #define LRU_BASE 0
254 #define LRU_ACTIVE 1
255 #define LRU_FILE 2
256 
257 enum lru_list {
258 	LRU_INACTIVE_ANON = LRU_BASE,
259 	LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
260 	LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
261 	LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
262 	LRU_UNEVICTABLE,
263 	NR_LRU_LISTS
264 };
265 
266 enum vmscan_throttle_state {
267 	VMSCAN_THROTTLE_WRITEBACK,
268 	VMSCAN_THROTTLE_ISOLATED,
269 	VMSCAN_THROTTLE_NOPROGRESS,
270 	VMSCAN_THROTTLE_CONGESTED,
271 	NR_VMSCAN_THROTTLE,
272 };
273 
274 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
275 
276 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
277 
278 static inline bool is_file_lru(enum lru_list lru)
279 {
280 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
281 }
282 
283 static inline bool is_active_lru(enum lru_list lru)
284 {
285 	return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
286 }
287 
288 #define WORKINGSET_ANON 0
289 #define WORKINGSET_FILE 1
290 #define ANON_AND_FILE 2
291 
292 enum lruvec_flags {
293 	LRUVEC_CONGESTED,		/* lruvec has many dirty pages
294 					 * backed by a congested BDI
295 					 */
296 };
297 
298 #endif /* !__GENERATING_BOUNDS_H */
299 
300 /*
301  * Evictable pages are divided into multiple generations. The youngest and the
302  * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
303  * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
304  * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
305  * corresponding generation. The gen counter in folio->flags stores gen+1 while
306  * a page is on one of lrugen->folios[]. Otherwise it stores 0.
307  *
308  * A page is added to the youngest generation on faulting. The aging needs to
309  * check the accessed bit at least twice before handing this page over to the
310  * eviction. The first check takes care of the accessed bit set on the initial
311  * fault; the second check makes sure this page hasn't been used since then.
312  * This process, AKA second chance, requires a minimum of two generations,
313  * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
314  * LRU, e.g., /proc/vmstat, these two generations are considered active; the
315  * rest of generations, if they exist, are considered inactive. See
316  * lru_gen_is_active().
317  *
318  * PG_active is always cleared while a page is on one of lrugen->folios[] so
319  * that the aging needs not to worry about it. And it's set again when a page
320  * considered active is isolated for non-reclaiming purposes, e.g., migration.
321  * See lru_gen_add_folio() and lru_gen_del_folio().
322  *
323  * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
324  * number of categories of the active/inactive LRU when keeping track of
325  * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
326  * in folio->flags.
327  */
328 #define MIN_NR_GENS		2U
329 #define MAX_NR_GENS		4U
330 
331 /*
332  * Each generation is divided into multiple tiers. A page accessed N times
333  * through file descriptors is in tier order_base_2(N). A page in the first tier
334  * (N=0,1) is marked by PG_referenced unless it was faulted in through page
335  * tables or read ahead. A page in any other tier (N>1) is marked by
336  * PG_referenced and PG_workingset. This implies a minimum of two tiers is
337  * supported without using additional bits in folio->flags.
338  *
339  * In contrast to moving across generations which requires the LRU lock, moving
340  * across tiers only involves atomic operations on folio->flags and therefore
341  * has a negligible cost in the buffered access path. In the eviction path,
342  * comparisons of refaulted/(evicted+protected) from the first tier and the
343  * rest infer whether pages accessed multiple times through file descriptors
344  * are statistically hot and thus worth protecting.
345  *
346  * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
347  * number of categories of the active/inactive LRU when keeping track of
348  * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
349  * folio->flags.
350  */
351 #define MAX_NR_TIERS		4U
352 
353 #ifndef __GENERATING_BOUNDS_H
354 
355 struct lruvec;
356 struct page_vma_mapped_walk;
357 
358 #define LRU_GEN_MASK		((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
359 #define LRU_REFS_MASK		((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
360 
361 #ifdef CONFIG_LRU_GEN
362 
363 enum {
364 	LRU_GEN_ANON,
365 	LRU_GEN_FILE,
366 };
367 
368 enum {
369 	LRU_GEN_CORE,
370 	LRU_GEN_MM_WALK,
371 	LRU_GEN_NONLEAF_YOUNG,
372 	NR_LRU_GEN_CAPS
373 };
374 
375 #define MIN_LRU_BATCH		BITS_PER_LONG
376 #define MAX_LRU_BATCH		(MIN_LRU_BATCH * 64)
377 
378 /* whether to keep historical stats from evicted generations */
379 #ifdef CONFIG_LRU_GEN_STATS
380 #define NR_HIST_GENS		MAX_NR_GENS
381 #else
382 #define NR_HIST_GENS		1U
383 #endif
384 
385 /*
386  * The youngest generation number is stored in max_seq for both anon and file
387  * types as they are aged on an equal footing. The oldest generation numbers are
388  * stored in min_seq[] separately for anon and file types as clean file pages
389  * can be evicted regardless of swap constraints.
390  *
391  * Normally anon and file min_seq are in sync. But if swapping is constrained,
392  * e.g., out of swap space, file min_seq is allowed to advance and leave anon
393  * min_seq behind.
394  *
395  * The number of pages in each generation is eventually consistent and therefore
396  * can be transiently negative when reset_batch_size() is pending.
397  */
398 struct lru_gen_folio {
399 	/* the aging increments the youngest generation number */
400 	unsigned long max_seq;
401 	/* the eviction increments the oldest generation numbers */
402 	unsigned long min_seq[ANON_AND_FILE];
403 	/* the birth time of each generation in jiffies */
404 	unsigned long timestamps[MAX_NR_GENS];
405 	/* the multi-gen LRU lists, lazily sorted on eviction */
406 	struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
407 	/* the multi-gen LRU sizes, eventually consistent */
408 	long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
409 	/* the exponential moving average of refaulted */
410 	unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
411 	/* the exponential moving average of evicted+protected */
412 	unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
413 	/* the first tier doesn't need protection, hence the minus one */
414 	unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
415 	/* can be modified without holding the LRU lock */
416 	atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
417 	atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
418 	/* whether the multi-gen LRU is enabled */
419 	bool enabled;
420 #ifdef CONFIG_MEMCG
421 	/* the memcg generation this lru_gen_folio belongs to */
422 	u8 gen;
423 	/* the list segment this lru_gen_folio belongs to */
424 	u8 seg;
425 	/* per-node lru_gen_folio list for global reclaim */
426 	struct hlist_nulls_node list;
427 #endif
428 };
429 
430 enum {
431 	MM_LEAF_TOTAL,		/* total leaf entries */
432 	MM_LEAF_OLD,		/* old leaf entries */
433 	MM_LEAF_YOUNG,		/* young leaf entries */
434 	MM_NONLEAF_TOTAL,	/* total non-leaf entries */
435 	MM_NONLEAF_FOUND,	/* non-leaf entries found in Bloom filters */
436 	MM_NONLEAF_ADDED,	/* non-leaf entries added to Bloom filters */
437 	NR_MM_STATS
438 };
439 
440 /* double-buffering Bloom filters */
441 #define NR_BLOOM_FILTERS	2
442 
443 struct lru_gen_mm_state {
444 	/* set to max_seq after each iteration */
445 	unsigned long seq;
446 	/* where the current iteration continues (inclusive) */
447 	struct list_head *head;
448 	/* where the last iteration ended (exclusive) */
449 	struct list_head *tail;
450 	/* to wait for the last page table walker to finish */
451 	struct wait_queue_head wait;
452 	/* Bloom filters flip after each iteration */
453 	unsigned long *filters[NR_BLOOM_FILTERS];
454 	/* the mm stats for debugging */
455 	unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
456 	/* the number of concurrent page table walkers */
457 	int nr_walkers;
458 };
459 
460 struct lru_gen_mm_walk {
461 	/* the lruvec under reclaim */
462 	struct lruvec *lruvec;
463 	/* unstable max_seq from lru_gen_folio */
464 	unsigned long max_seq;
465 	/* the next address within an mm to scan */
466 	unsigned long next_addr;
467 	/* to batch promoted pages */
468 	int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
469 	/* to batch the mm stats */
470 	int mm_stats[NR_MM_STATS];
471 	/* total batched items */
472 	int batched;
473 	bool can_swap;
474 	bool force_scan;
475 };
476 
477 void lru_gen_init_lruvec(struct lruvec *lruvec);
478 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
479 
480 #ifdef CONFIG_MEMCG
481 
482 /*
483  * For each node, memcgs are divided into two generations: the old and the
484  * young. For each generation, memcgs are randomly sharded into multiple bins
485  * to improve scalability. For each bin, the hlist_nulls is virtually divided
486  * into three segments: the head, the tail and the default.
487  *
488  * An onlining memcg is added to the tail of a random bin in the old generation.
489  * The eviction starts at the head of a random bin in the old generation. The
490  * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
491  * the old generation, is incremented when all its bins become empty.
492  *
493  * There are four operations:
494  * 1. MEMCG_LRU_HEAD, which moves an memcg to the head of a random bin in its
495  *    current generation (old or young) and updates its "seg" to "head";
496  * 2. MEMCG_LRU_TAIL, which moves an memcg to the tail of a random bin in its
497  *    current generation (old or young) and updates its "seg" to "tail";
498  * 3. MEMCG_LRU_OLD, which moves an memcg to the head of a random bin in the old
499  *    generation, updates its "gen" to "old" and resets its "seg" to "default";
500  * 4. MEMCG_LRU_YOUNG, which moves an memcg to the tail of a random bin in the
501  *    young generation, updates its "gen" to "young" and resets its "seg" to
502  *    "default".
503  *
504  * The events that trigger the above operations are:
505  * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
506  * 2. The first attempt to reclaim an memcg below low, which triggers
507  *    MEMCG_LRU_TAIL;
508  * 3. The first attempt to reclaim an memcg below reclaimable size threshold,
509  *    which triggers MEMCG_LRU_TAIL;
510  * 4. The second attempt to reclaim an memcg below reclaimable size threshold,
511  *    which triggers MEMCG_LRU_YOUNG;
512  * 5. Attempting to reclaim an memcg below min, which triggers MEMCG_LRU_YOUNG;
513  * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
514  * 7. Offlining an memcg, which triggers MEMCG_LRU_OLD.
515  *
516  * Note that memcg LRU only applies to global reclaim, and the round-robin
517  * incrementing of their max_seq counters ensures the eventual fairness to all
518  * eligible memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
519  */
520 #define MEMCG_NR_GENS	2
521 #define MEMCG_NR_BINS	8
522 
523 struct lru_gen_memcg {
524 	/* the per-node memcg generation counter */
525 	unsigned long seq;
526 	/* each memcg has one lru_gen_folio per node */
527 	unsigned long nr_memcgs[MEMCG_NR_GENS];
528 	/* per-node lru_gen_folio list for global reclaim */
529 	struct hlist_nulls_head	fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
530 	/* protects the above */
531 	spinlock_t lock;
532 };
533 
534 void lru_gen_init_pgdat(struct pglist_data *pgdat);
535 
536 void lru_gen_init_memcg(struct mem_cgroup *memcg);
537 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
538 void lru_gen_online_memcg(struct mem_cgroup *memcg);
539 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
540 void lru_gen_release_memcg(struct mem_cgroup *memcg);
541 void lru_gen_soft_reclaim(struct lruvec *lruvec);
542 
543 #else /* !CONFIG_MEMCG */
544 
545 #define MEMCG_NR_GENS	1
546 
547 struct lru_gen_memcg {
548 };
549 
550 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
551 {
552 }
553 
554 #endif /* CONFIG_MEMCG */
555 
556 #else /* !CONFIG_LRU_GEN */
557 
558 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
559 {
560 }
561 
562 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
563 {
564 }
565 
566 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
567 {
568 }
569 
570 #ifdef CONFIG_MEMCG
571 
572 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
573 {
574 }
575 
576 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
577 {
578 }
579 
580 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
581 {
582 }
583 
584 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
585 {
586 }
587 
588 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
589 {
590 }
591 
592 static inline void lru_gen_soft_reclaim(struct lruvec *lruvec)
593 {
594 }
595 
596 #endif /* CONFIG_MEMCG */
597 
598 #endif /* CONFIG_LRU_GEN */
599 
600 struct lruvec {
601 	struct list_head		lists[NR_LRU_LISTS];
602 	/* per lruvec lru_lock for memcg */
603 	spinlock_t			lru_lock;
604 	/*
605 	 * These track the cost of reclaiming one LRU - file or anon -
606 	 * over the other. As the observed cost of reclaiming one LRU
607 	 * increases, the reclaim scan balance tips toward the other.
608 	 */
609 	unsigned long			anon_cost;
610 	unsigned long			file_cost;
611 	/* Non-resident age, driven by LRU movement */
612 	atomic_long_t			nonresident_age;
613 	/* Refaults at the time of last reclaim cycle */
614 	unsigned long			refaults[ANON_AND_FILE];
615 	/* Various lruvec state flags (enum lruvec_flags) */
616 	unsigned long			flags;
617 #ifdef CONFIG_LRU_GEN
618 	/* evictable pages divided into generations */
619 	struct lru_gen_folio		lrugen;
620 	/* to concurrently iterate lru_gen_mm_list */
621 	struct lru_gen_mm_state		mm_state;
622 #endif
623 #ifdef CONFIG_MEMCG
624 	struct pglist_data *pgdat;
625 #endif
626 };
627 
628 /* Isolate unmapped pages */
629 #define ISOLATE_UNMAPPED	((__force isolate_mode_t)0x2)
630 /* Isolate for asynchronous migration */
631 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
632 /* Isolate unevictable pages */
633 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
634 
635 /* LRU Isolation modes. */
636 typedef unsigned __bitwise isolate_mode_t;
637 
638 enum zone_watermarks {
639 	WMARK_MIN,
640 	WMARK_LOW,
641 	WMARK_HIGH,
642 	WMARK_PROMO,
643 	NR_WMARK
644 };
645 
646 /*
647  * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
648  * for THP which will usually be GFP_MOVABLE. Even if it is another type,
649  * it should not contribute to serious fragmentation causing THP allocation
650  * failures.
651  */
652 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
653 #define NR_PCP_THP 1
654 #else
655 #define NR_PCP_THP 0
656 #endif
657 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
658 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
659 
660 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
661 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
662 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
663 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
664 
665 /* Fields and list protected by pagesets local_lock in page_alloc.c */
666 struct per_cpu_pages {
667 	spinlock_t lock;	/* Protects lists field */
668 	int count;		/* number of pages in the list */
669 	int high;		/* high watermark, emptying needed */
670 	int batch;		/* chunk size for buddy add/remove */
671 	short free_factor;	/* batch scaling factor during free */
672 #ifdef CONFIG_NUMA
673 	short expire;		/* When 0, remote pagesets are drained */
674 #endif
675 
676 	/* Lists of pages, one per migrate type stored on the pcp-lists */
677 	struct list_head lists[NR_PCP_LISTS];
678 } ____cacheline_aligned_in_smp;
679 
680 struct per_cpu_zonestat {
681 #ifdef CONFIG_SMP
682 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
683 	s8 stat_threshold;
684 #endif
685 #ifdef CONFIG_NUMA
686 	/*
687 	 * Low priority inaccurate counters that are only folded
688 	 * on demand. Use a large type to avoid the overhead of
689 	 * folding during refresh_cpu_vm_stats.
690 	 */
691 	unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
692 #endif
693 };
694 
695 struct per_cpu_nodestat {
696 	s8 stat_threshold;
697 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
698 };
699 
700 #endif /* !__GENERATING_BOUNDS.H */
701 
702 enum zone_type {
703 	/*
704 	 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
705 	 * to DMA to all of the addressable memory (ZONE_NORMAL).
706 	 * On architectures where this area covers the whole 32 bit address
707 	 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
708 	 * DMA addressing constraints. This distinction is important as a 32bit
709 	 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
710 	 * platforms may need both zones as they support peripherals with
711 	 * different DMA addressing limitations.
712 	 */
713 #ifdef CONFIG_ZONE_DMA
714 	ZONE_DMA,
715 #endif
716 #ifdef CONFIG_ZONE_DMA32
717 	ZONE_DMA32,
718 #endif
719 	/*
720 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
721 	 * performed on pages in ZONE_NORMAL if the DMA devices support
722 	 * transfers to all addressable memory.
723 	 */
724 	ZONE_NORMAL,
725 #ifdef CONFIG_HIGHMEM
726 	/*
727 	 * A memory area that is only addressable by the kernel through
728 	 * mapping portions into its own address space. This is for example
729 	 * used by i386 to allow the kernel to address the memory beyond
730 	 * 900MB. The kernel will set up special mappings (page
731 	 * table entries on i386) for each page that the kernel needs to
732 	 * access.
733 	 */
734 	ZONE_HIGHMEM,
735 #endif
736 	/*
737 	 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
738 	 * movable pages with few exceptional cases described below. Main use
739 	 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
740 	 * likely to succeed, and to locally limit unmovable allocations - e.g.,
741 	 * to increase the number of THP/huge pages. Notable special cases are:
742 	 *
743 	 * 1. Pinned pages: (long-term) pinning of movable pages might
744 	 *    essentially turn such pages unmovable. Therefore, we do not allow
745 	 *    pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
746 	 *    faulted, they come from the right zone right away. However, it is
747 	 *    still possible that address space already has pages in
748 	 *    ZONE_MOVABLE at the time when pages are pinned (i.e. user has
749 	 *    touches that memory before pinning). In such case we migrate them
750 	 *    to a different zone. When migration fails - pinning fails.
751 	 * 2. memblock allocations: kernelcore/movablecore setups might create
752 	 *    situations where ZONE_MOVABLE contains unmovable allocations
753 	 *    after boot. Memory offlining and allocations fail early.
754 	 * 3. Memory holes: kernelcore/movablecore setups might create very rare
755 	 *    situations where ZONE_MOVABLE contains memory holes after boot,
756 	 *    for example, if we have sections that are only partially
757 	 *    populated. Memory offlining and allocations fail early.
758 	 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
759 	 *    memory offlining, such pages cannot be allocated.
760 	 * 5. Unmovable PG_offline pages: in paravirtualized environments,
761 	 *    hotplugged memory blocks might only partially be managed by the
762 	 *    buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
763 	 *    parts not manged by the buddy are unmovable PG_offline pages. In
764 	 *    some cases (virtio-mem), such pages can be skipped during
765 	 *    memory offlining, however, cannot be moved/allocated. These
766 	 *    techniques might use alloc_contig_range() to hide previously
767 	 *    exposed pages from the buddy again (e.g., to implement some sort
768 	 *    of memory unplug in virtio-mem).
769 	 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
770 	 *    situations where ZERO_PAGE(0) which is allocated differently
771 	 *    on different platforms may end up in a movable zone. ZERO_PAGE(0)
772 	 *    cannot be migrated.
773 	 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
774 	 *    memory to the MOVABLE zone, the vmemmap pages are also placed in
775 	 *    such zone. Such pages cannot be really moved around as they are
776 	 *    self-stored in the range, but they are treated as movable when
777 	 *    the range they describe is about to be offlined.
778 	 *
779 	 * In general, no unmovable allocations that degrade memory offlining
780 	 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
781 	 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
782 	 * if has_unmovable_pages() states that there are no unmovable pages,
783 	 * there can be false negatives).
784 	 */
785 	ZONE_MOVABLE,
786 #ifdef CONFIG_ZONE_DEVICE
787 	ZONE_DEVICE,
788 #endif
789 	__MAX_NR_ZONES
790 
791 };
792 
793 #ifndef __GENERATING_BOUNDS_H
794 
795 #define ASYNC_AND_SYNC 2
796 
797 struct zone {
798 	/* Read-mostly fields */
799 
800 	/* zone watermarks, access with *_wmark_pages(zone) macros */
801 	unsigned long _watermark[NR_WMARK];
802 	unsigned long watermark_boost;
803 
804 	unsigned long nr_reserved_highatomic;
805 
806 	/*
807 	 * We don't know if the memory that we're going to allocate will be
808 	 * freeable or/and it will be released eventually, so to avoid totally
809 	 * wasting several GB of ram we must reserve some of the lower zone
810 	 * memory (otherwise we risk to run OOM on the lower zones despite
811 	 * there being tons of freeable ram on the higher zones).  This array is
812 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
813 	 * changes.
814 	 */
815 	long lowmem_reserve[MAX_NR_ZONES];
816 
817 #ifdef CONFIG_NUMA
818 	int node;
819 #endif
820 	struct pglist_data	*zone_pgdat;
821 	struct per_cpu_pages	__percpu *per_cpu_pageset;
822 	struct per_cpu_zonestat	__percpu *per_cpu_zonestats;
823 	/*
824 	 * the high and batch values are copied to individual pagesets for
825 	 * faster access
826 	 */
827 	int pageset_high;
828 	int pageset_batch;
829 
830 #ifndef CONFIG_SPARSEMEM
831 	/*
832 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
833 	 * In SPARSEMEM, this map is stored in struct mem_section
834 	 */
835 	unsigned long		*pageblock_flags;
836 #endif /* CONFIG_SPARSEMEM */
837 
838 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
839 	unsigned long		zone_start_pfn;
840 
841 	/*
842 	 * spanned_pages is the total pages spanned by the zone, including
843 	 * holes, which is calculated as:
844 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
845 	 *
846 	 * present_pages is physical pages existing within the zone, which
847 	 * is calculated as:
848 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
849 	 *
850 	 * present_early_pages is present pages existing within the zone
851 	 * located on memory available since early boot, excluding hotplugged
852 	 * memory.
853 	 *
854 	 * managed_pages is present pages managed by the buddy system, which
855 	 * is calculated as (reserved_pages includes pages allocated by the
856 	 * bootmem allocator):
857 	 *	managed_pages = present_pages - reserved_pages;
858 	 *
859 	 * cma pages is present pages that are assigned for CMA use
860 	 * (MIGRATE_CMA).
861 	 *
862 	 * So present_pages may be used by memory hotplug or memory power
863 	 * management logic to figure out unmanaged pages by checking
864 	 * (present_pages - managed_pages). And managed_pages should be used
865 	 * by page allocator and vm scanner to calculate all kinds of watermarks
866 	 * and thresholds.
867 	 *
868 	 * Locking rules:
869 	 *
870 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
871 	 * It is a seqlock because it has to be read outside of zone->lock,
872 	 * and it is done in the main allocator path.  But, it is written
873 	 * quite infrequently.
874 	 *
875 	 * The span_seq lock is declared along with zone->lock because it is
876 	 * frequently read in proximity to zone->lock.  It's good to
877 	 * give them a chance of being in the same cacheline.
878 	 *
879 	 * Write access to present_pages at runtime should be protected by
880 	 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
881 	 * present_pages should use get_online_mems() to get a stable value.
882 	 */
883 	atomic_long_t		managed_pages;
884 	unsigned long		spanned_pages;
885 	unsigned long		present_pages;
886 #if defined(CONFIG_MEMORY_HOTPLUG)
887 	unsigned long		present_early_pages;
888 #endif
889 #ifdef CONFIG_CMA
890 	unsigned long		cma_pages;
891 #endif
892 
893 	const char		*name;
894 
895 #ifdef CONFIG_MEMORY_ISOLATION
896 	/*
897 	 * Number of isolated pageblock. It is used to solve incorrect
898 	 * freepage counting problem due to racy retrieving migratetype
899 	 * of pageblock. Protected by zone->lock.
900 	 */
901 	unsigned long		nr_isolate_pageblock;
902 #endif
903 
904 #ifdef CONFIG_MEMORY_HOTPLUG
905 	/* see spanned/present_pages for more description */
906 	seqlock_t		span_seqlock;
907 #endif
908 
909 	int initialized;
910 
911 	/* Write-intensive fields used from the page allocator */
912 	CACHELINE_PADDING(_pad1_);
913 
914 	/* free areas of different sizes */
915 	struct free_area	free_area[MAX_ORDER + 1];
916 
917 	/* zone flags, see below */
918 	unsigned long		flags;
919 
920 	/* Primarily protects free_area */
921 	spinlock_t		lock;
922 
923 	/* Write-intensive fields used by compaction and vmstats. */
924 	CACHELINE_PADDING(_pad2_);
925 
926 	/*
927 	 * When free pages are below this point, additional steps are taken
928 	 * when reading the number of free pages to avoid per-cpu counter
929 	 * drift allowing watermarks to be breached
930 	 */
931 	unsigned long percpu_drift_mark;
932 
933 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
934 	/* pfn where compaction free scanner should start */
935 	unsigned long		compact_cached_free_pfn;
936 	/* pfn where compaction migration scanner should start */
937 	unsigned long		compact_cached_migrate_pfn[ASYNC_AND_SYNC];
938 	unsigned long		compact_init_migrate_pfn;
939 	unsigned long		compact_init_free_pfn;
940 #endif
941 
942 #ifdef CONFIG_COMPACTION
943 	/*
944 	 * On compaction failure, 1<<compact_defer_shift compactions
945 	 * are skipped before trying again. The number attempted since
946 	 * last failure is tracked with compact_considered.
947 	 * compact_order_failed is the minimum compaction failed order.
948 	 */
949 	unsigned int		compact_considered;
950 	unsigned int		compact_defer_shift;
951 	int			compact_order_failed;
952 #endif
953 
954 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
955 	/* Set to true when the PG_migrate_skip bits should be cleared */
956 	bool			compact_blockskip_flush;
957 #endif
958 
959 	bool			contiguous;
960 
961 	CACHELINE_PADDING(_pad3_);
962 	/* Zone statistics */
963 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
964 	atomic_long_t		vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
965 } ____cacheline_internodealigned_in_smp;
966 
967 enum pgdat_flags {
968 	PGDAT_DIRTY,			/* reclaim scanning has recently found
969 					 * many dirty file pages at the tail
970 					 * of the LRU.
971 					 */
972 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
973 					 * many pages under writeback
974 					 */
975 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
976 };
977 
978 enum zone_flags {
979 	ZONE_BOOSTED_WATERMARK,		/* zone recently boosted watermarks.
980 					 * Cleared when kswapd is woken.
981 					 */
982 	ZONE_RECLAIM_ACTIVE,		/* kswapd may be scanning the zone. */
983 };
984 
985 static inline unsigned long zone_managed_pages(struct zone *zone)
986 {
987 	return (unsigned long)atomic_long_read(&zone->managed_pages);
988 }
989 
990 static inline unsigned long zone_cma_pages(struct zone *zone)
991 {
992 #ifdef CONFIG_CMA
993 	return zone->cma_pages;
994 #else
995 	return 0;
996 #endif
997 }
998 
999 static inline unsigned long zone_end_pfn(const struct zone *zone)
1000 {
1001 	return zone->zone_start_pfn + zone->spanned_pages;
1002 }
1003 
1004 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1005 {
1006 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1007 }
1008 
1009 static inline bool zone_is_initialized(struct zone *zone)
1010 {
1011 	return zone->initialized;
1012 }
1013 
1014 static inline bool zone_is_empty(struct zone *zone)
1015 {
1016 	return zone->spanned_pages == 0;
1017 }
1018 
1019 #ifndef BUILD_VDSO32_64
1020 /*
1021  * The zone field is never updated after free_area_init_core()
1022  * sets it, so none of the operations on it need to be atomic.
1023  */
1024 
1025 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1026 #define SECTIONS_PGOFF		((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1027 #define NODES_PGOFF		(SECTIONS_PGOFF - NODES_WIDTH)
1028 #define ZONES_PGOFF		(NODES_PGOFF - ZONES_WIDTH)
1029 #define LAST_CPUPID_PGOFF	(ZONES_PGOFF - LAST_CPUPID_WIDTH)
1030 #define KASAN_TAG_PGOFF		(LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1031 #define LRU_GEN_PGOFF		(KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1032 #define LRU_REFS_PGOFF		(LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1033 
1034 /*
1035  * Define the bit shifts to access each section.  For non-existent
1036  * sections we define the shift as 0; that plus a 0 mask ensures
1037  * the compiler will optimise away reference to them.
1038  */
1039 #define SECTIONS_PGSHIFT	(SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1040 #define NODES_PGSHIFT		(NODES_PGOFF * (NODES_WIDTH != 0))
1041 #define ZONES_PGSHIFT		(ZONES_PGOFF * (ZONES_WIDTH != 0))
1042 #define LAST_CPUPID_PGSHIFT	(LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1043 #define KASAN_TAG_PGSHIFT	(KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1044 
1045 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1046 #ifdef NODE_NOT_IN_PAGE_FLAGS
1047 #define ZONEID_SHIFT		(SECTIONS_SHIFT + ZONES_SHIFT)
1048 #define ZONEID_PGOFF		((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1049 						SECTIONS_PGOFF : ZONES_PGOFF)
1050 #else
1051 #define ZONEID_SHIFT		(NODES_SHIFT + ZONES_SHIFT)
1052 #define ZONEID_PGOFF		((NODES_PGOFF < ZONES_PGOFF) ? \
1053 						NODES_PGOFF : ZONES_PGOFF)
1054 #endif
1055 
1056 #define ZONEID_PGSHIFT		(ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1057 
1058 #define ZONES_MASK		((1UL << ZONES_WIDTH) - 1)
1059 #define NODES_MASK		((1UL << NODES_WIDTH) - 1)
1060 #define SECTIONS_MASK		((1UL << SECTIONS_WIDTH) - 1)
1061 #define LAST_CPUPID_MASK	((1UL << LAST_CPUPID_SHIFT) - 1)
1062 #define KASAN_TAG_MASK		((1UL << KASAN_TAG_WIDTH) - 1)
1063 #define ZONEID_MASK		((1UL << ZONEID_SHIFT) - 1)
1064 
1065 static inline enum zone_type page_zonenum(const struct page *page)
1066 {
1067 	ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1068 	return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1069 }
1070 
1071 static inline enum zone_type folio_zonenum(const struct folio *folio)
1072 {
1073 	return page_zonenum(&folio->page);
1074 }
1075 
1076 #ifdef CONFIG_ZONE_DEVICE
1077 static inline bool is_zone_device_page(const struct page *page)
1078 {
1079 	return page_zonenum(page) == ZONE_DEVICE;
1080 }
1081 
1082 /*
1083  * Consecutive zone device pages should not be merged into the same sgl
1084  * or bvec segment with other types of pages or if they belong to different
1085  * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1086  * without scanning the entire segment. This helper returns true either if
1087  * both pages are not zone device pages or both pages are zone device pages
1088  * with the same pgmap.
1089  */
1090 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1091 						     const struct page *b)
1092 {
1093 	if (is_zone_device_page(a) != is_zone_device_page(b))
1094 		return false;
1095 	if (!is_zone_device_page(a))
1096 		return true;
1097 	return a->pgmap == b->pgmap;
1098 }
1099 
1100 extern void memmap_init_zone_device(struct zone *, unsigned long,
1101 				    unsigned long, struct dev_pagemap *);
1102 #else
1103 static inline bool is_zone_device_page(const struct page *page)
1104 {
1105 	return false;
1106 }
1107 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1108 						     const struct page *b)
1109 {
1110 	return true;
1111 }
1112 #endif
1113 
1114 static inline bool folio_is_zone_device(const struct folio *folio)
1115 {
1116 	return is_zone_device_page(&folio->page);
1117 }
1118 
1119 static inline bool is_zone_movable_page(const struct page *page)
1120 {
1121 	return page_zonenum(page) == ZONE_MOVABLE;
1122 }
1123 #endif
1124 
1125 /*
1126  * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1127  * intersection with the given zone
1128  */
1129 static inline bool zone_intersects(struct zone *zone,
1130 		unsigned long start_pfn, unsigned long nr_pages)
1131 {
1132 	if (zone_is_empty(zone))
1133 		return false;
1134 	if (start_pfn >= zone_end_pfn(zone) ||
1135 	    start_pfn + nr_pages <= zone->zone_start_pfn)
1136 		return false;
1137 
1138 	return true;
1139 }
1140 
1141 /*
1142  * The "priority" of VM scanning is how much of the queues we will scan in one
1143  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1144  * queues ("queue_length >> 12") during an aging round.
1145  */
1146 #define DEF_PRIORITY 12
1147 
1148 /* Maximum number of zones on a zonelist */
1149 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1150 
1151 enum {
1152 	ZONELIST_FALLBACK,	/* zonelist with fallback */
1153 #ifdef CONFIG_NUMA
1154 	/*
1155 	 * The NUMA zonelists are doubled because we need zonelists that
1156 	 * restrict the allocations to a single node for __GFP_THISNODE.
1157 	 */
1158 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
1159 #endif
1160 	MAX_ZONELISTS
1161 };
1162 
1163 /*
1164  * This struct contains information about a zone in a zonelist. It is stored
1165  * here to avoid dereferences into large structures and lookups of tables
1166  */
1167 struct zoneref {
1168 	struct zone *zone;	/* Pointer to actual zone */
1169 	int zone_idx;		/* zone_idx(zoneref->zone) */
1170 };
1171 
1172 /*
1173  * One allocation request operates on a zonelist. A zonelist
1174  * is a list of zones, the first one is the 'goal' of the
1175  * allocation, the other zones are fallback zones, in decreasing
1176  * priority.
1177  *
1178  * To speed the reading of the zonelist, the zonerefs contain the zone index
1179  * of the entry being read. Helper functions to access information given
1180  * a struct zoneref are
1181  *
1182  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
1183  * zonelist_zone_idx()	- Return the index of the zone for an entry
1184  * zonelist_node_idx()	- Return the index of the node for an entry
1185  */
1186 struct zonelist {
1187 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1188 };
1189 
1190 /*
1191  * The array of struct pages for flatmem.
1192  * It must be declared for SPARSEMEM as well because there are configurations
1193  * that rely on that.
1194  */
1195 extern struct page *mem_map;
1196 
1197 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1198 struct deferred_split {
1199 	spinlock_t split_queue_lock;
1200 	struct list_head split_queue;
1201 	unsigned long split_queue_len;
1202 };
1203 #endif
1204 
1205 #ifdef CONFIG_MEMORY_FAILURE
1206 /*
1207  * Per NUMA node memory failure handling statistics.
1208  */
1209 struct memory_failure_stats {
1210 	/*
1211 	 * Number of raw pages poisoned.
1212 	 * Cases not accounted: memory outside kernel control, offline page,
1213 	 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1214 	 * error events, and unpoison actions from hwpoison_unpoison.
1215 	 */
1216 	unsigned long total;
1217 	/*
1218 	 * Recovery results of poisoned raw pages handled by memory_failure,
1219 	 * in sync with mf_result.
1220 	 * total = ignored + failed + delayed + recovered.
1221 	 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1222 	 */
1223 	unsigned long ignored;
1224 	unsigned long failed;
1225 	unsigned long delayed;
1226 	unsigned long recovered;
1227 };
1228 #endif
1229 
1230 /*
1231  * On NUMA machines, each NUMA node would have a pg_data_t to describe
1232  * it's memory layout. On UMA machines there is a single pglist_data which
1233  * describes the whole memory.
1234  *
1235  * Memory statistics and page replacement data structures are maintained on a
1236  * per-zone basis.
1237  */
1238 typedef struct pglist_data {
1239 	/*
1240 	 * node_zones contains just the zones for THIS node. Not all of the
1241 	 * zones may be populated, but it is the full list. It is referenced by
1242 	 * this node's node_zonelists as well as other node's node_zonelists.
1243 	 */
1244 	struct zone node_zones[MAX_NR_ZONES];
1245 
1246 	/*
1247 	 * node_zonelists contains references to all zones in all nodes.
1248 	 * Generally the first zones will be references to this node's
1249 	 * node_zones.
1250 	 */
1251 	struct zonelist node_zonelists[MAX_ZONELISTS];
1252 
1253 	int nr_zones; /* number of populated zones in this node */
1254 #ifdef CONFIG_FLATMEM	/* means !SPARSEMEM */
1255 	struct page *node_mem_map;
1256 #ifdef CONFIG_PAGE_EXTENSION
1257 	struct page_ext *node_page_ext;
1258 #endif
1259 #endif
1260 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1261 	/*
1262 	 * Must be held any time you expect node_start_pfn,
1263 	 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1264 	 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1265 	 * init.
1266 	 *
1267 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1268 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1269 	 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1270 	 *
1271 	 * Nests above zone->lock and zone->span_seqlock
1272 	 */
1273 	spinlock_t node_size_lock;
1274 #endif
1275 	unsigned long node_start_pfn;
1276 	unsigned long node_present_pages; /* total number of physical pages */
1277 	unsigned long node_spanned_pages; /* total size of physical page
1278 					     range, including holes */
1279 	int node_id;
1280 	wait_queue_head_t kswapd_wait;
1281 	wait_queue_head_t pfmemalloc_wait;
1282 
1283 	/* workqueues for throttling reclaim for different reasons. */
1284 	wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1285 
1286 	atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1287 	unsigned long nr_reclaim_start;	/* nr pages written while throttled
1288 					 * when throttling started. */
1289 #ifdef CONFIG_MEMORY_HOTPLUG
1290 	struct mutex kswapd_lock;
1291 #endif
1292 	struct task_struct *kswapd;	/* Protected by kswapd_lock */
1293 	int kswapd_order;
1294 	enum zone_type kswapd_highest_zoneidx;
1295 
1296 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
1297 
1298 #ifdef CONFIG_COMPACTION
1299 	int kcompactd_max_order;
1300 	enum zone_type kcompactd_highest_zoneidx;
1301 	wait_queue_head_t kcompactd_wait;
1302 	struct task_struct *kcompactd;
1303 	bool proactive_compact_trigger;
1304 #endif
1305 	/*
1306 	 * This is a per-node reserve of pages that are not available
1307 	 * to userspace allocations.
1308 	 */
1309 	unsigned long		totalreserve_pages;
1310 
1311 #ifdef CONFIG_NUMA
1312 	/*
1313 	 * node reclaim becomes active if more unmapped pages exist.
1314 	 */
1315 	unsigned long		min_unmapped_pages;
1316 	unsigned long		min_slab_pages;
1317 #endif /* CONFIG_NUMA */
1318 
1319 	/* Write-intensive fields used by page reclaim */
1320 	CACHELINE_PADDING(_pad1_);
1321 
1322 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1323 	/*
1324 	 * If memory initialisation on large machines is deferred then this
1325 	 * is the first PFN that needs to be initialised.
1326 	 */
1327 	unsigned long first_deferred_pfn;
1328 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1329 
1330 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1331 	struct deferred_split deferred_split_queue;
1332 #endif
1333 
1334 #ifdef CONFIG_NUMA_BALANCING
1335 	/* start time in ms of current promote rate limit period */
1336 	unsigned int nbp_rl_start;
1337 	/* number of promote candidate pages at start time of current rate limit period */
1338 	unsigned long nbp_rl_nr_cand;
1339 	/* promote threshold in ms */
1340 	unsigned int nbp_threshold;
1341 	/* start time in ms of current promote threshold adjustment period */
1342 	unsigned int nbp_th_start;
1343 	/*
1344 	 * number of promote candidate pages at start time of current promote
1345 	 * threshold adjustment period
1346 	 */
1347 	unsigned long nbp_th_nr_cand;
1348 #endif
1349 	/* Fields commonly accessed by the page reclaim scanner */
1350 
1351 	/*
1352 	 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1353 	 *
1354 	 * Use mem_cgroup_lruvec() to look up lruvecs.
1355 	 */
1356 	struct lruvec		__lruvec;
1357 
1358 	unsigned long		flags;
1359 
1360 #ifdef CONFIG_LRU_GEN
1361 	/* kswap mm walk data */
1362 	struct lru_gen_mm_walk mm_walk;
1363 	/* lru_gen_folio list */
1364 	struct lru_gen_memcg memcg_lru;
1365 #endif
1366 
1367 	CACHELINE_PADDING(_pad2_);
1368 
1369 	/* Per-node vmstats */
1370 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1371 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
1372 #ifdef CONFIG_NUMA
1373 	struct memory_tier __rcu *memtier;
1374 #endif
1375 #ifdef CONFIG_MEMORY_FAILURE
1376 	struct memory_failure_stats mf_stats;
1377 #endif
1378 } pg_data_t;
1379 
1380 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
1381 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
1382 
1383 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
1384 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1385 
1386 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1387 {
1388 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1389 }
1390 
1391 #include <linux/memory_hotplug.h>
1392 
1393 void build_all_zonelists(pg_data_t *pgdat);
1394 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1395 		   enum zone_type highest_zoneidx);
1396 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1397 			 int highest_zoneidx, unsigned int alloc_flags,
1398 			 long free_pages);
1399 bool zone_watermark_ok(struct zone *z, unsigned int order,
1400 		unsigned long mark, int highest_zoneidx,
1401 		unsigned int alloc_flags);
1402 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1403 		unsigned long mark, int highest_zoneidx);
1404 /*
1405  * Memory initialization context, use to differentiate memory added by
1406  * the platform statically or via memory hotplug interface.
1407  */
1408 enum meminit_context {
1409 	MEMINIT_EARLY,
1410 	MEMINIT_HOTPLUG,
1411 };
1412 
1413 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1414 				     unsigned long size);
1415 
1416 extern void lruvec_init(struct lruvec *lruvec);
1417 
1418 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1419 {
1420 #ifdef CONFIG_MEMCG
1421 	return lruvec->pgdat;
1422 #else
1423 	return container_of(lruvec, struct pglist_data, __lruvec);
1424 #endif
1425 }
1426 
1427 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1428 int local_memory_node(int node_id);
1429 #else
1430 static inline int local_memory_node(int node_id) { return node_id; };
1431 #endif
1432 
1433 /*
1434  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1435  */
1436 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
1437 
1438 #ifdef CONFIG_ZONE_DEVICE
1439 static inline bool zone_is_zone_device(struct zone *zone)
1440 {
1441 	return zone_idx(zone) == ZONE_DEVICE;
1442 }
1443 #else
1444 static inline bool zone_is_zone_device(struct zone *zone)
1445 {
1446 	return false;
1447 }
1448 #endif
1449 
1450 /*
1451  * Returns true if a zone has pages managed by the buddy allocator.
1452  * All the reclaim decisions have to use this function rather than
1453  * populated_zone(). If the whole zone is reserved then we can easily
1454  * end up with populated_zone() && !managed_zone().
1455  */
1456 static inline bool managed_zone(struct zone *zone)
1457 {
1458 	return zone_managed_pages(zone);
1459 }
1460 
1461 /* Returns true if a zone has memory */
1462 static inline bool populated_zone(struct zone *zone)
1463 {
1464 	return zone->present_pages;
1465 }
1466 
1467 #ifdef CONFIG_NUMA
1468 static inline int zone_to_nid(struct zone *zone)
1469 {
1470 	return zone->node;
1471 }
1472 
1473 static inline void zone_set_nid(struct zone *zone, int nid)
1474 {
1475 	zone->node = nid;
1476 }
1477 #else
1478 static inline int zone_to_nid(struct zone *zone)
1479 {
1480 	return 0;
1481 }
1482 
1483 static inline void zone_set_nid(struct zone *zone, int nid) {}
1484 #endif
1485 
1486 extern int movable_zone;
1487 
1488 static inline int is_highmem_idx(enum zone_type idx)
1489 {
1490 #ifdef CONFIG_HIGHMEM
1491 	return (idx == ZONE_HIGHMEM ||
1492 		(idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1493 #else
1494 	return 0;
1495 #endif
1496 }
1497 
1498 /**
1499  * is_highmem - helper function to quickly check if a struct zone is a
1500  *              highmem zone or not.  This is an attempt to keep references
1501  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1502  * @zone: pointer to struct zone variable
1503  * Return: 1 for a highmem zone, 0 otherwise
1504  */
1505 static inline int is_highmem(struct zone *zone)
1506 {
1507 	return is_highmem_idx(zone_idx(zone));
1508 }
1509 
1510 #ifdef CONFIG_ZONE_DMA
1511 bool has_managed_dma(void);
1512 #else
1513 static inline bool has_managed_dma(void)
1514 {
1515 	return false;
1516 }
1517 #endif
1518 
1519 /* These two functions are used to setup the per zone pages min values */
1520 struct ctl_table;
1521 
1522 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1523 		loff_t *);
1524 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1525 		size_t *, loff_t *);
1526 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1527 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1528 		size_t *, loff_t *);
1529 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1530 		void *, size_t *, loff_t *);
1531 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1532 		void *, size_t *, loff_t *);
1533 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1534 		void *, size_t *, loff_t *);
1535 int numa_zonelist_order_handler(struct ctl_table *, int,
1536 		void *, size_t *, loff_t *);
1537 extern int percpu_pagelist_high_fraction;
1538 extern char numa_zonelist_order[];
1539 #define NUMA_ZONELIST_ORDER_LEN	16
1540 
1541 #ifndef CONFIG_NUMA
1542 
1543 extern struct pglist_data contig_page_data;
1544 static inline struct pglist_data *NODE_DATA(int nid)
1545 {
1546 	return &contig_page_data;
1547 }
1548 
1549 #else /* CONFIG_NUMA */
1550 
1551 #include <asm/mmzone.h>
1552 
1553 #endif /* !CONFIG_NUMA */
1554 
1555 extern struct pglist_data *first_online_pgdat(void);
1556 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1557 extern struct zone *next_zone(struct zone *zone);
1558 
1559 /**
1560  * for_each_online_pgdat - helper macro to iterate over all online nodes
1561  * @pgdat: pointer to a pg_data_t variable
1562  */
1563 #define for_each_online_pgdat(pgdat)			\
1564 	for (pgdat = first_online_pgdat();		\
1565 	     pgdat;					\
1566 	     pgdat = next_online_pgdat(pgdat))
1567 /**
1568  * for_each_zone - helper macro to iterate over all memory zones
1569  * @zone: pointer to struct zone variable
1570  *
1571  * The user only needs to declare the zone variable, for_each_zone
1572  * fills it in.
1573  */
1574 #define for_each_zone(zone)			        \
1575 	for (zone = (first_online_pgdat())->node_zones; \
1576 	     zone;					\
1577 	     zone = next_zone(zone))
1578 
1579 #define for_each_populated_zone(zone)		        \
1580 	for (zone = (first_online_pgdat())->node_zones; \
1581 	     zone;					\
1582 	     zone = next_zone(zone))			\
1583 		if (!populated_zone(zone))		\
1584 			; /* do nothing */		\
1585 		else
1586 
1587 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1588 {
1589 	return zoneref->zone;
1590 }
1591 
1592 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1593 {
1594 	return zoneref->zone_idx;
1595 }
1596 
1597 static inline int zonelist_node_idx(struct zoneref *zoneref)
1598 {
1599 	return zone_to_nid(zoneref->zone);
1600 }
1601 
1602 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1603 					enum zone_type highest_zoneidx,
1604 					nodemask_t *nodes);
1605 
1606 /**
1607  * 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
1608  * @z: The cursor used as a starting point for the search
1609  * @highest_zoneidx: The zone index of the highest zone to return
1610  * @nodes: An optional nodemask to filter the zonelist with
1611  *
1612  * This function returns the next zone at or below a given zone index that is
1613  * within the allowed nodemask using a cursor as the starting point for the
1614  * search. The zoneref returned is a cursor that represents the current zone
1615  * being examined. It should be advanced by one before calling
1616  * next_zones_zonelist again.
1617  *
1618  * Return: the next zone at or below highest_zoneidx within the allowed
1619  * nodemask using a cursor within a zonelist as a starting point
1620  */
1621 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1622 					enum zone_type highest_zoneidx,
1623 					nodemask_t *nodes)
1624 {
1625 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1626 		return z;
1627 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
1628 }
1629 
1630 /**
1631  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1632  * @zonelist: The zonelist to search for a suitable zone
1633  * @highest_zoneidx: The zone index of the highest zone to return
1634  * @nodes: An optional nodemask to filter the zonelist with
1635  *
1636  * This function returns the first zone at or below a given zone index that is
1637  * within the allowed nodemask. The zoneref returned is a cursor that can be
1638  * used to iterate the zonelist with next_zones_zonelist by advancing it by
1639  * one before calling.
1640  *
1641  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1642  * never NULL). This may happen either genuinely, or due to concurrent nodemask
1643  * update due to cpuset modification.
1644  *
1645  * Return: Zoneref pointer for the first suitable zone found
1646  */
1647 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1648 					enum zone_type highest_zoneidx,
1649 					nodemask_t *nodes)
1650 {
1651 	return next_zones_zonelist(zonelist->_zonerefs,
1652 							highest_zoneidx, nodes);
1653 }
1654 
1655 /**
1656  * 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
1657  * @zone: The current zone in the iterator
1658  * @z: The current pointer within zonelist->_zonerefs being iterated
1659  * @zlist: The zonelist being iterated
1660  * @highidx: The zone index of the highest zone to return
1661  * @nodemask: Nodemask allowed by the allocator
1662  *
1663  * This iterator iterates though all zones at or below a given zone index and
1664  * within a given nodemask
1665  */
1666 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1667 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1668 		zone;							\
1669 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1670 			zone = zonelist_zone(z))
1671 
1672 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1673 	for (zone = z->zone;	\
1674 		zone;							\
1675 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1676 			zone = zonelist_zone(z))
1677 
1678 
1679 /**
1680  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1681  * @zone: The current zone in the iterator
1682  * @z: The current pointer within zonelist->zones being iterated
1683  * @zlist: The zonelist being iterated
1684  * @highidx: The zone index of the highest zone to return
1685  *
1686  * This iterator iterates though all zones at or below a given zone index.
1687  */
1688 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1689 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1690 
1691 /* Whether the 'nodes' are all movable nodes */
1692 static inline bool movable_only_nodes(nodemask_t *nodes)
1693 {
1694 	struct zonelist *zonelist;
1695 	struct zoneref *z;
1696 	int nid;
1697 
1698 	if (nodes_empty(*nodes))
1699 		return false;
1700 
1701 	/*
1702 	 * We can chose arbitrary node from the nodemask to get a
1703 	 * zonelist as they are interlinked. We just need to find
1704 	 * at least one zone that can satisfy kernel allocations.
1705 	 */
1706 	nid = first_node(*nodes);
1707 	zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1708 	z = first_zones_zonelist(zonelist, ZONE_NORMAL,	nodes);
1709 	return (!z->zone) ? true : false;
1710 }
1711 
1712 
1713 #ifdef CONFIG_SPARSEMEM
1714 #include <asm/sparsemem.h>
1715 #endif
1716 
1717 #ifdef CONFIG_FLATMEM
1718 #define pfn_to_nid(pfn)		(0)
1719 #endif
1720 
1721 #ifdef CONFIG_SPARSEMEM
1722 
1723 /*
1724  * PA_SECTION_SHIFT		physical address to/from section number
1725  * PFN_SECTION_SHIFT		pfn to/from section number
1726  */
1727 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1728 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1729 
1730 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1731 
1732 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1733 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1734 
1735 #define SECTION_BLOCKFLAGS_BITS \
1736 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1737 
1738 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1739 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1740 #endif
1741 
1742 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1743 {
1744 	return pfn >> PFN_SECTION_SHIFT;
1745 }
1746 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1747 {
1748 	return sec << PFN_SECTION_SHIFT;
1749 }
1750 
1751 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1752 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1753 
1754 #define SUBSECTION_SHIFT 21
1755 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1756 
1757 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1758 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1759 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1760 
1761 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1762 #error Subsection size exceeds section size
1763 #else
1764 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1765 #endif
1766 
1767 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1768 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1769 
1770 struct mem_section_usage {
1771 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1772 	DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1773 #endif
1774 	/* See declaration of similar field in struct zone */
1775 	unsigned long pageblock_flags[0];
1776 };
1777 
1778 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1779 
1780 struct page;
1781 struct page_ext;
1782 struct mem_section {
1783 	/*
1784 	 * This is, logically, a pointer to an array of struct
1785 	 * pages.  However, it is stored with some other magic.
1786 	 * (see sparse.c::sparse_init_one_section())
1787 	 *
1788 	 * Additionally during early boot we encode node id of
1789 	 * the location of the section here to guide allocation.
1790 	 * (see sparse.c::memory_present())
1791 	 *
1792 	 * Making it a UL at least makes someone do a cast
1793 	 * before using it wrong.
1794 	 */
1795 	unsigned long section_mem_map;
1796 
1797 	struct mem_section_usage *usage;
1798 #ifdef CONFIG_PAGE_EXTENSION
1799 	/*
1800 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1801 	 * section. (see page_ext.h about this.)
1802 	 */
1803 	struct page_ext *page_ext;
1804 	unsigned long pad;
1805 #endif
1806 	/*
1807 	 * WARNING: mem_section must be a power-of-2 in size for the
1808 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1809 	 */
1810 };
1811 
1812 #ifdef CONFIG_SPARSEMEM_EXTREME
1813 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1814 #else
1815 #define SECTIONS_PER_ROOT	1
1816 #endif
1817 
1818 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1819 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1820 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1821 
1822 #ifdef CONFIG_SPARSEMEM_EXTREME
1823 extern struct mem_section **mem_section;
1824 #else
1825 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1826 #endif
1827 
1828 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1829 {
1830 	return ms->usage->pageblock_flags;
1831 }
1832 
1833 static inline struct mem_section *__nr_to_section(unsigned long nr)
1834 {
1835 	unsigned long root = SECTION_NR_TO_ROOT(nr);
1836 
1837 	if (unlikely(root >= NR_SECTION_ROOTS))
1838 		return NULL;
1839 
1840 #ifdef CONFIG_SPARSEMEM_EXTREME
1841 	if (!mem_section || !mem_section[root])
1842 		return NULL;
1843 #endif
1844 	return &mem_section[root][nr & SECTION_ROOT_MASK];
1845 }
1846 extern size_t mem_section_usage_size(void);
1847 
1848 /*
1849  * We use the lower bits of the mem_map pointer to store
1850  * a little bit of information.  The pointer is calculated
1851  * as mem_map - section_nr_to_pfn(pnum).  The result is
1852  * aligned to the minimum alignment of the two values:
1853  *   1. All mem_map arrays are page-aligned.
1854  *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1855  *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1856  *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1857  *      worst combination is powerpc with 256k pages,
1858  *      which results in PFN_SECTION_SHIFT equal 6.
1859  * To sum it up, at least 6 bits are available on all architectures.
1860  * However, we can exceed 6 bits on some other architectures except
1861  * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1862  * with the worst case of 64K pages on arm64) if we make sure the
1863  * exceeded bit is not applicable to powerpc.
1864  */
1865 enum {
1866 	SECTION_MARKED_PRESENT_BIT,
1867 	SECTION_HAS_MEM_MAP_BIT,
1868 	SECTION_IS_ONLINE_BIT,
1869 	SECTION_IS_EARLY_BIT,
1870 #ifdef CONFIG_ZONE_DEVICE
1871 	SECTION_TAINT_ZONE_DEVICE_BIT,
1872 #endif
1873 	SECTION_MAP_LAST_BIT,
1874 };
1875 
1876 #define SECTION_MARKED_PRESENT		BIT(SECTION_MARKED_PRESENT_BIT)
1877 #define SECTION_HAS_MEM_MAP		BIT(SECTION_HAS_MEM_MAP_BIT)
1878 #define SECTION_IS_ONLINE		BIT(SECTION_IS_ONLINE_BIT)
1879 #define SECTION_IS_EARLY		BIT(SECTION_IS_EARLY_BIT)
1880 #ifdef CONFIG_ZONE_DEVICE
1881 #define SECTION_TAINT_ZONE_DEVICE	BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1882 #endif
1883 #define SECTION_MAP_MASK		(~(BIT(SECTION_MAP_LAST_BIT) - 1))
1884 #define SECTION_NID_SHIFT		SECTION_MAP_LAST_BIT
1885 
1886 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1887 {
1888 	unsigned long map = section->section_mem_map;
1889 	map &= SECTION_MAP_MASK;
1890 	return (struct page *)map;
1891 }
1892 
1893 static inline int present_section(struct mem_section *section)
1894 {
1895 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1896 }
1897 
1898 static inline int present_section_nr(unsigned long nr)
1899 {
1900 	return present_section(__nr_to_section(nr));
1901 }
1902 
1903 static inline int valid_section(struct mem_section *section)
1904 {
1905 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1906 }
1907 
1908 static inline int early_section(struct mem_section *section)
1909 {
1910 	return (section && (section->section_mem_map & SECTION_IS_EARLY));
1911 }
1912 
1913 static inline int valid_section_nr(unsigned long nr)
1914 {
1915 	return valid_section(__nr_to_section(nr));
1916 }
1917 
1918 static inline int online_section(struct mem_section *section)
1919 {
1920 	return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1921 }
1922 
1923 #ifdef CONFIG_ZONE_DEVICE
1924 static inline int online_device_section(struct mem_section *section)
1925 {
1926 	unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1927 
1928 	return section && ((section->section_mem_map & flags) == flags);
1929 }
1930 #else
1931 static inline int online_device_section(struct mem_section *section)
1932 {
1933 	return 0;
1934 }
1935 #endif
1936 
1937 static inline int online_section_nr(unsigned long nr)
1938 {
1939 	return online_section(__nr_to_section(nr));
1940 }
1941 
1942 #ifdef CONFIG_MEMORY_HOTPLUG
1943 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1944 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1945 #endif
1946 
1947 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1948 {
1949 	return __nr_to_section(pfn_to_section_nr(pfn));
1950 }
1951 
1952 extern unsigned long __highest_present_section_nr;
1953 
1954 static inline int subsection_map_index(unsigned long pfn)
1955 {
1956 	return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1957 }
1958 
1959 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1960 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1961 {
1962 	int idx = subsection_map_index(pfn);
1963 
1964 	return test_bit(idx, ms->usage->subsection_map);
1965 }
1966 #else
1967 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1968 {
1969 	return 1;
1970 }
1971 #endif
1972 
1973 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1974 /**
1975  * pfn_valid - check if there is a valid memory map entry for a PFN
1976  * @pfn: the page frame number to check
1977  *
1978  * Check if there is a valid memory map entry aka struct page for the @pfn.
1979  * Note, that availability of the memory map entry does not imply that
1980  * there is actual usable memory at that @pfn. The struct page may
1981  * represent a hole or an unusable page frame.
1982  *
1983  * Return: 1 for PFNs that have memory map entries and 0 otherwise
1984  */
1985 static inline int pfn_valid(unsigned long pfn)
1986 {
1987 	struct mem_section *ms;
1988 
1989 	/*
1990 	 * Ensure the upper PAGE_SHIFT bits are clear in the
1991 	 * pfn. Else it might lead to false positives when
1992 	 * some of the upper bits are set, but the lower bits
1993 	 * match a valid pfn.
1994 	 */
1995 	if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
1996 		return 0;
1997 
1998 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1999 		return 0;
2000 	ms = __pfn_to_section(pfn);
2001 	if (!valid_section(ms))
2002 		return 0;
2003 	/*
2004 	 * Traditionally early sections always returned pfn_valid() for
2005 	 * the entire section-sized span.
2006 	 */
2007 	return early_section(ms) || pfn_section_valid(ms, pfn);
2008 }
2009 #endif
2010 
2011 static inline int pfn_in_present_section(unsigned long pfn)
2012 {
2013 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2014 		return 0;
2015 	return present_section(__pfn_to_section(pfn));
2016 }
2017 
2018 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2019 {
2020 	while (++section_nr <= __highest_present_section_nr) {
2021 		if (present_section_nr(section_nr))
2022 			return section_nr;
2023 	}
2024 
2025 	return -1;
2026 }
2027 
2028 /*
2029  * These are _only_ used during initialisation, therefore they
2030  * can use __initdata ...  They could have names to indicate
2031  * this restriction.
2032  */
2033 #ifdef CONFIG_NUMA
2034 #define pfn_to_nid(pfn)							\
2035 ({									\
2036 	unsigned long __pfn_to_nid_pfn = (pfn);				\
2037 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
2038 })
2039 #else
2040 #define pfn_to_nid(pfn)		(0)
2041 #endif
2042 
2043 void sparse_init(void);
2044 #else
2045 #define sparse_init()	do {} while (0)
2046 #define sparse_index_init(_sec, _nid)  do {} while (0)
2047 #define pfn_in_present_section pfn_valid
2048 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2049 #endif /* CONFIG_SPARSEMEM */
2050 
2051 #endif /* !__GENERATING_BOUNDS.H */
2052 #endif /* !__ASSEMBLY__ */
2053 #endif /* _LINUX_MMZONE_H */
2054