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_YOUNG,		/* young leaf entries */
462  	MM_NONLEAF_FOUND,	/* non-leaf entries found in Bloom filters */
463  	MM_NONLEAF_ADDED,	/* non-leaf entries added to Bloom filters */
464  	NR_MM_STATS
465  };
466  
467  /* double-buffering Bloom filters */
468  #define NR_BLOOM_FILTERS	2
469  
470  struct lru_gen_mm_state {
471  	/* synced with max_seq after each iteration */
472  	unsigned long seq;
473  	/* where the current iteration continues after */
474  	struct list_head *head;
475  	/* where the last iteration ended before */
476  	struct list_head *tail;
477  	/* Bloom filters flip after each iteration */
478  	unsigned long *filters[NR_BLOOM_FILTERS];
479  	/* the mm stats for debugging */
480  	unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
481  };
482  
483  struct lru_gen_mm_walk {
484  	/* the lruvec under reclaim */
485  	struct lruvec *lruvec;
486  	/* max_seq from lru_gen_folio: can be out of date */
487  	unsigned long seq;
488  	/* the next address within an mm to scan */
489  	unsigned long next_addr;
490  	/* to batch promoted pages */
491  	int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
492  	/* to batch the mm stats */
493  	int mm_stats[NR_MM_STATS];
494  	/* total batched items */
495  	int batched;
496  	bool can_swap;
497  	bool force_scan;
498  };
499  
500  /*
501   * For each node, memcgs are divided into two generations: the old and the
502   * young. For each generation, memcgs are randomly sharded into multiple bins
503   * to improve scalability. For each bin, the hlist_nulls is virtually divided
504   * into three segments: the head, the tail and the default.
505   *
506   * An onlining memcg is added to the tail of a random bin in the old generation.
507   * The eviction starts at the head of a random bin in the old generation. The
508   * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
509   * the old generation, is incremented when all its bins become empty.
510   *
511   * There are four operations:
512   * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
513   *    current generation (old or young) and updates its "seg" to "head";
514   * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
515   *    current generation (old or young) and updates its "seg" to "tail";
516   * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
517   *    generation, updates its "gen" to "old" and resets its "seg" to "default";
518   * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
519   *    young generation, updates its "gen" to "young" and resets its "seg" to
520   *    "default".
521   *
522   * The events that trigger the above operations are:
523   * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
524   * 2. The first attempt to reclaim a memcg below low, which triggers
525   *    MEMCG_LRU_TAIL;
526   * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
527   *    threshold, which triggers MEMCG_LRU_TAIL;
528   * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
529   *    threshold, which triggers MEMCG_LRU_YOUNG;
530   * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
531   * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
532   * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
533   *
534   * Notes:
535   * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
536   *    of their max_seq counters ensures the eventual fairness to all eligible
537   *    memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
538   * 2. There are only two valid generations: old (seq) and young (seq+1).
539   *    MEMCG_NR_GENS is set to three so that when reading the generation counter
540   *    locklessly, a stale value (seq-1) does not wraparound to young.
541   */
542  #define MEMCG_NR_GENS	3
543  #define MEMCG_NR_BINS	8
544  
545  struct lru_gen_memcg {
546  	/* the per-node memcg generation counter */
547  	unsigned long seq;
548  	/* each memcg has one lru_gen_folio per node */
549  	unsigned long nr_memcgs[MEMCG_NR_GENS];
550  	/* per-node lru_gen_folio list for global reclaim */
551  	struct hlist_nulls_head	fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
552  	/* protects the above */
553  	spinlock_t lock;
554  };
555  
556  void lru_gen_init_pgdat(struct pglist_data *pgdat);
557  void lru_gen_init_lruvec(struct lruvec *lruvec);
558  bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
559  
560  void lru_gen_init_memcg(struct mem_cgroup *memcg);
561  void lru_gen_exit_memcg(struct mem_cgroup *memcg);
562  void lru_gen_online_memcg(struct mem_cgroup *memcg);
563  void lru_gen_offline_memcg(struct mem_cgroup *memcg);
564  void lru_gen_release_memcg(struct mem_cgroup *memcg);
565  void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
566  
567  #else /* !CONFIG_LRU_GEN */
568  
lru_gen_init_pgdat(struct pglist_data * pgdat)569  static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
570  {
571  }
572  
lru_gen_init_lruvec(struct lruvec * lruvec)573  static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
574  {
575  }
576  
lru_gen_look_around(struct page_vma_mapped_walk * pvmw)577  static inline bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
578  {
579  	return false;
580  }
581  
lru_gen_init_memcg(struct mem_cgroup * memcg)582  static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
583  {
584  }
585  
lru_gen_exit_memcg(struct mem_cgroup * memcg)586  static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
587  {
588  }
589  
lru_gen_online_memcg(struct mem_cgroup * memcg)590  static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
591  {
592  }
593  
lru_gen_offline_memcg(struct mem_cgroup * memcg)594  static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
595  {
596  }
597  
lru_gen_release_memcg(struct mem_cgroup * memcg)598  static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
599  {
600  }
601  
lru_gen_soft_reclaim(struct mem_cgroup * memcg,int nid)602  static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
603  {
604  }
605  
606  #endif /* CONFIG_LRU_GEN */
607  
608  struct lruvec {
609  	struct list_head		lists[NR_LRU_LISTS];
610  	/* per lruvec lru_lock for memcg */
611  	spinlock_t			lru_lock;
612  	/*
613  	 * These track the cost of reclaiming one LRU - file or anon -
614  	 * over the other. As the observed cost of reclaiming one LRU
615  	 * increases, the reclaim scan balance tips toward the other.
616  	 */
617  	unsigned long			anon_cost;
618  	unsigned long			file_cost;
619  	/* Non-resident age, driven by LRU movement */
620  	atomic_long_t			nonresident_age;
621  	/* Refaults at the time of last reclaim cycle */
622  	unsigned long			refaults[ANON_AND_FILE];
623  	/* Various lruvec state flags (enum lruvec_flags) */
624  	unsigned long			flags;
625  #ifdef CONFIG_LRU_GEN
626  	/* evictable pages divided into generations */
627  	struct lru_gen_folio		lrugen;
628  #ifdef CONFIG_LRU_GEN_WALKS_MMU
629  	/* to concurrently iterate lru_gen_mm_list */
630  	struct lru_gen_mm_state		mm_state;
631  #endif
632  #endif /* CONFIG_LRU_GEN */
633  #ifdef CONFIG_MEMCG
634  	struct pglist_data *pgdat;
635  #endif
636  	struct zswap_lruvec_state zswap_lruvec_state;
637  };
638  
639  /* Isolate for asynchronous migration */
640  #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
641  /* Isolate unevictable pages */
642  #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
643  
644  /* LRU Isolation modes. */
645  typedef unsigned __bitwise isolate_mode_t;
646  
647  enum zone_watermarks {
648  	WMARK_MIN,
649  	WMARK_LOW,
650  	WMARK_HIGH,
651  	WMARK_PROMO,
652  	NR_WMARK
653  };
654  
655  /*
656   * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. Two additional lists
657   * are added for THP. One PCP list is used by GPF_MOVABLE, and the other PCP list
658   * is used by GFP_UNMOVABLE and GFP_RECLAIMABLE.
659   */
660  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
661  #define NR_PCP_THP 2
662  #else
663  #define NR_PCP_THP 0
664  #endif
665  #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
666  #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
667  
668  /*
669   * Flags used in pcp->flags field.
670   *
671   * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
672   * previous page freeing.  To avoid to drain PCP for an accident
673   * high-order page freeing.
674   *
675   * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
676   * draining PCP for consecutive high-order pages freeing without
677   * allocation if data cache slice of CPU is large enough.  To reduce
678   * zone lock contention and keep cache-hot pages reusing.
679   */
680  #define	PCPF_PREV_FREE_HIGH_ORDER	BIT(0)
681  #define	PCPF_FREE_HIGH_BATCH		BIT(1)
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 high_min;		/* min high watermark */
688  	int high_max;		/* max high watermark */
689  	int batch;		/* chunk size for buddy add/remove */
690  	u8 flags;		/* protected by pcp->lock */
691  	u8 alloc_factor;	/* batch scaling factor during allocate */
692  #ifdef CONFIG_NUMA
693  	u8 expire;		/* When 0, remote pagesets are drained */
694  #endif
695  	short free_count;	/* consecutive free count */
696  
697  	/* Lists of pages, one per migrate type stored on the pcp-lists */
698  	struct list_head lists[NR_PCP_LISTS];
699  } ____cacheline_aligned_in_smp;
700  
701  struct per_cpu_zonestat {
702  #ifdef CONFIG_SMP
703  	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
704  	s8 stat_threshold;
705  #endif
706  #ifdef CONFIG_NUMA
707  	/*
708  	 * Low priority inaccurate counters that are only folded
709  	 * on demand. Use a large type to avoid the overhead of
710  	 * folding during refresh_cpu_vm_stats.
711  	 */
712  	unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
713  #endif
714  };
715  
716  struct per_cpu_nodestat {
717  	s8 stat_threshold;
718  	s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
719  };
720  
721  #endif /* !__GENERATING_BOUNDS.H */
722  
723  enum zone_type {
724  	/*
725  	 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
726  	 * to DMA to all of the addressable memory (ZONE_NORMAL).
727  	 * On architectures where this area covers the whole 32 bit address
728  	 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
729  	 * DMA addressing constraints. This distinction is important as a 32bit
730  	 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
731  	 * platforms may need both zones as they support peripherals with
732  	 * different DMA addressing limitations.
733  	 */
734  #ifdef CONFIG_ZONE_DMA
735  	ZONE_DMA,
736  #endif
737  #ifdef CONFIG_ZONE_DMA32
738  	ZONE_DMA32,
739  #endif
740  	/*
741  	 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
742  	 * performed on pages in ZONE_NORMAL if the DMA devices support
743  	 * transfers to all addressable memory.
744  	 */
745  	ZONE_NORMAL,
746  #ifdef CONFIG_HIGHMEM
747  	/*
748  	 * A memory area that is only addressable by the kernel through
749  	 * mapping portions into its own address space. This is for example
750  	 * used by i386 to allow the kernel to address the memory beyond
751  	 * 900MB. The kernel will set up special mappings (page
752  	 * table entries on i386) for each page that the kernel needs to
753  	 * access.
754  	 */
755  	ZONE_HIGHMEM,
756  #endif
757  	/*
758  	 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
759  	 * movable pages with few exceptional cases described below. Main use
760  	 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
761  	 * likely to succeed, and to locally limit unmovable allocations - e.g.,
762  	 * to increase the number of THP/huge pages. Notable special cases are:
763  	 *
764  	 * 1. Pinned pages: (long-term) pinning of movable pages might
765  	 *    essentially turn such pages unmovable. Therefore, we do not allow
766  	 *    pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
767  	 *    faulted, they come from the right zone right away. However, it is
768  	 *    still possible that address space already has pages in
769  	 *    ZONE_MOVABLE at the time when pages are pinned (i.e. user has
770  	 *    touches that memory before pinning). In such case we migrate them
771  	 *    to a different zone. When migration fails - pinning fails.
772  	 * 2. memblock allocations: kernelcore/movablecore setups might create
773  	 *    situations where ZONE_MOVABLE contains unmovable allocations
774  	 *    after boot. Memory offlining and allocations fail early.
775  	 * 3. Memory holes: kernelcore/movablecore setups might create very rare
776  	 *    situations where ZONE_MOVABLE contains memory holes after boot,
777  	 *    for example, if we have sections that are only partially
778  	 *    populated. Memory offlining and allocations fail early.
779  	 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
780  	 *    memory offlining, such pages cannot be allocated.
781  	 * 5. Unmovable PG_offline pages: in paravirtualized environments,
782  	 *    hotplugged memory blocks might only partially be managed by the
783  	 *    buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
784  	 *    parts not manged by the buddy are unmovable PG_offline pages. In
785  	 *    some cases (virtio-mem), such pages can be skipped during
786  	 *    memory offlining, however, cannot be moved/allocated. These
787  	 *    techniques might use alloc_contig_range() to hide previously
788  	 *    exposed pages from the buddy again (e.g., to implement some sort
789  	 *    of memory unplug in virtio-mem).
790  	 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
791  	 *    situations where ZERO_PAGE(0) which is allocated differently
792  	 *    on different platforms may end up in a movable zone. ZERO_PAGE(0)
793  	 *    cannot be migrated.
794  	 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
795  	 *    memory to the MOVABLE zone, the vmemmap pages are also placed in
796  	 *    such zone. Such pages cannot be really moved around as they are
797  	 *    self-stored in the range, but they are treated as movable when
798  	 *    the range they describe is about to be offlined.
799  	 *
800  	 * In general, no unmovable allocations that degrade memory offlining
801  	 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
802  	 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
803  	 * if has_unmovable_pages() states that there are no unmovable pages,
804  	 * there can be false negatives).
805  	 */
806  	ZONE_MOVABLE,
807  #ifdef CONFIG_ZONE_DEVICE
808  	ZONE_DEVICE,
809  #endif
810  	__MAX_NR_ZONES
811  
812  };
813  
814  #ifndef __GENERATING_BOUNDS_H
815  
816  #define ASYNC_AND_SYNC 2
817  
818  struct zone {
819  	/* Read-mostly fields */
820  
821  	/* zone watermarks, access with *_wmark_pages(zone) macros */
822  	unsigned long _watermark[NR_WMARK];
823  	unsigned long watermark_boost;
824  
825  	unsigned long nr_reserved_highatomic;
826  	unsigned long nr_free_highatomic;
827  
828  	/*
829  	 * We don't know if the memory that we're going to allocate will be
830  	 * freeable or/and it will be released eventually, so to avoid totally
831  	 * wasting several GB of ram we must reserve some of the lower zone
832  	 * memory (otherwise we risk to run OOM on the lower zones despite
833  	 * there being tons of freeable ram on the higher zones).  This array is
834  	 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
835  	 * changes.
836  	 */
837  	long lowmem_reserve[MAX_NR_ZONES];
838  
839  #ifdef CONFIG_NUMA
840  	int node;
841  #endif
842  	struct pglist_data	*zone_pgdat;
843  	struct per_cpu_pages	__percpu *per_cpu_pageset;
844  	struct per_cpu_zonestat	__percpu *per_cpu_zonestats;
845  	/*
846  	 * the high and batch values are copied to individual pagesets for
847  	 * faster access
848  	 */
849  	int pageset_high_min;
850  	int pageset_high_max;
851  	int pageset_batch;
852  
853  #ifndef CONFIG_SPARSEMEM
854  	/*
855  	 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
856  	 * In SPARSEMEM, this map is stored in struct mem_section
857  	 */
858  	unsigned long		*pageblock_flags;
859  #endif /* CONFIG_SPARSEMEM */
860  
861  	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
862  	unsigned long		zone_start_pfn;
863  
864  	/*
865  	 * spanned_pages is the total pages spanned by the zone, including
866  	 * holes, which is calculated as:
867  	 * 	spanned_pages = zone_end_pfn - zone_start_pfn;
868  	 *
869  	 * present_pages is physical pages existing within the zone, which
870  	 * is calculated as:
871  	 *	present_pages = spanned_pages - absent_pages(pages in holes);
872  	 *
873  	 * present_early_pages is present pages existing within the zone
874  	 * located on memory available since early boot, excluding hotplugged
875  	 * memory.
876  	 *
877  	 * managed_pages is present pages managed by the buddy system, which
878  	 * is calculated as (reserved_pages includes pages allocated by the
879  	 * bootmem allocator):
880  	 *	managed_pages = present_pages - reserved_pages;
881  	 *
882  	 * cma pages is present pages that are assigned for CMA use
883  	 * (MIGRATE_CMA).
884  	 *
885  	 * So present_pages may be used by memory hotplug or memory power
886  	 * management logic to figure out unmanaged pages by checking
887  	 * (present_pages - managed_pages). And managed_pages should be used
888  	 * by page allocator and vm scanner to calculate all kinds of watermarks
889  	 * and thresholds.
890  	 *
891  	 * Locking rules:
892  	 *
893  	 * zone_start_pfn and spanned_pages are protected by span_seqlock.
894  	 * It is a seqlock because it has to be read outside of zone->lock,
895  	 * and it is done in the main allocator path.  But, it is written
896  	 * quite infrequently.
897  	 *
898  	 * The span_seq lock is declared along with zone->lock because it is
899  	 * frequently read in proximity to zone->lock.  It's good to
900  	 * give them a chance of being in the same cacheline.
901  	 *
902  	 * Write access to present_pages at runtime should be protected by
903  	 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
904  	 * present_pages should use get_online_mems() to get a stable value.
905  	 */
906  	atomic_long_t		managed_pages;
907  	unsigned long		spanned_pages;
908  	unsigned long		present_pages;
909  #if defined(CONFIG_MEMORY_HOTPLUG)
910  	unsigned long		present_early_pages;
911  #endif
912  #ifdef CONFIG_CMA
913  	unsigned long		cma_pages;
914  #endif
915  
916  	const char		*name;
917  
918  #ifdef CONFIG_MEMORY_ISOLATION
919  	/*
920  	 * Number of isolated pageblock. It is used to solve incorrect
921  	 * freepage counting problem due to racy retrieving migratetype
922  	 * of pageblock. Protected by zone->lock.
923  	 */
924  	unsigned long		nr_isolate_pageblock;
925  #endif
926  
927  #ifdef CONFIG_MEMORY_HOTPLUG
928  	/* see spanned/present_pages for more description */
929  	seqlock_t		span_seqlock;
930  #endif
931  
932  	int initialized;
933  
934  	/* Write-intensive fields used from the page allocator */
935  	CACHELINE_PADDING(_pad1_);
936  
937  	/* free areas of different sizes */
938  	struct free_area	free_area[NR_PAGE_ORDERS];
939  
940  #ifdef CONFIG_UNACCEPTED_MEMORY
941  	/* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
942  	struct list_head	unaccepted_pages;
943  #endif
944  
945  	/* zone flags, see below */
946  	unsigned long		flags;
947  
948  	/* Primarily protects free_area */
949  	spinlock_t		lock;
950  
951  	/* Write-intensive fields used by compaction and vmstats. */
952  	CACHELINE_PADDING(_pad2_);
953  
954  	/*
955  	 * When free pages are below this point, additional steps are taken
956  	 * when reading the number of free pages to avoid per-cpu counter
957  	 * drift allowing watermarks to be breached
958  	 */
959  	unsigned long percpu_drift_mark;
960  
961  #if defined CONFIG_COMPACTION || defined CONFIG_CMA
962  	/* pfn where compaction free scanner should start */
963  	unsigned long		compact_cached_free_pfn;
964  	/* pfn where compaction migration scanner should start */
965  	unsigned long		compact_cached_migrate_pfn[ASYNC_AND_SYNC];
966  	unsigned long		compact_init_migrate_pfn;
967  	unsigned long		compact_init_free_pfn;
968  #endif
969  
970  #ifdef CONFIG_COMPACTION
971  	/*
972  	 * On compaction failure, 1<<compact_defer_shift compactions
973  	 * are skipped before trying again. The number attempted since
974  	 * last failure is tracked with compact_considered.
975  	 * compact_order_failed is the minimum compaction failed order.
976  	 */
977  	unsigned int		compact_considered;
978  	unsigned int		compact_defer_shift;
979  	int			compact_order_failed;
980  #endif
981  
982  #if defined CONFIG_COMPACTION || defined CONFIG_CMA
983  	/* Set to true when the PG_migrate_skip bits should be cleared */
984  	bool			compact_blockskip_flush;
985  #endif
986  
987  	bool			contiguous;
988  
989  	CACHELINE_PADDING(_pad3_);
990  	/* Zone statistics */
991  	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
992  	atomic_long_t		vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
993  } ____cacheline_internodealigned_in_smp;
994  
995  enum pgdat_flags {
996  	PGDAT_DIRTY,			/* reclaim scanning has recently found
997  					 * many dirty file pages at the tail
998  					 * of the LRU.
999  					 */
1000  	PGDAT_WRITEBACK,		/* reclaim scanning has recently found
1001  					 * many pages under writeback
1002  					 */
1003  	PGDAT_RECLAIM_LOCKED,		/* prevents concurrent reclaim */
1004  };
1005  
1006  enum zone_flags {
1007  	ZONE_BOOSTED_WATERMARK,		/* zone recently boosted watermarks.
1008  					 * Cleared when kswapd is woken.
1009  					 */
1010  	ZONE_RECLAIM_ACTIVE,		/* kswapd may be scanning the zone. */
1011  	ZONE_BELOW_HIGH,		/* zone is below high watermark. */
1012  };
1013  
wmark_pages(const struct zone * z,enum zone_watermarks w)1014  static inline unsigned long wmark_pages(const struct zone *z,
1015  					enum zone_watermarks w)
1016  {
1017  	return z->_watermark[w] + z->watermark_boost;
1018  }
1019  
min_wmark_pages(const struct zone * z)1020  static inline unsigned long min_wmark_pages(const struct zone *z)
1021  {
1022  	return wmark_pages(z, WMARK_MIN);
1023  }
1024  
low_wmark_pages(const struct zone * z)1025  static inline unsigned long low_wmark_pages(const struct zone *z)
1026  {
1027  	return wmark_pages(z, WMARK_LOW);
1028  }
1029  
high_wmark_pages(const struct zone * z)1030  static inline unsigned long high_wmark_pages(const struct zone *z)
1031  {
1032  	return wmark_pages(z, WMARK_HIGH);
1033  }
1034  
promo_wmark_pages(const struct zone * z)1035  static inline unsigned long promo_wmark_pages(const struct zone *z)
1036  {
1037  	return wmark_pages(z, WMARK_PROMO);
1038  }
1039  
zone_managed_pages(struct zone * zone)1040  static inline unsigned long zone_managed_pages(struct zone *zone)
1041  {
1042  	return (unsigned long)atomic_long_read(&zone->managed_pages);
1043  }
1044  
zone_cma_pages(struct zone * zone)1045  static inline unsigned long zone_cma_pages(struct zone *zone)
1046  {
1047  #ifdef CONFIG_CMA
1048  	return zone->cma_pages;
1049  #else
1050  	return 0;
1051  #endif
1052  }
1053  
zone_end_pfn(const struct zone * zone)1054  static inline unsigned long zone_end_pfn(const struct zone *zone)
1055  {
1056  	return zone->zone_start_pfn + zone->spanned_pages;
1057  }
1058  
zone_spans_pfn(const struct zone * zone,unsigned long pfn)1059  static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1060  {
1061  	return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1062  }
1063  
zone_is_initialized(struct zone * zone)1064  static inline bool zone_is_initialized(struct zone *zone)
1065  {
1066  	return zone->initialized;
1067  }
1068  
zone_is_empty(struct zone * zone)1069  static inline bool zone_is_empty(struct zone *zone)
1070  {
1071  	return zone->spanned_pages == 0;
1072  }
1073  
1074  #ifndef BUILD_VDSO32_64
1075  /*
1076   * The zone field is never updated after free_area_init_core()
1077   * sets it, so none of the operations on it need to be atomic.
1078   */
1079  
1080  /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1081  #define SECTIONS_PGOFF		((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1082  #define NODES_PGOFF		(SECTIONS_PGOFF - NODES_WIDTH)
1083  #define ZONES_PGOFF		(NODES_PGOFF - ZONES_WIDTH)
1084  #define LAST_CPUPID_PGOFF	(ZONES_PGOFF - LAST_CPUPID_WIDTH)
1085  #define KASAN_TAG_PGOFF		(LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1086  #define LRU_GEN_PGOFF		(KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1087  #define LRU_REFS_PGOFF		(LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1088  
1089  /*
1090   * Define the bit shifts to access each section.  For non-existent
1091   * sections we define the shift as 0; that plus a 0 mask ensures
1092   * the compiler will optimise away reference to them.
1093   */
1094  #define SECTIONS_PGSHIFT	(SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1095  #define NODES_PGSHIFT		(NODES_PGOFF * (NODES_WIDTH != 0))
1096  #define ZONES_PGSHIFT		(ZONES_PGOFF * (ZONES_WIDTH != 0))
1097  #define LAST_CPUPID_PGSHIFT	(LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1098  #define KASAN_TAG_PGSHIFT	(KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1099  
1100  /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1101  #ifdef NODE_NOT_IN_PAGE_FLAGS
1102  #define ZONEID_SHIFT		(SECTIONS_SHIFT + ZONES_SHIFT)
1103  #define ZONEID_PGOFF		((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1104  						SECTIONS_PGOFF : ZONES_PGOFF)
1105  #else
1106  #define ZONEID_SHIFT		(NODES_SHIFT + ZONES_SHIFT)
1107  #define ZONEID_PGOFF		((NODES_PGOFF < ZONES_PGOFF) ? \
1108  						NODES_PGOFF : ZONES_PGOFF)
1109  #endif
1110  
1111  #define ZONEID_PGSHIFT		(ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1112  
1113  #define ZONES_MASK		((1UL << ZONES_WIDTH) - 1)
1114  #define NODES_MASK		((1UL << NODES_WIDTH) - 1)
1115  #define SECTIONS_MASK		((1UL << SECTIONS_WIDTH) - 1)
1116  #define LAST_CPUPID_MASK	((1UL << LAST_CPUPID_SHIFT) - 1)
1117  #define KASAN_TAG_MASK		((1UL << KASAN_TAG_WIDTH) - 1)
1118  #define ZONEID_MASK		((1UL << ZONEID_SHIFT) - 1)
1119  
page_zonenum(const struct page * page)1120  static inline enum zone_type page_zonenum(const struct page *page)
1121  {
1122  	ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1123  	return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1124  }
1125  
folio_zonenum(const struct folio * folio)1126  static inline enum zone_type folio_zonenum(const struct folio *folio)
1127  {
1128  	return page_zonenum(&folio->page);
1129  }
1130  
1131  #ifdef CONFIG_ZONE_DEVICE
is_zone_device_page(const struct page * page)1132  static inline bool is_zone_device_page(const struct page *page)
1133  {
1134  	return page_zonenum(page) == ZONE_DEVICE;
1135  }
1136  
1137  /*
1138   * Consecutive zone device pages should not be merged into the same sgl
1139   * or bvec segment with other types of pages or if they belong to different
1140   * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1141   * without scanning the entire segment. This helper returns true either if
1142   * both pages are not zone device pages or both pages are zone device pages
1143   * with the same pgmap.
1144   */
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1145  static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1146  						     const struct page *b)
1147  {
1148  	if (is_zone_device_page(a) != is_zone_device_page(b))
1149  		return false;
1150  	if (!is_zone_device_page(a))
1151  		return true;
1152  	return a->pgmap == b->pgmap;
1153  }
1154  
1155  extern void memmap_init_zone_device(struct zone *, unsigned long,
1156  				    unsigned long, struct dev_pagemap *);
1157  #else
is_zone_device_page(const struct page * page)1158  static inline bool is_zone_device_page(const struct page *page)
1159  {
1160  	return false;
1161  }
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1162  static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1163  						     const struct page *b)
1164  {
1165  	return true;
1166  }
1167  #endif
1168  
folio_is_zone_device(const struct folio * folio)1169  static inline bool folio_is_zone_device(const struct folio *folio)
1170  {
1171  	return is_zone_device_page(&folio->page);
1172  }
1173  
is_zone_movable_page(const struct page * page)1174  static inline bool is_zone_movable_page(const struct page *page)
1175  {
1176  	return page_zonenum(page) == ZONE_MOVABLE;
1177  }
1178  
folio_is_zone_movable(const struct folio * folio)1179  static inline bool folio_is_zone_movable(const struct folio *folio)
1180  {
1181  	return folio_zonenum(folio) == ZONE_MOVABLE;
1182  }
1183  #endif
1184  
1185  /*
1186   * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1187   * intersection with the given zone
1188   */
zone_intersects(struct zone * zone,unsigned long start_pfn,unsigned long nr_pages)1189  static inline bool zone_intersects(struct zone *zone,
1190  		unsigned long start_pfn, unsigned long nr_pages)
1191  {
1192  	if (zone_is_empty(zone))
1193  		return false;
1194  	if (start_pfn >= zone_end_pfn(zone) ||
1195  	    start_pfn + nr_pages <= zone->zone_start_pfn)
1196  		return false;
1197  
1198  	return true;
1199  }
1200  
1201  /*
1202   * The "priority" of VM scanning is how much of the queues we will scan in one
1203   * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1204   * queues ("queue_length >> 12") during an aging round.
1205   */
1206  #define DEF_PRIORITY 12
1207  
1208  /* Maximum number of zones on a zonelist */
1209  #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1210  
1211  enum {
1212  	ZONELIST_FALLBACK,	/* zonelist with fallback */
1213  #ifdef CONFIG_NUMA
1214  	/*
1215  	 * The NUMA zonelists are doubled because we need zonelists that
1216  	 * restrict the allocations to a single node for __GFP_THISNODE.
1217  	 */
1218  	ZONELIST_NOFALLBACK,	/* zonelist without fallback (__GFP_THISNODE) */
1219  #endif
1220  	MAX_ZONELISTS
1221  };
1222  
1223  /*
1224   * This struct contains information about a zone in a zonelist. It is stored
1225   * here to avoid dereferences into large structures and lookups of tables
1226   */
1227  struct zoneref {
1228  	struct zone *zone;	/* Pointer to actual zone */
1229  	int zone_idx;		/* zone_idx(zoneref->zone) */
1230  };
1231  
1232  /*
1233   * One allocation request operates on a zonelist. A zonelist
1234   * is a list of zones, the first one is the 'goal' of the
1235   * allocation, the other zones are fallback zones, in decreasing
1236   * priority.
1237   *
1238   * To speed the reading of the zonelist, the zonerefs contain the zone index
1239   * of the entry being read. Helper functions to access information given
1240   * a struct zoneref are
1241   *
1242   * zonelist_zone()	- Return the struct zone * for an entry in _zonerefs
1243   * zonelist_zone_idx()	- Return the index of the zone for an entry
1244   * zonelist_node_idx()	- Return the index of the node for an entry
1245   */
1246  struct zonelist {
1247  	struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1248  };
1249  
1250  /*
1251   * The array of struct pages for flatmem.
1252   * It must be declared for SPARSEMEM as well because there are configurations
1253   * that rely on that.
1254   */
1255  extern struct page *mem_map;
1256  
1257  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1258  struct deferred_split {
1259  	spinlock_t split_queue_lock;
1260  	struct list_head split_queue;
1261  	unsigned long split_queue_len;
1262  };
1263  #endif
1264  
1265  #ifdef CONFIG_MEMORY_FAILURE
1266  /*
1267   * Per NUMA node memory failure handling statistics.
1268   */
1269  struct memory_failure_stats {
1270  	/*
1271  	 * Number of raw pages poisoned.
1272  	 * Cases not accounted: memory outside kernel control, offline page,
1273  	 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1274  	 * error events, and unpoison actions from hwpoison_unpoison.
1275  	 */
1276  	unsigned long total;
1277  	/*
1278  	 * Recovery results of poisoned raw pages handled by memory_failure,
1279  	 * in sync with mf_result.
1280  	 * total = ignored + failed + delayed + recovered.
1281  	 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1282  	 */
1283  	unsigned long ignored;
1284  	unsigned long failed;
1285  	unsigned long delayed;
1286  	unsigned long recovered;
1287  };
1288  #endif
1289  
1290  /*
1291   * On NUMA machines, each NUMA node would have a pg_data_t to describe
1292   * it's memory layout. On UMA machines there is a single pglist_data which
1293   * describes the whole memory.
1294   *
1295   * Memory statistics and page replacement data structures are maintained on a
1296   * per-zone basis.
1297   */
1298  typedef struct pglist_data {
1299  	/*
1300  	 * node_zones contains just the zones for THIS node. Not all of the
1301  	 * zones may be populated, but it is the full list. It is referenced by
1302  	 * this node's node_zonelists as well as other node's node_zonelists.
1303  	 */
1304  	struct zone node_zones[MAX_NR_ZONES];
1305  
1306  	/*
1307  	 * node_zonelists contains references to all zones in all nodes.
1308  	 * Generally the first zones will be references to this node's
1309  	 * node_zones.
1310  	 */
1311  	struct zonelist node_zonelists[MAX_ZONELISTS];
1312  
1313  	int nr_zones; /* number of populated zones in this node */
1314  #ifdef CONFIG_FLATMEM	/* means !SPARSEMEM */
1315  	struct page *node_mem_map;
1316  #ifdef CONFIG_PAGE_EXTENSION
1317  	struct page_ext *node_page_ext;
1318  #endif
1319  #endif
1320  #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1321  	/*
1322  	 * Must be held any time you expect node_start_pfn,
1323  	 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1324  	 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1325  	 * init.
1326  	 *
1327  	 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1328  	 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1329  	 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1330  	 *
1331  	 * Nests above zone->lock and zone->span_seqlock
1332  	 */
1333  	spinlock_t node_size_lock;
1334  #endif
1335  	unsigned long node_start_pfn;
1336  	unsigned long node_present_pages; /* total number of physical pages */
1337  	unsigned long node_spanned_pages; /* total size of physical page
1338  					     range, including holes */
1339  	int node_id;
1340  	wait_queue_head_t kswapd_wait;
1341  	wait_queue_head_t pfmemalloc_wait;
1342  
1343  	/* workqueues for throttling reclaim for different reasons. */
1344  	wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1345  
1346  	atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1347  	unsigned long nr_reclaim_start;	/* nr pages written while throttled
1348  					 * when throttling started. */
1349  #ifdef CONFIG_MEMORY_HOTPLUG
1350  	struct mutex kswapd_lock;
1351  #endif
1352  	struct task_struct *kswapd;	/* Protected by kswapd_lock */
1353  	int kswapd_order;
1354  	enum zone_type kswapd_highest_zoneidx;
1355  
1356  	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
1357  
1358  #ifdef CONFIG_COMPACTION
1359  	int kcompactd_max_order;
1360  	enum zone_type kcompactd_highest_zoneidx;
1361  	wait_queue_head_t kcompactd_wait;
1362  	struct task_struct *kcompactd;
1363  	bool proactive_compact_trigger;
1364  #endif
1365  	/*
1366  	 * This is a per-node reserve of pages that are not available
1367  	 * to userspace allocations.
1368  	 */
1369  	unsigned long		totalreserve_pages;
1370  
1371  #ifdef CONFIG_NUMA
1372  	/*
1373  	 * node reclaim becomes active if more unmapped pages exist.
1374  	 */
1375  	unsigned long		min_unmapped_pages;
1376  	unsigned long		min_slab_pages;
1377  #endif /* CONFIG_NUMA */
1378  
1379  	/* Write-intensive fields used by page reclaim */
1380  	CACHELINE_PADDING(_pad1_);
1381  
1382  #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1383  	/*
1384  	 * If memory initialisation on large machines is deferred then this
1385  	 * is the first PFN that needs to be initialised.
1386  	 */
1387  	unsigned long first_deferred_pfn;
1388  #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1389  
1390  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1391  	struct deferred_split deferred_split_queue;
1392  #endif
1393  
1394  #ifdef CONFIG_NUMA_BALANCING
1395  	/* start time in ms of current promote rate limit period */
1396  	unsigned int nbp_rl_start;
1397  	/* number of promote candidate pages at start time of current rate limit period */
1398  	unsigned long nbp_rl_nr_cand;
1399  	/* promote threshold in ms */
1400  	unsigned int nbp_threshold;
1401  	/* start time in ms of current promote threshold adjustment period */
1402  	unsigned int nbp_th_start;
1403  	/*
1404  	 * number of promote candidate pages at start time of current promote
1405  	 * threshold adjustment period
1406  	 */
1407  	unsigned long nbp_th_nr_cand;
1408  #endif
1409  	/* Fields commonly accessed by the page reclaim scanner */
1410  
1411  	/*
1412  	 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1413  	 *
1414  	 * Use mem_cgroup_lruvec() to look up lruvecs.
1415  	 */
1416  	struct lruvec		__lruvec;
1417  
1418  	unsigned long		flags;
1419  
1420  #ifdef CONFIG_LRU_GEN
1421  	/* kswap mm walk data */
1422  	struct lru_gen_mm_walk mm_walk;
1423  	/* lru_gen_folio list */
1424  	struct lru_gen_memcg memcg_lru;
1425  #endif
1426  
1427  	CACHELINE_PADDING(_pad2_);
1428  
1429  	/* Per-node vmstats */
1430  	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1431  	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
1432  #ifdef CONFIG_NUMA
1433  	struct memory_tier __rcu *memtier;
1434  #endif
1435  #ifdef CONFIG_MEMORY_FAILURE
1436  	struct memory_failure_stats mf_stats;
1437  #endif
1438  } pg_data_t;
1439  
1440  #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
1441  #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
1442  
1443  #define node_start_pfn(nid)	(NODE_DATA(nid)->node_start_pfn)
1444  #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1445  
pgdat_end_pfn(pg_data_t * pgdat)1446  static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1447  {
1448  	return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1449  }
1450  
1451  #include <linux/memory_hotplug.h>
1452  
1453  void build_all_zonelists(pg_data_t *pgdat);
1454  void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1455  		   enum zone_type highest_zoneidx);
1456  bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1457  			 int highest_zoneidx, unsigned int alloc_flags,
1458  			 long free_pages);
1459  bool zone_watermark_ok(struct zone *z, unsigned int order,
1460  		unsigned long mark, int highest_zoneidx,
1461  		unsigned int alloc_flags);
1462  bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1463  		unsigned long mark, int highest_zoneidx);
1464  /*
1465   * Memory initialization context, use to differentiate memory added by
1466   * the platform statically or via memory hotplug interface.
1467   */
1468  enum meminit_context {
1469  	MEMINIT_EARLY,
1470  	MEMINIT_HOTPLUG,
1471  };
1472  
1473  extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1474  				     unsigned long size);
1475  
1476  extern void lruvec_init(struct lruvec *lruvec);
1477  
lruvec_pgdat(struct lruvec * lruvec)1478  static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1479  {
1480  #ifdef CONFIG_MEMCG
1481  	return lruvec->pgdat;
1482  #else
1483  	return container_of(lruvec, struct pglist_data, __lruvec);
1484  #endif
1485  }
1486  
1487  #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1488  int local_memory_node(int node_id);
1489  #else
local_memory_node(int node_id)1490  static inline int local_memory_node(int node_id) { return node_id; };
1491  #endif
1492  
1493  /*
1494   * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1495   */
1496  #define zone_idx(zone)		((zone) - (zone)->zone_pgdat->node_zones)
1497  
1498  #ifdef CONFIG_ZONE_DEVICE
zone_is_zone_device(struct zone * zone)1499  static inline bool zone_is_zone_device(struct zone *zone)
1500  {
1501  	return zone_idx(zone) == ZONE_DEVICE;
1502  }
1503  #else
zone_is_zone_device(struct zone * zone)1504  static inline bool zone_is_zone_device(struct zone *zone)
1505  {
1506  	return false;
1507  }
1508  #endif
1509  
1510  /*
1511   * Returns true if a zone has pages managed by the buddy allocator.
1512   * All the reclaim decisions have to use this function rather than
1513   * populated_zone(). If the whole zone is reserved then we can easily
1514   * end up with populated_zone() && !managed_zone().
1515   */
managed_zone(struct zone * zone)1516  static inline bool managed_zone(struct zone *zone)
1517  {
1518  	return zone_managed_pages(zone);
1519  }
1520  
1521  /* Returns true if a zone has memory */
populated_zone(struct zone * zone)1522  static inline bool populated_zone(struct zone *zone)
1523  {
1524  	return zone->present_pages;
1525  }
1526  
1527  #ifdef CONFIG_NUMA
zone_to_nid(struct zone * zone)1528  static inline int zone_to_nid(struct zone *zone)
1529  {
1530  	return zone->node;
1531  }
1532  
zone_set_nid(struct zone * zone,int nid)1533  static inline void zone_set_nid(struct zone *zone, int nid)
1534  {
1535  	zone->node = nid;
1536  }
1537  #else
zone_to_nid(struct zone * zone)1538  static inline int zone_to_nid(struct zone *zone)
1539  {
1540  	return 0;
1541  }
1542  
zone_set_nid(struct zone * zone,int nid)1543  static inline void zone_set_nid(struct zone *zone, int nid) {}
1544  #endif
1545  
1546  extern int movable_zone;
1547  
is_highmem_idx(enum zone_type idx)1548  static inline int is_highmem_idx(enum zone_type idx)
1549  {
1550  #ifdef CONFIG_HIGHMEM
1551  	return (idx == ZONE_HIGHMEM ||
1552  		(idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1553  #else
1554  	return 0;
1555  #endif
1556  }
1557  
1558  /**
1559   * is_highmem - helper function to quickly check if a struct zone is a
1560   *              highmem zone or not.  This is an attempt to keep references
1561   *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1562   * @zone: pointer to struct zone variable
1563   * Return: 1 for a highmem zone, 0 otherwise
1564   */
is_highmem(struct zone * zone)1565  static inline int is_highmem(struct zone *zone)
1566  {
1567  	return is_highmem_idx(zone_idx(zone));
1568  }
1569  
1570  #ifdef CONFIG_ZONE_DMA
1571  bool has_managed_dma(void);
1572  #else
has_managed_dma(void)1573  static inline bool has_managed_dma(void)
1574  {
1575  	return false;
1576  }
1577  #endif
1578  
1579  
1580  #ifndef CONFIG_NUMA
1581  
1582  extern struct pglist_data contig_page_data;
NODE_DATA(int nid)1583  static inline struct pglist_data *NODE_DATA(int nid)
1584  {
1585  	return &contig_page_data;
1586  }
1587  
1588  #else /* CONFIG_NUMA */
1589  
1590  #include <asm/mmzone.h>
1591  
1592  #endif /* !CONFIG_NUMA */
1593  
1594  extern struct pglist_data *first_online_pgdat(void);
1595  extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1596  extern struct zone *next_zone(struct zone *zone);
1597  
1598  /**
1599   * for_each_online_pgdat - helper macro to iterate over all online nodes
1600   * @pgdat: pointer to a pg_data_t variable
1601   */
1602  #define for_each_online_pgdat(pgdat)			\
1603  	for (pgdat = first_online_pgdat();		\
1604  	     pgdat;					\
1605  	     pgdat = next_online_pgdat(pgdat))
1606  /**
1607   * for_each_zone - helper macro to iterate over all memory zones
1608   * @zone: pointer to struct zone variable
1609   *
1610   * The user only needs to declare the zone variable, for_each_zone
1611   * fills it in.
1612   */
1613  #define for_each_zone(zone)			        \
1614  	for (zone = (first_online_pgdat())->node_zones; \
1615  	     zone;					\
1616  	     zone = next_zone(zone))
1617  
1618  #define for_each_populated_zone(zone)		        \
1619  	for (zone = (first_online_pgdat())->node_zones; \
1620  	     zone;					\
1621  	     zone = next_zone(zone))			\
1622  		if (!populated_zone(zone))		\
1623  			; /* do nothing */		\
1624  		else
1625  
zonelist_zone(struct zoneref * zoneref)1626  static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1627  {
1628  	return zoneref->zone;
1629  }
1630  
zonelist_zone_idx(struct zoneref * zoneref)1631  static inline int zonelist_zone_idx(struct zoneref *zoneref)
1632  {
1633  	return zoneref->zone_idx;
1634  }
1635  
zonelist_node_idx(struct zoneref * zoneref)1636  static inline int zonelist_node_idx(struct zoneref *zoneref)
1637  {
1638  	return zone_to_nid(zoneref->zone);
1639  }
1640  
1641  struct zoneref *__next_zones_zonelist(struct zoneref *z,
1642  					enum zone_type highest_zoneidx,
1643  					nodemask_t *nodes);
1644  
1645  /**
1646   * 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
1647   * @z: The cursor used as a starting point for the search
1648   * @highest_zoneidx: The zone index of the highest zone to return
1649   * @nodes: An optional nodemask to filter the zonelist with
1650   *
1651   * This function returns the next zone at or below a given zone index that is
1652   * within the allowed nodemask using a cursor as the starting point for the
1653   * search. The zoneref returned is a cursor that represents the current zone
1654   * being examined. It should be advanced by one before calling
1655   * next_zones_zonelist again.
1656   *
1657   * Return: the next zone at or below highest_zoneidx within the allowed
1658   * nodemask using a cursor within a zonelist as a starting point
1659   */
next_zones_zonelist(struct zoneref * z,enum zone_type highest_zoneidx,nodemask_t * nodes)1660  static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1661  					enum zone_type highest_zoneidx,
1662  					nodemask_t *nodes)
1663  {
1664  	if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1665  		return z;
1666  	return __next_zones_zonelist(z, highest_zoneidx, nodes);
1667  }
1668  
1669  /**
1670   * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1671   * @zonelist: The zonelist to search for a suitable zone
1672   * @highest_zoneidx: The zone index of the highest zone to return
1673   * @nodes: An optional nodemask to filter the zonelist with
1674   *
1675   * This function returns the first zone at or below a given zone index that is
1676   * within the allowed nodemask. The zoneref returned is a cursor that can be
1677   * used to iterate the zonelist with next_zones_zonelist by advancing it by
1678   * one before calling.
1679   *
1680   * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1681   * never NULL). This may happen either genuinely, or due to concurrent nodemask
1682   * update due to cpuset modification.
1683   *
1684   * Return: Zoneref pointer for the first suitable zone found
1685   */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes)1686  static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1687  					enum zone_type highest_zoneidx,
1688  					nodemask_t *nodes)
1689  {
1690  	return next_zones_zonelist(zonelist->_zonerefs,
1691  							highest_zoneidx, nodes);
1692  }
1693  
1694  /**
1695   * 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
1696   * @zone: The current zone in the iterator
1697   * @z: The current pointer within zonelist->_zonerefs being iterated
1698   * @zlist: The zonelist being iterated
1699   * @highidx: The zone index of the highest zone to return
1700   * @nodemask: Nodemask allowed by the allocator
1701   *
1702   * This iterator iterates though all zones at or below a given zone index and
1703   * within a given nodemask
1704   */
1705  #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1706  	for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);	\
1707  		zone;							\
1708  		z = next_zones_zonelist(++z, highidx, nodemask),	\
1709  			zone = zonelist_zone(z))
1710  
1711  #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1712  	for (zone = zonelist_zone(z);	\
1713  		zone;							\
1714  		z = next_zones_zonelist(++z, highidx, nodemask),	\
1715  			zone = zonelist_zone(z))
1716  
1717  
1718  /**
1719   * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1720   * @zone: The current zone in the iterator
1721   * @z: The current pointer within zonelist->zones being iterated
1722   * @zlist: The zonelist being iterated
1723   * @highidx: The zone index of the highest zone to return
1724   *
1725   * This iterator iterates though all zones at or below a given zone index.
1726   */
1727  #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1728  	for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1729  
1730  /* Whether the 'nodes' are all movable nodes */
movable_only_nodes(nodemask_t * nodes)1731  static inline bool movable_only_nodes(nodemask_t *nodes)
1732  {
1733  	struct zonelist *zonelist;
1734  	struct zoneref *z;
1735  	int nid;
1736  
1737  	if (nodes_empty(*nodes))
1738  		return false;
1739  
1740  	/*
1741  	 * We can chose arbitrary node from the nodemask to get a
1742  	 * zonelist as they are interlinked. We just need to find
1743  	 * at least one zone that can satisfy kernel allocations.
1744  	 */
1745  	nid = first_node(*nodes);
1746  	zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1747  	z = first_zones_zonelist(zonelist, ZONE_NORMAL,	nodes);
1748  	return (!zonelist_zone(z)) ? true : false;
1749  }
1750  
1751  
1752  #ifdef CONFIG_SPARSEMEM
1753  #include <asm/sparsemem.h>
1754  #endif
1755  
1756  #ifdef CONFIG_FLATMEM
1757  #define pfn_to_nid(pfn)		(0)
1758  #endif
1759  
1760  #ifdef CONFIG_SPARSEMEM
1761  
1762  /*
1763   * PA_SECTION_SHIFT		physical address to/from section number
1764   * PFN_SECTION_SHIFT		pfn to/from section number
1765   */
1766  #define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
1767  #define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
1768  
1769  #define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
1770  
1771  #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1772  #define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
1773  
1774  #define SECTION_BLOCKFLAGS_BITS \
1775  	((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1776  
1777  #if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1778  #error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1779  #endif
1780  
pfn_to_section_nr(unsigned long pfn)1781  static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1782  {
1783  	return pfn >> PFN_SECTION_SHIFT;
1784  }
section_nr_to_pfn(unsigned long sec)1785  static inline unsigned long section_nr_to_pfn(unsigned long sec)
1786  {
1787  	return sec << PFN_SECTION_SHIFT;
1788  }
1789  
1790  #define SECTION_ALIGN_UP(pfn)	(((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1791  #define SECTION_ALIGN_DOWN(pfn)	((pfn) & PAGE_SECTION_MASK)
1792  
1793  #define SUBSECTION_SHIFT 21
1794  #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1795  
1796  #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1797  #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1798  #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1799  
1800  #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1801  #error Subsection size exceeds section size
1802  #else
1803  #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1804  #endif
1805  
1806  #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1807  #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1808  
1809  struct mem_section_usage {
1810  	struct rcu_head rcu;
1811  #ifdef CONFIG_SPARSEMEM_VMEMMAP
1812  	DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1813  #endif
1814  	/* See declaration of similar field in struct zone */
1815  	unsigned long pageblock_flags[0];
1816  };
1817  
1818  void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1819  
1820  struct page;
1821  struct page_ext;
1822  struct mem_section {
1823  	/*
1824  	 * This is, logically, a pointer to an array of struct
1825  	 * pages.  However, it is stored with some other magic.
1826  	 * (see sparse.c::sparse_init_one_section())
1827  	 *
1828  	 * Additionally during early boot we encode node id of
1829  	 * the location of the section here to guide allocation.
1830  	 * (see sparse.c::memory_present())
1831  	 *
1832  	 * Making it a UL at least makes someone do a cast
1833  	 * before using it wrong.
1834  	 */
1835  	unsigned long section_mem_map;
1836  
1837  	struct mem_section_usage *usage;
1838  #ifdef CONFIG_PAGE_EXTENSION
1839  	/*
1840  	 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1841  	 * section. (see page_ext.h about this.)
1842  	 */
1843  	struct page_ext *page_ext;
1844  	unsigned long pad;
1845  #endif
1846  	/*
1847  	 * WARNING: mem_section must be a power-of-2 in size for the
1848  	 * calculation and use of SECTION_ROOT_MASK to make sense.
1849  	 */
1850  };
1851  
1852  #ifdef CONFIG_SPARSEMEM_EXTREME
1853  #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1854  #else
1855  #define SECTIONS_PER_ROOT	1
1856  #endif
1857  
1858  #define SECTION_NR_TO_ROOT(sec)	((sec) / SECTIONS_PER_ROOT)
1859  #define NR_SECTION_ROOTS	DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1860  #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
1861  
1862  #ifdef CONFIG_SPARSEMEM_EXTREME
1863  extern struct mem_section **mem_section;
1864  #else
1865  extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1866  #endif
1867  
section_to_usemap(struct mem_section * ms)1868  static inline unsigned long *section_to_usemap(struct mem_section *ms)
1869  {
1870  	return ms->usage->pageblock_flags;
1871  }
1872  
__nr_to_section(unsigned long nr)1873  static inline struct mem_section *__nr_to_section(unsigned long nr)
1874  {
1875  	unsigned long root = SECTION_NR_TO_ROOT(nr);
1876  
1877  	if (unlikely(root >= NR_SECTION_ROOTS))
1878  		return NULL;
1879  
1880  #ifdef CONFIG_SPARSEMEM_EXTREME
1881  	if (!mem_section || !mem_section[root])
1882  		return NULL;
1883  #endif
1884  	return &mem_section[root][nr & SECTION_ROOT_MASK];
1885  }
1886  extern size_t mem_section_usage_size(void);
1887  
1888  /*
1889   * We use the lower bits of the mem_map pointer to store
1890   * a little bit of information.  The pointer is calculated
1891   * as mem_map - section_nr_to_pfn(pnum).  The result is
1892   * aligned to the minimum alignment of the two values:
1893   *   1. All mem_map arrays are page-aligned.
1894   *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1895   *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1896   *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1897   *      worst combination is powerpc with 256k pages,
1898   *      which results in PFN_SECTION_SHIFT equal 6.
1899   * To sum it up, at least 6 bits are available on all architectures.
1900   * However, we can exceed 6 bits on some other architectures except
1901   * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1902   * with the worst case of 64K pages on arm64) if we make sure the
1903   * exceeded bit is not applicable to powerpc.
1904   */
1905  enum {
1906  	SECTION_MARKED_PRESENT_BIT,
1907  	SECTION_HAS_MEM_MAP_BIT,
1908  	SECTION_IS_ONLINE_BIT,
1909  	SECTION_IS_EARLY_BIT,
1910  #ifdef CONFIG_ZONE_DEVICE
1911  	SECTION_TAINT_ZONE_DEVICE_BIT,
1912  #endif
1913  	SECTION_MAP_LAST_BIT,
1914  };
1915  
1916  #define SECTION_MARKED_PRESENT		BIT(SECTION_MARKED_PRESENT_BIT)
1917  #define SECTION_HAS_MEM_MAP		BIT(SECTION_HAS_MEM_MAP_BIT)
1918  #define SECTION_IS_ONLINE		BIT(SECTION_IS_ONLINE_BIT)
1919  #define SECTION_IS_EARLY		BIT(SECTION_IS_EARLY_BIT)
1920  #ifdef CONFIG_ZONE_DEVICE
1921  #define SECTION_TAINT_ZONE_DEVICE	BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1922  #endif
1923  #define SECTION_MAP_MASK		(~(BIT(SECTION_MAP_LAST_BIT) - 1))
1924  #define SECTION_NID_SHIFT		SECTION_MAP_LAST_BIT
1925  
__section_mem_map_addr(struct mem_section * section)1926  static inline struct page *__section_mem_map_addr(struct mem_section *section)
1927  {
1928  	unsigned long map = section->section_mem_map;
1929  	map &= SECTION_MAP_MASK;
1930  	return (struct page *)map;
1931  }
1932  
present_section(struct mem_section * section)1933  static inline int present_section(struct mem_section *section)
1934  {
1935  	return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1936  }
1937  
present_section_nr(unsigned long nr)1938  static inline int present_section_nr(unsigned long nr)
1939  {
1940  	return present_section(__nr_to_section(nr));
1941  }
1942  
valid_section(struct mem_section * section)1943  static inline int valid_section(struct mem_section *section)
1944  {
1945  	return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1946  }
1947  
early_section(struct mem_section * section)1948  static inline int early_section(struct mem_section *section)
1949  {
1950  	return (section && (section->section_mem_map & SECTION_IS_EARLY));
1951  }
1952  
valid_section_nr(unsigned long nr)1953  static inline int valid_section_nr(unsigned long nr)
1954  {
1955  	return valid_section(__nr_to_section(nr));
1956  }
1957  
online_section(struct mem_section * section)1958  static inline int online_section(struct mem_section *section)
1959  {
1960  	return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1961  }
1962  
1963  #ifdef CONFIG_ZONE_DEVICE
online_device_section(struct mem_section * section)1964  static inline int online_device_section(struct mem_section *section)
1965  {
1966  	unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1967  
1968  	return section && ((section->section_mem_map & flags) == flags);
1969  }
1970  #else
online_device_section(struct mem_section * section)1971  static inline int online_device_section(struct mem_section *section)
1972  {
1973  	return 0;
1974  }
1975  #endif
1976  
online_section_nr(unsigned long nr)1977  static inline int online_section_nr(unsigned long nr)
1978  {
1979  	return online_section(__nr_to_section(nr));
1980  }
1981  
1982  #ifdef CONFIG_MEMORY_HOTPLUG
1983  void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1984  void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1985  #endif
1986  
__pfn_to_section(unsigned long pfn)1987  static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1988  {
1989  	return __nr_to_section(pfn_to_section_nr(pfn));
1990  }
1991  
1992  extern unsigned long __highest_present_section_nr;
1993  
subsection_map_index(unsigned long pfn)1994  static inline int subsection_map_index(unsigned long pfn)
1995  {
1996  	return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1997  }
1998  
1999  #ifdef CONFIG_SPARSEMEM_VMEMMAP
pfn_section_valid(struct mem_section * ms,unsigned long pfn)2000  static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
2001  {
2002  	int idx = subsection_map_index(pfn);
2003  	struct mem_section_usage *usage = READ_ONCE(ms->usage);
2004  
2005  	return usage ? test_bit(idx, usage->subsection_map) : 0;
2006  }
2007  #else
pfn_section_valid(struct mem_section * ms,unsigned long pfn)2008  static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
2009  {
2010  	return 1;
2011  }
2012  #endif
2013  
2014  #ifndef CONFIG_HAVE_ARCH_PFN_VALID
2015  /**
2016   * pfn_valid - check if there is a valid memory map entry for a PFN
2017   * @pfn: the page frame number to check
2018   *
2019   * Check if there is a valid memory map entry aka struct page for the @pfn.
2020   * Note, that availability of the memory map entry does not imply that
2021   * there is actual usable memory at that @pfn. The struct page may
2022   * represent a hole or an unusable page frame.
2023   *
2024   * Return: 1 for PFNs that have memory map entries and 0 otherwise
2025   */
pfn_valid(unsigned long pfn)2026  static inline int pfn_valid(unsigned long pfn)
2027  {
2028  	struct mem_section *ms;
2029  	int ret;
2030  
2031  	/*
2032  	 * Ensure the upper PAGE_SHIFT bits are clear in the
2033  	 * pfn. Else it might lead to false positives when
2034  	 * some of the upper bits are set, but the lower bits
2035  	 * match a valid pfn.
2036  	 */
2037  	if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2038  		return 0;
2039  
2040  	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2041  		return 0;
2042  	ms = __pfn_to_section(pfn);
2043  	rcu_read_lock_sched();
2044  	if (!valid_section(ms)) {
2045  		rcu_read_unlock_sched();
2046  		return 0;
2047  	}
2048  	/*
2049  	 * Traditionally early sections always returned pfn_valid() for
2050  	 * the entire section-sized span.
2051  	 */
2052  	ret = early_section(ms) || pfn_section_valid(ms, pfn);
2053  	rcu_read_unlock_sched();
2054  
2055  	return ret;
2056  }
2057  #endif
2058  
pfn_in_present_section(unsigned long pfn)2059  static inline int pfn_in_present_section(unsigned long pfn)
2060  {
2061  	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2062  		return 0;
2063  	return present_section(__pfn_to_section(pfn));
2064  }
2065  
next_present_section_nr(unsigned long section_nr)2066  static inline unsigned long next_present_section_nr(unsigned long section_nr)
2067  {
2068  	while (++section_nr <= __highest_present_section_nr) {
2069  		if (present_section_nr(section_nr))
2070  			return section_nr;
2071  	}
2072  
2073  	return -1;
2074  }
2075  
2076  /*
2077   * These are _only_ used during initialisation, therefore they
2078   * can use __initdata ...  They could have names to indicate
2079   * this restriction.
2080   */
2081  #ifdef CONFIG_NUMA
2082  #define pfn_to_nid(pfn)							\
2083  ({									\
2084  	unsigned long __pfn_to_nid_pfn = (pfn);				\
2085  	page_to_nid(pfn_to_page(__pfn_to_nid_pfn));			\
2086  })
2087  #else
2088  #define pfn_to_nid(pfn)		(0)
2089  #endif
2090  
2091  void sparse_init(void);
2092  #else
2093  #define sparse_init()	do {} while (0)
2094  #define sparse_index_init(_sec, _nid)  do {} while (0)
2095  #define pfn_in_present_section pfn_valid
2096  #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2097  #endif /* CONFIG_SPARSEMEM */
2098  
2099  #endif /* !__GENERATING_BOUNDS.H */
2100  #endif /* !__ASSEMBLY__ */
2101  #endif /* _LINUX_MMZONE_H */
2102