xref: /linux/include/linux/mmzone.h (revision 0cc6f45cecb46cefe89c17ec816dc8cd58a2229a)
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 <linux/zswap.h>
26 #include <asm/page.h>
27 
28 /* Free memory management - zoned buddy allocator.  */
29 #ifndef CONFIG_ARCH_FORCE_MAX_ORDER
30 #define MAX_PAGE_ORDER 10
31 #else
32 #define MAX_PAGE_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
33 #endif
34 #define MAX_ORDER_NR_PAGES (1 << MAX_PAGE_ORDER)
35 
36 #define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
37 
38 #define NR_PAGE_ORDERS (MAX_PAGE_ORDER + 1)
39 
40 /*
41  * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
42  * costly to service.  That is between allocation orders which should
43  * coalesce naturally under reasonable reclaim pressure and those which
44  * will not.
45  */
46 #define PAGE_ALLOC_COSTLY_ORDER 3
47 
48 enum migratetype {
49 	MIGRATE_UNMOVABLE,
50 	MIGRATE_MOVABLE,
51 	MIGRATE_RECLAIMABLE,
52 	MIGRATE_PCPTYPES,	/* the number of types on the pcp lists */
53 	MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
54 #ifdef CONFIG_CMA
55 	/*
56 	 * MIGRATE_CMA migration type is designed to mimic the way
57 	 * ZONE_MOVABLE works.  Only movable pages can be allocated
58 	 * from MIGRATE_CMA pageblocks and page allocator never
59 	 * implicitly change migration type of MIGRATE_CMA pageblock.
60 	 *
61 	 * The way to use it is to change migratetype of a range of
62 	 * pageblocks to MIGRATE_CMA which can be done by
63 	 * __free_pageblock_cma() function.
64 	 */
65 	MIGRATE_CMA,
66 #endif
67 #ifdef CONFIG_MEMORY_ISOLATION
68 	MIGRATE_ISOLATE,	/* can't allocate from here */
69 #endif
70 	MIGRATE_TYPES
71 };
72 
73 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
74 extern const char * const migratetype_names[MIGRATE_TYPES];
75 
76 #ifdef CONFIG_CMA
77 #  define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
78 #  define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
79 #  define is_migrate_cma_folio(folio, pfn)	(MIGRATE_CMA ==		\
80 	get_pfnblock_flags_mask(&folio->page, pfn, MIGRATETYPE_MASK))
81 #else
82 #  define is_migrate_cma(migratetype) false
83 #  define is_migrate_cma_page(_page) false
84 #  define is_migrate_cma_folio(folio, pfn) false
85 #endif
86 
is_migrate_movable(int mt)87 static inline bool is_migrate_movable(int mt)
88 {
89 	return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
90 }
91 
92 /*
93  * Check whether a migratetype can be merged with another migratetype.
94  *
95  * It is only mergeable when it can fall back to other migratetypes for
96  * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
97  */
migratetype_is_mergeable(int mt)98 static inline bool migratetype_is_mergeable(int mt)
99 {
100 	return mt < MIGRATE_PCPTYPES;
101 }
102 
103 #define for_each_migratetype_order(order, type) \
104 	for (order = 0; order < NR_PAGE_ORDERS; order++) \
105 		for (type = 0; type < MIGRATE_TYPES; type++)
106 
107 extern int page_group_by_mobility_disabled;
108 
109 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
110 
111 #define get_pageblock_migratetype(page)					\
112 	get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
113 
114 #define folio_migratetype(folio)				\
115 	get_pfnblock_flags_mask(&folio->page, folio_pfn(folio),		\
116 			MIGRATETYPE_MASK)
117 struct free_area {
118 	struct list_head	free_list[MIGRATE_TYPES];
119 	unsigned long		nr_free;
120 };
121 
122 struct pglist_data;
123 
124 #ifdef CONFIG_NUMA
125 enum numa_stat_item {
126 	NUMA_HIT,		/* allocated in intended node */
127 	NUMA_MISS,		/* allocated in non intended node */
128 	NUMA_FOREIGN,		/* was intended here, hit elsewhere */
129 	NUMA_INTERLEAVE_HIT,	/* interleaver preferred this zone */
130 	NUMA_LOCAL,		/* allocation from local node */
131 	NUMA_OTHER,		/* allocation from other node */
132 	NR_VM_NUMA_EVENT_ITEMS
133 };
134 #else
135 #define NR_VM_NUMA_EVENT_ITEMS 0
136 #endif
137 
138 enum zone_stat_item {
139 	/* First 128 byte cacheline (assuming 64 bit words) */
140 	NR_FREE_PAGES,
141 	NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
142 	NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
143 	NR_ZONE_ACTIVE_ANON,
144 	NR_ZONE_INACTIVE_FILE,
145 	NR_ZONE_ACTIVE_FILE,
146 	NR_ZONE_UNEVICTABLE,
147 	NR_ZONE_WRITE_PENDING,	/* Count of dirty, writeback and unstable pages */
148 	NR_MLOCK,		/* mlock()ed pages found and moved off LRU */
149 	/* Second 128 byte cacheline */
150 	NR_BOUNCE,
151 #if IS_ENABLED(CONFIG_ZSMALLOC)
152 	NR_ZSPAGES,		/* allocated in zsmalloc */
153 #endif
154 	NR_FREE_CMA_PAGES,
155 #ifdef CONFIG_UNACCEPTED_MEMORY
156 	NR_UNACCEPTED,
157 #endif
158 	NR_VM_ZONE_STAT_ITEMS };
159 
160 enum node_stat_item {
161 	NR_LRU_BASE,
162 	NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
163 	NR_ACTIVE_ANON,		/*  "     "     "   "       "         */
164 	NR_INACTIVE_FILE,	/*  "     "     "   "       "         */
165 	NR_ACTIVE_FILE,		/*  "     "     "   "       "         */
166 	NR_UNEVICTABLE,		/*  "     "     "   "       "         */
167 	NR_SLAB_RECLAIMABLE_B,
168 	NR_SLAB_UNRECLAIMABLE_B,
169 	NR_ISOLATED_ANON,	/* Temporary isolated pages from anon lru */
170 	NR_ISOLATED_FILE,	/* Temporary isolated pages from file lru */
171 	WORKINGSET_NODES,
172 	WORKINGSET_REFAULT_BASE,
173 	WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
174 	WORKINGSET_REFAULT_FILE,
175 	WORKINGSET_ACTIVATE_BASE,
176 	WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
177 	WORKINGSET_ACTIVATE_FILE,
178 	WORKINGSET_RESTORE_BASE,
179 	WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
180 	WORKINGSET_RESTORE_FILE,
181 	WORKINGSET_NODERECLAIM,
182 	NR_ANON_MAPPED,	/* Mapped anonymous pages */
183 	NR_FILE_MAPPED,	/* pagecache pages mapped into pagetables.
184 			   only modified from process context */
185 	NR_FILE_PAGES,
186 	NR_FILE_DIRTY,
187 	NR_WRITEBACK,
188 	NR_WRITEBACK_TEMP,	/* Writeback using temporary buffers */
189 	NR_SHMEM,		/* shmem pages (included tmpfs/GEM pages) */
190 	NR_SHMEM_THPS,
191 	NR_SHMEM_PMDMAPPED,
192 	NR_FILE_THPS,
193 	NR_FILE_PMDMAPPED,
194 	NR_ANON_THPS,
195 	NR_VMSCAN_WRITE,
196 	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
197 	NR_DIRTIED,		/* page dirtyings since bootup */
198 	NR_WRITTEN,		/* page writings since bootup */
199 	NR_THROTTLED_WRITTEN,	/* NR_WRITTEN while reclaim throttled */
200 	NR_KERNEL_MISC_RECLAIMABLE,	/* reclaimable non-slab kernel pages */
201 	NR_FOLL_PIN_ACQUIRED,	/* via: pin_user_page(), gup flag: FOLL_PIN */
202 	NR_FOLL_PIN_RELEASED,	/* pages returned via unpin_user_page() */
203 	NR_KERNEL_STACK_KB,	/* measured in KiB */
204 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
205 	NR_KERNEL_SCS_KB,	/* measured in KiB */
206 #endif
207 	NR_PAGETABLE,		/* used for pagetables */
208 	NR_SECONDARY_PAGETABLE, /* secondary pagetables, KVM & IOMMU */
209 #ifdef CONFIG_IOMMU_SUPPORT
210 	NR_IOMMU_PAGES,		/* # of pages allocated by IOMMU */
211 #endif
212 #ifdef CONFIG_SWAP
213 	NR_SWAPCACHE,
214 #endif
215 #ifdef CONFIG_NUMA_BALANCING
216 	PGPROMOTE_SUCCESS,	/* promote successfully */
217 	PGPROMOTE_CANDIDATE,	/* candidate pages to promote */
218 #endif
219 	/* PGDEMOTE_*: pages demoted */
220 	PGDEMOTE_KSWAPD,
221 	PGDEMOTE_DIRECT,
222 	PGDEMOTE_KHUGEPAGED,
223 	NR_VM_NODE_STAT_ITEMS
224 };
225 
226 /*
227  * Returns true if the item should be printed in THPs (/proc/vmstat
228  * currently prints number of anon, file and shmem THPs. But the item
229  * is charged in pages).
230  */
vmstat_item_print_in_thp(enum node_stat_item item)231 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
232 {
233 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
234 		return false;
235 
236 	return item == NR_ANON_THPS ||
237 	       item == NR_FILE_THPS ||
238 	       item == NR_SHMEM_THPS ||
239 	       item == NR_SHMEM_PMDMAPPED ||
240 	       item == NR_FILE_PMDMAPPED;
241 }
242 
243 /*
244  * Returns true if the value is measured in bytes (most vmstat values are
245  * measured in pages). This defines the API part, the internal representation
246  * might be different.
247  */
vmstat_item_in_bytes(int idx)248 static __always_inline bool vmstat_item_in_bytes(int idx)
249 {
250 	/*
251 	 * Global and per-node slab counters track slab pages.
252 	 * It's expected that changes are multiples of PAGE_SIZE.
253 	 * Internally values are stored in pages.
254 	 *
255 	 * Per-memcg and per-lruvec counters track memory, consumed
256 	 * by individual slab objects. These counters are actually
257 	 * byte-precise.
258 	 */
259 	return (idx == NR_SLAB_RECLAIMABLE_B ||
260 		idx == NR_SLAB_UNRECLAIMABLE_B);
261 }
262 
263 /*
264  * We do arithmetic on the LRU lists in various places in the code,
265  * so it is important to keep the active lists LRU_ACTIVE higher in
266  * the array than the corresponding inactive lists, and to keep
267  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
268  *
269  * This has to be kept in sync with the statistics in zone_stat_item
270  * above and the descriptions in vmstat_text in mm/vmstat.c
271  */
272 #define LRU_BASE 0
273 #define LRU_ACTIVE 1
274 #define LRU_FILE 2
275 
276 enum lru_list {
277 	LRU_INACTIVE_ANON = LRU_BASE,
278 	LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
279 	LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
280 	LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
281 	LRU_UNEVICTABLE,
282 	NR_LRU_LISTS
283 };
284 
285 enum vmscan_throttle_state {
286 	VMSCAN_THROTTLE_WRITEBACK,
287 	VMSCAN_THROTTLE_ISOLATED,
288 	VMSCAN_THROTTLE_NOPROGRESS,
289 	VMSCAN_THROTTLE_CONGESTED,
290 	NR_VMSCAN_THROTTLE,
291 };
292 
293 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
294 
295 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
296 
is_file_lru(enum lru_list lru)297 static inline bool is_file_lru(enum lru_list lru)
298 {
299 	return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
300 }
301 
is_active_lru(enum lru_list lru)302 static inline bool is_active_lru(enum lru_list lru)
303 {
304 	return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
305 }
306 
307 #define WORKINGSET_ANON 0
308 #define WORKINGSET_FILE 1
309 #define ANON_AND_FILE 2
310 
311 enum lruvec_flags {
312 	/*
313 	 * An lruvec has many dirty pages backed by a congested BDI:
314 	 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
315 	 *    It can be cleared by cgroup reclaim or kswapd.
316 	 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
317 	 *    It can only be cleared by kswapd.
318 	 *
319 	 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
320 	 * reclaim, but not vice versa. This only applies to the root cgroup.
321 	 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
322 	 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
323 	 * by kswapd).
324 	 */
325 	LRUVEC_CGROUP_CONGESTED,
326 	LRUVEC_NODE_CONGESTED,
327 };
328 
329 #endif /* !__GENERATING_BOUNDS_H */
330 
331 /*
332  * Evictable pages are divided into multiple generations. The youngest and the
333  * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
334  * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
335  * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
336  * corresponding generation. The gen counter in folio->flags stores gen+1 while
337  * a page is on one of lrugen->folios[]. Otherwise it stores 0.
338  *
339  * A page is added to the youngest generation on faulting. The aging needs to
340  * check the accessed bit at least twice before handing this page over to the
341  * eviction. The first check takes care of the accessed bit set on the initial
342  * fault; the second check makes sure this page hasn't been used since then.
343  * This process, AKA second chance, requires a minimum of two generations,
344  * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
345  * LRU, e.g., /proc/vmstat, these two generations are considered active; the
346  * rest of generations, if they exist, are considered inactive. See
347  * lru_gen_is_active().
348  *
349  * PG_active is always cleared while a page is on one of lrugen->folios[] so
350  * that the aging needs not to worry about it. And it's set again when a page
351  * considered active is isolated for non-reclaiming purposes, e.g., migration.
352  * See lru_gen_add_folio() and lru_gen_del_folio().
353  *
354  * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
355  * number of categories of the active/inactive LRU when keeping track of
356  * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
357  * in folio->flags.
358  */
359 #define MIN_NR_GENS		2U
360 #define MAX_NR_GENS		4U
361 
362 /*
363  * Each generation is divided into multiple tiers. A page accessed N times
364  * through file descriptors is in tier order_base_2(N). A page in the first tier
365  * (N=0,1) is marked by PG_referenced unless it was faulted in through page
366  * tables or read ahead. A page in any other tier (N>1) is marked by
367  * PG_referenced and PG_workingset. This implies a minimum of two tiers is
368  * supported without using additional bits in folio->flags.
369  *
370  * In contrast to moving across generations which requires the LRU lock, moving
371  * across tiers only involves atomic operations on folio->flags and therefore
372  * has a negligible cost in the buffered access path. In the eviction path,
373  * comparisons of refaulted/(evicted+protected) from the first tier and the
374  * rest infer whether pages accessed multiple times through file descriptors
375  * are statistically hot and thus worth protecting.
376  *
377  * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
378  * number of categories of the active/inactive LRU when keeping track of
379  * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
380  * folio->flags.
381  */
382 #define MAX_NR_TIERS		4U
383 
384 #ifndef __GENERATING_BOUNDS_H
385 
386 struct lruvec;
387 struct page_vma_mapped_walk;
388 
389 #define LRU_GEN_MASK		((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
390 #define LRU_REFS_MASK		((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
391 
392 #ifdef CONFIG_LRU_GEN
393 
394 enum {
395 	LRU_GEN_ANON,
396 	LRU_GEN_FILE,
397 };
398 
399 enum {
400 	LRU_GEN_CORE,
401 	LRU_GEN_MM_WALK,
402 	LRU_GEN_NONLEAF_YOUNG,
403 	NR_LRU_GEN_CAPS
404 };
405 
406 #define MIN_LRU_BATCH		BITS_PER_LONG
407 #define MAX_LRU_BATCH		(MIN_LRU_BATCH * 64)
408 
409 /* whether to keep historical stats from evicted generations */
410 #ifdef CONFIG_LRU_GEN_STATS
411 #define NR_HIST_GENS		MAX_NR_GENS
412 #else
413 #define NR_HIST_GENS		1U
414 #endif
415 
416 /*
417  * The youngest generation number is stored in max_seq for both anon and file
418  * types as they are aged on an equal footing. The oldest generation numbers are
419  * stored in min_seq[] separately for anon and file types as clean file pages
420  * can be evicted regardless of swap constraints.
421  *
422  * Normally anon and file min_seq are in sync. But if swapping is constrained,
423  * e.g., out of swap space, file min_seq is allowed to advance and leave anon
424  * min_seq behind.
425  *
426  * The number of pages in each generation is eventually consistent and therefore
427  * can be transiently negative when reset_batch_size() is pending.
428  */
429 struct lru_gen_folio {
430 	/* the aging increments the youngest generation number */
431 	unsigned long max_seq;
432 	/* the eviction increments the oldest generation numbers */
433 	unsigned long min_seq[ANON_AND_FILE];
434 	/* the birth time of each generation in jiffies */
435 	unsigned long timestamps[MAX_NR_GENS];
436 	/* the multi-gen LRU lists, lazily sorted on eviction */
437 	struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
438 	/* the multi-gen LRU sizes, eventually consistent */
439 	long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
440 	/* the exponential moving average of refaulted */
441 	unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
442 	/* the exponential moving average of evicted+protected */
443 	unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
444 	/* the first tier doesn't need protection, hence the minus one */
445 	unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
446 	/* can be modified without holding the LRU lock */
447 	atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
448 	atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
449 	/* whether the multi-gen LRU is enabled */
450 	bool enabled;
451 	/* the memcg generation this lru_gen_folio belongs to */
452 	u8 gen;
453 	/* the list segment this lru_gen_folio belongs to */
454 	u8 seg;
455 	/* per-node lru_gen_folio list for global reclaim */
456 	struct hlist_nulls_node list;
457 };
458 
459 enum {
460 	MM_LEAF_TOTAL,		/* total leaf entries */
461 	MM_LEAF_OLD,		/* old leaf entries */
462 	MM_LEAF_YOUNG,		/* young leaf entries */
463 	MM_NONLEAF_TOTAL,	/* total non-leaf entries */
464 	MM_NONLEAF_FOUND,	/* non-leaf entries found in Bloom filters */
465 	MM_NONLEAF_ADDED,	/* non-leaf entries added to Bloom filters */
466 	NR_MM_STATS
467 };
468 
469 /* double-buffering Bloom filters */
470 #define NR_BLOOM_FILTERS	2
471 
472 struct lru_gen_mm_state {
473 	/* synced with max_seq after each iteration */
474 	unsigned long seq;
475 	/* where the current iteration continues after */
476 	struct list_head *head;
477 	/* where the last iteration ended before */
478 	struct list_head *tail;
479 	/* Bloom filters flip after each iteration */
480 	unsigned long *filters[NR_BLOOM_FILTERS];
481 	/* the mm stats for debugging */
482 	unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
483 };
484 
485 struct lru_gen_mm_walk {
486 	/* the lruvec under reclaim */
487 	struct lruvec *lruvec;
488 	/* max_seq from lru_gen_folio: can be out of date */
489 	unsigned long seq;
490 	/* the next address within an mm to scan */
491 	unsigned long next_addr;
492 	/* to batch promoted pages */
493 	int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
494 	/* to batch the mm stats */
495 	int mm_stats[NR_MM_STATS];
496 	/* total batched items */
497 	int batched;
498 	bool can_swap;
499 	bool force_scan;
500 };
501 
502 /*
503  * For each node, memcgs are divided into two generations: the old and the
504  * young. For each generation, memcgs are randomly sharded into multiple bins
505  * to improve scalability. For each bin, the hlist_nulls is virtually divided
506  * into three segments: the head, the tail and the default.
507  *
508  * An onlining memcg is added to the tail of a random bin in the old generation.
509  * The eviction starts at the head of a random bin in the old generation. The
510  * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
511  * the old generation, is incremented when all its bins become empty.
512  *
513  * There are four operations:
514  * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
515  *    current generation (old or young) and updates its "seg" to "head";
516  * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
517  *    current generation (old or young) and updates its "seg" to "tail";
518  * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
519  *    generation, updates its "gen" to "old" and resets its "seg" to "default";
520  * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
521  *    young generation, updates its "gen" to "young" and resets its "seg" to
522  *    "default".
523  *
524  * The events that trigger the above operations are:
525  * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
526  * 2. The first attempt to reclaim a memcg below low, which triggers
527  *    MEMCG_LRU_TAIL;
528  * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
529  *    threshold, which triggers MEMCG_LRU_TAIL;
530  * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
531  *    threshold, which triggers MEMCG_LRU_YOUNG;
532  * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
533  * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
534  * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
535  *
536  * Notes:
537  * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
538  *    of their max_seq counters ensures the eventual fairness to all eligible
539  *    memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
540  * 2. There are only two valid generations: old (seq) and young (seq+1).
541  *    MEMCG_NR_GENS is set to three so that when reading the generation counter
542  *    locklessly, a stale value (seq-1) does not wraparound to young.
543  */
544 #define MEMCG_NR_GENS	3
545 #define MEMCG_NR_BINS	8
546 
547 struct lru_gen_memcg {
548 	/* the per-node memcg generation counter */
549 	unsigned long seq;
550 	/* each memcg has one lru_gen_folio per node */
551 	unsigned long nr_memcgs[MEMCG_NR_GENS];
552 	/* per-node lru_gen_folio list for global reclaim */
553 	struct hlist_nulls_head	fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
554 	/* protects the above */
555 	spinlock_t lock;
556 };
557 
558 void lru_gen_init_pgdat(struct pglist_data *pgdat);
559 void lru_gen_init_lruvec(struct lruvec *lruvec);
560 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
561 
562 void lru_gen_init_memcg(struct mem_cgroup *memcg);
563 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
564 void lru_gen_online_memcg(struct mem_cgroup *memcg);
565 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
566 void lru_gen_release_memcg(struct mem_cgroup *memcg);
567 void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
568 
569 #else /* !CONFIG_LRU_GEN */
570 
lru_gen_init_pgdat(struct pglist_data * pgdat)571 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
572 {
573 }
574 
lru_gen_init_lruvec(struct lruvec * lruvec)575 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
576 {
577 }
578 
lru_gen_look_around(struct page_vma_mapped_walk * pvmw)579 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
580 {
581 }
582 
lru_gen_init_memcg(struct mem_cgroup * memcg)583 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
584 {
585 }
586 
lru_gen_exit_memcg(struct mem_cgroup * memcg)587 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
588 {
589 }
590 
lru_gen_online_memcg(struct mem_cgroup * memcg)591 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
592 {
593 }
594 
lru_gen_offline_memcg(struct mem_cgroup * memcg)595 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
596 {
597 }
598 
lru_gen_release_memcg(struct mem_cgroup * memcg)599 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
600 {
601 }
602 
lru_gen_soft_reclaim(struct mem_cgroup * memcg,int nid)603 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
604 {
605 }
606 
607 #endif /* CONFIG_LRU_GEN */
608 
609 struct lruvec {
610 	struct list_head		lists[NR_LRU_LISTS];
611 	/* per lruvec lru_lock for memcg */
612 	spinlock_t			lru_lock;
613 	/*
614 	 * These track the cost of reclaiming one LRU - file or anon -
615 	 * over the other. As the observed cost of reclaiming one LRU
616 	 * increases, the reclaim scan balance tips toward the other.
617 	 */
618 	unsigned long			anon_cost;
619 	unsigned long			file_cost;
620 	/* Non-resident age, driven by LRU movement */
621 	atomic_long_t			nonresident_age;
622 	/* Refaults at the time of last reclaim cycle */
623 	unsigned long			refaults[ANON_AND_FILE];
624 	/* Various lruvec state flags (enum lruvec_flags) */
625 	unsigned long			flags;
626 #ifdef CONFIG_LRU_GEN
627 	/* evictable pages divided into generations */
628 	struct lru_gen_folio		lrugen;
629 #ifdef CONFIG_LRU_GEN_WALKS_MMU
630 	/* to concurrently iterate lru_gen_mm_list */
631 	struct lru_gen_mm_state		mm_state;
632 #endif
633 #endif /* CONFIG_LRU_GEN */
634 #ifdef CONFIG_MEMCG
635 	struct pglist_data *pgdat;
636 #endif
637 	struct zswap_lruvec_state zswap_lruvec_state;
638 };
639 
640 /* Isolate for asynchronous migration */
641 #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
642 /* Isolate unevictable pages */
643 #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
644 
645 /* LRU Isolation modes. */
646 typedef unsigned __bitwise isolate_mode_t;
647 
648 enum zone_watermarks {
649 	WMARK_MIN,
650 	WMARK_LOW,
651 	WMARK_HIGH,
652 	WMARK_PROMO,
653 	NR_WMARK
654 };
655 
656 /*
657  * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
658  * for THP which will usually be GFP_MOVABLE. Even if it is another type,
659  * it should not contribute to serious fragmentation causing THP allocation
660  * failures.
661  */
662 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
663 #define NR_PCP_THP 1
664 #else
665 #define NR_PCP_THP 0
666 #endif
667 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
668 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
669 
670 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
671 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
672 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
673 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
674 
675 /*
676  * Flags used in pcp->flags field.
677  *
678  * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
679  * previous page freeing.  To avoid to drain PCP for an accident
680  * high-order page freeing.
681  *
682  * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
683  * draining PCP for consecutive high-order pages freeing without
684  * allocation if data cache slice of CPU is large enough.  To reduce
685  * zone lock contention and keep cache-hot pages reusing.
686  */
687 #define	PCPF_PREV_FREE_HIGH_ORDER	BIT(0)
688 #define	PCPF_FREE_HIGH_BATCH		BIT(1)
689 
690 struct per_cpu_pages {
691 	spinlock_t lock;	/* Protects lists field */
692 	int count;		/* number of pages in the list */
693 	int high;		/* high watermark, emptying needed */
694 	int high_min;		/* min high watermark */
695 	int high_max;		/* max high watermark */
696 	int batch;		/* chunk size for buddy add/remove */
697 	u8 flags;		/* protected by pcp->lock */
698 	u8 alloc_factor;	/* batch scaling factor during allocate */
699 #ifdef CONFIG_NUMA
700 	u8 expire;		/* When 0, remote pagesets are drained */
701 #endif
702 	short free_count;	/* consecutive free count */
703 
704 	/* Lists of pages, one per migrate type stored on the pcp-lists */
705 	struct list_head lists[NR_PCP_LISTS];
706 } ____cacheline_aligned_in_smp;
707 
708 struct per_cpu_zonestat {
709 #ifdef CONFIG_SMP
710 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
711 	s8 stat_threshold;
712 #endif
713 #ifdef CONFIG_NUMA
714 	/*
715 	 * Low priority inaccurate counters that are only folded
716 	 * on demand. Use a large type to avoid the overhead of
717 	 * folding during refresh_cpu_vm_stats.
718 	 */
719 	unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
720 #endif
721 };
722 
723 struct per_cpu_nodestat {
724 	s8 stat_threshold;
725 	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
726 };
727 
728 #endif /* !__GENERATING_BOUNDS.H */
729 
730 enum zone_type {
731 	/*
732 	 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
733 	 * to DMA to all of the addressable memory (ZONE_NORMAL).
734 	 * On architectures where this area covers the whole 32 bit address
735 	 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
736 	 * DMA addressing constraints. This distinction is important as a 32bit
737 	 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
738 	 * platforms may need both zones as they support peripherals with
739 	 * different DMA addressing limitations.
740 	 */
741 #ifdef CONFIG_ZONE_DMA
742 	ZONE_DMA,
743 #endif
744 #ifdef CONFIG_ZONE_DMA32
745 	ZONE_DMA32,
746 #endif
747 	/*
748 	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
749 	 * performed on pages in ZONE_NORMAL if the DMA devices support
750 	 * transfers to all addressable memory.
751 	 */
752 	ZONE_NORMAL,
753 #ifdef CONFIG_HIGHMEM
754 	/*
755 	 * A memory area that is only addressable by the kernel through
756 	 * mapping portions into its own address space. This is for example
757 	 * used by i386 to allow the kernel to address the memory beyond
758 	 * 900MB. The kernel will set up special mappings (page
759 	 * table entries on i386) for each page that the kernel needs to
760 	 * access.
761 	 */
762 	ZONE_HIGHMEM,
763 #endif
764 	/*
765 	 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
766 	 * movable pages with few exceptional cases described below. Main use
767 	 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
768 	 * likely to succeed, and to locally limit unmovable allocations - e.g.,
769 	 * to increase the number of THP/huge pages. Notable special cases are:
770 	 *
771 	 * 1. Pinned pages: (long-term) pinning of movable pages might
772 	 *    essentially turn such pages unmovable. Therefore, we do not allow
773 	 *    pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
774 	 *    faulted, they come from the right zone right away. However, it is
775 	 *    still possible that address space already has pages in
776 	 *    ZONE_MOVABLE at the time when pages are pinned (i.e. user has
777 	 *    touches that memory before pinning). In such case we migrate them
778 	 *    to a different zone. When migration fails - pinning fails.
779 	 * 2. memblock allocations: kernelcore/movablecore setups might create
780 	 *    situations where ZONE_MOVABLE contains unmovable allocations
781 	 *    after boot. Memory offlining and allocations fail early.
782 	 * 3. Memory holes: kernelcore/movablecore setups might create very rare
783 	 *    situations where ZONE_MOVABLE contains memory holes after boot,
784 	 *    for example, if we have sections that are only partially
785 	 *    populated. Memory offlining and allocations fail early.
786 	 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
787 	 *    memory offlining, such pages cannot be allocated.
788 	 * 5. Unmovable PG_offline pages: in paravirtualized environments,
789 	 *    hotplugged memory blocks might only partially be managed by the
790 	 *    buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
791 	 *    parts not manged by the buddy are unmovable PG_offline pages. In
792 	 *    some cases (virtio-mem), such pages can be skipped during
793 	 *    memory offlining, however, cannot be moved/allocated. These
794 	 *    techniques might use alloc_contig_range() to hide previously
795 	 *    exposed pages from the buddy again (e.g., to implement some sort
796 	 *    of memory unplug in virtio-mem).
797 	 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
798 	 *    situations where ZERO_PAGE(0) which is allocated differently
799 	 *    on different platforms may end up in a movable zone. ZERO_PAGE(0)
800 	 *    cannot be migrated.
801 	 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
802 	 *    memory to the MOVABLE zone, the vmemmap pages are also placed in
803 	 *    such zone. Such pages cannot be really moved around as they are
804 	 *    self-stored in the range, but they are treated as movable when
805 	 *    the range they describe is about to be offlined.
806 	 *
807 	 * In general, no unmovable allocations that degrade memory offlining
808 	 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
809 	 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
810 	 * if has_unmovable_pages() states that there are no unmovable pages,
811 	 * there can be false negatives).
812 	 */
813 	ZONE_MOVABLE,
814 #ifdef CONFIG_ZONE_DEVICE
815 	ZONE_DEVICE,
816 #endif
817 	__MAX_NR_ZONES
818 
819 };
820 
821 #ifndef __GENERATING_BOUNDS_H
822 
823 #define ASYNC_AND_SYNC 2
824 
825 struct zone {
826 	/* Read-mostly fields */
827 
828 	/* zone watermarks, access with *_wmark_pages(zone) macros */
829 	unsigned long _watermark[NR_WMARK];
830 	unsigned long watermark_boost;
831 
832 	unsigned long nr_reserved_highatomic;
833 
834 	/*
835 	 * We don't know if the memory that we're going to allocate will be
836 	 * freeable or/and it will be released eventually, so to avoid totally
837 	 * wasting several GB of ram we must reserve some of the lower zone
838 	 * memory (otherwise we risk to run OOM on the lower zones despite
839 	 * there being tons of freeable ram on the higher zones).  This array is
840 	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
841 	 * changes.
842 	 */
843 	long lowmem_reserve[MAX_NR_ZONES];
844 
845 #ifdef CONFIG_NUMA
846 	int node;
847 #endif
848 	struct pglist_data	*zone_pgdat;
849 	struct per_cpu_pages	__percpu *per_cpu_pageset;
850 	struct per_cpu_zonestat	__percpu *per_cpu_zonestats;
851 	/*
852 	 * the high and batch values are copied to individual pagesets for
853 	 * faster access
854 	 */
855 	int pageset_high_min;
856 	int pageset_high_max;
857 	int pageset_batch;
858 
859 #ifndef CONFIG_SPARSEMEM
860 	/*
861 	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
862 	 * In SPARSEMEM, this map is stored in struct mem_section
863 	 */
864 	unsigned long		*pageblock_flags;
865 #endif /* CONFIG_SPARSEMEM */
866 
867 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
868 	unsigned long		zone_start_pfn;
869 
870 	/*
871 	 * spanned_pages is the total pages spanned by the zone, including
872 	 * holes, which is calculated as:
873 	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
874 	 *
875 	 * present_pages is physical pages existing within the zone, which
876 	 * is calculated as:
877 	 *	present_pages = spanned_pages - absent_pages(pages in holes);
878 	 *
879 	 * present_early_pages is present pages existing within the zone
880 	 * located on memory available since early boot, excluding hotplugged
881 	 * memory.
882 	 *
883 	 * managed_pages is present pages managed by the buddy system, which
884 	 * is calculated as (reserved_pages includes pages allocated by the
885 	 * bootmem allocator):
886 	 *	managed_pages = present_pages - reserved_pages;
887 	 *
888 	 * cma pages is present pages that are assigned for CMA use
889 	 * (MIGRATE_CMA).
890 	 *
891 	 * So present_pages may be used by memory hotplug or memory power
892 	 * management logic to figure out unmanaged pages by checking
893 	 * (present_pages - managed_pages). And managed_pages should be used
894 	 * by page allocator and vm scanner to calculate all kinds of watermarks
895 	 * and thresholds.
896 	 *
897 	 * Locking rules:
898 	 *
899 	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
900 	 * It is a seqlock because it has to be read outside of zone->lock,
901 	 * and it is done in the main allocator path.  But, it is written
902 	 * quite infrequently.
903 	 *
904 	 * The span_seq lock is declared along with zone->lock because it is
905 	 * frequently read in proximity to zone->lock.  It's good to
906 	 * give them a chance of being in the same cacheline.
907 	 *
908 	 * Write access to present_pages at runtime should be protected by
909 	 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
910 	 * present_pages should use get_online_mems() to get a stable value.
911 	 */
912 	atomic_long_t		managed_pages;
913 	unsigned long		spanned_pages;
914 	unsigned long		present_pages;
915 #if defined(CONFIG_MEMORY_HOTPLUG)
916 	unsigned long		present_early_pages;
917 #endif
918 #ifdef CONFIG_CMA
919 	unsigned long		cma_pages;
920 #endif
921 
922 	const char		*name;
923 
924 #ifdef CONFIG_MEMORY_ISOLATION
925 	/*
926 	 * Number of isolated pageblock. It is used to solve incorrect
927 	 * freepage counting problem due to racy retrieving migratetype
928 	 * of pageblock. Protected by zone->lock.
929 	 */
930 	unsigned long		nr_isolate_pageblock;
931 #endif
932 
933 #ifdef CONFIG_MEMORY_HOTPLUG
934 	/* see spanned/present_pages for more description */
935 	seqlock_t		span_seqlock;
936 #endif
937 
938 	int initialized;
939 
940 	/* Write-intensive fields used from the page allocator */
941 	CACHELINE_PADDING(_pad1_);
942 
943 	/* free areas of different sizes */
944 	struct free_area	free_area[NR_PAGE_ORDERS];
945 
946 #ifdef CONFIG_UNACCEPTED_MEMORY
947 	/* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
948 	struct list_head	unaccepted_pages;
949 #endif
950 
951 	/* zone flags, see below */
952 	unsigned long		flags;
953 
954 	/* Primarily protects free_area */
955 	spinlock_t		lock;
956 
957 	/* Write-intensive fields used by compaction and vmstats. */
958 	CACHELINE_PADDING(_pad2_);
959 
960 	/*
961 	 * When free pages are below this point, additional steps are taken
962 	 * when reading the number of free pages to avoid per-cpu counter
963 	 * drift allowing watermarks to be breached
964 	 */
965 	unsigned long percpu_drift_mark;
966 
967 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
968 	/* pfn where compaction free scanner should start */
969 	unsigned long		compact_cached_free_pfn;
970 	/* pfn where compaction migration scanner should start */
971 	unsigned long		compact_cached_migrate_pfn[ASYNC_AND_SYNC];
972 	unsigned long		compact_init_migrate_pfn;
973 	unsigned long		compact_init_free_pfn;
974 #endif
975 
976 #ifdef CONFIG_COMPACTION
977 	/*
978 	 * On compaction failure, 1<<compact_defer_shift compactions
979 	 * are skipped before trying again. The number attempted since
980 	 * last failure is tracked with compact_considered.
981 	 * compact_order_failed is the minimum compaction failed order.
982 	 */
983 	unsigned int		compact_considered;
984 	unsigned int		compact_defer_shift;
985 	int			compact_order_failed;
986 #endif
987 
988 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
989 	/* Set to true when the PG_migrate_skip bits should be cleared */
990 	bool			compact_blockskip_flush;
991 #endif
992 
993 	bool			contiguous;
994 
995 	CACHELINE_PADDING(_pad3_);
996 	/* Zone statistics */
997 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
998 	atomic_long_t		vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
999 } ____cacheline_internodealigned_in_smp;
1000 
1001 enum pgdat_flags {
1002 	PGDAT_DIRTY,			/* reclaim scanning has recently found
1003 					 * many dirty file pages at the tail
1004 					 * of the LRU.
1005 					 */
1006 	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
1007 					 * many pages under writeback
1008 					 */
1009 	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
1010 };
1011 
1012 enum zone_flags {
1013 	ZONE_BOOSTED_WATERMARK,		/* zone recently boosted watermarks.
1014 					 * Cleared when kswapd is woken.
1015 					 */
1016 	ZONE_RECLAIM_ACTIVE,		/* kswapd may be scanning the zone. */
1017 	ZONE_BELOW_HIGH,		/* zone is below high watermark. */
1018 };
1019 
zone_managed_pages(struct zone * zone)1020 static inline unsigned long zone_managed_pages(struct zone *zone)
1021 {
1022 	return (unsigned long)atomic_long_read(&zone->managed_pages);
1023 }
1024 
zone_cma_pages(struct zone * zone)1025 static inline unsigned long zone_cma_pages(struct zone *zone)
1026 {
1027 #ifdef CONFIG_CMA
1028 	return zone->cma_pages;
1029 #else
1030 	return 0;
1031 #endif
1032 }
1033 
zone_end_pfn(const struct zone * zone)1034 static inline unsigned long zone_end_pfn(const struct zone *zone)
1035 {
1036 	return zone->zone_start_pfn + zone->spanned_pages;
1037 }
1038 
zone_spans_pfn(const struct zone * zone,unsigned long pfn)1039 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1040 {
1041 	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1042 }
1043 
zone_is_initialized(struct zone * zone)1044 static inline bool zone_is_initialized(struct zone *zone)
1045 {
1046 	return zone->initialized;
1047 }
1048 
zone_is_empty(struct zone * zone)1049 static inline bool zone_is_empty(struct zone *zone)
1050 {
1051 	return zone->spanned_pages == 0;
1052 }
1053 
1054 #ifndef BUILD_VDSO32_64
1055 /*
1056  * The zone field is never updated after free_area_init_core()
1057  * sets it, so none of the operations on it need to be atomic.
1058  */
1059 
1060 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1061 #define SECTIONS_PGOFF		((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1062 #define NODES_PGOFF		(SECTIONS_PGOFF - NODES_WIDTH)
1063 #define ZONES_PGOFF		(NODES_PGOFF - ZONES_WIDTH)
1064 #define LAST_CPUPID_PGOFF	(ZONES_PGOFF - LAST_CPUPID_WIDTH)
1065 #define KASAN_TAG_PGOFF		(LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1066 #define LRU_GEN_PGOFF		(KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1067 #define LRU_REFS_PGOFF		(LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1068 
1069 /*
1070  * Define the bit shifts to access each section.  For non-existent
1071  * sections we define the shift as 0; that plus a 0 mask ensures
1072  * the compiler will optimise away reference to them.
1073  */
1074 #define SECTIONS_PGSHIFT	(SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1075 #define NODES_PGSHIFT		(NODES_PGOFF * (NODES_WIDTH != 0))
1076 #define ZONES_PGSHIFT		(ZONES_PGOFF * (ZONES_WIDTH != 0))
1077 #define LAST_CPUPID_PGSHIFT	(LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1078 #define KASAN_TAG_PGSHIFT	(KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1079 
1080 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1081 #ifdef NODE_NOT_IN_PAGE_FLAGS
1082 #define ZONEID_SHIFT		(SECTIONS_SHIFT + ZONES_SHIFT)
1083 #define ZONEID_PGOFF		((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1084 						SECTIONS_PGOFF : ZONES_PGOFF)
1085 #else
1086 #define ZONEID_SHIFT		(NODES_SHIFT + ZONES_SHIFT)
1087 #define ZONEID_PGOFF		((NODES_PGOFF < ZONES_PGOFF) ? \
1088 						NODES_PGOFF : ZONES_PGOFF)
1089 #endif
1090 
1091 #define ZONEID_PGSHIFT		(ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1092 
1093 #define ZONES_MASK		((1UL << ZONES_WIDTH) - 1)
1094 #define NODES_MASK		((1UL << NODES_WIDTH) - 1)
1095 #define SECTIONS_MASK		((1UL << SECTIONS_WIDTH) - 1)
1096 #define LAST_CPUPID_MASK	((1UL << LAST_CPUPID_SHIFT) - 1)
1097 #define KASAN_TAG_MASK		((1UL << KASAN_TAG_WIDTH) - 1)
1098 #define ZONEID_MASK		((1UL << ZONEID_SHIFT) - 1)
1099 
page_zonenum(const struct page * page)1100 static inline enum zone_type page_zonenum(const struct page *page)
1101 {
1102 	ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1103 	return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1104 }
1105 
folio_zonenum(const struct folio * folio)1106 static inline enum zone_type folio_zonenum(const struct folio *folio)
1107 {
1108 	return page_zonenum(&folio->page);
1109 }
1110 
1111 #ifdef CONFIG_ZONE_DEVICE
is_zone_device_page(const struct page * page)1112 static inline bool is_zone_device_page(const struct page *page)
1113 {
1114 	return page_zonenum(page) == ZONE_DEVICE;
1115 }
1116 
1117 /*
1118  * Consecutive zone device pages should not be merged into the same sgl
1119  * or bvec segment with other types of pages or if they belong to different
1120  * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1121  * without scanning the entire segment. This helper returns true either if
1122  * both pages are not zone device pages or both pages are zone device pages
1123  * with the same pgmap.
1124  */
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1125 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1126 						     const struct page *b)
1127 {
1128 	if (is_zone_device_page(a) != is_zone_device_page(b))
1129 		return false;
1130 	if (!is_zone_device_page(a))
1131 		return true;
1132 	return a->pgmap == b->pgmap;
1133 }
1134 
1135 extern void memmap_init_zone_device(struct zone *, unsigned long,
1136 				    unsigned long, struct dev_pagemap *);
1137 #else
is_zone_device_page(const struct page * page)1138 static inline bool is_zone_device_page(const struct page *page)
1139 {
1140 	return false;
1141 }
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1142 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1143 						     const struct page *b)
1144 {
1145 	return true;
1146 }
1147 #endif
1148 
folio_is_zone_device(const struct folio * folio)1149 static inline bool folio_is_zone_device(const struct folio *folio)
1150 {
1151 	return is_zone_device_page(&folio->page);
1152 }
1153 
is_zone_movable_page(const struct page * page)1154 static inline bool is_zone_movable_page(const struct page *page)
1155 {
1156 	return page_zonenum(page) == ZONE_MOVABLE;
1157 }
1158 
folio_is_zone_movable(const struct folio * folio)1159 static inline bool folio_is_zone_movable(const struct folio *folio)
1160 {
1161 	return folio_zonenum(folio) == ZONE_MOVABLE;
1162 }
1163 #endif
1164 
1165 /*
1166  * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1167  * intersection with the given zone
1168  */
zone_intersects(struct zone * zone,unsigned long start_pfn,unsigned long nr_pages)1169 static inline bool zone_intersects(struct zone *zone,
1170 		unsigned long start_pfn, unsigned long nr_pages)
1171 {
1172 	if (zone_is_empty(zone))
1173 		return false;
1174 	if (start_pfn >= zone_end_pfn(zone) ||
1175 	    start_pfn + nr_pages <= zone->zone_start_pfn)
1176 		return false;
1177 
1178 	return true;
1179 }
1180 
1181 /*
1182  * The "priority" of VM scanning is how much of the queues we will scan in one
1183  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1184  * queues ("queue_length >> 12") during an aging round.
1185  */
1186 #define DEF_PRIORITY 12
1187 
1188 /* Maximum number of zones on a zonelist */
1189 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1190 
1191 enum {
1192 	ZONELIST_FALLBACK,	/* zonelist with fallback */
1193 #ifdef CONFIG_NUMA
1194 	/*
1195 	 * The NUMA zonelists are doubled because we need zonelists that
1196 	 * restrict the allocations to a single node for __GFP_THISNODE.
1197 	 */
1198 	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
1199 #endif
1200 	MAX_ZONELISTS
1201 };
1202 
1203 /*
1204  * This struct contains information about a zone in a zonelist. It is stored
1205  * here to avoid dereferences into large structures and lookups of tables
1206  */
1207 struct zoneref {
1208 	struct zone *zone;	/* Pointer to actual zone */
1209 	int zone_idx;		/* zone_idx(zoneref->zone) */
1210 };
1211 
1212 /*
1213  * One allocation request operates on a zonelist. A zonelist
1214  * is a list of zones, the first one is the 'goal' of the
1215  * allocation, the other zones are fallback zones, in decreasing
1216  * priority.
1217  *
1218  * To speed the reading of the zonelist, the zonerefs contain the zone index
1219  * of the entry being read. Helper functions to access information given
1220  * a struct zoneref are
1221  *
1222  * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
1223  * zonelist_zone_idx()	- Return the index of the zone for an entry
1224  * zonelist_node_idx()	- Return the index of the node for an entry
1225  */
1226 struct zonelist {
1227 	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1228 };
1229 
1230 /*
1231  * The array of struct pages for flatmem.
1232  * It must be declared for SPARSEMEM as well because there are configurations
1233  * that rely on that.
1234  */
1235 extern struct page *mem_map;
1236 
1237 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1238 struct deferred_split {
1239 	spinlock_t split_queue_lock;
1240 	struct list_head split_queue;
1241 	unsigned long split_queue_len;
1242 };
1243 #endif
1244 
1245 #ifdef CONFIG_MEMORY_FAILURE
1246 /*
1247  * Per NUMA node memory failure handling statistics.
1248  */
1249 struct memory_failure_stats {
1250 	/*
1251 	 * Number of raw pages poisoned.
1252 	 * Cases not accounted: memory outside kernel control, offline page,
1253 	 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1254 	 * error events, and unpoison actions from hwpoison_unpoison.
1255 	 */
1256 	unsigned long total;
1257 	/*
1258 	 * Recovery results of poisoned raw pages handled by memory_failure,
1259 	 * in sync with mf_result.
1260 	 * total = ignored + failed + delayed + recovered.
1261 	 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1262 	 */
1263 	unsigned long ignored;
1264 	unsigned long failed;
1265 	unsigned long delayed;
1266 	unsigned long recovered;
1267 };
1268 #endif
1269 
1270 /*
1271  * On NUMA machines, each NUMA node would have a pg_data_t to describe
1272  * it's memory layout. On UMA machines there is a single pglist_data which
1273  * describes the whole memory.
1274  *
1275  * Memory statistics and page replacement data structures are maintained on a
1276  * per-zone basis.
1277  */
1278 typedef struct pglist_data {
1279 	/*
1280 	 * node_zones contains just the zones for THIS node. Not all of the
1281 	 * zones may be populated, but it is the full list. It is referenced by
1282 	 * this node's node_zonelists as well as other node's node_zonelists.
1283 	 */
1284 	struct zone node_zones[MAX_NR_ZONES];
1285 
1286 	/*
1287 	 * node_zonelists contains references to all zones in all nodes.
1288 	 * Generally the first zones will be references to this node's
1289 	 * node_zones.
1290 	 */
1291 	struct zonelist node_zonelists[MAX_ZONELISTS];
1292 
1293 	int nr_zones; /* number of populated zones in this node */
1294 #ifdef CONFIG_FLATMEM	/* means !SPARSEMEM */
1295 	struct page *node_mem_map;
1296 #ifdef CONFIG_PAGE_EXTENSION
1297 	struct page_ext *node_page_ext;
1298 #endif
1299 #endif
1300 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1301 	/*
1302 	 * Must be held any time you expect node_start_pfn,
1303 	 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1304 	 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1305 	 * init.
1306 	 *
1307 	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1308 	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1309 	 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1310 	 *
1311 	 * Nests above zone->lock and zone->span_seqlock
1312 	 */
1313 	spinlock_t node_size_lock;
1314 #endif
1315 	unsigned long node_start_pfn;
1316 	unsigned long node_present_pages; /* total number of physical pages */
1317 	unsigned long node_spanned_pages; /* total size of physical page
1318 					     range, including holes */
1319 	int node_id;
1320 	wait_queue_head_t kswapd_wait;
1321 	wait_queue_head_t pfmemalloc_wait;
1322 
1323 	/* workqueues for throttling reclaim for different reasons. */
1324 	wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1325 
1326 	atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1327 	unsigned long nr_reclaim_start;	/* nr pages written while throttled
1328 					 * when throttling started. */
1329 #ifdef CONFIG_MEMORY_HOTPLUG
1330 	struct mutex kswapd_lock;
1331 #endif
1332 	struct task_struct *kswapd;	/* Protected by kswapd_lock */
1333 	int kswapd_order;
1334 	enum zone_type kswapd_highest_zoneidx;
1335 
1336 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
1337 
1338 #ifdef CONFIG_COMPACTION
1339 	int kcompactd_max_order;
1340 	enum zone_type kcompactd_highest_zoneidx;
1341 	wait_queue_head_t kcompactd_wait;
1342 	struct task_struct *kcompactd;
1343 	bool proactive_compact_trigger;
1344 #endif
1345 	/*
1346 	 * This is a per-node reserve of pages that are not available
1347 	 * to userspace allocations.
1348 	 */
1349 	unsigned long		totalreserve_pages;
1350 
1351 #ifdef CONFIG_NUMA
1352 	/*
1353 	 * node reclaim becomes active if more unmapped pages exist.
1354 	 */
1355 	unsigned long		min_unmapped_pages;
1356 	unsigned long		min_slab_pages;
1357 #endif /* CONFIG_NUMA */
1358 
1359 	/* Write-intensive fields used by page reclaim */
1360 	CACHELINE_PADDING(_pad1_);
1361 
1362 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1363 	/*
1364 	 * If memory initialisation on large machines is deferred then this
1365 	 * is the first PFN that needs to be initialised.
1366 	 */
1367 	unsigned long first_deferred_pfn;
1368 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1369 
1370 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1371 	struct deferred_split deferred_split_queue;
1372 #endif
1373 
1374 #ifdef CONFIG_NUMA_BALANCING
1375 	/* start time in ms of current promote rate limit period */
1376 	unsigned int nbp_rl_start;
1377 	/* number of promote candidate pages at start time of current rate limit period */
1378 	unsigned long nbp_rl_nr_cand;
1379 	/* promote threshold in ms */
1380 	unsigned int nbp_threshold;
1381 	/* start time in ms of current promote threshold adjustment period */
1382 	unsigned int nbp_th_start;
1383 	/*
1384 	 * number of promote candidate pages at start time of current promote
1385 	 * threshold adjustment period
1386 	 */
1387 	unsigned long nbp_th_nr_cand;
1388 #endif
1389 	/* Fields commonly accessed by the page reclaim scanner */
1390 
1391 	/*
1392 	 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1393 	 *
1394 	 * Use mem_cgroup_lruvec() to look up lruvecs.
1395 	 */
1396 	struct lruvec		__lruvec;
1397 
1398 	unsigned long		flags;
1399 
1400 #ifdef CONFIG_LRU_GEN
1401 	/* kswap mm walk data */
1402 	struct lru_gen_mm_walk mm_walk;
1403 	/* lru_gen_folio list */
1404 	struct lru_gen_memcg memcg_lru;
1405 #endif
1406 
1407 	CACHELINE_PADDING(_pad2_);
1408 
1409 	/* Per-node vmstats */
1410 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1411 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
1412 #ifdef CONFIG_NUMA
1413 	struct memory_tier __rcu *memtier;
1414 #endif
1415 #ifdef CONFIG_MEMORY_FAILURE
1416 	struct memory_failure_stats mf_stats;
1417 #endif
1418 } pg_data_t;
1419 
1420 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
1421 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
1422 
1423 #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
1424 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1425 
pgdat_end_pfn(pg_data_t * pgdat)1426 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1427 {
1428 	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1429 }
1430 
1431 #include <linux/memory_hotplug.h>
1432 
1433 void build_all_zonelists(pg_data_t *pgdat);
1434 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1435 		   enum zone_type highest_zoneidx);
1436 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1437 			 int highest_zoneidx, unsigned int alloc_flags,
1438 			 long free_pages);
1439 bool zone_watermark_ok(struct zone *z, unsigned int order,
1440 		unsigned long mark, int highest_zoneidx,
1441 		unsigned int alloc_flags);
1442 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1443 		unsigned long mark, int highest_zoneidx);
1444 /*
1445  * Memory initialization context, use to differentiate memory added by
1446  * the platform statically or via memory hotplug interface.
1447  */
1448 enum meminit_context {
1449 	MEMINIT_EARLY,
1450 	MEMINIT_HOTPLUG,
1451 };
1452 
1453 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1454 				     unsigned long size);
1455 
1456 extern void lruvec_init(struct lruvec *lruvec);
1457 
lruvec_pgdat(struct lruvec * lruvec)1458 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1459 {
1460 #ifdef CONFIG_MEMCG
1461 	return lruvec->pgdat;
1462 #else
1463 	return container_of(lruvec, struct pglist_data, __lruvec);
1464 #endif
1465 }
1466 
1467 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1468 int local_memory_node(int node_id);
1469 #else
local_memory_node(int node_id)1470 static inline int local_memory_node(int node_id) { return node_id; };
1471 #endif
1472 
1473 /*
1474  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1475  */
1476 #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
1477 
1478 #ifdef CONFIG_ZONE_DEVICE
zone_is_zone_device(struct zone * zone)1479 static inline bool zone_is_zone_device(struct zone *zone)
1480 {
1481 	return zone_idx(zone) == ZONE_DEVICE;
1482 }
1483 #else
zone_is_zone_device(struct zone * zone)1484 static inline bool zone_is_zone_device(struct zone *zone)
1485 {
1486 	return false;
1487 }
1488 #endif
1489 
1490 /*
1491  * Returns true if a zone has pages managed by the buddy allocator.
1492  * All the reclaim decisions have to use this function rather than
1493  * populated_zone(). If the whole zone is reserved then we can easily
1494  * end up with populated_zone() && !managed_zone().
1495  */
managed_zone(struct zone * zone)1496 static inline bool managed_zone(struct zone *zone)
1497 {
1498 	return zone_managed_pages(zone);
1499 }
1500 
1501 /* Returns true if a zone has memory */
populated_zone(struct zone * zone)1502 static inline bool populated_zone(struct zone *zone)
1503 {
1504 	return zone->present_pages;
1505 }
1506 
1507 #ifdef CONFIG_NUMA
zone_to_nid(struct zone * zone)1508 static inline int zone_to_nid(struct zone *zone)
1509 {
1510 	return zone->node;
1511 }
1512 
zone_set_nid(struct zone * zone,int nid)1513 static inline void zone_set_nid(struct zone *zone, int nid)
1514 {
1515 	zone->node = nid;
1516 }
1517 #else
zone_to_nid(struct zone * zone)1518 static inline int zone_to_nid(struct zone *zone)
1519 {
1520 	return 0;
1521 }
1522 
zone_set_nid(struct zone * zone,int nid)1523 static inline void zone_set_nid(struct zone *zone, int nid) {}
1524 #endif
1525 
1526 extern int movable_zone;
1527 
is_highmem_idx(enum zone_type idx)1528 static inline int is_highmem_idx(enum zone_type idx)
1529 {
1530 #ifdef CONFIG_HIGHMEM
1531 	return (idx == ZONE_HIGHMEM ||
1532 		(idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1533 #else
1534 	return 0;
1535 #endif
1536 }
1537 
1538 /**
1539  * is_highmem - helper function to quickly check if a struct zone is a
1540  *              highmem zone or not.  This is an attempt to keep references
1541  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1542  * @zone: pointer to struct zone variable
1543  * Return: 1 for a highmem zone, 0 otherwise
1544  */
is_highmem(struct zone * zone)1545 static inline int is_highmem(struct zone *zone)
1546 {
1547 	return is_highmem_idx(zone_idx(zone));
1548 }
1549 
1550 #ifdef CONFIG_ZONE_DMA
1551 bool has_managed_dma(void);
1552 #else
has_managed_dma(void)1553 static inline bool has_managed_dma(void)
1554 {
1555 	return false;
1556 }
1557 #endif
1558 
1559 
1560 #ifndef CONFIG_NUMA
1561 
1562 extern struct pglist_data contig_page_data;
NODE_DATA(int nid)1563 static inline struct pglist_data *NODE_DATA(int nid)
1564 {
1565 	return &contig_page_data;
1566 }
1567 
1568 #else /* CONFIG_NUMA */
1569 
1570 #include <asm/mmzone.h>
1571 
1572 #endif /* !CONFIG_NUMA */
1573 
1574 extern struct pglist_data *first_online_pgdat(void);
1575 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1576 extern struct zone *next_zone(struct zone *zone);
1577 
1578 /**
1579  * for_each_online_pgdat - helper macro to iterate over all online nodes
1580  * @pgdat: pointer to a pg_data_t variable
1581  */
1582 #define for_each_online_pgdat(pgdat)			\
1583 	for (pgdat = first_online_pgdat();		\
1584 	     pgdat;					\
1585 	     pgdat = next_online_pgdat(pgdat))
1586 /**
1587  * for_each_zone - helper macro to iterate over all memory zones
1588  * @zone: pointer to struct zone variable
1589  *
1590  * The user only needs to declare the zone variable, for_each_zone
1591  * fills it in.
1592  */
1593 #define for_each_zone(zone)			        \
1594 	for (zone = (first_online_pgdat())->node_zones; \
1595 	     zone;					\
1596 	     zone = next_zone(zone))
1597 
1598 #define for_each_populated_zone(zone)		        \
1599 	for (zone = (first_online_pgdat())->node_zones; \
1600 	     zone;					\
1601 	     zone = next_zone(zone))			\
1602 		if (!populated_zone(zone))		\
1603 			; /* do nothing */		\
1604 		else
1605 
zonelist_zone(struct zoneref * zoneref)1606 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1607 {
1608 	return zoneref->zone;
1609 }
1610 
zonelist_zone_idx(struct zoneref * zoneref)1611 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1612 {
1613 	return zoneref->zone_idx;
1614 }
1615 
zonelist_node_idx(struct zoneref * zoneref)1616 static inline int zonelist_node_idx(struct zoneref *zoneref)
1617 {
1618 	return zone_to_nid(zoneref->zone);
1619 }
1620 
1621 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1622 					enum zone_type highest_zoneidx,
1623 					nodemask_t *nodes);
1624 
1625 /**
1626  * 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
1627  * @z: The cursor used as a starting point for the search
1628  * @highest_zoneidx: The zone index of the highest zone to return
1629  * @nodes: An optional nodemask to filter the zonelist with
1630  *
1631  * This function returns the next zone at or below a given zone index that is
1632  * within the allowed nodemask using a cursor as the starting point for the
1633  * search. The zoneref returned is a cursor that represents the current zone
1634  * being examined. It should be advanced by one before calling
1635  * next_zones_zonelist again.
1636  *
1637  * Return: the next zone at or below highest_zoneidx within the allowed
1638  * nodemask using a cursor within a zonelist as a starting point
1639  */
next_zones_zonelist(struct zoneref * z,enum zone_type highest_zoneidx,nodemask_t * nodes)1640 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1641 					enum zone_type highest_zoneidx,
1642 					nodemask_t *nodes)
1643 {
1644 	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1645 		return z;
1646 	return __next_zones_zonelist(z, highest_zoneidx, nodes);
1647 }
1648 
1649 /**
1650  * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1651  * @zonelist: The zonelist to search for a suitable zone
1652  * @highest_zoneidx: The zone index of the highest zone to return
1653  * @nodes: An optional nodemask to filter the zonelist with
1654  *
1655  * This function returns the first zone at or below a given zone index that is
1656  * within the allowed nodemask. The zoneref returned is a cursor that can be
1657  * used to iterate the zonelist with next_zones_zonelist by advancing it by
1658  * one before calling.
1659  *
1660  * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1661  * never NULL). This may happen either genuinely, or due to concurrent nodemask
1662  * update due to cpuset modification.
1663  *
1664  * Return: Zoneref pointer for the first suitable zone found
1665  */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes)1666 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1667 					enum zone_type highest_zoneidx,
1668 					nodemask_t *nodes)
1669 {
1670 	return next_zones_zonelist(zonelist->_zonerefs,
1671 							highest_zoneidx, nodes);
1672 }
1673 
1674 /**
1675  * 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
1676  * @zone: The current zone in the iterator
1677  * @z: The current pointer within zonelist->_zonerefs being iterated
1678  * @zlist: The zonelist being iterated
1679  * @highidx: The zone index of the highest zone to return
1680  * @nodemask: Nodemask allowed by the allocator
1681  *
1682  * This iterator iterates though all zones at or below a given zone index and
1683  * within a given nodemask
1684  */
1685 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1686 	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1687 		zone;							\
1688 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1689 			zone = zonelist_zone(z))
1690 
1691 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1692 	for (zone = z->zone;	\
1693 		zone;							\
1694 		z = next_zones_zonelist(++z, highidx, nodemask),	\
1695 			zone = zonelist_zone(z))
1696 
1697 
1698 /**
1699  * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1700  * @zone: The current zone in the iterator
1701  * @z: The current pointer within zonelist->zones being iterated
1702  * @zlist: The zonelist being iterated
1703  * @highidx: The zone index of the highest zone to return
1704  *
1705  * This iterator iterates though all zones at or below a given zone index.
1706  */
1707 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1708 	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1709 
1710 /* Whether the 'nodes' are all movable nodes */
movable_only_nodes(nodemask_t * nodes)1711 static inline bool movable_only_nodes(nodemask_t *nodes)
1712 {
1713 	struct zonelist *zonelist;
1714 	struct zoneref *z;
1715 	int nid;
1716 
1717 	if (nodes_empty(*nodes))
1718 		return false;
1719 
1720 	/*
1721 	 * We can chose arbitrary node from the nodemask to get a
1722 	 * zonelist as they are interlinked. We just need to find
1723 	 * at least one zone that can satisfy kernel allocations.
1724 	 */
1725 	nid = first_node(*nodes);
1726 	zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1727 	z = first_zones_zonelist(zonelist, ZONE_NORMAL,	nodes);
1728 	return (!z->zone) ? true : false;
1729 }
1730 
1731 
1732 #ifdef CONFIG_SPARSEMEM
1733 #include <asm/sparsemem.h>
1734 #endif
1735 
1736 #ifdef CONFIG_FLATMEM
1737 #define pfn_to_nid(pfn)		(0)
1738 #endif
1739 
1740 #ifdef CONFIG_SPARSEMEM
1741 
1742 /*
1743  * PA_SECTION_SHIFT		physical address to/from section number
1744  * PFN_SECTION_SHIFT		pfn to/from section number
1745  */
1746 #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1747 #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1748 
1749 #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1750 
1751 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1752 #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1753 
1754 #define SECTION_BLOCKFLAGS_BITS \
1755 	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1756 
1757 #if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1758 #error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1759 #endif
1760 
pfn_to_section_nr(unsigned long pfn)1761 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1762 {
1763 	return pfn >> PFN_SECTION_SHIFT;
1764 }
section_nr_to_pfn(unsigned long sec)1765 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1766 {
1767 	return sec << PFN_SECTION_SHIFT;
1768 }
1769 
1770 #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1771 #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1772 
1773 #define SUBSECTION_SHIFT 21
1774 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1775 
1776 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1777 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1778 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1779 
1780 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1781 #error Subsection size exceeds section size
1782 #else
1783 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1784 #endif
1785 
1786 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1787 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1788 
1789 struct mem_section_usage {
1790 	struct rcu_head rcu;
1791 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1792 	DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1793 #endif
1794 	/* See declaration of similar field in struct zone */
1795 	unsigned long pageblock_flags[0];
1796 };
1797 
1798 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1799 
1800 struct page;
1801 struct page_ext;
1802 struct mem_section {
1803 	/*
1804 	 * This is, logically, a pointer to an array of struct
1805 	 * pages.  However, it is stored with some other magic.
1806 	 * (see sparse.c::sparse_init_one_section())
1807 	 *
1808 	 * Additionally during early boot we encode node id of
1809 	 * the location of the section here to guide allocation.
1810 	 * (see sparse.c::memory_present())
1811 	 *
1812 	 * Making it a UL at least makes someone do a cast
1813 	 * before using it wrong.
1814 	 */
1815 	unsigned long section_mem_map;
1816 
1817 	struct mem_section_usage *usage;
1818 #ifdef CONFIG_PAGE_EXTENSION
1819 	/*
1820 	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1821 	 * section. (see page_ext.h about this.)
1822 	 */
1823 	struct page_ext *page_ext;
1824 	unsigned long pad;
1825 #endif
1826 	/*
1827 	 * WARNING: mem_section must be a power-of-2 in size for the
1828 	 * calculation and use of SECTION_ROOT_MASK to make sense.
1829 	 */
1830 };
1831 
1832 #ifdef CONFIG_SPARSEMEM_EXTREME
1833 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1834 #else
1835 #define SECTIONS_PER_ROOT	1
1836 #endif
1837 
1838 #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1839 #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1840 #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1841 
1842 #ifdef CONFIG_SPARSEMEM_EXTREME
1843 extern struct mem_section **mem_section;
1844 #else
1845 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1846 #endif
1847 
section_to_usemap(struct mem_section * ms)1848 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1849 {
1850 	return ms->usage->pageblock_flags;
1851 }
1852 
__nr_to_section(unsigned long nr)1853 static inline struct mem_section *__nr_to_section(unsigned long nr)
1854 {
1855 	unsigned long root = SECTION_NR_TO_ROOT(nr);
1856 
1857 	if (unlikely(root >= NR_SECTION_ROOTS))
1858 		return NULL;
1859 
1860 #ifdef CONFIG_SPARSEMEM_EXTREME
1861 	if (!mem_section || !mem_section[root])
1862 		return NULL;
1863 #endif
1864 	return &mem_section[root][nr & SECTION_ROOT_MASK];
1865 }
1866 extern size_t mem_section_usage_size(void);
1867 
1868 /*
1869  * We use the lower bits of the mem_map pointer to store
1870  * a little bit of information.  The pointer is calculated
1871  * as mem_map - section_nr_to_pfn(pnum).  The result is
1872  * aligned to the minimum alignment of the two values:
1873  *   1. All mem_map arrays are page-aligned.
1874  *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1875  *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1876  *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1877  *      worst combination is powerpc with 256k pages,
1878  *      which results in PFN_SECTION_SHIFT equal 6.
1879  * To sum it up, at least 6 bits are available on all architectures.
1880  * However, we can exceed 6 bits on some other architectures except
1881  * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1882  * with the worst case of 64K pages on arm64) if we make sure the
1883  * exceeded bit is not applicable to powerpc.
1884  */
1885 enum {
1886 	SECTION_MARKED_PRESENT_BIT,
1887 	SECTION_HAS_MEM_MAP_BIT,
1888 	SECTION_IS_ONLINE_BIT,
1889 	SECTION_IS_EARLY_BIT,
1890 #ifdef CONFIG_ZONE_DEVICE
1891 	SECTION_TAINT_ZONE_DEVICE_BIT,
1892 #endif
1893 	SECTION_MAP_LAST_BIT,
1894 };
1895 
1896 #define SECTION_MARKED_PRESENT		BIT(SECTION_MARKED_PRESENT_BIT)
1897 #define SECTION_HAS_MEM_MAP		BIT(SECTION_HAS_MEM_MAP_BIT)
1898 #define SECTION_IS_ONLINE		BIT(SECTION_IS_ONLINE_BIT)
1899 #define SECTION_IS_EARLY		BIT(SECTION_IS_EARLY_BIT)
1900 #ifdef CONFIG_ZONE_DEVICE
1901 #define SECTION_TAINT_ZONE_DEVICE	BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1902 #endif
1903 #define SECTION_MAP_MASK		(~(BIT(SECTION_MAP_LAST_BIT) - 1))
1904 #define SECTION_NID_SHIFT		SECTION_MAP_LAST_BIT
1905 
__section_mem_map_addr(struct mem_section * section)1906 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1907 {
1908 	unsigned long map = section->section_mem_map;
1909 	map &= SECTION_MAP_MASK;
1910 	return (struct page *)map;
1911 }
1912 
present_section(struct mem_section * section)1913 static inline int present_section(struct mem_section *section)
1914 {
1915 	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1916 }
1917 
present_section_nr(unsigned long nr)1918 static inline int present_section_nr(unsigned long nr)
1919 {
1920 	return present_section(__nr_to_section(nr));
1921 }
1922 
valid_section(struct mem_section * section)1923 static inline int valid_section(struct mem_section *section)
1924 {
1925 	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1926 }
1927 
early_section(struct mem_section * section)1928 static inline int early_section(struct mem_section *section)
1929 {
1930 	return (section && (section->section_mem_map & SECTION_IS_EARLY));
1931 }
1932 
valid_section_nr(unsigned long nr)1933 static inline int valid_section_nr(unsigned long nr)
1934 {
1935 	return valid_section(__nr_to_section(nr));
1936 }
1937 
online_section(struct mem_section * section)1938 static inline int online_section(struct mem_section *section)
1939 {
1940 	return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1941 }
1942 
1943 #ifdef CONFIG_ZONE_DEVICE
online_device_section(struct mem_section * section)1944 static inline int online_device_section(struct mem_section *section)
1945 {
1946 	unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1947 
1948 	return section && ((section->section_mem_map & flags) == flags);
1949 }
1950 #else
online_device_section(struct mem_section * section)1951 static inline int online_device_section(struct mem_section *section)
1952 {
1953 	return 0;
1954 }
1955 #endif
1956 
online_section_nr(unsigned long nr)1957 static inline int online_section_nr(unsigned long nr)
1958 {
1959 	return online_section(__nr_to_section(nr));
1960 }
1961 
1962 #ifdef CONFIG_MEMORY_HOTPLUG
1963 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1964 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1965 #endif
1966 
__pfn_to_section(unsigned long pfn)1967 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1968 {
1969 	return __nr_to_section(pfn_to_section_nr(pfn));
1970 }
1971 
1972 extern unsigned long __highest_present_section_nr;
1973 
subsection_map_index(unsigned long pfn)1974 static inline int subsection_map_index(unsigned long pfn)
1975 {
1976 	return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1977 }
1978 
1979 #ifdef CONFIG_SPARSEMEM_VMEMMAP
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1980 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1981 {
1982 	int idx = subsection_map_index(pfn);
1983 
1984 	return test_bit(idx, READ_ONCE(ms->usage)->subsection_map);
1985 }
1986 #else
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1987 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1988 {
1989 	return 1;
1990 }
1991 #endif
1992 
1993 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1994 /**
1995  * pfn_valid - check if there is a valid memory map entry for a PFN
1996  * @pfn: the page frame number to check
1997  *
1998  * Check if there is a valid memory map entry aka struct page for the @pfn.
1999  * Note, that availability of the memory map entry does not imply that
2000  * there is actual usable memory at that @pfn. The struct page may
2001  * represent a hole or an unusable page frame.
2002  *
2003  * Return: 1 for PFNs that have memory map entries and 0 otherwise
2004  */
pfn_valid(unsigned long pfn)2005 static inline int pfn_valid(unsigned long pfn)
2006 {
2007 	struct mem_section *ms;
2008 	int ret;
2009 
2010 	/*
2011 	 * Ensure the upper PAGE_SHIFT bits are clear in the
2012 	 * pfn. Else it might lead to false positives when
2013 	 * some of the upper bits are set, but the lower bits
2014 	 * match a valid pfn.
2015 	 */
2016 	if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2017 		return 0;
2018 
2019 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2020 		return 0;
2021 	ms = __pfn_to_section(pfn);
2022 	rcu_read_lock_sched();
2023 	if (!valid_section(ms)) {
2024 		rcu_read_unlock_sched();
2025 		return 0;
2026 	}
2027 	/*
2028 	 * Traditionally early sections always returned pfn_valid() for
2029 	 * the entire section-sized span.
2030 	 */
2031 	ret = early_section(ms) || pfn_section_valid(ms, pfn);
2032 	rcu_read_unlock_sched();
2033 
2034 	return ret;
2035 }
2036 #endif
2037 
pfn_in_present_section(unsigned long pfn)2038 static inline int pfn_in_present_section(unsigned long pfn)
2039 {
2040 	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2041 		return 0;
2042 	return present_section(__pfn_to_section(pfn));
2043 }
2044 
next_present_section_nr(unsigned long section_nr)2045 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2046 {
2047 	while (++section_nr <= __highest_present_section_nr) {
2048 		if (present_section_nr(section_nr))
2049 			return section_nr;
2050 	}
2051 
2052 	return -1;
2053 }
2054 
2055 /*
2056  * These are _only_ used during initialisation, therefore they
2057  * can use __initdata ...  They could have names to indicate
2058  * this restriction.
2059  */
2060 #ifdef CONFIG_NUMA
2061 #define pfn_to_nid(pfn)							\
2062 ({									\
2063 	unsigned long __pfn_to_nid_pfn = (pfn);				\
2064 	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
2065 })
2066 #else
2067 #define pfn_to_nid(pfn)		(0)
2068 #endif
2069 
2070 void sparse_init(void);
2071 #else
2072 #define sparse_init()	do {} while (0)
2073 #define sparse_index_init(_sec, _nid)  do {} while (0)
2074 #define pfn_in_present_section pfn_valid
2075 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2076 #endif /* CONFIG_SPARSEMEM */
2077 
2078 #endif /* !__GENERATING_BOUNDS.H */
2079 #endif /* !__ASSEMBLY__ */
2080 #endif /* _LINUX_MMZONE_H */
2081