Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 <linux/sizes.h>
27#include <asm/page.h>
28
29/* Free memory management - zoned buddy allocator. */
30#ifndef CONFIG_ARCH_FORCE_MAX_ORDER
31#define MAX_PAGE_ORDER 10
32#else
33#define MAX_PAGE_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
34#endif
35#define MAX_ORDER_NR_PAGES (1 << MAX_PAGE_ORDER)
36
37#define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
38
39#define NR_PAGE_ORDERS (MAX_PAGE_ORDER + 1)
40
41/* Defines the order for the number of pages that have a migrate type. */
42#ifndef CONFIG_PAGE_BLOCK_MAX_ORDER
43#define PAGE_BLOCK_MAX_ORDER MAX_PAGE_ORDER
44#else
45#define PAGE_BLOCK_MAX_ORDER CONFIG_PAGE_BLOCK_MAX_ORDER
46#endif /* CONFIG_PAGE_BLOCK_MAX_ORDER */
47
48/*
49 * The MAX_PAGE_ORDER, which defines the max order of pages to be allocated
50 * by the buddy allocator, has to be larger or equal to the PAGE_BLOCK_MAX_ORDER,
51 * which defines the order for the number of pages that can have a migrate type
52 */
53#if (PAGE_BLOCK_MAX_ORDER > MAX_PAGE_ORDER)
54#error MAX_PAGE_ORDER must be >= PAGE_BLOCK_MAX_ORDER
55#endif
56
57/*
58 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
59 * costly to service. That is between allocation orders which should
60 * coalesce naturally under reasonable reclaim pressure and those which
61 * will not.
62 */
63#define PAGE_ALLOC_COSTLY_ORDER 3
64
65#if !defined(CONFIG_HAVE_GIGANTIC_FOLIOS)
66/*
67 * We don't expect any folios that exceed buddy sizes (and consequently
68 * memory sections).
69 */
70#define MAX_FOLIO_ORDER MAX_PAGE_ORDER
71#elif defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
72/*
73 * Only pages within a single memory section are guaranteed to be
74 * contiguous. By limiting folios to a single memory section, all folio
75 * pages are guaranteed to be contiguous.
76 */
77#define MAX_FOLIO_ORDER PFN_SECTION_SHIFT
78#elif defined(CONFIG_HUGETLB_PAGE)
79/*
80 * There is no real limit on the folio size. We limit them to the maximum we
81 * currently expect (see CONFIG_HAVE_GIGANTIC_FOLIOS): with hugetlb, we expect
82 * no folios larger than 16 GiB on 64bit and 1 GiB on 32bit.
83 */
84#ifdef CONFIG_64BIT
85#define MAX_FOLIO_ORDER (ilog2(SZ_16G) - PAGE_SHIFT)
86#else
87#define MAX_FOLIO_ORDER (ilog2(SZ_1G) - PAGE_SHIFT)
88#endif
89#else
90/*
91 * Without hugetlb, gigantic folios that are bigger than a single PUD are
92 * currently impossible.
93 */
94#define MAX_FOLIO_ORDER (PUD_SHIFT - PAGE_SHIFT)
95#endif
96
97#define MAX_FOLIO_NR_PAGES (1UL << MAX_FOLIO_ORDER)
98
99/*
100 * HugeTLB Vmemmap Optimization (HVO) requires struct pages of the head page to
101 * be naturally aligned with regard to the folio size.
102 *
103 * HVO which is only active if the size of struct page is a power of 2.
104 */
105#define MAX_FOLIO_VMEMMAP_ALIGN \
106 (IS_ENABLED(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP) && \
107 is_power_of_2(sizeof(struct page)) ? \
108 MAX_FOLIO_NR_PAGES * sizeof(struct page) : 0)
109
110/*
111 * vmemmap optimization (like HVO) is only possible for page orders that fill
112 * two or more pages with struct pages.
113 */
114#define VMEMMAP_TAIL_MIN_ORDER (ilog2(2 * PAGE_SIZE / sizeof(struct page)))
115#define __NR_VMEMMAP_TAILS (MAX_FOLIO_ORDER - VMEMMAP_TAIL_MIN_ORDER + 1)
116#define NR_VMEMMAP_TAILS (__NR_VMEMMAP_TAILS > 0 ? __NR_VMEMMAP_TAILS : 0)
117
118enum migratetype {
119 MIGRATE_UNMOVABLE,
120 MIGRATE_MOVABLE,
121 MIGRATE_RECLAIMABLE,
122 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
123 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
124#ifdef CONFIG_CMA
125 /*
126 * MIGRATE_CMA migration type is designed to mimic the way
127 * ZONE_MOVABLE works. Only movable pages can be allocated
128 * from MIGRATE_CMA pageblocks and page allocator never
129 * implicitly change migration type of MIGRATE_CMA pageblock.
130 *
131 * The way to use it is to change migratetype of a range of
132 * pageblocks to MIGRATE_CMA which can be done by
133 * __free_pageblock_cma() function.
134 */
135 MIGRATE_CMA,
136 __MIGRATE_TYPE_END = MIGRATE_CMA,
137#else
138 __MIGRATE_TYPE_END = MIGRATE_HIGHATOMIC,
139#endif
140#ifdef CONFIG_MEMORY_ISOLATION
141 MIGRATE_ISOLATE, /* can't allocate from here */
142#endif
143 MIGRATE_TYPES
144};
145
146/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
147extern const char * const migratetype_names[MIGRATE_TYPES];
148
149#ifdef CONFIG_CMA
150# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
151# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
152/*
153 * __dump_folio() in mm/debug.c passes a folio pointer to on-stack struct folio,
154 * so folio_pfn() cannot be used and pfn is needed.
155 */
156# define is_migrate_cma_folio(folio, pfn) \
157 (get_pfnblock_migratetype(&folio->page, pfn) == MIGRATE_CMA)
158#else
159# define is_migrate_cma(migratetype) false
160# define is_migrate_cma_page(_page) false
161# define is_migrate_cma_folio(folio, pfn) false
162#endif
163
164static inline bool is_migrate_movable(int mt)
165{
166 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
167}
168
169/*
170 * Check whether a migratetype can be merged with another migratetype.
171 *
172 * It is only mergeable when it can fall back to other migratetypes for
173 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
174 */
175static inline bool migratetype_is_mergeable(int mt)
176{
177 return mt < MIGRATE_PCPTYPES;
178}
179
180#define for_each_migratetype_order(order, type) \
181 for (order = 0; order < NR_PAGE_ORDERS; order++) \
182 for (type = 0; type < MIGRATE_TYPES; type++)
183
184extern int page_group_by_mobility_disabled;
185
186#define get_pageblock_migratetype(page) \
187 get_pfnblock_migratetype(page, page_to_pfn(page))
188
189#define folio_migratetype(folio) \
190 get_pageblock_migratetype(&folio->page)
191
192struct free_area {
193 struct list_head free_list[MIGRATE_TYPES];
194 unsigned long nr_free;
195};
196
197struct pglist_data;
198
199#ifdef CONFIG_NUMA
200enum numa_stat_item {
201 NUMA_HIT, /* allocated in intended node */
202 NUMA_MISS, /* allocated in non intended node */
203 NUMA_FOREIGN, /* was intended here, hit elsewhere */
204 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
205 NUMA_LOCAL, /* allocation from local node */
206 NUMA_OTHER, /* allocation from other node */
207 NR_VM_NUMA_EVENT_ITEMS
208};
209#else
210#define NR_VM_NUMA_EVENT_ITEMS 0
211#endif
212
213enum zone_stat_item {
214 /* First 128 byte cacheline (assuming 64 bit words) */
215 NR_FREE_PAGES,
216 NR_FREE_PAGES_BLOCKS,
217 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
218 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
219 NR_ZONE_ACTIVE_ANON,
220 NR_ZONE_INACTIVE_FILE,
221 NR_ZONE_ACTIVE_FILE,
222 NR_ZONE_UNEVICTABLE,
223 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
224 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
225 /* Second 128 byte cacheline */
226#if IS_ENABLED(CONFIG_ZSMALLOC)
227 NR_ZSPAGES, /* allocated in zsmalloc */
228#endif
229 NR_FREE_CMA_PAGES,
230#ifdef CONFIG_UNACCEPTED_MEMORY
231 NR_UNACCEPTED,
232#endif
233 NR_VM_ZONE_STAT_ITEMS };
234
235enum node_stat_item {
236 NR_LRU_BASE,
237 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
238 NR_ACTIVE_ANON, /* " " " " " */
239 NR_INACTIVE_FILE, /* " " " " " */
240 NR_ACTIVE_FILE, /* " " " " " */
241 NR_UNEVICTABLE, /* " " " " " */
242 NR_SLAB_RECLAIMABLE_B,
243 NR_SLAB_UNRECLAIMABLE_B,
244 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
245 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
246 WORKINGSET_NODES,
247 WORKINGSET_REFAULT_BASE,
248 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
249 WORKINGSET_REFAULT_FILE,
250 WORKINGSET_ACTIVATE_BASE,
251 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
252 WORKINGSET_ACTIVATE_FILE,
253 WORKINGSET_RESTORE_BASE,
254 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
255 WORKINGSET_RESTORE_FILE,
256 WORKINGSET_NODERECLAIM,
257 NR_ANON_MAPPED, /* Mapped anonymous pages */
258 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
259 only modified from process context */
260 NR_FILE_PAGES,
261 NR_FILE_DIRTY,
262 NR_WRITEBACK,
263 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
264 NR_SHMEM_THPS,
265 NR_SHMEM_PMDMAPPED,
266 NR_FILE_THPS,
267 NR_FILE_PMDMAPPED,
268 NR_ANON_THPS,
269 NR_VMSCAN_WRITE,
270 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
271 NR_DIRTIED, /* page dirtyings since bootup */
272 NR_WRITTEN, /* page writings since bootup */
273 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
274 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
275 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
276 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
277 NR_VMALLOC,
278 NR_KERNEL_STACK_KB, /* measured in KiB */
279#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
280 NR_KERNEL_SCS_KB, /* measured in KiB */
281#endif
282 NR_PAGETABLE, /* used for pagetables */
283 NR_SECONDARY_PAGETABLE, /* secondary pagetables, KVM & IOMMU */
284#ifdef CONFIG_IOMMU_SUPPORT
285 NR_IOMMU_PAGES, /* # of pages allocated by IOMMU */
286#endif
287#ifdef CONFIG_SWAP
288 NR_SWAPCACHE,
289#endif
290#ifdef CONFIG_NUMA_BALANCING
291 PGPROMOTE_SUCCESS, /* promote successfully */
292 /**
293 * Candidate pages for promotion based on hint fault latency. This
294 * counter is used to control the promotion rate and adjust the hot
295 * threshold.
296 */
297 PGPROMOTE_CANDIDATE,
298 /**
299 * Not rate-limited (NRL) candidate pages for those can be promoted
300 * without considering hot threshold because of enough free pages in
301 * fast-tier node. These promotions bypass the regular hotness checks
302 * and do NOT influence the promotion rate-limiter or
303 * threshold-adjustment logic.
304 * This is for statistics/monitoring purposes.
305 */
306 PGPROMOTE_CANDIDATE_NRL,
307#endif
308 /* PGDEMOTE_*: pages demoted */
309 PGDEMOTE_KSWAPD,
310 PGDEMOTE_DIRECT,
311 PGDEMOTE_KHUGEPAGED,
312 PGDEMOTE_PROACTIVE,
313 PGSTEAL_KSWAPD,
314 PGSTEAL_DIRECT,
315 PGSTEAL_KHUGEPAGED,
316 PGSTEAL_PROACTIVE,
317 PGSTEAL_ANON,
318 PGSTEAL_FILE,
319 PGSCAN_KSWAPD,
320 PGSCAN_DIRECT,
321 PGSCAN_KHUGEPAGED,
322 PGSCAN_PROACTIVE,
323 PGSCAN_ANON,
324 PGSCAN_FILE,
325 PGREFILL,
326#ifdef CONFIG_HUGETLB_PAGE
327 NR_HUGETLB,
328#endif
329 NR_BALLOON_PAGES,
330 NR_KERNEL_FILE_PAGES,
331 NR_GPU_ACTIVE, /* Pages assigned to GPU objects */
332 NR_GPU_RECLAIM, /* Pages in shrinkable GPU pools */
333 NR_VM_NODE_STAT_ITEMS
334};
335
336/*
337 * Returns true if the item should be printed in THPs (/proc/vmstat
338 * currently prints number of anon, file and shmem THPs. But the item
339 * is charged in pages).
340 */
341static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
342{
343 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
344 return false;
345
346 return item == NR_ANON_THPS ||
347 item == NR_FILE_THPS ||
348 item == NR_SHMEM_THPS ||
349 item == NR_SHMEM_PMDMAPPED ||
350 item == NR_FILE_PMDMAPPED;
351}
352
353/*
354 * Returns true if the value is measured in bytes (most vmstat values are
355 * measured in pages). This defines the API part, the internal representation
356 * might be different.
357 */
358static __always_inline bool vmstat_item_in_bytes(int idx)
359{
360 /*
361 * Global and per-node slab counters track slab pages.
362 * It's expected that changes are multiples of PAGE_SIZE.
363 * Internally values are stored in pages.
364 *
365 * Per-memcg and per-lruvec counters track memory, consumed
366 * by individual slab objects. These counters are actually
367 * byte-precise.
368 */
369 return (idx == NR_SLAB_RECLAIMABLE_B ||
370 idx == NR_SLAB_UNRECLAIMABLE_B);
371}
372
373/*
374 * We do arithmetic on the LRU lists in various places in the code,
375 * so it is important to keep the active lists LRU_ACTIVE higher in
376 * the array than the corresponding inactive lists, and to keep
377 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
378 *
379 * This has to be kept in sync with the statistics in zone_stat_item
380 * above and the descriptions in vmstat_text in mm/vmstat.c
381 */
382#define LRU_BASE 0
383#define LRU_ACTIVE 1
384#define LRU_FILE 2
385
386enum lru_list {
387 LRU_INACTIVE_ANON = LRU_BASE,
388 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
389 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
390 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
391 LRU_UNEVICTABLE,
392 NR_LRU_LISTS
393};
394
395enum vmscan_throttle_state {
396 VMSCAN_THROTTLE_WRITEBACK,
397 VMSCAN_THROTTLE_ISOLATED,
398 VMSCAN_THROTTLE_NOPROGRESS,
399 VMSCAN_THROTTLE_CONGESTED,
400 NR_VMSCAN_THROTTLE,
401};
402
403#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
404
405#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
406
407static inline bool is_file_lru(enum lru_list lru)
408{
409 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
410}
411
412static inline bool is_active_lru(enum lru_list lru)
413{
414 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
415}
416
417#define WORKINGSET_ANON 0
418#define WORKINGSET_FILE 1
419#define ANON_AND_FILE 2
420
421enum lruvec_flags {
422 /*
423 * An lruvec has many dirty pages backed by a congested BDI:
424 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
425 * It can be cleared by cgroup reclaim or kswapd.
426 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
427 * It can only be cleared by kswapd.
428 *
429 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
430 * reclaim, but not vice versa. This only applies to the root cgroup.
431 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
432 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
433 * by kswapd).
434 */
435 LRUVEC_CGROUP_CONGESTED,
436 LRUVEC_NODE_CONGESTED,
437};
438
439#endif /* !__GENERATING_BOUNDS_H */
440
441/*
442 * Evictable folios are divided into multiple generations. The youngest and the
443 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
444 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
445 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
446 * corresponding generation. The gen counter in folio->flags stores gen+1 while
447 * a folio is on one of lrugen->folios[]. Otherwise it stores 0.
448 *
449 * After a folio is faulted in, the aging needs to check the accessed bit at
450 * least twice before handing this folio over to the eviction. The first check
451 * clears the accessed bit from the initial fault; the second check makes sure
452 * this folio hasn't been used since then. This process, AKA second chance,
453 * requires a minimum of two generations, hence MIN_NR_GENS. And to maintain ABI
454 * compatibility with the active/inactive LRU, e.g., /proc/vmstat, these two
455 * generations are considered active; the rest of generations, if they exist,
456 * are considered inactive. See lru_gen_is_active().
457 *
458 * PG_active is always cleared while a folio is on one of lrugen->folios[] so
459 * that the sliding window needs not to worry about it. And it's set again when
460 * a folio considered active is isolated for non-reclaiming purposes, e.g.,
461 * migration. See lru_gen_add_folio() and lru_gen_del_folio().
462 *
463 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
464 * number of categories of the active/inactive LRU when keeping track of
465 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
466 * in folio->flags, masked by LRU_GEN_MASK.
467 */
468#define MIN_NR_GENS 2U
469#define MAX_NR_GENS 4U
470
471/*
472 * Each generation is divided into multiple tiers. A folio accessed N times
473 * through file descriptors is in tier order_base_2(N). A folio in the first
474 * tier (N=0,1) is marked by PG_referenced unless it was faulted in through page
475 * tables or read ahead. A folio in the last tier (MAX_NR_TIERS-1) is marked by
476 * PG_workingset. A folio in any other tier (1<N<5) between the first and last
477 * is marked by additional bits of LRU_REFS_WIDTH in folio->flags.
478 *
479 * In contrast to moving across generations which requires the LRU lock, moving
480 * across tiers only involves atomic operations on folio->flags and therefore
481 * has a negligible cost in the buffered access path. In the eviction path,
482 * comparisons of refaulted/(evicted+protected) from the first tier and the rest
483 * infer whether folios accessed multiple times through file descriptors are
484 * statistically hot and thus worth protecting.
485 *
486 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
487 * number of categories of the active/inactive LRU when keeping track of
488 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
489 * folio->flags, masked by LRU_REFS_MASK.
490 */
491#define MAX_NR_TIERS 4U
492
493#ifndef __GENERATING_BOUNDS_H
494
495#define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
496#define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
497
498/*
499 * For folios accessed multiple times through file descriptors,
500 * lru_gen_inc_refs() sets additional bits of LRU_REFS_WIDTH in folio->flags
501 * after PG_referenced, then PG_workingset after LRU_REFS_WIDTH. After all its
502 * bits are set, i.e., LRU_REFS_FLAGS|BIT(PG_workingset), a folio is lazily
503 * promoted into the second oldest generation in the eviction path. And when
504 * folio_inc_gen() does that, it clears LRU_REFS_FLAGS so that
505 * lru_gen_inc_refs() can start over. Note that for this case, LRU_REFS_MASK is
506 * only valid when PG_referenced is set.
507 *
508 * For folios accessed multiple times through page tables, folio_update_gen()
509 * from a page table walk or lru_gen_set_refs() from a rmap walk sets
510 * PG_referenced after the accessed bit is cleared for the first time.
511 * Thereafter, those two paths set PG_workingset and promote folios to the
512 * youngest generation. Like folio_inc_gen(), folio_update_gen() also clears
513 * PG_referenced. Note that for this case, LRU_REFS_MASK is not used.
514 *
515 * For both cases above, after PG_workingset is set on a folio, it remains until
516 * this folio is either reclaimed, or "deactivated" by lru_gen_clear_refs(). It
517 * can be set again if lru_gen_test_recent() returns true upon a refault.
518 */
519#define LRU_REFS_FLAGS (LRU_REFS_MASK | BIT(PG_referenced))
520
521struct lruvec;
522struct page_vma_mapped_walk;
523
524#ifdef CONFIG_LRU_GEN
525
526enum {
527 LRU_GEN_ANON,
528 LRU_GEN_FILE,
529};
530
531enum {
532 LRU_GEN_CORE,
533 LRU_GEN_MM_WALK,
534 LRU_GEN_NONLEAF_YOUNG,
535 NR_LRU_GEN_CAPS
536};
537
538#define MIN_LRU_BATCH BITS_PER_LONG
539#define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
540
541/* whether to keep historical stats from evicted generations */
542#ifdef CONFIG_LRU_GEN_STATS
543#define NR_HIST_GENS MAX_NR_GENS
544#else
545#define NR_HIST_GENS 1U
546#endif
547
548/*
549 * The youngest generation number is stored in max_seq for both anon and file
550 * types as they are aged on an equal footing. The oldest generation numbers are
551 * stored in min_seq[] separately for anon and file types so that they can be
552 * incremented independently. Ideally min_seq[] are kept in sync when both anon
553 * and file types are evictable. However, to adapt to situations like extreme
554 * swappiness, they are allowed to be out of sync by at most
555 * MAX_NR_GENS-MIN_NR_GENS-1.
556 *
557 * The number of pages in each generation is eventually consistent and therefore
558 * can be transiently negative when reset_batch_size() is pending.
559 */
560struct lru_gen_folio {
561 /* the aging increments the youngest generation number */
562 unsigned long max_seq;
563 /* the eviction increments the oldest generation numbers */
564 unsigned long min_seq[ANON_AND_FILE];
565 /* the birth time of each generation in jiffies */
566 unsigned long timestamps[MAX_NR_GENS];
567 /* the multi-gen LRU lists, lazily sorted on eviction */
568 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
569 /* the multi-gen LRU sizes, eventually consistent */
570 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
571 /* the exponential moving average of refaulted */
572 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
573 /* the exponential moving average of evicted+protected */
574 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
575 /* can only be modified under the LRU lock */
576 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
577 /* can be modified without holding the LRU lock */
578 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
579 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
580 /* whether the multi-gen LRU is enabled */
581 bool enabled;
582 /* the memcg generation this lru_gen_folio belongs to */
583 u8 gen;
584 /* the list segment this lru_gen_folio belongs to */
585 u8 seg;
586 /* per-node lru_gen_folio list for global reclaim */
587 struct hlist_nulls_node list;
588};
589
590enum {
591 MM_LEAF_TOTAL, /* total leaf entries */
592 MM_LEAF_YOUNG, /* young leaf entries */
593 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
594 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
595 NR_MM_STATS
596};
597
598/* double-buffering Bloom filters */
599#define NR_BLOOM_FILTERS 2
600
601struct lru_gen_mm_state {
602 /* synced with max_seq after each iteration */
603 unsigned long seq;
604 /* where the current iteration continues after */
605 struct list_head *head;
606 /* where the last iteration ended before */
607 struct list_head *tail;
608 /* Bloom filters flip after each iteration */
609 unsigned long *filters[NR_BLOOM_FILTERS];
610 /* the mm stats for debugging */
611 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
612};
613
614struct lru_gen_mm_walk {
615 /* the lruvec under reclaim */
616 struct lruvec *lruvec;
617 /* max_seq from lru_gen_folio: can be out of date */
618 unsigned long seq;
619 /* the next address within an mm to scan */
620 unsigned long next_addr;
621 /* to batch promoted pages */
622 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
623 /* to batch the mm stats */
624 int mm_stats[NR_MM_STATS];
625 /* total batched items */
626 int batched;
627 int swappiness;
628 bool force_scan;
629};
630
631/*
632 * For each node, memcgs are divided into two generations: the old and the
633 * young. For each generation, memcgs are randomly sharded into multiple bins
634 * to improve scalability. For each bin, the hlist_nulls is virtually divided
635 * into three segments: the head, the tail and the default.
636 *
637 * An onlining memcg is added to the tail of a random bin in the old generation.
638 * The eviction starts at the head of a random bin in the old generation. The
639 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
640 * the old generation, is incremented when all its bins become empty.
641 *
642 * There are four operations:
643 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
644 * current generation (old or young) and updates its "seg" to "head";
645 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
646 * current generation (old or young) and updates its "seg" to "tail";
647 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
648 * generation, updates its "gen" to "old" and resets its "seg" to "default";
649 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
650 * young generation, updates its "gen" to "young" and resets its "seg" to
651 * "default".
652 *
653 * The events that trigger the above operations are:
654 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
655 * 2. The first attempt to reclaim a memcg below low, which triggers
656 * MEMCG_LRU_TAIL;
657 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
658 * threshold, which triggers MEMCG_LRU_TAIL;
659 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
660 * threshold, which triggers MEMCG_LRU_YOUNG;
661 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
662 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
663 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
664 *
665 * Notes:
666 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
667 * of their max_seq counters ensures the eventual fairness to all eligible
668 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
669 * 2. There are only two valid generations: old (seq) and young (seq+1).
670 * MEMCG_NR_GENS is set to three so that when reading the generation counter
671 * locklessly, a stale value (seq-1) does not wraparound to young.
672 */
673#define MEMCG_NR_GENS 3
674#define MEMCG_NR_BINS 8
675
676struct lru_gen_memcg {
677 /* the per-node memcg generation counter */
678 unsigned long seq;
679 /* each memcg has one lru_gen_folio per node */
680 unsigned long nr_memcgs[MEMCG_NR_GENS];
681 /* per-node lru_gen_folio list for global reclaim */
682 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
683 /* protects the above */
684 spinlock_t lock;
685};
686
687void lru_gen_init_pgdat(struct pglist_data *pgdat);
688void lru_gen_init_lruvec(struct lruvec *lruvec);
689bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw, unsigned int nr);
690
691void lru_gen_init_memcg(struct mem_cgroup *memcg);
692void lru_gen_exit_memcg(struct mem_cgroup *memcg);
693void lru_gen_online_memcg(struct mem_cgroup *memcg);
694void lru_gen_offline_memcg(struct mem_cgroup *memcg);
695void lru_gen_release_memcg(struct mem_cgroup *memcg);
696void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
697
698#else /* !CONFIG_LRU_GEN */
699
700static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
701{
702}
703
704static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
705{
706}
707
708static inline bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw,
709 unsigned int nr)
710{
711 return false;
712}
713
714static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
715{
716}
717
718static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
719{
720}
721
722static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
723{
724}
725
726static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
727{
728}
729
730static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
731{
732}
733
734static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
735{
736}
737
738#endif /* CONFIG_LRU_GEN */
739
740struct lruvec {
741 struct list_head lists[NR_LRU_LISTS];
742 /* per lruvec lru_lock for memcg */
743 spinlock_t lru_lock;
744 /*
745 * These track the cost of reclaiming one LRU - file or anon -
746 * over the other. As the observed cost of reclaiming one LRU
747 * increases, the reclaim scan balance tips toward the other.
748 */
749 unsigned long anon_cost;
750 unsigned long file_cost;
751 /* Non-resident age, driven by LRU movement */
752 atomic_long_t nonresident_age;
753 /* Refaults at the time of last reclaim cycle */
754 unsigned long refaults[ANON_AND_FILE];
755 /* Various lruvec state flags (enum lruvec_flags) */
756 unsigned long flags;
757#ifdef CONFIG_LRU_GEN
758 /* evictable pages divided into generations */
759 struct lru_gen_folio lrugen;
760#ifdef CONFIG_LRU_GEN_WALKS_MMU
761 /* to concurrently iterate lru_gen_mm_list */
762 struct lru_gen_mm_state mm_state;
763#endif
764#endif /* CONFIG_LRU_GEN */
765#ifdef CONFIG_MEMCG
766 struct pglist_data *pgdat;
767#endif
768 struct zswap_lruvec_state zswap_lruvec_state;
769};
770
771/* Isolate for asynchronous migration */
772#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
773/* Isolate unevictable pages */
774#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
775
776/* LRU Isolation modes. */
777typedef unsigned __bitwise isolate_mode_t;
778
779enum zone_watermarks {
780 WMARK_MIN,
781 WMARK_LOW,
782 WMARK_HIGH,
783 WMARK_PROMO,
784 NR_WMARK
785};
786
787/*
788 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. Two additional lists
789 * are added for THP. One PCP list is used by GPF_MOVABLE, and the other PCP list
790 * is used by GFP_UNMOVABLE and GFP_RECLAIMABLE.
791 */
792#ifdef CONFIG_TRANSPARENT_HUGEPAGE
793#define NR_PCP_THP 2
794#else
795#define NR_PCP_THP 0
796#endif
797#define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
798#define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
799
800/*
801 * Flags used in pcp->flags field.
802 *
803 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
804 * previous page freeing. To avoid to drain PCP for an accident
805 * high-order page freeing.
806 *
807 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
808 * draining PCP for consecutive high-order pages freeing without
809 * allocation if data cache slice of CPU is large enough. To reduce
810 * zone lock contention and keep cache-hot pages reusing.
811 */
812#define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
813#define PCPF_FREE_HIGH_BATCH BIT(1)
814
815struct per_cpu_pages {
816 spinlock_t lock; /* Protects lists field */
817 int count; /* number of pages in the list */
818 int high; /* high watermark, emptying needed */
819 int high_min; /* min high watermark */
820 int high_max; /* max high watermark */
821 int batch; /* chunk size for buddy add/remove */
822 u8 flags; /* protected by pcp->lock */
823 u8 alloc_factor; /* batch scaling factor during allocate */
824#ifdef CONFIG_NUMA
825 u8 expire; /* When 0, remote pagesets are drained */
826#endif
827 short free_count; /* consecutive free count */
828
829 /* Lists of pages, one per migrate type stored on the pcp-lists */
830 struct list_head lists[NR_PCP_LISTS];
831} ____cacheline_aligned_in_smp;
832
833struct per_cpu_zonestat {
834#ifdef CONFIG_SMP
835 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
836 s8 stat_threshold;
837#endif
838#ifdef CONFIG_NUMA
839 /*
840 * Low priority inaccurate counters that are only folded
841 * on demand. Use a large type to avoid the overhead of
842 * folding during refresh_cpu_vm_stats.
843 */
844 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
845#endif
846};
847
848struct per_cpu_nodestat {
849 s8 stat_threshold;
850 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
851};
852
853#endif /* !__GENERATING_BOUNDS.H */
854
855enum zone_type {
856 /*
857 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
858 * to DMA to all of the addressable memory (ZONE_NORMAL).
859 * On architectures where this area covers the whole 32 bit address
860 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
861 * DMA addressing constraints. This distinction is important as a 32bit
862 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
863 * platforms may need both zones as they support peripherals with
864 * different DMA addressing limitations.
865 */
866#ifdef CONFIG_ZONE_DMA
867 ZONE_DMA,
868#endif
869#ifdef CONFIG_ZONE_DMA32
870 ZONE_DMA32,
871#endif
872 /*
873 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
874 * performed on pages in ZONE_NORMAL if the DMA devices support
875 * transfers to all addressable memory.
876 */
877 ZONE_NORMAL,
878#ifdef CONFIG_HIGHMEM
879 /*
880 * A memory area that is only addressable by the kernel through
881 * mapping portions into its own address space. This is for example
882 * used by i386 to allow the kernel to address the memory beyond
883 * 900MB. The kernel will set up special mappings (page
884 * table entries on i386) for each page that the kernel needs to
885 * access.
886 */
887 ZONE_HIGHMEM,
888#endif
889 /*
890 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
891 * movable pages with few exceptional cases described below. Main use
892 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
893 * likely to succeed, and to locally limit unmovable allocations - e.g.,
894 * to increase the number of THP/huge pages. Notable special cases are:
895 *
896 * 1. Pinned pages: (long-term) pinning of movable pages might
897 * essentially turn such pages unmovable. Therefore, we do not allow
898 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
899 * faulted, they come from the right zone right away. However, it is
900 * still possible that address space already has pages in
901 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
902 * touches that memory before pinning). In such case we migrate them
903 * to a different zone. When migration fails - pinning fails.
904 * 2. memblock allocations: kernelcore/movablecore setups might create
905 * situations where ZONE_MOVABLE contains unmovable allocations
906 * after boot. Memory offlining and allocations fail early.
907 * 3. Memory holes: kernelcore/movablecore setups might create very rare
908 * situations where ZONE_MOVABLE contains memory holes after boot,
909 * for example, if we have sections that are only partially
910 * populated. Memory offlining and allocations fail early.
911 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
912 * memory offlining, such pages cannot be allocated.
913 * 5. Unmovable PG_offline pages: in paravirtualized environments,
914 * hotplugged memory blocks might only partially be managed by the
915 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
916 * parts not manged by the buddy are unmovable PG_offline pages. In
917 * some cases (virtio-mem), such pages can be skipped during
918 * memory offlining, however, cannot be moved/allocated. These
919 * techniques might use alloc_contig_range() to hide previously
920 * exposed pages from the buddy again (e.g., to implement some sort
921 * of memory unplug in virtio-mem).
922 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
923 * situations where ZERO_PAGE(0) which is allocated differently
924 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
925 * cannot be migrated.
926 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
927 * memory to the MOVABLE zone, the vmemmap pages are also placed in
928 * such zone. Such pages cannot be really moved around as they are
929 * self-stored in the range, but they are treated as movable when
930 * the range they describe is about to be offlined.
931 *
932 * In general, no unmovable allocations that degrade memory offlining
933 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
934 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
935 * if has_unmovable_pages() states that there are no unmovable pages,
936 * there can be false negatives).
937 */
938 ZONE_MOVABLE,
939#ifdef CONFIG_ZONE_DEVICE
940 ZONE_DEVICE,
941#endif
942 __MAX_NR_ZONES
943
944};
945
946#ifndef __GENERATING_BOUNDS_H
947
948#define ASYNC_AND_SYNC 2
949
950struct zone {
951 /* Read-mostly fields */
952
953 /* zone watermarks, access with *_wmark_pages(zone) macros */
954 unsigned long _watermark[NR_WMARK];
955 unsigned long watermark_boost;
956
957 unsigned long nr_reserved_highatomic;
958 unsigned long nr_free_highatomic;
959
960 /*
961 * We don't know if the memory that we're going to allocate will be
962 * freeable or/and it will be released eventually, so to avoid totally
963 * wasting several GB of ram we must reserve some of the lower zone
964 * memory (otherwise we risk to run OOM on the lower zones despite
965 * there being tons of freeable ram on the higher zones). This array is
966 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
967 * changes.
968 */
969 long lowmem_reserve[MAX_NR_ZONES];
970
971#ifdef CONFIG_NUMA
972 int node;
973#endif
974 struct pglist_data *zone_pgdat;
975 struct per_cpu_pages __percpu *per_cpu_pageset;
976 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
977 /*
978 * the high and batch values are copied to individual pagesets for
979 * faster access
980 */
981 int pageset_high_min;
982 int pageset_high_max;
983 int pageset_batch;
984
985#ifndef CONFIG_SPARSEMEM
986 /*
987 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
988 * In SPARSEMEM, this map is stored in struct mem_section
989 */
990 unsigned long *pageblock_flags;
991#endif /* CONFIG_SPARSEMEM */
992
993 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
994 unsigned long zone_start_pfn;
995
996 /*
997 * spanned_pages is the total pages spanned by the zone, including
998 * holes, which is calculated as:
999 * spanned_pages = zone_end_pfn - zone_start_pfn;
1000 *
1001 * present_pages is physical pages existing within the zone, which
1002 * is calculated as:
1003 * present_pages = spanned_pages - absent_pages(pages in holes);
1004 *
1005 * present_early_pages is present pages existing within the zone
1006 * located on memory available since early boot, excluding hotplugged
1007 * memory.
1008 *
1009 * managed_pages is present pages managed by the buddy system, which
1010 * is calculated as (reserved_pages includes pages allocated by the
1011 * bootmem allocator):
1012 * managed_pages = present_pages - reserved_pages;
1013 *
1014 * cma pages is present pages that are assigned for CMA use
1015 * (MIGRATE_CMA).
1016 *
1017 * So present_pages may be used by memory hotplug or memory power
1018 * management logic to figure out unmanaged pages by checking
1019 * (present_pages - managed_pages). And managed_pages should be used
1020 * by page allocator and vm scanner to calculate all kinds of watermarks
1021 * and thresholds.
1022 *
1023 * Locking rules:
1024 *
1025 * zone_start_pfn and spanned_pages are protected by span_seqlock.
1026 * It is a seqlock because it has to be read outside of zone->lock,
1027 * and it is done in the main allocator path. But, it is written
1028 * quite infrequently.
1029 *
1030 * The span_seq lock is declared along with zone->lock because it is
1031 * frequently read in proximity to zone->lock. It's good to
1032 * give them a chance of being in the same cacheline.
1033 *
1034 * Write access to present_pages at runtime should be protected by
1035 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
1036 * present_pages should use get_online_mems() to get a stable value.
1037 */
1038 atomic_long_t managed_pages;
1039 unsigned long spanned_pages;
1040 unsigned long present_pages;
1041#if defined(CONFIG_MEMORY_HOTPLUG)
1042 unsigned long present_early_pages;
1043#endif
1044#ifdef CONFIG_CMA
1045 unsigned long cma_pages;
1046#endif
1047
1048 const char *name;
1049
1050#ifdef CONFIG_MEMORY_ISOLATION
1051 /*
1052 * Number of isolated pageblock. It is used to solve incorrect
1053 * freepage counting problem due to racy retrieving migratetype
1054 * of pageblock. Protected by zone->lock.
1055 */
1056 unsigned long nr_isolate_pageblock;
1057#endif
1058
1059#ifdef CONFIG_MEMORY_HOTPLUG
1060 /* see spanned/present_pages for more description */
1061 seqlock_t span_seqlock;
1062#endif
1063
1064 int initialized;
1065
1066 /* Write-intensive fields used from the page allocator */
1067 CACHELINE_PADDING(_pad1_);
1068
1069 /* free areas of different sizes */
1070 struct free_area free_area[NR_PAGE_ORDERS];
1071
1072#ifdef CONFIG_UNACCEPTED_MEMORY
1073 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
1074 struct list_head unaccepted_pages;
1075
1076 /* To be called once the last page in the zone is accepted */
1077 struct work_struct unaccepted_cleanup;
1078#endif
1079
1080 /* zone flags, see below */
1081 unsigned long flags;
1082
1083 /* Primarily protects free_area */
1084 spinlock_t lock;
1085
1086 /* Pages to be freed when next trylock succeeds */
1087 struct llist_head trylock_free_pages;
1088
1089 /* Write-intensive fields used by compaction and vmstats. */
1090 CACHELINE_PADDING(_pad2_);
1091
1092 /*
1093 * When free pages are below this point, additional steps are taken
1094 * when reading the number of free pages to avoid per-cpu counter
1095 * drift allowing watermarks to be breached
1096 */
1097 unsigned long percpu_drift_mark;
1098
1099#if defined CONFIG_COMPACTION || defined CONFIG_CMA
1100 /* pfn where compaction free scanner should start */
1101 unsigned long compact_cached_free_pfn;
1102 /* pfn where compaction migration scanner should start */
1103 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
1104 unsigned long compact_init_migrate_pfn;
1105 unsigned long compact_init_free_pfn;
1106#endif
1107
1108#ifdef CONFIG_COMPACTION
1109 /*
1110 * On compaction failure, 1<<compact_defer_shift compactions
1111 * are skipped before trying again. The number attempted since
1112 * last failure is tracked with compact_considered.
1113 * compact_order_failed is the minimum compaction failed order.
1114 */
1115 unsigned int compact_considered;
1116 unsigned int compact_defer_shift;
1117 int compact_order_failed;
1118#endif
1119
1120#if defined CONFIG_COMPACTION || defined CONFIG_CMA
1121 /* Set to true when the PG_migrate_skip bits should be cleared */
1122 bool compact_blockskip_flush;
1123#endif
1124
1125 bool contiguous;
1126
1127 CACHELINE_PADDING(_pad3_);
1128 /* Zone statistics */
1129 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
1130 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
1131#ifdef CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP
1132 struct page *vmemmap_tails[NR_VMEMMAP_TAILS];
1133#endif
1134} ____cacheline_internodealigned_in_smp;
1135
1136enum pgdat_flags {
1137 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1138 * many pages under writeback
1139 */
1140 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1141};
1142
1143enum zone_flags {
1144 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1145 * Cleared when kswapd is woken.
1146 */
1147 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1148 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1149};
1150
1151static inline unsigned long wmark_pages(const struct zone *z,
1152 enum zone_watermarks w)
1153{
1154 return z->_watermark[w] + z->watermark_boost;
1155}
1156
1157static inline unsigned long min_wmark_pages(const struct zone *z)
1158{
1159 return wmark_pages(z, WMARK_MIN);
1160}
1161
1162static inline unsigned long low_wmark_pages(const struct zone *z)
1163{
1164 return wmark_pages(z, WMARK_LOW);
1165}
1166
1167static inline unsigned long high_wmark_pages(const struct zone *z)
1168{
1169 return wmark_pages(z, WMARK_HIGH);
1170}
1171
1172static inline unsigned long promo_wmark_pages(const struct zone *z)
1173{
1174 return wmark_pages(z, WMARK_PROMO);
1175}
1176
1177static inline unsigned long zone_managed_pages(const struct zone *zone)
1178{
1179 return (unsigned long)atomic_long_read(&zone->managed_pages);
1180}
1181
1182static inline unsigned long zone_cma_pages(struct zone *zone)
1183{
1184#ifdef CONFIG_CMA
1185 return zone->cma_pages;
1186#else
1187 return 0;
1188#endif
1189}
1190
1191static inline unsigned long zone_end_pfn(const struct zone *zone)
1192{
1193 return zone->zone_start_pfn + zone->spanned_pages;
1194}
1195
1196static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1197{
1198 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1199}
1200
1201static inline bool zone_is_initialized(const struct zone *zone)
1202{
1203 return zone->initialized;
1204}
1205
1206static inline bool zone_is_empty(const struct zone *zone)
1207{
1208 return zone->spanned_pages == 0;
1209}
1210
1211#ifndef BUILD_VDSO32_64
1212/*
1213 * The zone field is never updated after free_area_init_core()
1214 * sets it, so none of the operations on it need to be atomic.
1215 */
1216
1217/* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1218#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1219#define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1220#define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1221#define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1222#define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1223#define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1224#define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1225
1226/*
1227 * Define the bit shifts to access each section. For non-existent
1228 * sections we define the shift as 0; that plus a 0 mask ensures
1229 * the compiler will optimise away reference to them.
1230 */
1231#define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1232#define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1233#define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1234#define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1235#define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1236
1237/* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1238#ifdef NODE_NOT_IN_PAGE_FLAGS
1239#define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1240#define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1241 SECTIONS_PGOFF : ZONES_PGOFF)
1242#else
1243#define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1244#define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1245 NODES_PGOFF : ZONES_PGOFF)
1246#endif
1247
1248#define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1249
1250#define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1251#define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1252#define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1253#define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1254#define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1255#define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1256
1257static inline enum zone_type memdesc_zonenum(memdesc_flags_t flags)
1258{
1259 ASSERT_EXCLUSIVE_BITS(flags.f, ZONES_MASK << ZONES_PGSHIFT);
1260 return (flags.f >> ZONES_PGSHIFT) & ZONES_MASK;
1261}
1262
1263static inline enum zone_type page_zonenum(const struct page *page)
1264{
1265 return memdesc_zonenum(page->flags);
1266}
1267
1268static inline enum zone_type folio_zonenum(const struct folio *folio)
1269{
1270 return memdesc_zonenum(folio->flags);
1271}
1272
1273#ifdef CONFIG_ZONE_DEVICE
1274static inline bool memdesc_is_zone_device(memdesc_flags_t mdf)
1275{
1276 return memdesc_zonenum(mdf) == ZONE_DEVICE;
1277}
1278
1279static inline struct dev_pagemap *page_pgmap(const struct page *page)
1280{
1281 VM_WARN_ON_ONCE_PAGE(!memdesc_is_zone_device(page->flags), page);
1282 return page_folio(page)->pgmap;
1283}
1284
1285/*
1286 * Consecutive zone device pages should not be merged into the same sgl
1287 * or bvec segment with other types of pages or if they belong to different
1288 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1289 * without scanning the entire segment. This helper returns true either if
1290 * both pages are not zone device pages or both pages are zone device pages
1291 * with the same pgmap.
1292 */
1293static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1294 const struct page *b)
1295{
1296 if (memdesc_is_zone_device(a->flags) != memdesc_is_zone_device(b->flags))
1297 return false;
1298 if (!memdesc_is_zone_device(a->flags))
1299 return true;
1300 return page_pgmap(a) == page_pgmap(b);
1301}
1302
1303extern void memmap_init_zone_device(struct zone *, unsigned long,
1304 unsigned long, struct dev_pagemap *);
1305#else
1306static inline bool memdesc_is_zone_device(memdesc_flags_t mdf)
1307{
1308 return false;
1309}
1310static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1311 const struct page *b)
1312{
1313 return true;
1314}
1315static inline struct dev_pagemap *page_pgmap(const struct page *page)
1316{
1317 return NULL;
1318}
1319#endif
1320
1321static inline bool is_zone_device_page(const struct page *page)
1322{
1323 return memdesc_is_zone_device(page->flags);
1324}
1325
1326static inline bool folio_is_zone_device(const struct folio *folio)
1327{
1328 return memdesc_is_zone_device(folio->flags);
1329}
1330
1331static inline bool is_zone_movable_page(const struct page *page)
1332{
1333 return page_zonenum(page) == ZONE_MOVABLE;
1334}
1335
1336static inline bool folio_is_zone_movable(const struct folio *folio)
1337{
1338 return folio_zonenum(folio) == ZONE_MOVABLE;
1339}
1340#endif
1341
1342/*
1343 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1344 * intersection with the given zone
1345 */
1346static inline bool zone_intersects(const struct zone *zone,
1347 unsigned long start_pfn, unsigned long nr_pages)
1348{
1349 if (zone_is_empty(zone))
1350 return false;
1351 if (start_pfn >= zone_end_pfn(zone) ||
1352 start_pfn + nr_pages <= zone->zone_start_pfn)
1353 return false;
1354
1355 return true;
1356}
1357
1358/*
1359 * The "priority" of VM scanning is how much of the queues we will scan in one
1360 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1361 * queues ("queue_length >> 12") during an aging round.
1362 */
1363#define DEF_PRIORITY 12
1364
1365/* Maximum number of zones on a zonelist */
1366#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1367
1368enum {
1369 ZONELIST_FALLBACK, /* zonelist with fallback */
1370#ifdef CONFIG_NUMA
1371 /*
1372 * The NUMA zonelists are doubled because we need zonelists that
1373 * restrict the allocations to a single node for __GFP_THISNODE.
1374 */
1375 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1376#endif
1377 MAX_ZONELISTS
1378};
1379
1380/*
1381 * This struct contains information about a zone in a zonelist. It is stored
1382 * here to avoid dereferences into large structures and lookups of tables
1383 */
1384struct zoneref {
1385 struct zone *zone; /* Pointer to actual zone */
1386 int zone_idx; /* zone_idx(zoneref->zone) */
1387};
1388
1389/*
1390 * One allocation request operates on a zonelist. A zonelist
1391 * is a list of zones, the first one is the 'goal' of the
1392 * allocation, the other zones are fallback zones, in decreasing
1393 * priority.
1394 *
1395 * To speed the reading of the zonelist, the zonerefs contain the zone index
1396 * of the entry being read. Helper functions to access information given
1397 * a struct zoneref are
1398 *
1399 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1400 * zonelist_zone_idx() - Return the index of the zone for an entry
1401 * zonelist_node_idx() - Return the index of the node for an entry
1402 */
1403struct zonelist {
1404 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1405};
1406
1407/*
1408 * The array of struct pages for flatmem.
1409 * It must be declared for SPARSEMEM as well because there are configurations
1410 * that rely on that.
1411 */
1412extern struct page *mem_map;
1413
1414#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1415struct deferred_split {
1416 spinlock_t split_queue_lock;
1417 struct list_head split_queue;
1418 unsigned long split_queue_len;
1419};
1420#endif
1421
1422#ifdef CONFIG_MEMORY_FAILURE
1423/*
1424 * Per NUMA node memory failure handling statistics.
1425 */
1426struct memory_failure_stats {
1427 /*
1428 * Number of raw pages poisoned.
1429 * Cases not accounted: memory outside kernel control, offline page,
1430 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1431 * error events, and unpoison actions from hwpoison_unpoison.
1432 */
1433 unsigned long total;
1434 /*
1435 * Recovery results of poisoned raw pages handled by memory_failure,
1436 * in sync with mf_result.
1437 * total = ignored + failed + delayed + recovered.
1438 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1439 */
1440 unsigned long ignored;
1441 unsigned long failed;
1442 unsigned long delayed;
1443 unsigned long recovered;
1444};
1445#endif
1446
1447/*
1448 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1449 * it's memory layout. On UMA machines there is a single pglist_data which
1450 * describes the whole memory.
1451 *
1452 * Memory statistics and page replacement data structures are maintained on a
1453 * per-zone basis.
1454 */
1455typedef struct pglist_data {
1456 /*
1457 * node_zones contains just the zones for THIS node. Not all of the
1458 * zones may be populated, but it is the full list. It is referenced by
1459 * this node's node_zonelists as well as other node's node_zonelists.
1460 */
1461 struct zone node_zones[MAX_NR_ZONES];
1462
1463 /*
1464 * node_zonelists contains references to all zones in all nodes.
1465 * Generally the first zones will be references to this node's
1466 * node_zones.
1467 */
1468 struct zonelist node_zonelists[MAX_ZONELISTS];
1469
1470 int nr_zones; /* number of populated zones in this node */
1471#ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1472 struct page *node_mem_map;
1473#ifdef CONFIG_PAGE_EXTENSION
1474 struct page_ext *node_page_ext;
1475#endif
1476#endif
1477#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1478 /*
1479 * Must be held any time you expect node_start_pfn,
1480 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1481 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1482 * init.
1483 *
1484 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1485 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1486 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1487 *
1488 * Nests above zone->lock and zone->span_seqlock
1489 */
1490 spinlock_t node_size_lock;
1491#endif
1492 unsigned long node_start_pfn;
1493 unsigned long node_present_pages; /* total number of physical pages */
1494 unsigned long node_spanned_pages; /* total size of physical page
1495 range, including holes */
1496 int node_id;
1497 wait_queue_head_t kswapd_wait;
1498 wait_queue_head_t pfmemalloc_wait;
1499
1500 /* workqueues for throttling reclaim for different reasons. */
1501 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1502
1503 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1504 unsigned long nr_reclaim_start; /* nr pages written while throttled
1505 * when throttling started. */
1506#ifdef CONFIG_MEMORY_HOTPLUG
1507 struct mutex kswapd_lock;
1508#endif
1509 struct task_struct *kswapd; /* Protected by kswapd_lock */
1510 int kswapd_order;
1511 enum zone_type kswapd_highest_zoneidx;
1512
1513 atomic_t kswapd_failures; /* Number of 'reclaimed == 0' runs */
1514
1515#ifdef CONFIG_COMPACTION
1516 int kcompactd_max_order;
1517 enum zone_type kcompactd_highest_zoneidx;
1518 wait_queue_head_t kcompactd_wait;
1519 struct task_struct *kcompactd;
1520 bool proactive_compact_trigger;
1521#endif
1522 /*
1523 * This is a per-node reserve of pages that are not available
1524 * to userspace allocations.
1525 */
1526 unsigned long totalreserve_pages;
1527
1528#ifdef CONFIG_NUMA
1529 /*
1530 * node reclaim becomes active if more unmapped pages exist.
1531 */
1532 unsigned long min_unmapped_pages;
1533 unsigned long min_slab_pages;
1534#endif /* CONFIG_NUMA */
1535
1536 /* Write-intensive fields used by page reclaim */
1537 CACHELINE_PADDING(_pad1_);
1538
1539#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1540 /*
1541 * If memory initialisation on large machines is deferred then this
1542 * is the first PFN that needs to be initialised.
1543 */
1544 unsigned long first_deferred_pfn;
1545#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1546
1547#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1548 struct deferred_split deferred_split_queue;
1549#endif
1550
1551#ifdef CONFIG_NUMA_BALANCING
1552 /* start time in ms of current promote rate limit period */
1553 unsigned int nbp_rl_start;
1554 /* number of promote candidate pages at start time of current rate limit period */
1555 unsigned long nbp_rl_nr_cand;
1556 /* promote threshold in ms */
1557 unsigned int nbp_threshold;
1558 /* start time in ms of current promote threshold adjustment period */
1559 unsigned int nbp_th_start;
1560 /*
1561 * number of promote candidate pages at start time of current promote
1562 * threshold adjustment period
1563 */
1564 unsigned long nbp_th_nr_cand;
1565#endif
1566 /* Fields commonly accessed by the page reclaim scanner */
1567
1568 /*
1569 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1570 *
1571 * Use mem_cgroup_lruvec() to look up lruvecs.
1572 */
1573 struct lruvec __lruvec;
1574
1575 unsigned long flags;
1576
1577#ifdef CONFIG_LRU_GEN
1578 /* kswap mm walk data */
1579 struct lru_gen_mm_walk mm_walk;
1580 /* lru_gen_folio list */
1581 struct lru_gen_memcg memcg_lru;
1582#endif
1583
1584 CACHELINE_PADDING(_pad2_);
1585
1586 /* Per-node vmstats */
1587 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1588 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1589#ifdef CONFIG_NUMA
1590 struct memory_tier __rcu *memtier;
1591#endif
1592#ifdef CONFIG_MEMORY_FAILURE
1593 struct memory_failure_stats mf_stats;
1594#endif
1595} pg_data_t;
1596
1597#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1598#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1599
1600#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1601#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1602
1603static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1604{
1605 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1606}
1607
1608#include <linux/memory_hotplug.h>
1609
1610void build_all_zonelists(pg_data_t *pgdat);
1611bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1612 int highest_zoneidx, unsigned int alloc_flags,
1613 long free_pages);
1614bool zone_watermark_ok(struct zone *z, unsigned int order,
1615 unsigned long mark, int highest_zoneidx,
1616 unsigned int alloc_flags);
1617
1618enum kswapd_clear_hopeless_reason {
1619 KSWAPD_CLEAR_HOPELESS_OTHER = 0,
1620 KSWAPD_CLEAR_HOPELESS_KSWAPD,
1621 KSWAPD_CLEAR_HOPELESS_DIRECT,
1622 KSWAPD_CLEAR_HOPELESS_PCP,
1623};
1624
1625void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1626 enum zone_type highest_zoneidx);
1627void kswapd_try_clear_hopeless(struct pglist_data *pgdat,
1628 unsigned int order, int highest_zoneidx);
1629void kswapd_clear_hopeless(pg_data_t *pgdat, enum kswapd_clear_hopeless_reason reason);
1630bool kswapd_test_hopeless(pg_data_t *pgdat);
1631
1632/*
1633 * Memory initialization context, use to differentiate memory added by
1634 * the platform statically or via memory hotplug interface.
1635 */
1636enum meminit_context {
1637 MEMINIT_EARLY,
1638 MEMINIT_HOTPLUG,
1639};
1640
1641extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1642 unsigned long size);
1643
1644extern void lruvec_init(struct lruvec *lruvec);
1645
1646static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1647{
1648#ifdef CONFIG_MEMCG
1649 return lruvec->pgdat;
1650#else
1651 return container_of(lruvec, struct pglist_data, __lruvec);
1652#endif
1653}
1654
1655#ifdef CONFIG_HAVE_MEMORYLESS_NODES
1656int local_memory_node(int node_id);
1657#else
1658static inline int local_memory_node(int node_id) { return node_id; };
1659#endif
1660
1661/*
1662 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1663 */
1664#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1665
1666#ifdef CONFIG_ZONE_DEVICE
1667static inline bool zone_is_zone_device(const struct zone *zone)
1668{
1669 return zone_idx(zone) == ZONE_DEVICE;
1670}
1671#else
1672static inline bool zone_is_zone_device(const struct zone *zone)
1673{
1674 return false;
1675}
1676#endif
1677
1678/*
1679 * Returns true if a zone has pages managed by the buddy allocator.
1680 * All the reclaim decisions have to use this function rather than
1681 * populated_zone(). If the whole zone is reserved then we can easily
1682 * end up with populated_zone() && !managed_zone().
1683 */
1684static inline bool managed_zone(const struct zone *zone)
1685{
1686 return zone_managed_pages(zone);
1687}
1688
1689/* Returns true if a zone has memory */
1690static inline bool populated_zone(const struct zone *zone)
1691{
1692 return zone->present_pages;
1693}
1694
1695#ifdef CONFIG_NUMA
1696static inline int zone_to_nid(const struct zone *zone)
1697{
1698 return zone->node;
1699}
1700
1701static inline void zone_set_nid(struct zone *zone, int nid)
1702{
1703 zone->node = nid;
1704}
1705#else
1706static inline int zone_to_nid(const struct zone *zone)
1707{
1708 return 0;
1709}
1710
1711static inline void zone_set_nid(struct zone *zone, int nid) {}
1712#endif
1713
1714extern int movable_zone;
1715
1716static inline int is_highmem_idx(enum zone_type idx)
1717{
1718#ifdef CONFIG_HIGHMEM
1719 return (idx == ZONE_HIGHMEM ||
1720 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1721#else
1722 return 0;
1723#endif
1724}
1725
1726/**
1727 * is_highmem - helper function to quickly check if a struct zone is a
1728 * highmem zone or not. This is an attempt to keep references
1729 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1730 * @zone: pointer to struct zone variable
1731 * Return: 1 for a highmem zone, 0 otherwise
1732 */
1733static inline int is_highmem(const struct zone *zone)
1734{
1735 return is_highmem_idx(zone_idx(zone));
1736}
1737
1738bool has_managed_zone(enum zone_type zone);
1739static inline bool has_managed_dma(void)
1740{
1741#ifdef CONFIG_ZONE_DMA
1742 return has_managed_zone(ZONE_DMA);
1743#else
1744 return false;
1745#endif
1746}
1747
1748
1749#ifndef CONFIG_NUMA
1750
1751extern struct pglist_data contig_page_data;
1752static inline struct pglist_data *NODE_DATA(int nid)
1753{
1754 return &contig_page_data;
1755}
1756
1757#else /* CONFIG_NUMA */
1758
1759#include <asm/mmzone.h>
1760
1761#endif /* !CONFIG_NUMA */
1762
1763extern struct pglist_data *first_online_pgdat(void);
1764extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1765extern struct zone *next_zone(struct zone *zone);
1766
1767/**
1768 * for_each_online_pgdat - helper macro to iterate over all online nodes
1769 * @pgdat: pointer to a pg_data_t variable
1770 */
1771#define for_each_online_pgdat(pgdat) \
1772 for (pgdat = first_online_pgdat(); \
1773 pgdat; \
1774 pgdat = next_online_pgdat(pgdat))
1775/**
1776 * for_each_zone - helper macro to iterate over all memory zones
1777 * @zone: pointer to struct zone variable
1778 *
1779 * The user only needs to declare the zone variable, for_each_zone
1780 * fills it in.
1781 */
1782#define for_each_zone(zone) \
1783 for (zone = (first_online_pgdat())->node_zones; \
1784 zone; \
1785 zone = next_zone(zone))
1786
1787#define for_each_populated_zone(zone) \
1788 for (zone = (first_online_pgdat())->node_zones; \
1789 zone; \
1790 zone = next_zone(zone)) \
1791 if (!populated_zone(zone)) \
1792 ; /* do nothing */ \
1793 else
1794
1795static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1796{
1797 return zoneref->zone;
1798}
1799
1800static inline int zonelist_zone_idx(const struct zoneref *zoneref)
1801{
1802 return zoneref->zone_idx;
1803}
1804
1805static inline int zonelist_node_idx(const struct zoneref *zoneref)
1806{
1807 return zone_to_nid(zoneref->zone);
1808}
1809
1810struct zoneref *__next_zones_zonelist(struct zoneref *z,
1811 enum zone_type highest_zoneidx,
1812 nodemask_t *nodes);
1813
1814/**
1815 * 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
1816 * @z: The cursor used as a starting point for the search
1817 * @highest_zoneidx: The zone index of the highest zone to return
1818 * @nodes: An optional nodemask to filter the zonelist with
1819 *
1820 * This function returns the next zone at or below a given zone index that is
1821 * within the allowed nodemask using a cursor as the starting point for the
1822 * search. The zoneref returned is a cursor that represents the current zone
1823 * being examined. It should be advanced by one before calling
1824 * next_zones_zonelist again.
1825 *
1826 * Return: the next zone at or below highest_zoneidx within the allowed
1827 * nodemask using a cursor within a zonelist as a starting point
1828 */
1829static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1830 enum zone_type highest_zoneidx,
1831 nodemask_t *nodes)
1832{
1833 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1834 return z;
1835 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1836}
1837
1838/**
1839 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1840 * @zonelist: The zonelist to search for a suitable zone
1841 * @highest_zoneidx: The zone index of the highest zone to return
1842 * @nodes: An optional nodemask to filter the zonelist with
1843 *
1844 * This function returns the first zone at or below a given zone index that is
1845 * within the allowed nodemask. The zoneref returned is a cursor that can be
1846 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1847 * one before calling.
1848 *
1849 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1850 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1851 * update due to cpuset modification.
1852 *
1853 * Return: Zoneref pointer for the first suitable zone found
1854 */
1855static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1856 enum zone_type highest_zoneidx,
1857 nodemask_t *nodes)
1858{
1859 return next_zones_zonelist(zonelist->_zonerefs,
1860 highest_zoneidx, nodes);
1861}
1862
1863/**
1864 * 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
1865 * @zone: The current zone in the iterator
1866 * @z: The current pointer within zonelist->_zonerefs being iterated
1867 * @zlist: The zonelist being iterated
1868 * @highidx: The zone index of the highest zone to return
1869 * @nodemask: Nodemask allowed by the allocator
1870 *
1871 * This iterator iterates though all zones at or below a given zone index and
1872 * within a given nodemask
1873 */
1874#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1875 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1876 zone; \
1877 z = next_zones_zonelist(++z, highidx, nodemask), \
1878 zone = zonelist_zone(z))
1879
1880#define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1881 for (zone = zonelist_zone(z); \
1882 zone; \
1883 z = next_zones_zonelist(++z, highidx, nodemask), \
1884 zone = zonelist_zone(z))
1885
1886
1887/**
1888 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1889 * @zone: The current zone in the iterator
1890 * @z: The current pointer within zonelist->zones being iterated
1891 * @zlist: The zonelist being iterated
1892 * @highidx: The zone index of the highest zone to return
1893 *
1894 * This iterator iterates though all zones at or below a given zone index.
1895 */
1896#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1897 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1898
1899/* Whether the 'nodes' are all movable nodes */
1900static inline bool movable_only_nodes(nodemask_t *nodes)
1901{
1902 struct zonelist *zonelist;
1903 struct zoneref *z;
1904 int nid;
1905
1906 if (nodes_empty(*nodes))
1907 return false;
1908
1909 /*
1910 * We can chose arbitrary node from the nodemask to get a
1911 * zonelist as they are interlinked. We just need to find
1912 * at least one zone that can satisfy kernel allocations.
1913 */
1914 nid = first_node(*nodes);
1915 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1916 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1917 return (!zonelist_zone(z)) ? true : false;
1918}
1919
1920
1921#ifdef CONFIG_SPARSEMEM
1922#include <asm/sparsemem.h>
1923#endif
1924
1925#ifdef CONFIG_FLATMEM
1926#define pfn_to_nid(pfn) (0)
1927#endif
1928
1929#ifdef CONFIG_SPARSEMEM
1930
1931/*
1932 * PA_SECTION_SHIFT physical address to/from section number
1933 * PFN_SECTION_SHIFT pfn to/from section number
1934 */
1935#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1936#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1937
1938#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1939
1940#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1941#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1942
1943#define SECTION_BLOCKFLAGS_BITS \
1944 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1945
1946#if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1947#error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1948#endif
1949
1950static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1951{
1952 return pfn >> PFN_SECTION_SHIFT;
1953}
1954static inline unsigned long section_nr_to_pfn(unsigned long sec)
1955{
1956 return sec << PFN_SECTION_SHIFT;
1957}
1958
1959#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1960#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1961
1962#define SUBSECTION_SHIFT 21
1963#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1964
1965#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1966#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1967#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1968
1969#if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1970#error Subsection size exceeds section size
1971#else
1972#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1973#endif
1974
1975#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1976#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1977
1978struct mem_section_usage {
1979 struct rcu_head rcu;
1980#ifdef CONFIG_SPARSEMEM_VMEMMAP
1981 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1982#endif
1983 /* See declaration of similar field in struct zone */
1984 unsigned long pageblock_flags[0];
1985};
1986
1987struct page;
1988struct page_ext;
1989struct mem_section {
1990 /*
1991 * This is, logically, a pointer to an array of struct
1992 * pages. However, it is stored with some other magic.
1993 * (see sparse_init_one_section())
1994 *
1995 * Additionally during early boot we encode node id of
1996 * the location of the section here to guide allocation.
1997 * (see sparse.c::memory_present())
1998 *
1999 * Making it a UL at least makes someone do a cast
2000 * before using it wrong.
2001 */
2002 unsigned long section_mem_map;
2003
2004 struct mem_section_usage *usage;
2005#ifdef CONFIG_PAGE_EXTENSION
2006 /*
2007 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
2008 * section. (see page_ext.h about this.)
2009 */
2010 struct page_ext *page_ext;
2011 unsigned long pad;
2012#endif
2013 /*
2014 * WARNING: mem_section must be a power-of-2 in size for the
2015 * calculation and use of SECTION_ROOT_MASK to make sense.
2016 */
2017};
2018
2019#ifdef CONFIG_SPARSEMEM_EXTREME
2020#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
2021#else
2022#define SECTIONS_PER_ROOT 1
2023#endif
2024
2025#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
2026#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
2027#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
2028
2029#ifdef CONFIG_SPARSEMEM_EXTREME
2030extern struct mem_section **mem_section;
2031#else
2032extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
2033#endif
2034
2035static inline unsigned long *section_to_usemap(struct mem_section *ms)
2036{
2037 return ms->usage->pageblock_flags;
2038}
2039
2040static inline struct mem_section *__nr_to_section(unsigned long nr)
2041{
2042 unsigned long root = SECTION_NR_TO_ROOT(nr);
2043
2044 if (unlikely(root >= NR_SECTION_ROOTS))
2045 return NULL;
2046
2047#ifdef CONFIG_SPARSEMEM_EXTREME
2048 if (!mem_section || !mem_section[root])
2049 return NULL;
2050#endif
2051 return &mem_section[root][nr & SECTION_ROOT_MASK];
2052}
2053extern size_t mem_section_usage_size(void);
2054
2055/*
2056 * We use the lower bits of the mem_map pointer to store
2057 * a little bit of information. The pointer is calculated
2058 * as mem_map - section_nr_to_pfn(pnum). The result is
2059 * aligned to the minimum alignment of the two values:
2060 * 1. All mem_map arrays are page-aligned.
2061 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
2062 * lowest bits. PFN_SECTION_SHIFT is arch-specific
2063 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
2064 * worst combination is powerpc with 256k pages,
2065 * which results in PFN_SECTION_SHIFT equal 6.
2066 * To sum it up, at least 6 bits are available on all architectures.
2067 * However, we can exceed 6 bits on some other architectures except
2068 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
2069 * with the worst case of 64K pages on arm64) if we make sure the
2070 * exceeded bit is not applicable to powerpc.
2071 */
2072enum {
2073 SECTION_MARKED_PRESENT_BIT,
2074 SECTION_HAS_MEM_MAP_BIT,
2075 SECTION_IS_ONLINE_BIT,
2076 SECTION_IS_EARLY_BIT,
2077#ifdef CONFIG_ZONE_DEVICE
2078 SECTION_TAINT_ZONE_DEVICE_BIT,
2079#endif
2080#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2081 SECTION_IS_VMEMMAP_PREINIT_BIT,
2082#endif
2083 SECTION_MAP_LAST_BIT,
2084};
2085
2086#define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
2087#define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
2088#define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
2089#define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
2090#ifdef CONFIG_ZONE_DEVICE
2091#define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
2092#endif
2093#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2094#define SECTION_IS_VMEMMAP_PREINIT BIT(SECTION_IS_VMEMMAP_PREINIT_BIT)
2095#endif
2096#define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
2097#define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
2098
2099static inline struct page *__section_mem_map_addr(struct mem_section *section)
2100{
2101 unsigned long map = section->section_mem_map;
2102 map &= SECTION_MAP_MASK;
2103 return (struct page *)map;
2104}
2105
2106static inline int present_section(const struct mem_section *section)
2107{
2108 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
2109}
2110
2111static inline int present_section_nr(unsigned long nr)
2112{
2113 return present_section(__nr_to_section(nr));
2114}
2115
2116static inline int valid_section(const struct mem_section *section)
2117{
2118 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
2119}
2120
2121static inline int early_section(const struct mem_section *section)
2122{
2123 return (section && (section->section_mem_map & SECTION_IS_EARLY));
2124}
2125
2126static inline int valid_section_nr(unsigned long nr)
2127{
2128 return valid_section(__nr_to_section(nr));
2129}
2130
2131static inline int online_section(const struct mem_section *section)
2132{
2133 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
2134}
2135
2136#ifdef CONFIG_ZONE_DEVICE
2137static inline int online_device_section(const struct mem_section *section)
2138{
2139 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
2140
2141 return section && ((section->section_mem_map & flags) == flags);
2142}
2143#else
2144static inline int online_device_section(const struct mem_section *section)
2145{
2146 return 0;
2147}
2148#endif
2149
2150#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2151static inline int preinited_vmemmap_section(const struct mem_section *section)
2152{
2153 return (section &&
2154 (section->section_mem_map & SECTION_IS_VMEMMAP_PREINIT));
2155}
2156
2157void sparse_vmemmap_init_nid_early(int nid);
2158void sparse_vmemmap_init_nid_late(int nid);
2159
2160#else
2161static inline int preinited_vmemmap_section(const struct mem_section *section)
2162{
2163 return 0;
2164}
2165static inline void sparse_vmemmap_init_nid_early(int nid)
2166{
2167}
2168
2169static inline void sparse_vmemmap_init_nid_late(int nid)
2170{
2171}
2172#endif
2173
2174static inline int online_section_nr(unsigned long nr)
2175{
2176 return online_section(__nr_to_section(nr));
2177}
2178
2179#ifdef CONFIG_MEMORY_HOTPLUG
2180void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
2181void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
2182#endif
2183
2184static inline struct mem_section *__pfn_to_section(unsigned long pfn)
2185{
2186 return __nr_to_section(pfn_to_section_nr(pfn));
2187}
2188
2189extern unsigned long __highest_present_section_nr;
2190
2191static inline int subsection_map_index(unsigned long pfn)
2192{
2193 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
2194}
2195
2196#ifdef CONFIG_SPARSEMEM_VMEMMAP
2197static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
2198{
2199 int idx = subsection_map_index(pfn);
2200 struct mem_section_usage *usage = READ_ONCE(ms->usage);
2201
2202 return usage ? test_bit(idx, usage->subsection_map) : 0;
2203}
2204
2205static inline bool pfn_section_first_valid(struct mem_section *ms, unsigned long *pfn)
2206{
2207 struct mem_section_usage *usage = READ_ONCE(ms->usage);
2208 int idx = subsection_map_index(*pfn);
2209 unsigned long bit;
2210
2211 if (!usage)
2212 return false;
2213
2214 if (test_bit(idx, usage->subsection_map))
2215 return true;
2216
2217 /* Find the next subsection that exists */
2218 bit = find_next_bit(usage->subsection_map, SUBSECTIONS_PER_SECTION, idx);
2219 if (bit == SUBSECTIONS_PER_SECTION)
2220 return false;
2221
2222 *pfn = (*pfn & PAGE_SECTION_MASK) + (bit * PAGES_PER_SUBSECTION);
2223 return true;
2224}
2225#else
2226static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
2227{
2228 return 1;
2229}
2230
2231static inline bool pfn_section_first_valid(struct mem_section *ms, unsigned long *pfn)
2232{
2233 return true;
2234}
2235#endif
2236
2237void sparse_init_early_section(int nid, struct page *map, unsigned long pnum,
2238 unsigned long flags);
2239
2240#ifndef CONFIG_HAVE_ARCH_PFN_VALID
2241/**
2242 * pfn_valid - check if there is a valid memory map entry for a PFN
2243 * @pfn: the page frame number to check
2244 *
2245 * Check if there is a valid memory map entry aka struct page for the @pfn.
2246 * Note, that availability of the memory map entry does not imply that
2247 * there is actual usable memory at that @pfn. The struct page may
2248 * represent a hole or an unusable page frame.
2249 *
2250 * Return: 1 for PFNs that have memory map entries and 0 otherwise
2251 */
2252static inline int pfn_valid(unsigned long pfn)
2253{
2254 struct mem_section *ms;
2255 int ret;
2256
2257 /*
2258 * Ensure the upper PAGE_SHIFT bits are clear in the
2259 * pfn. Else it might lead to false positives when
2260 * some of the upper bits are set, but the lower bits
2261 * match a valid pfn.
2262 */
2263 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2264 return 0;
2265
2266 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2267 return 0;
2268 ms = __pfn_to_section(pfn);
2269 rcu_read_lock_sched();
2270 if (!valid_section(ms)) {
2271 rcu_read_unlock_sched();
2272 return 0;
2273 }
2274 /*
2275 * Traditionally early sections always returned pfn_valid() for
2276 * the entire section-sized span.
2277 */
2278 ret = early_section(ms) || pfn_section_valid(ms, pfn);
2279 rcu_read_unlock_sched();
2280
2281 return ret;
2282}
2283
2284/* Returns end_pfn or higher if no valid PFN remaining in range */
2285static inline unsigned long first_valid_pfn(unsigned long pfn, unsigned long end_pfn)
2286{
2287 unsigned long nr = pfn_to_section_nr(pfn);
2288
2289 rcu_read_lock_sched();
2290
2291 while (nr <= __highest_present_section_nr && pfn < end_pfn) {
2292 struct mem_section *ms = __pfn_to_section(pfn);
2293
2294 if (valid_section(ms) &&
2295 (early_section(ms) || pfn_section_first_valid(ms, &pfn))) {
2296 rcu_read_unlock_sched();
2297 return pfn;
2298 }
2299
2300 /* Nothing left in this section? Skip to next section */
2301 nr++;
2302 pfn = section_nr_to_pfn(nr);
2303 }
2304
2305 rcu_read_unlock_sched();
2306 return end_pfn;
2307}
2308
2309static inline unsigned long next_valid_pfn(unsigned long pfn, unsigned long end_pfn)
2310{
2311 pfn++;
2312
2313 if (pfn >= end_pfn)
2314 return end_pfn;
2315
2316 /*
2317 * Either every PFN within the section (or subsection for VMEMMAP) is
2318 * valid, or none of them are. So there's no point repeating the check
2319 * for every PFN; only call first_valid_pfn() again when crossing a
2320 * (sub)section boundary (i.e. !(pfn & ~PAGE_{SUB,}SECTION_MASK)).
2321 */
2322 if (pfn & ~(IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) ?
2323 PAGE_SUBSECTION_MASK : PAGE_SECTION_MASK))
2324 return pfn;
2325
2326 return first_valid_pfn(pfn, end_pfn);
2327}
2328
2329
2330#define for_each_valid_pfn(_pfn, _start_pfn, _end_pfn) \
2331 for ((_pfn) = first_valid_pfn((_start_pfn), (_end_pfn)); \
2332 (_pfn) < (_end_pfn); \
2333 (_pfn) = next_valid_pfn((_pfn), (_end_pfn)))
2334
2335#endif
2336
2337static inline int pfn_in_present_section(unsigned long pfn)
2338{
2339 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2340 return 0;
2341 return present_section(__pfn_to_section(pfn));
2342}
2343
2344static inline unsigned long next_present_section_nr(unsigned long section_nr)
2345{
2346 while (++section_nr <= __highest_present_section_nr) {
2347 if (present_section_nr(section_nr))
2348 return section_nr;
2349 }
2350
2351 return -1;
2352}
2353
2354#define for_each_present_section_nr(start, section_nr) \
2355 for (section_nr = next_present_section_nr(start - 1); \
2356 section_nr != -1; \
2357 section_nr = next_present_section_nr(section_nr))
2358
2359/*
2360 * These are _only_ used during initialisation, therefore they
2361 * can use __initdata ... They could have names to indicate
2362 * this restriction.
2363 */
2364#ifdef CONFIG_NUMA
2365#define pfn_to_nid(pfn) \
2366({ \
2367 unsigned long __pfn_to_nid_pfn = (pfn); \
2368 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2369})
2370#else
2371#define pfn_to_nid(pfn) (0)
2372#endif
2373
2374#else
2375#define sparse_vmemmap_init_nid_early(_nid) do {} while (0)
2376#define sparse_vmemmap_init_nid_late(_nid) do {} while (0)
2377#define pfn_in_present_section pfn_valid
2378#endif /* CONFIG_SPARSEMEM */
2379
2380/*
2381 * Fallback case for when the architecture provides its own pfn_valid() but
2382 * not a corresponding for_each_valid_pfn().
2383 */
2384#ifndef for_each_valid_pfn
2385#define for_each_valid_pfn(_pfn, _start_pfn, _end_pfn) \
2386 for ((_pfn) = (_start_pfn); (_pfn) < (_end_pfn); (_pfn)++) \
2387 if (pfn_valid(_pfn))
2388#endif
2389
2390#endif /* !__GENERATING_BOUNDS.H */
2391#endif /* !__ASSEMBLY__ */
2392#endif /* _LINUX_MMZONE_H */