xref: /linux/mm/internal.h (revision 2363088eba2ecccfb643725e4864af73c4226a04)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* internal.h: mm/ internal definitions
3  *
4  * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 #ifndef __MM_INTERNAL_H
8 #define __MM_INTERNAL_H
9 
10 #include <linux/fs.h>
11 #include <linux/mm.h>
12 #include <linux/pagemap.h>
13 #include <linux/rmap.h>
14 #include <linux/tracepoint-defs.h>
15 
16 struct folio_batch;
17 
18 /*
19  * The set of flags that only affect watermark checking and reclaim
20  * behaviour. This is used by the MM to obey the caller constraints
21  * about IO, FS and watermark checking while ignoring placement
22  * hints such as HIGHMEM usage.
23  */
24 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
25 			__GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
26 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
27 			__GFP_NOLOCKDEP)
28 
29 /* The GFP flags allowed during early boot */
30 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
31 
32 /* Control allocation cpuset and node placement constraints */
33 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
34 
35 /* Do not use these with a slab allocator */
36 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
37 
38 /*
39  * Different from WARN_ON_ONCE(), no warning will be issued
40  * when we specify __GFP_NOWARN.
41  */
42 #define WARN_ON_ONCE_GFP(cond, gfp)	({				\
43 	static bool __section(".data.once") __warned;			\
44 	int __ret_warn_once = !!(cond);					\
45 									\
46 	if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
47 		__warned = true;					\
48 		WARN_ON(1);						\
49 	}								\
50 	unlikely(__ret_warn_once);					\
51 })
52 
53 void page_writeback_init(void);
54 
55 /*
56  * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
57  * its nr_pages_mapped would be 0x400000: choose the COMPOUND_MAPPED bit
58  * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE).  Hugetlb currently
59  * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
60  */
61 #define COMPOUND_MAPPED		0x800000
62 #define FOLIO_PAGES_MAPPED	(COMPOUND_MAPPED - 1)
63 
64 /*
65  * How many individual pages have an elevated _mapcount.  Excludes
66  * the folio's entire_mapcount.
67  */
68 static inline int folio_nr_pages_mapped(struct folio *folio)
69 {
70 	return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
71 }
72 
73 static inline void *folio_raw_mapping(struct folio *folio)
74 {
75 	unsigned long mapping = (unsigned long)folio->mapping;
76 
77 	return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
78 }
79 
80 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
81 						int nr_throttled);
82 static inline void acct_reclaim_writeback(struct folio *folio)
83 {
84 	pg_data_t *pgdat = folio_pgdat(folio);
85 	int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
86 
87 	if (nr_throttled)
88 		__acct_reclaim_writeback(pgdat, folio, nr_throttled);
89 }
90 
91 static inline void wake_throttle_isolated(pg_data_t *pgdat)
92 {
93 	wait_queue_head_t *wqh;
94 
95 	wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
96 	if (waitqueue_active(wqh))
97 		wake_up(wqh);
98 }
99 
100 vm_fault_t do_swap_page(struct vm_fault *vmf);
101 void folio_rotate_reclaimable(struct folio *folio);
102 bool __folio_end_writeback(struct folio *folio);
103 void deactivate_file_folio(struct folio *folio);
104 void folio_activate(struct folio *folio);
105 
106 void free_pgtables(struct mmu_gather *tlb, struct maple_tree *mt,
107 		   struct vm_area_struct *start_vma, unsigned long floor,
108 		   unsigned long ceiling, bool mm_wr_locked);
109 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
110 
111 struct zap_details;
112 void unmap_page_range(struct mmu_gather *tlb,
113 			     struct vm_area_struct *vma,
114 			     unsigned long addr, unsigned long end,
115 			     struct zap_details *details);
116 
117 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *,
118 		unsigned int order);
119 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
120 static inline void force_page_cache_readahead(struct address_space *mapping,
121 		struct file *file, pgoff_t index, unsigned long nr_to_read)
122 {
123 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
124 	force_page_cache_ra(&ractl, nr_to_read);
125 }
126 
127 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
128 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
129 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
130 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
131 void filemap_free_folio(struct address_space *mapping, struct folio *folio);
132 int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
133 bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
134 		loff_t end);
135 long invalidate_inode_page(struct page *page);
136 unsigned long mapping_try_invalidate(struct address_space *mapping,
137 		pgoff_t start, pgoff_t end, unsigned long *nr_failed);
138 
139 /**
140  * folio_evictable - Test whether a folio is evictable.
141  * @folio: The folio to test.
142  *
143  * Test whether @folio is evictable -- i.e., should be placed on
144  * active/inactive lists vs unevictable list.
145  *
146  * Reasons folio might not be evictable:
147  * 1. folio's mapping marked unevictable
148  * 2. One of the pages in the folio is part of an mlocked VMA
149  */
150 static inline bool folio_evictable(struct folio *folio)
151 {
152 	bool ret;
153 
154 	/* Prevent address_space of inode and swap cache from being freed */
155 	rcu_read_lock();
156 	ret = !mapping_unevictable(folio_mapping(folio)) &&
157 			!folio_test_mlocked(folio);
158 	rcu_read_unlock();
159 	return ret;
160 }
161 
162 /*
163  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
164  * a count of one.
165  */
166 static inline void set_page_refcounted(struct page *page)
167 {
168 	VM_BUG_ON_PAGE(PageTail(page), page);
169 	VM_BUG_ON_PAGE(page_ref_count(page), page);
170 	set_page_count(page, 1);
171 }
172 
173 extern unsigned long highest_memmap_pfn;
174 
175 /*
176  * Maximum number of reclaim retries without progress before the OOM
177  * killer is consider the only way forward.
178  */
179 #define MAX_RECLAIM_RETRIES 16
180 
181 /*
182  * in mm/vmscan.c:
183  */
184 bool isolate_lru_page(struct page *page);
185 bool folio_isolate_lru(struct folio *folio);
186 void putback_lru_page(struct page *page);
187 void folio_putback_lru(struct folio *folio);
188 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
189 
190 /*
191  * in mm/rmap.c:
192  */
193 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
194 
195 /*
196  * in mm/page_alloc.c
197  */
198 #define K(x) ((x) << (PAGE_SHIFT-10))
199 
200 extern char * const zone_names[MAX_NR_ZONES];
201 
202 /* perform sanity checks on struct pages being allocated or freed */
203 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
204 
205 extern int min_free_kbytes;
206 
207 void setup_per_zone_wmarks(void);
208 void calculate_min_free_kbytes(void);
209 int __meminit init_per_zone_wmark_min(void);
210 void page_alloc_sysctl_init(void);
211 
212 /*
213  * Structure for holding the mostly immutable allocation parameters passed
214  * between functions involved in allocations, including the alloc_pages*
215  * family of functions.
216  *
217  * nodemask, migratetype and highest_zoneidx are initialized only once in
218  * __alloc_pages() and then never change.
219  *
220  * zonelist, preferred_zone and highest_zoneidx are set first in
221  * __alloc_pages() for the fast path, and might be later changed
222  * in __alloc_pages_slowpath(). All other functions pass the whole structure
223  * by a const pointer.
224  */
225 struct alloc_context {
226 	struct zonelist *zonelist;
227 	nodemask_t *nodemask;
228 	struct zoneref *preferred_zoneref;
229 	int migratetype;
230 
231 	/*
232 	 * highest_zoneidx represents highest usable zone index of
233 	 * the allocation request. Due to the nature of the zone,
234 	 * memory on lower zone than the highest_zoneidx will be
235 	 * protected by lowmem_reserve[highest_zoneidx].
236 	 *
237 	 * highest_zoneidx is also used by reclaim/compaction to limit
238 	 * the target zone since higher zone than this index cannot be
239 	 * usable for this allocation request.
240 	 */
241 	enum zone_type highest_zoneidx;
242 	bool spread_dirty_pages;
243 };
244 
245 /*
246  * This function returns the order of a free page in the buddy system. In
247  * general, page_zone(page)->lock must be held by the caller to prevent the
248  * page from being allocated in parallel and returning garbage as the order.
249  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
250  * page cannot be allocated or merged in parallel. Alternatively, it must
251  * handle invalid values gracefully, and use buddy_order_unsafe() below.
252  */
253 static inline unsigned int buddy_order(struct page *page)
254 {
255 	/* PageBuddy() must be checked by the caller */
256 	return page_private(page);
257 }
258 
259 /*
260  * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
261  * PageBuddy() should be checked first by the caller to minimize race window,
262  * and invalid values must be handled gracefully.
263  *
264  * READ_ONCE is used so that if the caller assigns the result into a local
265  * variable and e.g. tests it for valid range before using, the compiler cannot
266  * decide to remove the variable and inline the page_private(page) multiple
267  * times, potentially observing different values in the tests and the actual
268  * use of the result.
269  */
270 #define buddy_order_unsafe(page)	READ_ONCE(page_private(page))
271 
272 /*
273  * This function checks whether a page is free && is the buddy
274  * we can coalesce a page and its buddy if
275  * (a) the buddy is not in a hole (check before calling!) &&
276  * (b) the buddy is in the buddy system &&
277  * (c) a page and its buddy have the same order &&
278  * (d) a page and its buddy are in the same zone.
279  *
280  * For recording whether a page is in the buddy system, we set PageBuddy.
281  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
282  *
283  * For recording page's order, we use page_private(page).
284  */
285 static inline bool page_is_buddy(struct page *page, struct page *buddy,
286 				 unsigned int order)
287 {
288 	if (!page_is_guard(buddy) && !PageBuddy(buddy))
289 		return false;
290 
291 	if (buddy_order(buddy) != order)
292 		return false;
293 
294 	/*
295 	 * zone check is done late to avoid uselessly calculating
296 	 * zone/node ids for pages that could never merge.
297 	 */
298 	if (page_zone_id(page) != page_zone_id(buddy))
299 		return false;
300 
301 	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
302 
303 	return true;
304 }
305 
306 /*
307  * Locate the struct page for both the matching buddy in our
308  * pair (buddy1) and the combined O(n+1) page they form (page).
309  *
310  * 1) Any buddy B1 will have an order O twin B2 which satisfies
311  * the following equation:
312  *     B2 = B1 ^ (1 << O)
313  * For example, if the starting buddy (buddy2) is #8 its order
314  * 1 buddy is #10:
315  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
316  *
317  * 2) Any buddy B will have an order O+1 parent P which
318  * satisfies the following equation:
319  *     P = B & ~(1 << O)
320  *
321  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
322  */
323 static inline unsigned long
324 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
325 {
326 	return page_pfn ^ (1 << order);
327 }
328 
329 /*
330  * Find the buddy of @page and validate it.
331  * @page: The input page
332  * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
333  *       function is used in the performance-critical __free_one_page().
334  * @order: The order of the page
335  * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
336  *             page_to_pfn().
337  *
338  * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
339  * not the same as @page. The validation is necessary before use it.
340  *
341  * Return: the found buddy page or NULL if not found.
342  */
343 static inline struct page *find_buddy_page_pfn(struct page *page,
344 			unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
345 {
346 	unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
347 	struct page *buddy;
348 
349 	buddy = page + (__buddy_pfn - pfn);
350 	if (buddy_pfn)
351 		*buddy_pfn = __buddy_pfn;
352 
353 	if (page_is_buddy(page, buddy, order))
354 		return buddy;
355 	return NULL;
356 }
357 
358 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
359 				unsigned long end_pfn, struct zone *zone);
360 
361 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
362 				unsigned long end_pfn, struct zone *zone)
363 {
364 	if (zone->contiguous)
365 		return pfn_to_page(start_pfn);
366 
367 	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
368 }
369 
370 void set_zone_contiguous(struct zone *zone);
371 
372 static inline void clear_zone_contiguous(struct zone *zone)
373 {
374 	zone->contiguous = false;
375 }
376 
377 extern int __isolate_free_page(struct page *page, unsigned int order);
378 extern void __putback_isolated_page(struct page *page, unsigned int order,
379 				    int mt);
380 extern void memblock_free_pages(struct page *page, unsigned long pfn,
381 					unsigned int order);
382 extern void __free_pages_core(struct page *page, unsigned int order);
383 
384 /*
385  * This will have no effect, other than possibly generating a warning, if the
386  * caller passes in a non-large folio.
387  */
388 static inline void folio_set_order(struct folio *folio, unsigned int order)
389 {
390 	if (WARN_ON_ONCE(!order || !folio_test_large(folio)))
391 		return;
392 
393 	folio->_folio_order = order;
394 #ifdef CONFIG_64BIT
395 	folio->_folio_nr_pages = 1U << order;
396 #endif
397 }
398 
399 static inline void prep_compound_head(struct page *page, unsigned int order)
400 {
401 	struct folio *folio = (struct folio *)page;
402 
403 	folio_set_compound_dtor(folio, COMPOUND_PAGE_DTOR);
404 	folio_set_order(folio, order);
405 	atomic_set(&folio->_entire_mapcount, -1);
406 	atomic_set(&folio->_nr_pages_mapped, 0);
407 	atomic_set(&folio->_pincount, 0);
408 }
409 
410 static inline void prep_compound_tail(struct page *head, int tail_idx)
411 {
412 	struct page *p = head + tail_idx;
413 
414 	p->mapping = TAIL_MAPPING;
415 	set_compound_head(p, head);
416 	set_page_private(p, 0);
417 }
418 
419 extern void prep_compound_page(struct page *page, unsigned int order);
420 
421 extern void post_alloc_hook(struct page *page, unsigned int order,
422 					gfp_t gfp_flags);
423 extern int user_min_free_kbytes;
424 
425 extern void free_unref_page(struct page *page, unsigned int order);
426 extern void free_unref_page_list(struct list_head *list);
427 
428 extern void zone_pcp_reset(struct zone *zone);
429 extern void zone_pcp_disable(struct zone *zone);
430 extern void zone_pcp_enable(struct zone *zone);
431 extern void zone_pcp_init(struct zone *zone);
432 
433 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
434 			  phys_addr_t min_addr,
435 			  int nid, bool exact_nid);
436 
437 void memmap_init_range(unsigned long, int, unsigned long, unsigned long,
438 		unsigned long, enum meminit_context, struct vmem_altmap *, int);
439 
440 
441 int split_free_page(struct page *free_page,
442 			unsigned int order, unsigned long split_pfn_offset);
443 
444 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
445 
446 /*
447  * in mm/compaction.c
448  */
449 /*
450  * compact_control is used to track pages being migrated and the free pages
451  * they are being migrated to during memory compaction. The free_pfn starts
452  * at the end of a zone and migrate_pfn begins at the start. Movable pages
453  * are moved to the end of a zone during a compaction run and the run
454  * completes when free_pfn <= migrate_pfn
455  */
456 struct compact_control {
457 	struct list_head freepages;	/* List of free pages to migrate to */
458 	struct list_head migratepages;	/* List of pages being migrated */
459 	unsigned int nr_freepages;	/* Number of isolated free pages */
460 	unsigned int nr_migratepages;	/* Number of pages to migrate */
461 	unsigned long free_pfn;		/* isolate_freepages search base */
462 	/*
463 	 * Acts as an in/out parameter to page isolation for migration.
464 	 * isolate_migratepages uses it as a search base.
465 	 * isolate_migratepages_block will update the value to the next pfn
466 	 * after the last isolated one.
467 	 */
468 	unsigned long migrate_pfn;
469 	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
470 	struct zone *zone;
471 	unsigned long total_migrate_scanned;
472 	unsigned long total_free_scanned;
473 	unsigned short fast_search_fail;/* failures to use free list searches */
474 	short search_order;		/* order to start a fast search at */
475 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
476 	int order;			/* order a direct compactor needs */
477 	int migratetype;		/* migratetype of direct compactor */
478 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
479 	const int highest_zoneidx;	/* zone index of a direct compactor */
480 	enum migrate_mode mode;		/* Async or sync migration mode */
481 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
482 	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
483 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
484 	bool direct_compaction;		/* False from kcompactd or /proc/... */
485 	bool proactive_compaction;	/* kcompactd proactive compaction */
486 	bool whole_zone;		/* Whole zone should/has been scanned */
487 	bool contended;			/* Signal lock contention */
488 	bool finish_pageblock;		/* Scan the remainder of a pageblock. Used
489 					 * when there are potentially transient
490 					 * isolation or migration failures to
491 					 * ensure forward progress.
492 					 */
493 	bool alloc_contig;		/* alloc_contig_range allocation */
494 };
495 
496 /*
497  * Used in direct compaction when a page should be taken from the freelists
498  * immediately when one is created during the free path.
499  */
500 struct capture_control {
501 	struct compact_control *cc;
502 	struct page *page;
503 };
504 
505 unsigned long
506 isolate_freepages_range(struct compact_control *cc,
507 			unsigned long start_pfn, unsigned long end_pfn);
508 int
509 isolate_migratepages_range(struct compact_control *cc,
510 			   unsigned long low_pfn, unsigned long end_pfn);
511 
512 int __alloc_contig_migrate_range(struct compact_control *cc,
513 					unsigned long start, unsigned long end);
514 
515 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
516 void init_cma_reserved_pageblock(struct page *page);
517 
518 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
519 
520 int find_suitable_fallback(struct free_area *area, unsigned int order,
521 			int migratetype, bool only_stealable, bool *can_steal);
522 
523 static inline bool free_area_empty(struct free_area *area, int migratetype)
524 {
525 	return list_empty(&area->free_list[migratetype]);
526 }
527 
528 /*
529  * These three helpers classifies VMAs for virtual memory accounting.
530  */
531 
532 /*
533  * Executable code area - executable, not writable, not stack
534  */
535 static inline bool is_exec_mapping(vm_flags_t flags)
536 {
537 	return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
538 }
539 
540 /*
541  * Stack area - automatically grows in one direction
542  *
543  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
544  * do_mmap() forbids all other combinations.
545  */
546 static inline bool is_stack_mapping(vm_flags_t flags)
547 {
548 	return (flags & VM_STACK) == VM_STACK;
549 }
550 
551 /*
552  * Data area - private, writable, not stack
553  */
554 static inline bool is_data_mapping(vm_flags_t flags)
555 {
556 	return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
557 }
558 
559 /* mm/util.c */
560 struct anon_vma *folio_anon_vma(struct folio *folio);
561 
562 #ifdef CONFIG_MMU
563 void unmap_mapping_folio(struct folio *folio);
564 extern long populate_vma_page_range(struct vm_area_struct *vma,
565 		unsigned long start, unsigned long end, int *locked);
566 extern long faultin_vma_page_range(struct vm_area_struct *vma,
567 				   unsigned long start, unsigned long end,
568 				   bool write, int *locked);
569 extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags,
570 			       unsigned long bytes);
571 /*
572  * mlock_vma_folio() and munlock_vma_folio():
573  * should be called with vma's mmap_lock held for read or write,
574  * under page table lock for the pte/pmd being added or removed.
575  *
576  * mlock is usually called at the end of page_add_*_rmap(), munlock at
577  * the end of page_remove_rmap(); but new anon folios are managed by
578  * folio_add_lru_vma() calling mlock_new_folio().
579  *
580  * @compound is used to include pmd mappings of THPs, but filter out
581  * pte mappings of THPs, which cannot be consistently counted: a pte
582  * mapping of the THP head cannot be distinguished by the page alone.
583  */
584 void mlock_folio(struct folio *folio);
585 static inline void mlock_vma_folio(struct folio *folio,
586 			struct vm_area_struct *vma, bool compound)
587 {
588 	/*
589 	 * The VM_SPECIAL check here serves two purposes.
590 	 * 1) VM_IO check prevents migration from double-counting during mlock.
591 	 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
592 	 *    is never left set on a VM_SPECIAL vma, there is an interval while
593 	 *    file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
594 	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
595 	 */
596 	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) &&
597 	    (compound || !folio_test_large(folio)))
598 		mlock_folio(folio);
599 }
600 
601 void munlock_folio(struct folio *folio);
602 static inline void munlock_vma_folio(struct folio *folio,
603 			struct vm_area_struct *vma, bool compound)
604 {
605 	if (unlikely(vma->vm_flags & VM_LOCKED) &&
606 	    (compound || !folio_test_large(folio)))
607 		munlock_folio(folio);
608 }
609 
610 void mlock_new_folio(struct folio *folio);
611 bool need_mlock_drain(int cpu);
612 void mlock_drain_local(void);
613 void mlock_drain_remote(int cpu);
614 
615 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
616 
617 /*
618  * Return the start of user virtual address at the specific offset within
619  * a vma.
620  */
621 static inline unsigned long
622 vma_pgoff_address(pgoff_t pgoff, unsigned long nr_pages,
623 		  struct vm_area_struct *vma)
624 {
625 	unsigned long address;
626 
627 	if (pgoff >= vma->vm_pgoff) {
628 		address = vma->vm_start +
629 			((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
630 		/* Check for address beyond vma (or wrapped through 0?) */
631 		if (address < vma->vm_start || address >= vma->vm_end)
632 			address = -EFAULT;
633 	} else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
634 		/* Test above avoids possibility of wrap to 0 on 32-bit */
635 		address = vma->vm_start;
636 	} else {
637 		address = -EFAULT;
638 	}
639 	return address;
640 }
641 
642 /*
643  * Return the start of user virtual address of a page within a vma.
644  * Returns -EFAULT if all of the page is outside the range of vma.
645  * If page is a compound head, the entire compound page is considered.
646  */
647 static inline unsigned long
648 vma_address(struct page *page, struct vm_area_struct *vma)
649 {
650 	VM_BUG_ON_PAGE(PageKsm(page), page);	/* KSM page->index unusable */
651 	return vma_pgoff_address(page_to_pgoff(page), compound_nr(page), vma);
652 }
653 
654 /*
655  * Then at what user virtual address will none of the range be found in vma?
656  * Assumes that vma_address() already returned a good starting address.
657  */
658 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
659 {
660 	struct vm_area_struct *vma = pvmw->vma;
661 	pgoff_t pgoff;
662 	unsigned long address;
663 
664 	/* Common case, plus ->pgoff is invalid for KSM */
665 	if (pvmw->nr_pages == 1)
666 		return pvmw->address + PAGE_SIZE;
667 
668 	pgoff = pvmw->pgoff + pvmw->nr_pages;
669 	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
670 	/* Check for address beyond vma (or wrapped through 0?) */
671 	if (address < vma->vm_start || address > vma->vm_end)
672 		address = vma->vm_end;
673 	return address;
674 }
675 
676 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
677 						    struct file *fpin)
678 {
679 	int flags = vmf->flags;
680 
681 	if (fpin)
682 		return fpin;
683 
684 	/*
685 	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
686 	 * anything, so we only pin the file and drop the mmap_lock if only
687 	 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
688 	 */
689 	if (fault_flag_allow_retry_first(flags) &&
690 	    !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
691 		fpin = get_file(vmf->vma->vm_file);
692 		mmap_read_unlock(vmf->vma->vm_mm);
693 	}
694 	return fpin;
695 }
696 #else /* !CONFIG_MMU */
697 static inline void unmap_mapping_folio(struct folio *folio) { }
698 static inline void mlock_new_folio(struct folio *folio) { }
699 static inline bool need_mlock_drain(int cpu) { return false; }
700 static inline void mlock_drain_local(void) { }
701 static inline void mlock_drain_remote(int cpu) { }
702 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
703 {
704 }
705 #endif /* !CONFIG_MMU */
706 
707 /* Memory initialisation debug and verification */
708 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
709 DECLARE_STATIC_KEY_TRUE(deferred_pages);
710 
711 bool __init deferred_grow_zone(struct zone *zone, unsigned int order);
712 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
713 
714 enum mminit_level {
715 	MMINIT_WARNING,
716 	MMINIT_VERIFY,
717 	MMINIT_TRACE
718 };
719 
720 #ifdef CONFIG_DEBUG_MEMORY_INIT
721 
722 extern int mminit_loglevel;
723 
724 #define mminit_dprintk(level, prefix, fmt, arg...) \
725 do { \
726 	if (level < mminit_loglevel) { \
727 		if (level <= MMINIT_WARNING) \
728 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
729 		else \
730 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
731 	} \
732 } while (0)
733 
734 extern void mminit_verify_pageflags_layout(void);
735 extern void mminit_verify_zonelist(void);
736 #else
737 
738 static inline void mminit_dprintk(enum mminit_level level,
739 				const char *prefix, const char *fmt, ...)
740 {
741 }
742 
743 static inline void mminit_verify_pageflags_layout(void)
744 {
745 }
746 
747 static inline void mminit_verify_zonelist(void)
748 {
749 }
750 #endif /* CONFIG_DEBUG_MEMORY_INIT */
751 
752 #define NODE_RECLAIM_NOSCAN	-2
753 #define NODE_RECLAIM_FULL	-1
754 #define NODE_RECLAIM_SOME	0
755 #define NODE_RECLAIM_SUCCESS	1
756 
757 #ifdef CONFIG_NUMA
758 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
759 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
760 #else
761 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
762 				unsigned int order)
763 {
764 	return NODE_RECLAIM_NOSCAN;
765 }
766 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
767 {
768 	return NUMA_NO_NODE;
769 }
770 #endif
771 
772 /*
773  * mm/memory-failure.c
774  */
775 extern int hwpoison_filter(struct page *p);
776 
777 extern u32 hwpoison_filter_dev_major;
778 extern u32 hwpoison_filter_dev_minor;
779 extern u64 hwpoison_filter_flags_mask;
780 extern u64 hwpoison_filter_flags_value;
781 extern u64 hwpoison_filter_memcg;
782 extern u32 hwpoison_filter_enable;
783 
784 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
785         unsigned long, unsigned long,
786         unsigned long, unsigned long);
787 
788 extern void set_pageblock_order(void);
789 unsigned long reclaim_pages(struct list_head *folio_list);
790 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
791 					    struct list_head *folio_list);
792 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
793 #define ALLOC_WMARK_MIN		WMARK_MIN
794 #define ALLOC_WMARK_LOW		WMARK_LOW
795 #define ALLOC_WMARK_HIGH	WMARK_HIGH
796 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
797 
798 /* Mask to get the watermark bits */
799 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
800 
801 /*
802  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
803  * cannot assume a reduced access to memory reserves is sufficient for
804  * !MMU
805  */
806 #ifdef CONFIG_MMU
807 #define ALLOC_OOM		0x08
808 #else
809 #define ALLOC_OOM		ALLOC_NO_WATERMARKS
810 #endif
811 
812 #define ALLOC_NON_BLOCK		 0x10 /* Caller cannot block. Allow access
813 				       * to 25% of the min watermark or
814 				       * 62.5% if __GFP_HIGH is set.
815 				       */
816 #define ALLOC_MIN_RESERVE	 0x20 /* __GFP_HIGH set. Allow access to 50%
817 				       * of the min watermark.
818 				       */
819 #define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
820 #define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
821 #ifdef CONFIG_ZONE_DMA32
822 #define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
823 #else
824 #define ALLOC_NOFRAGMENT	  0x0
825 #endif
826 #define ALLOC_HIGHATOMIC	0x200 /* Allows access to MIGRATE_HIGHATOMIC */
827 #define ALLOC_KSWAPD		0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
828 
829 /* Flags that allow allocations below the min watermark. */
830 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
831 
832 enum ttu_flags;
833 struct tlbflush_unmap_batch;
834 
835 
836 /*
837  * only for MM internal work items which do not depend on
838  * any allocations or locks which might depend on allocations
839  */
840 extern struct workqueue_struct *mm_percpu_wq;
841 
842 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
843 void try_to_unmap_flush(void);
844 void try_to_unmap_flush_dirty(void);
845 void flush_tlb_batched_pending(struct mm_struct *mm);
846 #else
847 static inline void try_to_unmap_flush(void)
848 {
849 }
850 static inline void try_to_unmap_flush_dirty(void)
851 {
852 }
853 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
854 {
855 }
856 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
857 
858 extern const struct trace_print_flags pageflag_names[];
859 extern const struct trace_print_flags pagetype_names[];
860 extern const struct trace_print_flags vmaflag_names[];
861 extern const struct trace_print_flags gfpflag_names[];
862 
863 static inline bool is_migrate_highatomic(enum migratetype migratetype)
864 {
865 	return migratetype == MIGRATE_HIGHATOMIC;
866 }
867 
868 static inline bool is_migrate_highatomic_page(struct page *page)
869 {
870 	return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
871 }
872 
873 void setup_zone_pageset(struct zone *zone);
874 
875 struct migration_target_control {
876 	int nid;		/* preferred node id */
877 	nodemask_t *nmask;
878 	gfp_t gfp_mask;
879 };
880 
881 /*
882  * mm/filemap.c
883  */
884 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
885 			      struct folio *folio, loff_t fpos, size_t size);
886 
887 /*
888  * mm/vmalloc.c
889  */
890 #ifdef CONFIG_MMU
891 void __init vmalloc_init(void);
892 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
893                 pgprot_t prot, struct page **pages, unsigned int page_shift);
894 #else
895 static inline void vmalloc_init(void)
896 {
897 }
898 
899 static inline
900 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
901                 pgprot_t prot, struct page **pages, unsigned int page_shift)
902 {
903 	return -EINVAL;
904 }
905 #endif
906 
907 int __must_check __vmap_pages_range_noflush(unsigned long addr,
908 			       unsigned long end, pgprot_t prot,
909 			       struct page **pages, unsigned int page_shift);
910 
911 void vunmap_range_noflush(unsigned long start, unsigned long end);
912 
913 void __vunmap_range_noflush(unsigned long start, unsigned long end);
914 
915 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
916 		      unsigned long addr, int page_nid, int *flags);
917 
918 void free_zone_device_page(struct page *page);
919 int migrate_device_coherent_page(struct page *page);
920 
921 /*
922  * mm/gup.c
923  */
924 struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags);
925 int __must_check try_grab_page(struct page *page, unsigned int flags);
926 
927 /*
928  * mm/huge_memory.c
929  */
930 struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
931 				   unsigned long addr, pmd_t *pmd,
932 				   unsigned int flags);
933 
934 enum {
935 	/* mark page accessed */
936 	FOLL_TOUCH = 1 << 16,
937 	/* a retry, previous pass started an IO */
938 	FOLL_TRIED = 1 << 17,
939 	/* we are working on non-current tsk/mm */
940 	FOLL_REMOTE = 1 << 18,
941 	/* pages must be released via unpin_user_page */
942 	FOLL_PIN = 1 << 19,
943 	/* gup_fast: prevent fall-back to slow gup */
944 	FOLL_FAST_ONLY = 1 << 20,
945 	/* allow unlocking the mmap lock */
946 	FOLL_UNLOCKABLE = 1 << 21,
947 };
948 
949 /*
950  * Indicates for which pages that are write-protected in the page table,
951  * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
952  * GUP pin will remain consistent with the pages mapped into the page tables
953  * of the MM.
954  *
955  * Temporary unmapping of PageAnonExclusive() pages or clearing of
956  * PageAnonExclusive() has to protect against concurrent GUP:
957  * * Ordinary GUP: Using the PT lock
958  * * GUP-fast and fork(): mm->write_protect_seq
959  * * GUP-fast and KSM or temporary unmapping (swap, migration): see
960  *    page_try_share_anon_rmap()
961  *
962  * Must be called with the (sub)page that's actually referenced via the
963  * page table entry, which might not necessarily be the head page for a
964  * PTE-mapped THP.
965  *
966  * If the vma is NULL, we're coming from the GUP-fast path and might have
967  * to fallback to the slow path just to lookup the vma.
968  */
969 static inline bool gup_must_unshare(struct vm_area_struct *vma,
970 				    unsigned int flags, struct page *page)
971 {
972 	/*
973 	 * FOLL_WRITE is implicitly handled correctly as the page table entry
974 	 * has to be writable -- and if it references (part of) an anonymous
975 	 * folio, that part is required to be marked exclusive.
976 	 */
977 	if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
978 		return false;
979 	/*
980 	 * Note: PageAnon(page) is stable until the page is actually getting
981 	 * freed.
982 	 */
983 	if (!PageAnon(page)) {
984 		/*
985 		 * We only care about R/O long-term pining: R/O short-term
986 		 * pinning does not have the semantics to observe successive
987 		 * changes through the process page tables.
988 		 */
989 		if (!(flags & FOLL_LONGTERM))
990 			return false;
991 
992 		/* We really need the vma ... */
993 		if (!vma)
994 			return true;
995 
996 		/*
997 		 * ... because we only care about writable private ("COW")
998 		 * mappings where we have to break COW early.
999 		 */
1000 		return is_cow_mapping(vma->vm_flags);
1001 	}
1002 
1003 	/* Paired with a memory barrier in page_try_share_anon_rmap(). */
1004 	if (IS_ENABLED(CONFIG_HAVE_FAST_GUP))
1005 		smp_rmb();
1006 
1007 	/*
1008 	 * During GUP-fast we might not get called on the head page for a
1009 	 * hugetlb page that is mapped using cont-PTE, because GUP-fast does
1010 	 * not work with the abstracted hugetlb PTEs that always point at the
1011 	 * head page. For hugetlb, PageAnonExclusive only applies on the head
1012 	 * page (as it cannot be partially COW-shared), so lookup the head page.
1013 	 */
1014 	if (unlikely(!PageHead(page) && PageHuge(page)))
1015 		page = compound_head(page);
1016 
1017 	/*
1018 	 * Note that PageKsm() pages cannot be exclusive, and consequently,
1019 	 * cannot get pinned.
1020 	 */
1021 	return !PageAnonExclusive(page);
1022 }
1023 
1024 extern bool mirrored_kernelcore;
1025 
1026 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1027 {
1028 	/*
1029 	 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1030 	 * enablements, because when without soft-dirty being compiled in,
1031 	 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1032 	 * will be constantly true.
1033 	 */
1034 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
1035 		return false;
1036 
1037 	/*
1038 	 * Soft-dirty is kind of special: its tracking is enabled when the
1039 	 * vma flags not set.
1040 	 */
1041 	return !(vma->vm_flags & VM_SOFTDIRTY);
1042 }
1043 
1044 /*
1045  * VMA Iterator functions shared between nommu and mmap
1046  */
1047 static inline int vma_iter_prealloc(struct vma_iterator *vmi)
1048 {
1049 	return mas_preallocate(&vmi->mas, GFP_KERNEL);
1050 }
1051 
1052 static inline void vma_iter_clear(struct vma_iterator *vmi,
1053 				  unsigned long start, unsigned long end)
1054 {
1055 	mas_set_range(&vmi->mas, start, end - 1);
1056 	mas_store_prealloc(&vmi->mas, NULL);
1057 }
1058 
1059 static inline struct vm_area_struct *vma_iter_load(struct vma_iterator *vmi)
1060 {
1061 	return mas_walk(&vmi->mas);
1062 }
1063 
1064 /* Store a VMA with preallocated memory */
1065 static inline void vma_iter_store(struct vma_iterator *vmi,
1066 				  struct vm_area_struct *vma)
1067 {
1068 
1069 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
1070 	if (MAS_WARN_ON(&vmi->mas, vmi->mas.node != MAS_START &&
1071 			vmi->mas.index > vma->vm_start)) {
1072 		pr_warn("%lx > %lx\n store vma %lx-%lx\n into slot %lx-%lx\n",
1073 			vmi->mas.index, vma->vm_start, vma->vm_start,
1074 			vma->vm_end, vmi->mas.index, vmi->mas.last);
1075 	}
1076 	if (MAS_WARN_ON(&vmi->mas, vmi->mas.node != MAS_START &&
1077 			vmi->mas.last <  vma->vm_start)) {
1078 		pr_warn("%lx < %lx\nstore vma %lx-%lx\ninto slot %lx-%lx\n",
1079 		       vmi->mas.last, vma->vm_start, vma->vm_start, vma->vm_end,
1080 		       vmi->mas.index, vmi->mas.last);
1081 	}
1082 #endif
1083 
1084 	if (vmi->mas.node != MAS_START &&
1085 	    ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
1086 		vma_iter_invalidate(vmi);
1087 
1088 	vmi->mas.index = vma->vm_start;
1089 	vmi->mas.last = vma->vm_end - 1;
1090 	mas_store_prealloc(&vmi->mas, vma);
1091 }
1092 
1093 static inline int vma_iter_store_gfp(struct vma_iterator *vmi,
1094 			struct vm_area_struct *vma, gfp_t gfp)
1095 {
1096 	if (vmi->mas.node != MAS_START &&
1097 	    ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
1098 		vma_iter_invalidate(vmi);
1099 
1100 	vmi->mas.index = vma->vm_start;
1101 	vmi->mas.last = vma->vm_end - 1;
1102 	mas_store_gfp(&vmi->mas, vma, gfp);
1103 	if (unlikely(mas_is_err(&vmi->mas)))
1104 		return -ENOMEM;
1105 
1106 	return 0;
1107 }
1108 
1109 /*
1110  * VMA lock generalization
1111  */
1112 struct vma_prepare {
1113 	struct vm_area_struct *vma;
1114 	struct vm_area_struct *adj_next;
1115 	struct file *file;
1116 	struct address_space *mapping;
1117 	struct anon_vma *anon_vma;
1118 	struct vm_area_struct *insert;
1119 	struct vm_area_struct *remove;
1120 	struct vm_area_struct *remove2;
1121 };
1122 #endif	/* __MM_INTERNAL_H */
1123