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 
is_migrate_movable(int mt)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  */
migratetype_is_mergeable(int mt)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 #define folio_migratetype(folio)				\
109 	get_pfnblock_flags_mask(&folio->page, folio_pfn(folio),		\
110 			MIGRATETYPE_MASK)
111 struct free_area {
112 	struct list_head	free_list[MIGRATE_TYPES];
113 	unsigned long		nr_free;
114 };
115 
116 struct pglist_data;
117 
118 #ifdef CONFIG_NUMA
119 enum 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 
132 enum 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 
154 enum 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  */
vmstat_item_print_in_thp(enum node_stat_item item)218 static __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  */
vmstat_item_in_bytes(int idx)235 static __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 
263 enum 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 
272 enum 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 
is_file_lru(enum lru_list lru)284 static inline bool is_file_lru(enum lru_list lru)
285 {
286 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
287 }
288 
is_active_lru(enum lru_list lru)289 static 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 
298 enum 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 
373 struct lruvec;
374 struct 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 
381 enum {
382 	LRU_GEN_ANON,
383 	LRU_GEN_FILE,
384 };
385 
386 enum {
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  */
416 struct 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 
448 enum {
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 
461 struct 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 
474 struct 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 
491 void lru_gen_init_lruvec(struct lruvec *lruvec);
492 void 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 a memcg to the head of a random bin in its
509  *    current generation (old or young) and updates its "seg" to "head";
510  * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
511  *    current generation (old or young) and updates its "seg" to "tail";
512  * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
513  *    generation, updates its "gen" to "old" and resets its "seg" to "default";
514  * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
515  *    young generation, updates its "gen" to "young" and resets its "seg" to
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 a memcg below low, which triggers
521  *    MEMCG_LRU_TAIL;
522  * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
523  *    threshold, which triggers MEMCG_LRU_TAIL;
524  * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
525  *    threshold, which triggers MEMCG_LRU_YOUNG;
526  * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
527  * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
528  * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
529  *
530  * Notes:
531  * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
532  *    of their max_seq counters ensures the eventual fairness to all eligible
533  *    memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
534  * 2. There are only two valid generations: old (seq) and young (seq+1).
535  *    MEMCG_NR_GENS is set to three so that when reading the generation counter
536  *    locklessly, a stale value (seq-1) does not wraparound to young.
537  */
538 #define MEMCG_NR_GENS	3
539 #define MEMCG_NR_BINS	8
540 
541 struct lru_gen_memcg {
542 	/* the per-node memcg generation counter */
543 	unsigned long seq;
544 	/* each memcg has one lru_gen_folio per node */
545 	unsigned long nr_memcgs[MEMCG_NR_GENS];
546 	/* per-node lru_gen_folio list for global reclaim */
547 	struct hlist_nulls_head	fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
548 	/* protects the above */
549 	spinlock_t lock;
550 };
551 
552 void lru_gen_init_pgdat(struct pglist_data *pgdat);
553 
554 void lru_gen_init_memcg(struct mem_cgroup *memcg);
555 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
556 void lru_gen_online_memcg(struct mem_cgroup *memcg);
557 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
558 void lru_gen_release_memcg(struct mem_cgroup *memcg);
559 void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
560 
561 #else /* !CONFIG_MEMCG */
562 
563 #define MEMCG_NR_GENS	1
564 
565 struct lru_gen_memcg {
566 };
567 
lru_gen_init_pgdat(struct pglist_data * pgdat)568 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
569 {
570 }
571 
572 #endif /* CONFIG_MEMCG */
573 
574 #else /* !CONFIG_LRU_GEN */
575 
lru_gen_init_pgdat(struct pglist_data * pgdat)576 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
577 {
578 }
579 
lru_gen_init_lruvec(struct lruvec * lruvec)580 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
581 {
582 }
583 
lru_gen_look_around(struct page_vma_mapped_walk * pvmw)584 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
585 {
586 }
587 
588 #ifdef CONFIG_MEMCG
589 
lru_gen_init_memcg(struct mem_cgroup * memcg)590 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
591 {
592 }
593 
lru_gen_exit_memcg(struct mem_cgroup * memcg)594 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
595 {
596 }
597 
lru_gen_online_memcg(struct mem_cgroup * memcg)598 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
599 {
600 }
601 
lru_gen_offline_memcg(struct mem_cgroup * memcg)602 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
603 {
604 }
605 
lru_gen_release_memcg(struct mem_cgroup * memcg)606 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
607 {
608 }
609 
lru_gen_soft_reclaim(struct mem_cgroup * memcg,int nid)610 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
611 {
612 }
613 
614 #endif /* CONFIG_MEMCG */
615 
616 #endif /* CONFIG_LRU_GEN */
617 
618 struct lruvec {
619 	struct list_head		lists[NR_LRU_LISTS];
620 	/* per lruvec lru_lock for memcg */
621 	spinlock_t			lru_lock;
622 	/*
623 	 * These track the cost of reclaiming one LRU - file or anon -
624 	 * over the other. As the observed cost of reclaiming one LRU
625 	 * increases, the reclaim scan balance tips toward the other.
626 	 */
627 	unsigned long			anon_cost;
628 	unsigned long			file_cost;
629 	/* Non-resident age, driven by LRU movement */
630 	atomic_long_t			nonresident_age;
631 	/* Refaults at the time of last reclaim cycle */
632 	unsigned long			refaults[ANON_AND_FILE];
633 	/* Various lruvec state flags (enum lruvec_flags) */
634 	unsigned long			flags;
635 #ifdef CONFIG_LRU_GEN
636 	/* evictable pages divided into generations */
637 	struct lru_gen_folio		lrugen;
638 	/* to concurrently iterate lru_gen_mm_list */
639 	struct lru_gen_mm_state		mm_state;
640 #endif
641 #ifdef CONFIG_MEMCG
642 	struct pglist_data *pgdat;
643 #endif
644 };
645 
646 /* Isolate unmapped pages */
647 #define ISOLATE_UNMAPPED	((__force isolate_mode_t)0x2)
648 /* Isolate for asynchronous migration */
649 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
650 /* Isolate unevictable pages */
651 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
652 
653 /* LRU Isolation modes. */
654 typedef unsigned __bitwise isolate_mode_t;
655 
656 enum zone_watermarks {
657 	WMARK_MIN,
658 	WMARK_LOW,
659 	WMARK_HIGH,
660 	WMARK_PROMO,
661 	NR_WMARK
662 };
663 
664 /*
665  * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
666  * for THP which will usually be GFP_MOVABLE. Even if it is another type,
667  * it should not contribute to serious fragmentation causing THP allocation
668  * failures.
669  */
670 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
671 #define NR_PCP_THP 1
672 #else
673 #define NR_PCP_THP 0
674 #endif
675 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
676 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
677 
678 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
679 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
680 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
681 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
682 
683 struct per_cpu_pages {
684 	spinlock_t lock;	/* Protects lists field */
685 	int count;		/* number of pages in the list */
686 	int high;		/* high watermark, emptying needed */
687 	int batch;		/* chunk size for buddy add/remove */
688 	short free_factor;	/* batch scaling factor during free */
689 #ifdef CONFIG_NUMA
690 	short expire;		/* When 0, remote pagesets are drained */
691 #endif
692 
693 	/* Lists of pages, one per migrate type stored on the pcp-lists */
694 	struct list_head lists[NR_PCP_LISTS];
695 } ____cacheline_aligned_in_smp;
696 
697 struct per_cpu_zonestat {
698 #ifdef CONFIG_SMP
699 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
700 	s8 stat_threshold;
701 #endif
702 #ifdef CONFIG_NUMA
703 	/*
704 	 * Low priority inaccurate counters that are only folded
705 	 * on demand. Use a large type to avoid the overhead of
706 	 * folding during refresh_cpu_vm_stats.
707 	 */
708 	unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
709 #endif
710 };
711 
712 struct per_cpu_nodestat {
713 	s8 stat_threshold;
714 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
715 };
716 
717 #endif /* !__GENERATING_BOUNDS.H */
718 
719 enum zone_type {
720 	/*
721 	 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
722 	 * to DMA to all of the addressable memory (ZONE_NORMAL).
723 	 * On architectures where this area covers the whole 32 bit address
724 	 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
725 	 * DMA addressing constraints. This distinction is important as a 32bit
726 	 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
727 	 * platforms may need both zones as they support peripherals with
728 	 * different DMA addressing limitations.
729 	 */
730 #ifdef CONFIG_ZONE_DMA
731 	ZONE_DMA,
732 #endif
733 #ifdef CONFIG_ZONE_DMA32
734 	ZONE_DMA32,
735 #endif
736 	/*
737 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
738 	 * performed on pages in ZONE_NORMAL if the DMA devices support
739 	 * transfers to all addressable memory.
740 	 */
741 	ZONE_NORMAL,
742 #ifdef CONFIG_HIGHMEM
743 	/*
744 	 * A memory area that is only addressable by the kernel through
745 	 * mapping portions into its own address space. This is for example
746 	 * used by i386 to allow the kernel to address the memory beyond
747 	 * 900MB. The kernel will set up special mappings (page
748 	 * table entries on i386) for each page that the kernel needs to
749 	 * access.
750 	 */
751 	ZONE_HIGHMEM,
752 #endif
753 	/*
754 	 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
755 	 * movable pages with few exceptional cases described below. Main use
756 	 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
757 	 * likely to succeed, and to locally limit unmovable allocations - e.g.,
758 	 * to increase the number of THP/huge pages. Notable special cases are:
759 	 *
760 	 * 1. Pinned pages: (long-term) pinning of movable pages might
761 	 *    essentially turn such pages unmovable. Therefore, we do not allow
762 	 *    pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
763 	 *    faulted, they come from the right zone right away. However, it is
764 	 *    still possible that address space already has pages in
765 	 *    ZONE_MOVABLE at the time when pages are pinned (i.e. user has
766 	 *    touches that memory before pinning). In such case we migrate them
767 	 *    to a different zone. When migration fails - pinning fails.
768 	 * 2. memblock allocations: kernelcore/movablecore setups might create
769 	 *    situations where ZONE_MOVABLE contains unmovable allocations
770 	 *    after boot. Memory offlining and allocations fail early.
771 	 * 3. Memory holes: kernelcore/movablecore setups might create very rare
772 	 *    situations where ZONE_MOVABLE contains memory holes after boot,
773 	 *    for example, if we have sections that are only partially
774 	 *    populated. Memory offlining and allocations fail early.
775 	 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
776 	 *    memory offlining, such pages cannot be allocated.
777 	 * 5. Unmovable PG_offline pages: in paravirtualized environments,
778 	 *    hotplugged memory blocks might only partially be managed by the
779 	 *    buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
780 	 *    parts not manged by the buddy are unmovable PG_offline pages. In
781 	 *    some cases (virtio-mem), such pages can be skipped during
782 	 *    memory offlining, however, cannot be moved/allocated. These
783 	 *    techniques might use alloc_contig_range() to hide previously
784 	 *    exposed pages from the buddy again (e.g., to implement some sort
785 	 *    of memory unplug in virtio-mem).
786 	 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
787 	 *    situations where ZERO_PAGE(0) which is allocated differently
788 	 *    on different platforms may end up in a movable zone. ZERO_PAGE(0)
789 	 *    cannot be migrated.
790 	 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
791 	 *    memory to the MOVABLE zone, the vmemmap pages are also placed in
792 	 *    such zone. Such pages cannot be really moved around as they are
793 	 *    self-stored in the range, but they are treated as movable when
794 	 *    the range they describe is about to be offlined.
795 	 *
796 	 * In general, no unmovable allocations that degrade memory offlining
797 	 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
798 	 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
799 	 * if has_unmovable_pages() states that there are no unmovable pages,
800 	 * there can be false negatives).
801 	 */
802 	ZONE_MOVABLE,
803 #ifdef CONFIG_ZONE_DEVICE
804 	ZONE_DEVICE,
805 #endif
806 	__MAX_NR_ZONES
807 
808 };
809 
810 #ifndef __GENERATING_BOUNDS_H
811 
812 #define ASYNC_AND_SYNC 2
813 
814 struct zone {
815 	/* Read-mostly fields */
816 
817 	/* zone watermarks, access with *_wmark_pages(zone) macros */
818 	unsigned long _watermark[NR_WMARK];
819 	unsigned long watermark_boost;
820 
821 	unsigned long nr_reserved_highatomic;
822 
823 	/*
824 	 * We don't know if the memory that we're going to allocate will be
825 	 * freeable or/and it will be released eventually, so to avoid totally
826 	 * wasting several GB of ram we must reserve some of the lower zone
827 	 * memory (otherwise we risk to run OOM on the lower zones despite
828 	 * there being tons of freeable ram on the higher zones).  This array is
829 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
830 	 * changes.
831 	 */
832 	long lowmem_reserve[MAX_NR_ZONES];
833 
834 #ifdef CONFIG_NUMA
835 	int node;
836 #endif
837 	struct pglist_data	*zone_pgdat;
838 	struct per_cpu_pages	__percpu *per_cpu_pageset;
839 	struct per_cpu_zonestat	__percpu *per_cpu_zonestats;
840 	/*
841 	 * the high and batch values are copied to individual pagesets for
842 	 * faster access
843 	 */
844 	int pageset_high;
845 	int pageset_batch;
846 
847 #ifndef CONFIG_SPARSEMEM
848 	/*
849 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
850 	 * In SPARSEMEM, this map is stored in struct mem_section
851 	 */
852 	unsigned long		*pageblock_flags;
853 #endif /* CONFIG_SPARSEMEM */
854 
855 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
856 	unsigned long		zone_start_pfn;
857 
858 	/*
859 	 * spanned_pages is the total pages spanned by the zone, including
860 	 * holes, which is calculated as:
861 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
862 	 *
863 	 * present_pages is physical pages existing within the zone, which
864 	 * is calculated as:
865 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
866 	 *
867 	 * present_early_pages is present pages existing within the zone
868 	 * located on memory available since early boot, excluding hotplugged
869 	 * memory.
870 	 *
871 	 * managed_pages is present pages managed by the buddy system, which
872 	 * is calculated as (reserved_pages includes pages allocated by the
873 	 * bootmem allocator):
874 	 *	managed_pages = present_pages - reserved_pages;
875 	 *
876 	 * cma pages is present pages that are assigned for CMA use
877 	 * (MIGRATE_CMA).
878 	 *
879 	 * So present_pages may be used by memory hotplug or memory power
880 	 * management logic to figure out unmanaged pages by checking
881 	 * (present_pages - managed_pages). And managed_pages should be used
882 	 * by page allocator and vm scanner to calculate all kinds of watermarks
883 	 * and thresholds.
884 	 *
885 	 * Locking rules:
886 	 *
887 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
888 	 * It is a seqlock because it has to be read outside of zone->lock,
889 	 * and it is done in the main allocator path.  But, it is written
890 	 * quite infrequently.
891 	 *
892 	 * The span_seq lock is declared along with zone->lock because it is
893 	 * frequently read in proximity to zone->lock.  It's good to
894 	 * give them a chance of being in the same cacheline.
895 	 *
896 	 * Write access to present_pages at runtime should be protected by
897 	 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
898 	 * present_pages should use get_online_mems() to get a stable value.
899 	 */
900 	atomic_long_t		managed_pages;
901 	unsigned long		spanned_pages;
902 	unsigned long		present_pages;
903 #if defined(CONFIG_MEMORY_HOTPLUG)
904 	unsigned long		present_early_pages;
905 #endif
906 #ifdef CONFIG_CMA
907 	unsigned long		cma_pages;
908 #endif
909 
910 	const char		*name;
911 
912 #ifdef CONFIG_MEMORY_ISOLATION
913 	/*
914 	 * Number of isolated pageblock. It is used to solve incorrect
915 	 * freepage counting problem due to racy retrieving migratetype
916 	 * of pageblock. Protected by zone->lock.
917 	 */
918 	unsigned long		nr_isolate_pageblock;
919 #endif
920 
921 #ifdef CONFIG_MEMORY_HOTPLUG
922 	/* see spanned/present_pages for more description */
923 	seqlock_t		span_seqlock;
924 #endif
925 
926 	int initialized;
927 
928 	/* Write-intensive fields used from the page allocator */
929 	CACHELINE_PADDING(_pad1_);
930 
931 	/* free areas of different sizes */
932 	struct free_area	free_area[MAX_ORDER + 1];
933 
934 #ifdef CONFIG_UNACCEPTED_MEMORY
935 	/* Pages to be accepted. All pages on the list are MAX_ORDER */
936 	struct list_head	unaccepted_pages;
937 #endif
938 
939 	/* zone flags, see below */
940 	unsigned long		flags;
941 
942 	/* Primarily protects free_area */
943 	spinlock_t		lock;
944 
945 	/* Write-intensive fields used by compaction and vmstats. */
946 	CACHELINE_PADDING(_pad2_);
947 
948 	/*
949 	 * When free pages are below this point, additional steps are taken
950 	 * when reading the number of free pages to avoid per-cpu counter
951 	 * drift allowing watermarks to be breached
952 	 */
953 	unsigned long percpu_drift_mark;
954 
955 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
956 	/* pfn where compaction free scanner should start */
957 	unsigned long		compact_cached_free_pfn;
958 	/* pfn where compaction migration scanner should start */
959 	unsigned long		compact_cached_migrate_pfn[ASYNC_AND_SYNC];
960 	unsigned long		compact_init_migrate_pfn;
961 	unsigned long		compact_init_free_pfn;
962 #endif
963 
964 #ifdef CONFIG_COMPACTION
965 	/*
966 	 * On compaction failure, 1<<compact_defer_shift compactions
967 	 * are skipped before trying again. The number attempted since
968 	 * last failure is tracked with compact_considered.
969 	 * compact_order_failed is the minimum compaction failed order.
970 	 */
971 	unsigned int		compact_considered;
972 	unsigned int		compact_defer_shift;
973 	int			compact_order_failed;
974 #endif
975 
976 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
977 	/* Set to true when the PG_migrate_skip bits should be cleared */
978 	bool			compact_blockskip_flush;
979 #endif
980 
981 	bool			contiguous;
982 
983 	CACHELINE_PADDING(_pad3_);
984 	/* Zone statistics */
985 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
986 	atomic_long_t		vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
987 } ____cacheline_internodealigned_in_smp;
988 
989 enum pgdat_flags {
990 	PGDAT_DIRTY,			/* reclaim scanning has recently found
991 					 * many dirty file pages at the tail
992 					 * of the LRU.
993 					 */
994 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
995 					 * many pages under writeback
996 					 */
997 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
998 };
999 
1000 enum zone_flags {
1001 	ZONE_BOOSTED_WATERMARK,		/* zone recently boosted watermarks.
1002 					 * Cleared when kswapd is woken.
1003 					 */
1004 	ZONE_RECLAIM_ACTIVE,		/* kswapd may be scanning the zone. */
1005 };
1006 
zone_managed_pages(struct zone * zone)1007 static inline unsigned long zone_managed_pages(struct zone *zone)
1008 {
1009 	return (unsigned long)atomic_long_read(&zone->managed_pages);
1010 }
1011 
zone_cma_pages(struct zone * zone)1012 static inline unsigned long zone_cma_pages(struct zone *zone)
1013 {
1014 #ifdef CONFIG_CMA
1015 	return zone->cma_pages;
1016 #else
1017 	return 0;
1018 #endif
1019 }
1020 
zone_end_pfn(const struct zone * zone)1021 static inline unsigned long zone_end_pfn(const struct zone *zone)
1022 {
1023 	return zone->zone_start_pfn + zone->spanned_pages;
1024 }
1025 
zone_spans_pfn(const struct zone * zone,unsigned long pfn)1026 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1027 {
1028 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1029 }
1030 
zone_is_initialized(struct zone * zone)1031 static inline bool zone_is_initialized(struct zone *zone)
1032 {
1033 	return zone->initialized;
1034 }
1035 
zone_is_empty(struct zone * zone)1036 static inline bool zone_is_empty(struct zone *zone)
1037 {
1038 	return zone->spanned_pages == 0;
1039 }
1040 
1041 #ifndef BUILD_VDSO32_64
1042 /*
1043  * The zone field is never updated after free_area_init_core()
1044  * sets it, so none of the operations on it need to be atomic.
1045  */
1046 
1047 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1048 #define SECTIONS_PGOFF		((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1049 #define NODES_PGOFF		(SECTIONS_PGOFF - NODES_WIDTH)
1050 #define ZONES_PGOFF		(NODES_PGOFF - ZONES_WIDTH)
1051 #define LAST_CPUPID_PGOFF	(ZONES_PGOFF - LAST_CPUPID_WIDTH)
1052 #define KASAN_TAG_PGOFF		(LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1053 #define LRU_GEN_PGOFF		(KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1054 #define LRU_REFS_PGOFF		(LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1055 
1056 /*
1057  * Define the bit shifts to access each section.  For non-existent
1058  * sections we define the shift as 0; that plus a 0 mask ensures
1059  * the compiler will optimise away reference to them.
1060  */
1061 #define SECTIONS_PGSHIFT	(SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1062 #define NODES_PGSHIFT		(NODES_PGOFF * (NODES_WIDTH != 0))
1063 #define ZONES_PGSHIFT		(ZONES_PGOFF * (ZONES_WIDTH != 0))
1064 #define LAST_CPUPID_PGSHIFT	(LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1065 #define KASAN_TAG_PGSHIFT	(KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1066 
1067 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1068 #ifdef NODE_NOT_IN_PAGE_FLAGS
1069 #define ZONEID_SHIFT		(SECTIONS_SHIFT + ZONES_SHIFT)
1070 #define ZONEID_PGOFF		((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1071 						SECTIONS_PGOFF : ZONES_PGOFF)
1072 #else
1073 #define ZONEID_SHIFT		(NODES_SHIFT + ZONES_SHIFT)
1074 #define ZONEID_PGOFF		((NODES_PGOFF < ZONES_PGOFF) ? \
1075 						NODES_PGOFF : ZONES_PGOFF)
1076 #endif
1077 
1078 #define ZONEID_PGSHIFT		(ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1079 
1080 #define ZONES_MASK		((1UL << ZONES_WIDTH) - 1)
1081 #define NODES_MASK		((1UL << NODES_WIDTH) - 1)
1082 #define SECTIONS_MASK		((1UL << SECTIONS_WIDTH) - 1)
1083 #define LAST_CPUPID_MASK	((1UL << LAST_CPUPID_SHIFT) - 1)
1084 #define KASAN_TAG_MASK		((1UL << KASAN_TAG_WIDTH) - 1)
1085 #define ZONEID_MASK		((1UL << ZONEID_SHIFT) - 1)
1086 
page_zonenum(const struct page * page)1087 static inline enum zone_type page_zonenum(const struct page *page)
1088 {
1089 	ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1090 	return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1091 }
1092 
folio_zonenum(const struct folio * folio)1093 static inline enum zone_type folio_zonenum(const struct folio *folio)
1094 {
1095 	return page_zonenum(&folio->page);
1096 }
1097 
1098 #ifdef CONFIG_ZONE_DEVICE
is_zone_device_page(const struct page * page)1099 static inline bool is_zone_device_page(const struct page *page)
1100 {
1101 	return page_zonenum(page) == ZONE_DEVICE;
1102 }
1103 
1104 /*
1105  * Consecutive zone device pages should not be merged into the same sgl
1106  * or bvec segment with other types of pages or if they belong to different
1107  * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1108  * without scanning the entire segment. This helper returns true either if
1109  * both pages are not zone device pages or both pages are zone device pages
1110  * with the same pgmap.
1111  */
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1112 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1113 						     const struct page *b)
1114 {
1115 	if (is_zone_device_page(a) != is_zone_device_page(b))
1116 		return false;
1117 	if (!is_zone_device_page(a))
1118 		return true;
1119 	return a->pgmap == b->pgmap;
1120 }
1121 
1122 extern void memmap_init_zone_device(struct zone *, unsigned long,
1123 				    unsigned long, struct dev_pagemap *);
1124 #else
is_zone_device_page(const struct page * page)1125 static inline bool is_zone_device_page(const struct page *page)
1126 {
1127 	return false;
1128 }
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1129 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1130 						     const struct page *b)
1131 {
1132 	return true;
1133 }
1134 #endif
1135 
folio_is_zone_device(const struct folio * folio)1136 static inline bool folio_is_zone_device(const struct folio *folio)
1137 {
1138 	return is_zone_device_page(&folio->page);
1139 }
1140 
is_zone_movable_page(const struct page * page)1141 static inline bool is_zone_movable_page(const struct page *page)
1142 {
1143 	return page_zonenum(page) == ZONE_MOVABLE;
1144 }
1145 
folio_is_zone_movable(const struct folio * folio)1146 static inline bool folio_is_zone_movable(const struct folio *folio)
1147 {
1148 	return folio_zonenum(folio) == ZONE_MOVABLE;
1149 }
1150 #endif
1151 
1152 /*
1153  * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1154  * intersection with the given zone
1155  */
zone_intersects(struct zone * zone,unsigned long start_pfn,unsigned long nr_pages)1156 static inline bool zone_intersects(struct zone *zone,
1157 		unsigned long start_pfn, unsigned long nr_pages)
1158 {
1159 	if (zone_is_empty(zone))
1160 		return false;
1161 	if (start_pfn >= zone_end_pfn(zone) ||
1162 	    start_pfn + nr_pages <= zone->zone_start_pfn)
1163 		return false;
1164 
1165 	return true;
1166 }
1167 
1168 /*
1169  * The "priority" of VM scanning is how much of the queues we will scan in one
1170  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1171  * queues ("queue_length >> 12") during an aging round.
1172  */
1173 #define DEF_PRIORITY 12
1174 
1175 /* Maximum number of zones on a zonelist */
1176 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1177 
1178 enum {
1179 	ZONELIST_FALLBACK,	/* zonelist with fallback */
1180 #ifdef CONFIG_NUMA
1181 	/*
1182 	 * The NUMA zonelists are doubled because we need zonelists that
1183 	 * restrict the allocations to a single node for __GFP_THISNODE.
1184 	 */
1185 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
1186 #endif
1187 	MAX_ZONELISTS
1188 };
1189 
1190 /*
1191  * This struct contains information about a zone in a zonelist. It is stored
1192  * here to avoid dereferences into large structures and lookups of tables
1193  */
1194 struct zoneref {
1195 	struct zone *zone;	/* Pointer to actual zone */
1196 	int zone_idx;		/* zone_idx(zoneref->zone) */
1197 };
1198 
1199 /*
1200  * One allocation request operates on a zonelist. A zonelist
1201  * is a list of zones, the first one is the 'goal' of the
1202  * allocation, the other zones are fallback zones, in decreasing
1203  * priority.
1204  *
1205  * To speed the reading of the zonelist, the zonerefs contain the zone index
1206  * of the entry being read. Helper functions to access information given
1207  * a struct zoneref are
1208  *
1209  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
1210  * zonelist_zone_idx()	- Return the index of the zone for an entry
1211  * zonelist_node_idx()	- Return the index of the node for an entry
1212  */
1213 struct zonelist {
1214 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1215 };
1216 
1217 /*
1218  * The array of struct pages for flatmem.
1219  * It must be declared for SPARSEMEM as well because there are configurations
1220  * that rely on that.
1221  */
1222 extern struct page *mem_map;
1223 
1224 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1225 struct deferred_split {
1226 	spinlock_t split_queue_lock;
1227 	struct list_head split_queue;
1228 	unsigned long split_queue_len;
1229 };
1230 #endif
1231 
1232 #ifdef CONFIG_MEMORY_FAILURE
1233 /*
1234  * Per NUMA node memory failure handling statistics.
1235  */
1236 struct memory_failure_stats {
1237 	/*
1238 	 * Number of raw pages poisoned.
1239 	 * Cases not accounted: memory outside kernel control, offline page,
1240 	 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1241 	 * error events, and unpoison actions from hwpoison_unpoison.
1242 	 */
1243 	unsigned long total;
1244 	/*
1245 	 * Recovery results of poisoned raw pages handled by memory_failure,
1246 	 * in sync with mf_result.
1247 	 * total = ignored + failed + delayed + recovered.
1248 	 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1249 	 */
1250 	unsigned long ignored;
1251 	unsigned long failed;
1252 	unsigned long delayed;
1253 	unsigned long recovered;
1254 };
1255 #endif
1256 
1257 /*
1258  * On NUMA machines, each NUMA node would have a pg_data_t to describe
1259  * it's memory layout. On UMA machines there is a single pglist_data which
1260  * describes the whole memory.
1261  *
1262  * Memory statistics and page replacement data structures are maintained on a
1263  * per-zone basis.
1264  */
1265 typedef struct pglist_data {
1266 	/*
1267 	 * node_zones contains just the zones for THIS node. Not all of the
1268 	 * zones may be populated, but it is the full list. It is referenced by
1269 	 * this node's node_zonelists as well as other node's node_zonelists.
1270 	 */
1271 	struct zone node_zones[MAX_NR_ZONES];
1272 
1273 	/*
1274 	 * node_zonelists contains references to all zones in all nodes.
1275 	 * Generally the first zones will be references to this node's
1276 	 * node_zones.
1277 	 */
1278 	struct zonelist node_zonelists[MAX_ZONELISTS];
1279 
1280 	int nr_zones; /* number of populated zones in this node */
1281 #ifdef CONFIG_FLATMEM	/* means !SPARSEMEM */
1282 	struct page *node_mem_map;
1283 #ifdef CONFIG_PAGE_EXTENSION
1284 	struct page_ext *node_page_ext;
1285 #endif
1286 #endif
1287 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1288 	/*
1289 	 * Must be held any time you expect node_start_pfn,
1290 	 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1291 	 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1292 	 * init.
1293 	 *
1294 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1295 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1296 	 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1297 	 *
1298 	 * Nests above zone->lock and zone->span_seqlock
1299 	 */
1300 	spinlock_t node_size_lock;
1301 #endif
1302 	unsigned long node_start_pfn;
1303 	unsigned long node_present_pages; /* total number of physical pages */
1304 	unsigned long node_spanned_pages; /* total size of physical page
1305 					     range, including holes */
1306 	int node_id;
1307 	wait_queue_head_t kswapd_wait;
1308 	wait_queue_head_t pfmemalloc_wait;
1309 
1310 	/* workqueues for throttling reclaim for different reasons. */
1311 	wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1312 
1313 	atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1314 	unsigned long nr_reclaim_start;	/* nr pages written while throttled
1315 					 * when throttling started. */
1316 #ifdef CONFIG_MEMORY_HOTPLUG
1317 	struct mutex kswapd_lock;
1318 #endif
1319 	struct task_struct *kswapd;	/* Protected by kswapd_lock */
1320 	int kswapd_order;
1321 	enum zone_type kswapd_highest_zoneidx;
1322 
1323 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
1324 
1325 #ifdef CONFIG_COMPACTION
1326 	int kcompactd_max_order;
1327 	enum zone_type kcompactd_highest_zoneidx;
1328 	wait_queue_head_t kcompactd_wait;
1329 	struct task_struct *kcompactd;
1330 	bool proactive_compact_trigger;
1331 #endif
1332 	/*
1333 	 * This is a per-node reserve of pages that are not available
1334 	 * to userspace allocations.
1335 	 */
1336 	unsigned long		totalreserve_pages;
1337 
1338 #ifdef CONFIG_NUMA
1339 	/*
1340 	 * node reclaim becomes active if more unmapped pages exist.
1341 	 */
1342 	unsigned long		min_unmapped_pages;
1343 	unsigned long		min_slab_pages;
1344 #endif /* CONFIG_NUMA */
1345 
1346 	/* Write-intensive fields used by page reclaim */
1347 	CACHELINE_PADDING(_pad1_);
1348 
1349 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1350 	/*
1351 	 * If memory initialisation on large machines is deferred then this
1352 	 * is the first PFN that needs to be initialised.
1353 	 */
1354 	unsigned long first_deferred_pfn;
1355 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1356 
1357 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1358 	struct deferred_split deferred_split_queue;
1359 #endif
1360 
1361 #ifdef CONFIG_NUMA_BALANCING
1362 	/* start time in ms of current promote rate limit period */
1363 	unsigned int nbp_rl_start;
1364 	/* number of promote candidate pages at start time of current rate limit period */
1365 	unsigned long nbp_rl_nr_cand;
1366 	/* promote threshold in ms */
1367 	unsigned int nbp_threshold;
1368 	/* start time in ms of current promote threshold adjustment period */
1369 	unsigned int nbp_th_start;
1370 	/*
1371 	 * number of promote candidate pages at start time of current promote
1372 	 * threshold adjustment period
1373 	 */
1374 	unsigned long nbp_th_nr_cand;
1375 #endif
1376 	/* Fields commonly accessed by the page reclaim scanner */
1377 
1378 	/*
1379 	 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1380 	 *
1381 	 * Use mem_cgroup_lruvec() to look up lruvecs.
1382 	 */
1383 	struct lruvec		__lruvec;
1384 
1385 	unsigned long		flags;
1386 
1387 #ifdef CONFIG_LRU_GEN
1388 	/* kswap mm walk data */
1389 	struct lru_gen_mm_walk mm_walk;
1390 	/* lru_gen_folio list */
1391 	struct lru_gen_memcg memcg_lru;
1392 #endif
1393 
1394 	CACHELINE_PADDING(_pad2_);
1395 
1396 	/* Per-node vmstats */
1397 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1398 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
1399 #ifdef CONFIG_NUMA
1400 	struct memory_tier __rcu *memtier;
1401 #endif
1402 #ifdef CONFIG_MEMORY_FAILURE
1403 	struct memory_failure_stats mf_stats;
1404 #endif
1405 } pg_data_t;
1406 
1407 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
1408 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
1409 
1410 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
1411 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1412 
pgdat_end_pfn(pg_data_t * pgdat)1413 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1414 {
1415 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1416 }
1417 
1418 #include <linux/memory_hotplug.h>
1419 
1420 void build_all_zonelists(pg_data_t *pgdat);
1421 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1422 		   enum zone_type highest_zoneidx);
1423 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1424 			 int highest_zoneidx, unsigned int alloc_flags,
1425 			 long free_pages);
1426 bool zone_watermark_ok(struct zone *z, unsigned int order,
1427 		unsigned long mark, int highest_zoneidx,
1428 		unsigned int alloc_flags);
1429 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1430 		unsigned long mark, int highest_zoneidx);
1431 /*
1432  * Memory initialization context, use to differentiate memory added by
1433  * the platform statically or via memory hotplug interface.
1434  */
1435 enum meminit_context {
1436 	MEMINIT_EARLY,
1437 	MEMINIT_HOTPLUG,
1438 };
1439 
1440 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1441 				     unsigned long size);
1442 
1443 extern void lruvec_init(struct lruvec *lruvec);
1444 
lruvec_pgdat(struct lruvec * lruvec)1445 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1446 {
1447 #ifdef CONFIG_MEMCG
1448 	return lruvec->pgdat;
1449 #else
1450 	return container_of(lruvec, struct pglist_data, __lruvec);
1451 #endif
1452 }
1453 
1454 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1455 int local_memory_node(int node_id);
1456 #else
local_memory_node(int node_id)1457 static inline int local_memory_node(int node_id) { return node_id; };
1458 #endif
1459 
1460 /*
1461  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1462  */
1463 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
1464 
1465 #ifdef CONFIG_ZONE_DEVICE
zone_is_zone_device(struct zone * zone)1466 static inline bool zone_is_zone_device(struct zone *zone)
1467 {
1468 	return zone_idx(zone) == ZONE_DEVICE;
1469 }
1470 #else
zone_is_zone_device(struct zone * zone)1471 static inline bool zone_is_zone_device(struct zone *zone)
1472 {
1473 	return false;
1474 }
1475 #endif
1476 
1477 /*
1478  * Returns true if a zone has pages managed by the buddy allocator.
1479  * All the reclaim decisions have to use this function rather than
1480  * populated_zone(). If the whole zone is reserved then we can easily
1481  * end up with populated_zone() && !managed_zone().
1482  */
managed_zone(struct zone * zone)1483 static inline bool managed_zone(struct zone *zone)
1484 {
1485 	return zone_managed_pages(zone);
1486 }
1487 
1488 /* Returns true if a zone has memory */
populated_zone(struct zone * zone)1489 static inline bool populated_zone(struct zone *zone)
1490 {
1491 	return zone->present_pages;
1492 }
1493 
1494 #ifdef CONFIG_NUMA
zone_to_nid(struct zone * zone)1495 static inline int zone_to_nid(struct zone *zone)
1496 {
1497 	return zone->node;
1498 }
1499 
zone_set_nid(struct zone * zone,int nid)1500 static inline void zone_set_nid(struct zone *zone, int nid)
1501 {
1502 	zone->node = nid;
1503 }
1504 #else
zone_to_nid(struct zone * zone)1505 static inline int zone_to_nid(struct zone *zone)
1506 {
1507 	return 0;
1508 }
1509 
zone_set_nid(struct zone * zone,int nid)1510 static inline void zone_set_nid(struct zone *zone, int nid) {}
1511 #endif
1512 
1513 extern int movable_zone;
1514 
is_highmem_idx(enum zone_type idx)1515 static inline int is_highmem_idx(enum zone_type idx)
1516 {
1517 #ifdef CONFIG_HIGHMEM
1518 	return (idx == ZONE_HIGHMEM ||
1519 		(idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1520 #else
1521 	return 0;
1522 #endif
1523 }
1524 
1525 /**
1526  * is_highmem - helper function to quickly check if a struct zone is a
1527  *              highmem zone or not.  This is an attempt to keep references
1528  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1529  * @zone: pointer to struct zone variable
1530  * Return: 1 for a highmem zone, 0 otherwise
1531  */
is_highmem(struct zone * zone)1532 static inline int is_highmem(struct zone *zone)
1533 {
1534 	return is_highmem_idx(zone_idx(zone));
1535 }
1536 
1537 #ifdef CONFIG_ZONE_DMA
1538 bool has_managed_dma(void);
1539 #else
has_managed_dma(void)1540 static inline bool has_managed_dma(void)
1541 {
1542 	return false;
1543 }
1544 #endif
1545 
1546 
1547 #ifndef CONFIG_NUMA
1548 
1549 extern struct pglist_data contig_page_data;
NODE_DATA(int nid)1550 static inline struct pglist_data *NODE_DATA(int nid)
1551 {
1552 	return &contig_page_data;
1553 }
1554 
1555 #else /* CONFIG_NUMA */
1556 
1557 #include <asm/mmzone.h>
1558 
1559 #endif /* !CONFIG_NUMA */
1560 
1561 extern struct pglist_data *first_online_pgdat(void);
1562 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1563 extern struct zone *next_zone(struct zone *zone);
1564 
1565 /**
1566  * for_each_online_pgdat - helper macro to iterate over all online nodes
1567  * @pgdat: pointer to a pg_data_t variable
1568  */
1569 #define for_each_online_pgdat(pgdat)			\
1570 	for (pgdat = first_online_pgdat();		\
1571 	     pgdat;					\
1572 	     pgdat = next_online_pgdat(pgdat))
1573 /**
1574  * for_each_zone - helper macro to iterate over all memory zones
1575  * @zone: pointer to struct zone variable
1576  *
1577  * The user only needs to declare the zone variable, for_each_zone
1578  * fills it in.
1579  */
1580 #define for_each_zone(zone)			        \
1581 	for (zone = (first_online_pgdat())->node_zones; \
1582 	     zone;					\
1583 	     zone = next_zone(zone))
1584 
1585 #define for_each_populated_zone(zone)		        \
1586 	for (zone = (first_online_pgdat())->node_zones; \
1587 	     zone;					\
1588 	     zone = next_zone(zone))			\
1589 		if (!populated_zone(zone))		\
1590 			; /* do nothing */		\
1591 		else
1592 
zonelist_zone(struct zoneref * zoneref)1593 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1594 {
1595 	return zoneref->zone;
1596 }
1597 
zonelist_zone_idx(struct zoneref * zoneref)1598 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1599 {
1600 	return zoneref->zone_idx;
1601 }
1602 
zonelist_node_idx(struct zoneref * zoneref)1603 static inline int zonelist_node_idx(struct zoneref *zoneref)
1604 {
1605 	return zone_to_nid(zoneref->zone);
1606 }
1607 
1608 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1609 					enum zone_type highest_zoneidx,
1610 					nodemask_t *nodes);
1611 
1612 /**
1613  * 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
1614  * @z: The cursor used as a starting point for the search
1615  * @highest_zoneidx: The zone index of the highest zone to return
1616  * @nodes: An optional nodemask to filter the zonelist with
1617  *
1618  * This function returns the next zone at or below a given zone index that is
1619  * within the allowed nodemask using a cursor as the starting point for the
1620  * search. The zoneref returned is a cursor that represents the current zone
1621  * being examined. It should be advanced by one before calling
1622  * next_zones_zonelist again.
1623  *
1624  * Return: the next zone at or below highest_zoneidx within the allowed
1625  * nodemask using a cursor within a zonelist as a starting point
1626  */
next_zones_zonelist(struct zoneref * z,enum zone_type highest_zoneidx,nodemask_t * nodes)1627 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1628 					enum zone_type highest_zoneidx,
1629 					nodemask_t *nodes)
1630 {
1631 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1632 		return z;
1633 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
1634 }
1635 
1636 /**
1637  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1638  * @zonelist: The zonelist to search for a suitable zone
1639  * @highest_zoneidx: The zone index of the highest zone to return
1640  * @nodes: An optional nodemask to filter the zonelist with
1641  *
1642  * This function returns the first zone at or below a given zone index that is
1643  * within the allowed nodemask. The zoneref returned is a cursor that can be
1644  * used to iterate the zonelist with next_zones_zonelist by advancing it by
1645  * one before calling.
1646  *
1647  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1648  * never NULL). This may happen either genuinely, or due to concurrent nodemask
1649  * update due to cpuset modification.
1650  *
1651  * Return: Zoneref pointer for the first suitable zone found
1652  */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes)1653 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1654 					enum zone_type highest_zoneidx,
1655 					nodemask_t *nodes)
1656 {
1657 	return next_zones_zonelist(zonelist->_zonerefs,
1658 							highest_zoneidx, nodes);
1659 }
1660 
1661 /**
1662  * 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
1663  * @zone: The current zone in the iterator
1664  * @z: The current pointer within zonelist->_zonerefs being iterated
1665  * @zlist: The zonelist being iterated
1666  * @highidx: The zone index of the highest zone to return
1667  * @nodemask: Nodemask allowed by the allocator
1668  *
1669  * This iterator iterates though all zones at or below a given zone index and
1670  * within a given nodemask
1671  */
1672 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1673 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1674 		zone;							\
1675 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1676 			zone = zonelist_zone(z))
1677 
1678 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1679 	for (zone = z->zone;	\
1680 		zone;							\
1681 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1682 			zone = zonelist_zone(z))
1683 
1684 
1685 /**
1686  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1687  * @zone: The current zone in the iterator
1688  * @z: The current pointer within zonelist->zones being iterated
1689  * @zlist: The zonelist being iterated
1690  * @highidx: The zone index of the highest zone to return
1691  *
1692  * This iterator iterates though all zones at or below a given zone index.
1693  */
1694 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1695 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1696 
1697 /* Whether the 'nodes' are all movable nodes */
movable_only_nodes(nodemask_t * nodes)1698 static inline bool movable_only_nodes(nodemask_t *nodes)
1699 {
1700 	struct zonelist *zonelist;
1701 	struct zoneref *z;
1702 	int nid;
1703 
1704 	if (nodes_empty(*nodes))
1705 		return false;
1706 
1707 	/*
1708 	 * We can chose arbitrary node from the nodemask to get a
1709 	 * zonelist as they are interlinked. We just need to find
1710 	 * at least one zone that can satisfy kernel allocations.
1711 	 */
1712 	nid = first_node(*nodes);
1713 	zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1714 	z = first_zones_zonelist(zonelist, ZONE_NORMAL,	nodes);
1715 	return (!z->zone) ? true : false;
1716 }
1717 
1718 
1719 #ifdef CONFIG_SPARSEMEM
1720 #include <asm/sparsemem.h>
1721 #endif
1722 
1723 #ifdef CONFIG_FLATMEM
1724 #define pfn_to_nid(pfn)		(0)
1725 #endif
1726 
1727 #ifdef CONFIG_SPARSEMEM
1728 
1729 /*
1730  * PA_SECTION_SHIFT		physical address to/from section number
1731  * PFN_SECTION_SHIFT		pfn to/from section number
1732  */
1733 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1734 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1735 
1736 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1737 
1738 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1739 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1740 
1741 #define SECTION_BLOCKFLAGS_BITS \
1742 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1743 
1744 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1745 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1746 #endif
1747 
pfn_to_section_nr(unsigned long pfn)1748 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1749 {
1750 	return pfn >> PFN_SECTION_SHIFT;
1751 }
section_nr_to_pfn(unsigned long sec)1752 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1753 {
1754 	return sec << PFN_SECTION_SHIFT;
1755 }
1756 
1757 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1758 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1759 
1760 #define SUBSECTION_SHIFT 21
1761 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1762 
1763 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1764 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1765 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1766 
1767 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1768 #error Subsection size exceeds section size
1769 #else
1770 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1771 #endif
1772 
1773 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1774 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1775 
1776 struct mem_section_usage {
1777 	struct rcu_head rcu;
1778 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1779 	DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1780 #endif
1781 	/* See declaration of similar field in struct zone */
1782 	unsigned long pageblock_flags[0];
1783 };
1784 
1785 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1786 
1787 struct page;
1788 struct page_ext;
1789 struct mem_section {
1790 	/*
1791 	 * This is, logically, a pointer to an array of struct
1792 	 * pages.  However, it is stored with some other magic.
1793 	 * (see sparse.c::sparse_init_one_section())
1794 	 *
1795 	 * Additionally during early boot we encode node id of
1796 	 * the location of the section here to guide allocation.
1797 	 * (see sparse.c::memory_present())
1798 	 *
1799 	 * Making it a UL at least makes someone do a cast
1800 	 * before using it wrong.
1801 	 */
1802 	unsigned long section_mem_map;
1803 
1804 	struct mem_section_usage *usage;
1805 #ifdef CONFIG_PAGE_EXTENSION
1806 	/*
1807 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1808 	 * section. (see page_ext.h about this.)
1809 	 */
1810 	struct page_ext *page_ext;
1811 	unsigned long pad;
1812 #endif
1813 	/*
1814 	 * WARNING: mem_section must be a power-of-2 in size for the
1815 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1816 	 */
1817 };
1818 
1819 #ifdef CONFIG_SPARSEMEM_EXTREME
1820 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1821 #else
1822 #define SECTIONS_PER_ROOT	1
1823 #endif
1824 
1825 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1826 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1827 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1828 
1829 #ifdef CONFIG_SPARSEMEM_EXTREME
1830 extern struct mem_section **mem_section;
1831 #else
1832 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1833 #endif
1834 
section_to_usemap(struct mem_section * ms)1835 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1836 {
1837 	return ms->usage->pageblock_flags;
1838 }
1839 
__nr_to_section(unsigned long nr)1840 static inline struct mem_section *__nr_to_section(unsigned long nr)
1841 {
1842 	unsigned long root = SECTION_NR_TO_ROOT(nr);
1843 
1844 	if (unlikely(root >= NR_SECTION_ROOTS))
1845 		return NULL;
1846 
1847 #ifdef CONFIG_SPARSEMEM_EXTREME
1848 	if (!mem_section || !mem_section[root])
1849 		return NULL;
1850 #endif
1851 	return &mem_section[root][nr & SECTION_ROOT_MASK];
1852 }
1853 extern size_t mem_section_usage_size(void);
1854 
1855 /*
1856  * We use the lower bits of the mem_map pointer to store
1857  * a little bit of information.  The pointer is calculated
1858  * as mem_map - section_nr_to_pfn(pnum).  The result is
1859  * aligned to the minimum alignment of the two values:
1860  *   1. All mem_map arrays are page-aligned.
1861  *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1862  *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1863  *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1864  *      worst combination is powerpc with 256k pages,
1865  *      which results in PFN_SECTION_SHIFT equal 6.
1866  * To sum it up, at least 6 bits are available on all architectures.
1867  * However, we can exceed 6 bits on some other architectures except
1868  * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1869  * with the worst case of 64K pages on arm64) if we make sure the
1870  * exceeded bit is not applicable to powerpc.
1871  */
1872 enum {
1873 	SECTION_MARKED_PRESENT_BIT,
1874 	SECTION_HAS_MEM_MAP_BIT,
1875 	SECTION_IS_ONLINE_BIT,
1876 	SECTION_IS_EARLY_BIT,
1877 #ifdef CONFIG_ZONE_DEVICE
1878 	SECTION_TAINT_ZONE_DEVICE_BIT,
1879 #endif
1880 	SECTION_MAP_LAST_BIT,
1881 };
1882 
1883 #define SECTION_MARKED_PRESENT		BIT(SECTION_MARKED_PRESENT_BIT)
1884 #define SECTION_HAS_MEM_MAP		BIT(SECTION_HAS_MEM_MAP_BIT)
1885 #define SECTION_IS_ONLINE		BIT(SECTION_IS_ONLINE_BIT)
1886 #define SECTION_IS_EARLY		BIT(SECTION_IS_EARLY_BIT)
1887 #ifdef CONFIG_ZONE_DEVICE
1888 #define SECTION_TAINT_ZONE_DEVICE	BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1889 #endif
1890 #define SECTION_MAP_MASK		(~(BIT(SECTION_MAP_LAST_BIT) - 1))
1891 #define SECTION_NID_SHIFT		SECTION_MAP_LAST_BIT
1892 
__section_mem_map_addr(struct mem_section * section)1893 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1894 {
1895 	unsigned long map = section->section_mem_map;
1896 	map &= SECTION_MAP_MASK;
1897 	return (struct page *)map;
1898 }
1899 
present_section(struct mem_section * section)1900 static inline int present_section(struct mem_section *section)
1901 {
1902 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1903 }
1904 
present_section_nr(unsigned long nr)1905 static inline int present_section_nr(unsigned long nr)
1906 {
1907 	return present_section(__nr_to_section(nr));
1908 }
1909 
valid_section(struct mem_section * section)1910 static inline int valid_section(struct mem_section *section)
1911 {
1912 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1913 }
1914 
early_section(struct mem_section * section)1915 static inline int early_section(struct mem_section *section)
1916 {
1917 	return (section && (section->section_mem_map & SECTION_IS_EARLY));
1918 }
1919 
valid_section_nr(unsigned long nr)1920 static inline int valid_section_nr(unsigned long nr)
1921 {
1922 	return valid_section(__nr_to_section(nr));
1923 }
1924 
online_section(struct mem_section * section)1925 static inline int online_section(struct mem_section *section)
1926 {
1927 	return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1928 }
1929 
1930 #ifdef CONFIG_ZONE_DEVICE
online_device_section(struct mem_section * section)1931 static inline int online_device_section(struct mem_section *section)
1932 {
1933 	unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1934 
1935 	return section && ((section->section_mem_map & flags) == flags);
1936 }
1937 #else
online_device_section(struct mem_section * section)1938 static inline int online_device_section(struct mem_section *section)
1939 {
1940 	return 0;
1941 }
1942 #endif
1943 
online_section_nr(unsigned long nr)1944 static inline int online_section_nr(unsigned long nr)
1945 {
1946 	return online_section(__nr_to_section(nr));
1947 }
1948 
1949 #ifdef CONFIG_MEMORY_HOTPLUG
1950 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1951 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1952 #endif
1953 
__pfn_to_section(unsigned long pfn)1954 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1955 {
1956 	return __nr_to_section(pfn_to_section_nr(pfn));
1957 }
1958 
1959 extern unsigned long __highest_present_section_nr;
1960 
subsection_map_index(unsigned long pfn)1961 static inline int subsection_map_index(unsigned long pfn)
1962 {
1963 	return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1964 }
1965 
1966 #ifdef CONFIG_SPARSEMEM_VMEMMAP
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1967 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1968 {
1969 	int idx = subsection_map_index(pfn);
1970 
1971 	return test_bit(idx, READ_ONCE(ms->usage)->subsection_map);
1972 }
1973 #else
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1974 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1975 {
1976 	return 1;
1977 }
1978 #endif
1979 
1980 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1981 /**
1982  * pfn_valid - check if there is a valid memory map entry for a PFN
1983  * @pfn: the page frame number to check
1984  *
1985  * Check if there is a valid memory map entry aka struct page for the @pfn.
1986  * Note, that availability of the memory map entry does not imply that
1987  * there is actual usable memory at that @pfn. The struct page may
1988  * represent a hole or an unusable page frame.
1989  *
1990  * Return: 1 for PFNs that have memory map entries and 0 otherwise
1991  */
pfn_valid(unsigned long pfn)1992 static inline int pfn_valid(unsigned long pfn)
1993 {
1994 	struct mem_section *ms;
1995 	int ret;
1996 
1997 	/*
1998 	 * Ensure the upper PAGE_SHIFT bits are clear in the
1999 	 * pfn. Else it might lead to false positives when
2000 	 * some of the upper bits are set, but the lower bits
2001 	 * match a valid pfn.
2002 	 */
2003 	if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2004 		return 0;
2005 
2006 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2007 		return 0;
2008 	ms = __pfn_to_section(pfn);
2009 	rcu_read_lock_sched();
2010 	if (!valid_section(ms)) {
2011 		rcu_read_unlock_sched();
2012 		return 0;
2013 	}
2014 	/*
2015 	 * Traditionally early sections always returned pfn_valid() for
2016 	 * the entire section-sized span.
2017 	 */
2018 	ret = early_section(ms) || pfn_section_valid(ms, pfn);
2019 	rcu_read_unlock_sched();
2020 
2021 	return ret;
2022 }
2023 #endif
2024 
pfn_in_present_section(unsigned long pfn)2025 static inline int pfn_in_present_section(unsigned long pfn)
2026 {
2027 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2028 		return 0;
2029 	return present_section(__pfn_to_section(pfn));
2030 }
2031 
next_present_section_nr(unsigned long section_nr)2032 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2033 {
2034 	while (++section_nr <= __highest_present_section_nr) {
2035 		if (present_section_nr(section_nr))
2036 			return section_nr;
2037 	}
2038 
2039 	return -1;
2040 }
2041 
2042 /*
2043  * These are _only_ used during initialisation, therefore they
2044  * can use __initdata ...  They could have names to indicate
2045  * this restriction.
2046  */
2047 #ifdef CONFIG_NUMA
2048 #define pfn_to_nid(pfn)							\
2049 ({									\
2050 	unsigned long __pfn_to_nid_pfn = (pfn);				\
2051 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
2052 })
2053 #else
2054 #define pfn_to_nid(pfn)		(0)
2055 #endif
2056 
2057 void sparse_init(void);
2058 #else
2059 #define sparse_init()	do {} while (0)
2060 #define sparse_index_init(_sec, _nid)  do {} while (0)
2061 #define pfn_in_present_section pfn_valid
2062 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2063 #endif /* CONFIG_SPARSEMEM */
2064 
2065 #endif /* !__GENERATING_BOUNDS.H */
2066 #endif /* !__ASSEMBLY__ */
2067 #endif /* _LINUX_MMZONE_H */
2068