xref: /linux/mm/internal.h (revision 49b1b8d6f6831026cb105b0eafa18f13db612d86)
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/khugepaged.h>
12 #include <linux/mm.h>
13 #include <linux/mm_inline.h>
14 #include <linux/pagemap.h>
15 #include <linux/rmap.h>
16 #include <linux/swap.h>
17 #include <linux/swapops.h>
18 #include <linux/tracepoint-defs.h>
19 
20 /* Internal core VMA manipulation functions. */
21 #include "vma.h"
22 
23 struct folio_batch;
24 
25 /*
26  * The set of flags that only affect watermark checking and reclaim
27  * behaviour. This is used by the MM to obey the caller constraints
28  * about IO, FS and watermark checking while ignoring placement
29  * hints such as HIGHMEM usage.
30  */
31 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
32 			__GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
33 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
34 			__GFP_NOLOCKDEP)
35 
36 /* The GFP flags allowed during early boot */
37 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
38 
39 /* Control allocation cpuset and node placement constraints */
40 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
41 
42 /* Do not use these with a slab allocator */
43 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
44 
45 /*
46  * Different from WARN_ON_ONCE(), no warning will be issued
47  * when we specify __GFP_NOWARN.
48  */
49 #define WARN_ON_ONCE_GFP(cond, gfp)	({				\
50 	static bool __section(".data.once") __warned;			\
51 	int __ret_warn_once = !!(cond);					\
52 									\
53 	if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
54 		__warned = true;					\
55 		WARN_ON(1);						\
56 	}								\
57 	unlikely(__ret_warn_once);					\
58 })
59 
60 void page_writeback_init(void);
61 
62 /*
63  * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
64  * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit
65  * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE).  Hugetlb currently
66  * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
67  */
68 #define ENTIRELY_MAPPED		0x800000
69 #define FOLIO_PAGES_MAPPED	(ENTIRELY_MAPPED - 1)
70 
71 /*
72  * Flags passed to __show_mem() and show_free_areas() to suppress output in
73  * various contexts.
74  */
75 #define SHOW_MEM_FILTER_NODES		(0x0001u)	/* disallowed nodes */
76 
77 /*
78  * How many individual pages have an elevated _mapcount.  Excludes
79  * the folio's entire_mapcount.
80  *
81  * Don't use this function outside of debugging code.
82  */
83 static inline int folio_nr_pages_mapped(const struct folio *folio)
84 {
85 	return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
86 }
87 
88 /*
89  * Retrieve the first entry of a folio based on a provided entry within the
90  * folio. We cannot rely on folio->swap as there is no guarantee that it has
91  * been initialized. Used for calling arch_swap_restore()
92  */
93 static inline swp_entry_t folio_swap(swp_entry_t entry,
94 		const struct folio *folio)
95 {
96 	swp_entry_t swap = {
97 		.val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)),
98 	};
99 
100 	return swap;
101 }
102 
103 static inline void *folio_raw_mapping(const struct folio *folio)
104 {
105 	unsigned long mapping = (unsigned long)folio->mapping;
106 
107 	return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
108 }
109 
110 #ifdef CONFIG_MMU
111 
112 /* Flags for folio_pte_batch(). */
113 typedef int __bitwise fpb_t;
114 
115 /* Compare PTEs after pte_mkclean(), ignoring the dirty bit. */
116 #define FPB_IGNORE_DIRTY		((__force fpb_t)BIT(0))
117 
118 /* Compare PTEs after pte_clear_soft_dirty(), ignoring the soft-dirty bit. */
119 #define FPB_IGNORE_SOFT_DIRTY		((__force fpb_t)BIT(1))
120 
121 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
122 {
123 	if (flags & FPB_IGNORE_DIRTY)
124 		pte = pte_mkclean(pte);
125 	if (likely(flags & FPB_IGNORE_SOFT_DIRTY))
126 		pte = pte_clear_soft_dirty(pte);
127 	return pte_wrprotect(pte_mkold(pte));
128 }
129 
130 /**
131  * folio_pte_batch - detect a PTE batch for a large folio
132  * @folio: The large folio to detect a PTE batch for.
133  * @addr: The user virtual address the first page is mapped at.
134  * @start_ptep: Page table pointer for the first entry.
135  * @pte: Page table entry for the first page.
136  * @max_nr: The maximum number of table entries to consider.
137  * @flags: Flags to modify the PTE batch semantics.
138  * @any_writable: Optional pointer to indicate whether any entry except the
139  *		  first one is writable.
140  * @any_young: Optional pointer to indicate whether any entry except the
141  *		  first one is young.
142  * @any_dirty: Optional pointer to indicate whether any entry except the
143  *		  first one is dirty.
144  *
145  * Detect a PTE batch: consecutive (present) PTEs that map consecutive
146  * pages of the same large folio.
147  *
148  * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN,
149  * the accessed bit, writable bit, dirty bit (with FPB_IGNORE_DIRTY) and
150  * soft-dirty bit (with FPB_IGNORE_SOFT_DIRTY).
151  *
152  * start_ptep must map any page of the folio. max_nr must be at least one and
153  * must be limited by the caller so scanning cannot exceed a single page table.
154  *
155  * Return: the number of table entries in the batch.
156  */
157 static inline int folio_pte_batch(struct folio *folio, unsigned long addr,
158 		pte_t *start_ptep, pte_t pte, int max_nr, fpb_t flags,
159 		bool *any_writable, bool *any_young, bool *any_dirty)
160 {
161 	unsigned long folio_end_pfn = folio_pfn(folio) + folio_nr_pages(folio);
162 	const pte_t *end_ptep = start_ptep + max_nr;
163 	pte_t expected_pte, *ptep;
164 	bool writable, young, dirty;
165 	int nr;
166 
167 	if (any_writable)
168 		*any_writable = false;
169 	if (any_young)
170 		*any_young = false;
171 	if (any_dirty)
172 		*any_dirty = false;
173 
174 	VM_WARN_ON_FOLIO(!pte_present(pte), folio);
175 	VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio);
176 	VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio);
177 
178 	nr = pte_batch_hint(start_ptep, pte);
179 	expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags);
180 	ptep = start_ptep + nr;
181 
182 	while (ptep < end_ptep) {
183 		pte = ptep_get(ptep);
184 		if (any_writable)
185 			writable = !!pte_write(pte);
186 		if (any_young)
187 			young = !!pte_young(pte);
188 		if (any_dirty)
189 			dirty = !!pte_dirty(pte);
190 		pte = __pte_batch_clear_ignored(pte, flags);
191 
192 		if (!pte_same(pte, expected_pte))
193 			break;
194 
195 		/*
196 		 * Stop immediately once we reached the end of the folio. In
197 		 * corner cases the next PFN might fall into a different
198 		 * folio.
199 		 */
200 		if (pte_pfn(pte) >= folio_end_pfn)
201 			break;
202 
203 		if (any_writable)
204 			*any_writable |= writable;
205 		if (any_young)
206 			*any_young |= young;
207 		if (any_dirty)
208 			*any_dirty |= dirty;
209 
210 		nr = pte_batch_hint(ptep, pte);
211 		expected_pte = pte_advance_pfn(expected_pte, nr);
212 		ptep += nr;
213 	}
214 
215 	return min(ptep - start_ptep, max_nr);
216 }
217 
218 /**
219  * pte_move_swp_offset - Move the swap entry offset field of a swap pte
220  *	 forward or backward by delta
221  * @pte: The initial pte state; is_swap_pte(pte) must be true and
222  *	 non_swap_entry() must be false.
223  * @delta: The direction and the offset we are moving; forward if delta
224  *	 is positive; backward if delta is negative
225  *
226  * Moves the swap offset, while maintaining all other fields, including
227  * swap type, and any swp pte bits. The resulting pte is returned.
228  */
229 static inline pte_t pte_move_swp_offset(pte_t pte, long delta)
230 {
231 	swp_entry_t entry = pte_to_swp_entry(pte);
232 	pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry),
233 						   (swp_offset(entry) + delta)));
234 
235 	if (pte_swp_soft_dirty(pte))
236 		new = pte_swp_mksoft_dirty(new);
237 	if (pte_swp_exclusive(pte))
238 		new = pte_swp_mkexclusive(new);
239 	if (pte_swp_uffd_wp(pte))
240 		new = pte_swp_mkuffd_wp(new);
241 
242 	return new;
243 }
244 
245 
246 /**
247  * pte_next_swp_offset - Increment the swap entry offset field of a swap pte.
248  * @pte: The initial pte state; is_swap_pte(pte) must be true and
249  *	 non_swap_entry() must be false.
250  *
251  * Increments the swap offset, while maintaining all other fields, including
252  * swap type, and any swp pte bits. The resulting pte is returned.
253  */
254 static inline pte_t pte_next_swp_offset(pte_t pte)
255 {
256 	return pte_move_swp_offset(pte, 1);
257 }
258 
259 /**
260  * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries
261  * @start_ptep: Page table pointer for the first entry.
262  * @max_nr: The maximum number of table entries to consider.
263  * @pte: Page table entry for the first entry.
264  *
265  * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs
266  * containing swap entries all with consecutive offsets and targeting the same
267  * swap type, all with matching swp pte bits.
268  *
269  * max_nr must be at least one and must be limited by the caller so scanning
270  * cannot exceed a single page table.
271  *
272  * Return: the number of table entries in the batch.
273  */
274 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte)
275 {
276 	pte_t expected_pte = pte_next_swp_offset(pte);
277 	const pte_t *end_ptep = start_ptep + max_nr;
278 	pte_t *ptep = start_ptep + 1;
279 
280 	VM_WARN_ON(max_nr < 1);
281 	VM_WARN_ON(!is_swap_pte(pte));
282 	VM_WARN_ON(non_swap_entry(pte_to_swp_entry(pte)));
283 
284 	while (ptep < end_ptep) {
285 		pte = ptep_get(ptep);
286 
287 		if (!pte_same(pte, expected_pte))
288 			break;
289 
290 		expected_pte = pte_next_swp_offset(expected_pte);
291 		ptep++;
292 	}
293 
294 	return ptep - start_ptep;
295 }
296 #endif /* CONFIG_MMU */
297 
298 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
299 						int nr_throttled);
300 static inline void acct_reclaim_writeback(struct folio *folio)
301 {
302 	pg_data_t *pgdat = folio_pgdat(folio);
303 	int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
304 
305 	if (nr_throttled)
306 		__acct_reclaim_writeback(pgdat, folio, nr_throttled);
307 }
308 
309 static inline void wake_throttle_isolated(pg_data_t *pgdat)
310 {
311 	wait_queue_head_t *wqh;
312 
313 	wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
314 	if (waitqueue_active(wqh))
315 		wake_up(wqh);
316 }
317 
318 vm_fault_t vmf_anon_prepare(struct vm_fault *vmf);
319 vm_fault_t do_swap_page(struct vm_fault *vmf);
320 void folio_rotate_reclaimable(struct folio *folio);
321 bool __folio_end_writeback(struct folio *folio);
322 void deactivate_file_folio(struct folio *folio);
323 void folio_activate(struct folio *folio);
324 
325 void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas,
326 		   struct vm_area_struct *start_vma, unsigned long floor,
327 		   unsigned long ceiling, bool mm_wr_locked);
328 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
329 
330 struct zap_details;
331 void unmap_page_range(struct mmu_gather *tlb,
332 			     struct vm_area_struct *vma,
333 			     unsigned long addr, unsigned long end,
334 			     struct zap_details *details);
335 
336 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *,
337 		unsigned int order);
338 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
339 static inline void force_page_cache_readahead(struct address_space *mapping,
340 		struct file *file, pgoff_t index, unsigned long nr_to_read)
341 {
342 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
343 	force_page_cache_ra(&ractl, nr_to_read);
344 }
345 
346 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
347 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
348 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
349 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
350 void filemap_free_folio(struct address_space *mapping, struct folio *folio);
351 int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
352 bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
353 		loff_t end);
354 long mapping_evict_folio(struct address_space *mapping, struct folio *folio);
355 unsigned long mapping_try_invalidate(struct address_space *mapping,
356 		pgoff_t start, pgoff_t end, unsigned long *nr_failed);
357 
358 /**
359  * folio_evictable - Test whether a folio is evictable.
360  * @folio: The folio to test.
361  *
362  * Test whether @folio is evictable -- i.e., should be placed on
363  * active/inactive lists vs unevictable list.
364  *
365  * Reasons folio might not be evictable:
366  * 1. folio's mapping marked unevictable
367  * 2. One of the pages in the folio is part of an mlocked VMA
368  */
369 static inline bool folio_evictable(struct folio *folio)
370 {
371 	bool ret;
372 
373 	/* Prevent address_space of inode and swap cache from being freed */
374 	rcu_read_lock();
375 	ret = !mapping_unevictable(folio_mapping(folio)) &&
376 			!folio_test_mlocked(folio);
377 	rcu_read_unlock();
378 	return ret;
379 }
380 
381 /*
382  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
383  * a count of one.
384  */
385 static inline void set_page_refcounted(struct page *page)
386 {
387 	VM_BUG_ON_PAGE(PageTail(page), page);
388 	VM_BUG_ON_PAGE(page_ref_count(page), page);
389 	set_page_count(page, 1);
390 }
391 
392 /*
393  * Return true if a folio needs ->release_folio() calling upon it.
394  */
395 static inline bool folio_needs_release(struct folio *folio)
396 {
397 	struct address_space *mapping = folio_mapping(folio);
398 
399 	return folio_has_private(folio) ||
400 		(mapping && mapping_release_always(mapping));
401 }
402 
403 extern unsigned long highest_memmap_pfn;
404 
405 /*
406  * Maximum number of reclaim retries without progress before the OOM
407  * killer is consider the only way forward.
408  */
409 #define MAX_RECLAIM_RETRIES 16
410 
411 /*
412  * in mm/vmscan.c:
413  */
414 bool isolate_lru_page(struct page *page);
415 bool folio_isolate_lru(struct folio *folio);
416 void putback_lru_page(struct page *page);
417 void folio_putback_lru(struct folio *folio);
418 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
419 
420 /*
421  * in mm/rmap.c:
422  */
423 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
424 
425 /*
426  * in mm/page_alloc.c
427  */
428 #define K(x) ((x) << (PAGE_SHIFT-10))
429 
430 extern char * const zone_names[MAX_NR_ZONES];
431 
432 /* perform sanity checks on struct pages being allocated or freed */
433 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
434 
435 extern int min_free_kbytes;
436 
437 void setup_per_zone_wmarks(void);
438 void calculate_min_free_kbytes(void);
439 int __meminit init_per_zone_wmark_min(void);
440 void page_alloc_sysctl_init(void);
441 
442 /*
443  * Structure for holding the mostly immutable allocation parameters passed
444  * between functions involved in allocations, including the alloc_pages*
445  * family of functions.
446  *
447  * nodemask, migratetype and highest_zoneidx are initialized only once in
448  * __alloc_pages() and then never change.
449  *
450  * zonelist, preferred_zone and highest_zoneidx are set first in
451  * __alloc_pages() for the fast path, and might be later changed
452  * in __alloc_pages_slowpath(). All other functions pass the whole structure
453  * by a const pointer.
454  */
455 struct alloc_context {
456 	struct zonelist *zonelist;
457 	nodemask_t *nodemask;
458 	struct zoneref *preferred_zoneref;
459 	int migratetype;
460 
461 	/*
462 	 * highest_zoneidx represents highest usable zone index of
463 	 * the allocation request. Due to the nature of the zone,
464 	 * memory on lower zone than the highest_zoneidx will be
465 	 * protected by lowmem_reserve[highest_zoneidx].
466 	 *
467 	 * highest_zoneidx is also used by reclaim/compaction to limit
468 	 * the target zone since higher zone than this index cannot be
469 	 * usable for this allocation request.
470 	 */
471 	enum zone_type highest_zoneidx;
472 	bool spread_dirty_pages;
473 };
474 
475 /*
476  * This function returns the order of a free page in the buddy system. In
477  * general, page_zone(page)->lock must be held by the caller to prevent the
478  * page from being allocated in parallel and returning garbage as the order.
479  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
480  * page cannot be allocated or merged in parallel. Alternatively, it must
481  * handle invalid values gracefully, and use buddy_order_unsafe() below.
482  */
483 static inline unsigned int buddy_order(struct page *page)
484 {
485 	/* PageBuddy() must be checked by the caller */
486 	return page_private(page);
487 }
488 
489 /*
490  * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
491  * PageBuddy() should be checked first by the caller to minimize race window,
492  * and invalid values must be handled gracefully.
493  *
494  * READ_ONCE is used so that if the caller assigns the result into a local
495  * variable and e.g. tests it for valid range before using, the compiler cannot
496  * decide to remove the variable and inline the page_private(page) multiple
497  * times, potentially observing different values in the tests and the actual
498  * use of the result.
499  */
500 #define buddy_order_unsafe(page)	READ_ONCE(page_private(page))
501 
502 /*
503  * This function checks whether a page is free && is the buddy
504  * we can coalesce a page and its buddy if
505  * (a) the buddy is not in a hole (check before calling!) &&
506  * (b) the buddy is in the buddy system &&
507  * (c) a page and its buddy have the same order &&
508  * (d) a page and its buddy are in the same zone.
509  *
510  * For recording whether a page is in the buddy system, we set PageBuddy.
511  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
512  *
513  * For recording page's order, we use page_private(page).
514  */
515 static inline bool page_is_buddy(struct page *page, struct page *buddy,
516 				 unsigned int order)
517 {
518 	if (!page_is_guard(buddy) && !PageBuddy(buddy))
519 		return false;
520 
521 	if (buddy_order(buddy) != order)
522 		return false;
523 
524 	/*
525 	 * zone check is done late to avoid uselessly calculating
526 	 * zone/node ids for pages that could never merge.
527 	 */
528 	if (page_zone_id(page) != page_zone_id(buddy))
529 		return false;
530 
531 	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
532 
533 	return true;
534 }
535 
536 /*
537  * Locate the struct page for both the matching buddy in our
538  * pair (buddy1) and the combined O(n+1) page they form (page).
539  *
540  * 1) Any buddy B1 will have an order O twin B2 which satisfies
541  * the following equation:
542  *     B2 = B1 ^ (1 << O)
543  * For example, if the starting buddy (buddy2) is #8 its order
544  * 1 buddy is #10:
545  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
546  *
547  * 2) Any buddy B will have an order O+1 parent P which
548  * satisfies the following equation:
549  *     P = B & ~(1 << O)
550  *
551  * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER
552  */
553 static inline unsigned long
554 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
555 {
556 	return page_pfn ^ (1 << order);
557 }
558 
559 /*
560  * Find the buddy of @page and validate it.
561  * @page: The input page
562  * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
563  *       function is used in the performance-critical __free_one_page().
564  * @order: The order of the page
565  * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
566  *             page_to_pfn().
567  *
568  * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
569  * not the same as @page. The validation is necessary before use it.
570  *
571  * Return: the found buddy page or NULL if not found.
572  */
573 static inline struct page *find_buddy_page_pfn(struct page *page,
574 			unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
575 {
576 	unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
577 	struct page *buddy;
578 
579 	buddy = page + (__buddy_pfn - pfn);
580 	if (buddy_pfn)
581 		*buddy_pfn = __buddy_pfn;
582 
583 	if (page_is_buddy(page, buddy, order))
584 		return buddy;
585 	return NULL;
586 }
587 
588 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
589 				unsigned long end_pfn, struct zone *zone);
590 
591 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
592 				unsigned long end_pfn, struct zone *zone)
593 {
594 	if (zone->contiguous)
595 		return pfn_to_page(start_pfn);
596 
597 	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
598 }
599 
600 void set_zone_contiguous(struct zone *zone);
601 
602 static inline void clear_zone_contiguous(struct zone *zone)
603 {
604 	zone->contiguous = false;
605 }
606 
607 extern int __isolate_free_page(struct page *page, unsigned int order);
608 extern void __putback_isolated_page(struct page *page, unsigned int order,
609 				    int mt);
610 extern void memblock_free_pages(struct page *page, unsigned long pfn,
611 					unsigned int order);
612 extern void __free_pages_core(struct page *page, unsigned int order,
613 		enum meminit_context context);
614 
615 /*
616  * This will have no effect, other than possibly generating a warning, if the
617  * caller passes in a non-large folio.
618  */
619 static inline void folio_set_order(struct folio *folio, unsigned int order)
620 {
621 	if (WARN_ON_ONCE(!order || !folio_test_large(folio)))
622 		return;
623 
624 	folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order;
625 #ifdef CONFIG_64BIT
626 	folio->_folio_nr_pages = 1U << order;
627 #endif
628 }
629 
630 void __folio_undo_large_rmappable(struct folio *folio);
631 static inline void folio_undo_large_rmappable(struct folio *folio)
632 {
633 	if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio))
634 		return;
635 
636 	/*
637 	 * At this point, there is no one trying to add the folio to
638 	 * deferred_list. If folio is not in deferred_list, it's safe
639 	 * to check without acquiring the split_queue_lock.
640 	 */
641 	if (data_race(list_empty(&folio->_deferred_list)))
642 		return;
643 
644 	__folio_undo_large_rmappable(folio);
645 }
646 
647 static inline struct folio *page_rmappable_folio(struct page *page)
648 {
649 	struct folio *folio = (struct folio *)page;
650 
651 	if (folio && folio_test_large(folio))
652 		folio_set_large_rmappable(folio);
653 	return folio;
654 }
655 
656 static inline void prep_compound_head(struct page *page, unsigned int order)
657 {
658 	struct folio *folio = (struct folio *)page;
659 
660 	folio_set_order(folio, order);
661 	atomic_set(&folio->_large_mapcount, -1);
662 	atomic_set(&folio->_entire_mapcount, -1);
663 	atomic_set(&folio->_nr_pages_mapped, 0);
664 	atomic_set(&folio->_pincount, 0);
665 	if (order > 1)
666 		INIT_LIST_HEAD(&folio->_deferred_list);
667 }
668 
669 static inline void prep_compound_tail(struct page *head, int tail_idx)
670 {
671 	struct page *p = head + tail_idx;
672 
673 	p->mapping = TAIL_MAPPING;
674 	set_compound_head(p, head);
675 	set_page_private(p, 0);
676 }
677 
678 extern void prep_compound_page(struct page *page, unsigned int order);
679 
680 extern void post_alloc_hook(struct page *page, unsigned int order,
681 					gfp_t gfp_flags);
682 extern bool free_pages_prepare(struct page *page, unsigned int order);
683 
684 extern int user_min_free_kbytes;
685 
686 void free_unref_page(struct page *page, unsigned int order);
687 void free_unref_folios(struct folio_batch *fbatch);
688 
689 extern void zone_pcp_reset(struct zone *zone);
690 extern void zone_pcp_disable(struct zone *zone);
691 extern void zone_pcp_enable(struct zone *zone);
692 extern void zone_pcp_init(struct zone *zone);
693 
694 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
695 			  phys_addr_t min_addr,
696 			  int nid, bool exact_nid);
697 
698 void memmap_init_range(unsigned long, int, unsigned long, unsigned long,
699 		unsigned long, enum meminit_context, struct vmem_altmap *, int);
700 
701 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
702 
703 /*
704  * in mm/compaction.c
705  */
706 /*
707  * compact_control is used to track pages being migrated and the free pages
708  * they are being migrated to during memory compaction. The free_pfn starts
709  * at the end of a zone and migrate_pfn begins at the start. Movable pages
710  * are moved to the end of a zone during a compaction run and the run
711  * completes when free_pfn <= migrate_pfn
712  */
713 struct compact_control {
714 	struct list_head freepages[NR_PAGE_ORDERS];	/* List of free pages to migrate to */
715 	struct list_head migratepages;	/* List of pages being migrated */
716 	unsigned int nr_freepages;	/* Number of isolated free pages */
717 	unsigned int nr_migratepages;	/* Number of pages to migrate */
718 	unsigned long free_pfn;		/* isolate_freepages search base */
719 	/*
720 	 * Acts as an in/out parameter to page isolation for migration.
721 	 * isolate_migratepages uses it as a search base.
722 	 * isolate_migratepages_block will update the value to the next pfn
723 	 * after the last isolated one.
724 	 */
725 	unsigned long migrate_pfn;
726 	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
727 	struct zone *zone;
728 	unsigned long total_migrate_scanned;
729 	unsigned long total_free_scanned;
730 	unsigned short fast_search_fail;/* failures to use free list searches */
731 	short search_order;		/* order to start a fast search at */
732 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
733 	int order;			/* order a direct compactor needs */
734 	int migratetype;		/* migratetype of direct compactor */
735 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
736 	const int highest_zoneidx;	/* zone index of a direct compactor */
737 	enum migrate_mode mode;		/* Async or sync migration mode */
738 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
739 	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
740 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
741 	bool direct_compaction;		/* False from kcompactd or /proc/... */
742 	bool proactive_compaction;	/* kcompactd proactive compaction */
743 	bool whole_zone;		/* Whole zone should/has been scanned */
744 	bool contended;			/* Signal lock contention */
745 	bool finish_pageblock;		/* Scan the remainder of a pageblock. Used
746 					 * when there are potentially transient
747 					 * isolation or migration failures to
748 					 * ensure forward progress.
749 					 */
750 	bool alloc_contig;		/* alloc_contig_range allocation */
751 };
752 
753 /*
754  * Used in direct compaction when a page should be taken from the freelists
755  * immediately when one is created during the free path.
756  */
757 struct capture_control {
758 	struct compact_control *cc;
759 	struct page *page;
760 };
761 
762 unsigned long
763 isolate_freepages_range(struct compact_control *cc,
764 			unsigned long start_pfn, unsigned long end_pfn);
765 int
766 isolate_migratepages_range(struct compact_control *cc,
767 			   unsigned long low_pfn, unsigned long end_pfn);
768 
769 int __alloc_contig_migrate_range(struct compact_control *cc,
770 					unsigned long start, unsigned long end,
771 					int migratetype);
772 
773 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
774 void init_cma_reserved_pageblock(struct page *page);
775 
776 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
777 
778 int find_suitable_fallback(struct free_area *area, unsigned int order,
779 			int migratetype, bool only_stealable, bool *can_steal);
780 
781 static inline bool free_area_empty(struct free_area *area, int migratetype)
782 {
783 	return list_empty(&area->free_list[migratetype]);
784 }
785 
786 /* mm/util.c */
787 struct anon_vma *folio_anon_vma(struct folio *folio);
788 
789 #ifdef CONFIG_MMU
790 void unmap_mapping_folio(struct folio *folio);
791 extern long populate_vma_page_range(struct vm_area_struct *vma,
792 		unsigned long start, unsigned long end, int *locked);
793 extern long faultin_page_range(struct mm_struct *mm, unsigned long start,
794 		unsigned long end, bool write, int *locked);
795 extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags,
796 			       unsigned long bytes);
797 
798 /*
799  * NOTE: This function can't tell whether the folio is "fully mapped" in the
800  * range.
801  * "fully mapped" means all the pages of folio is associated with the page
802  * table of range while this function just check whether the folio range is
803  * within the range [start, end). Function caller needs to do page table
804  * check if it cares about the page table association.
805  *
806  * Typical usage (like mlock or madvise) is:
807  * Caller knows at least 1 page of folio is associated with page table of VMA
808  * and the range [start, end) is intersect with the VMA range. Caller wants
809  * to know whether the folio is fully associated with the range. It calls
810  * this function to check whether the folio is in the range first. Then checks
811  * the page table to know whether the folio is fully mapped to the range.
812  */
813 static inline bool
814 folio_within_range(struct folio *folio, struct vm_area_struct *vma,
815 		unsigned long start, unsigned long end)
816 {
817 	pgoff_t pgoff, addr;
818 	unsigned long vma_pglen = vma_pages(vma);
819 
820 	VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio);
821 	if (start > end)
822 		return false;
823 
824 	if (start < vma->vm_start)
825 		start = vma->vm_start;
826 
827 	if (end > vma->vm_end)
828 		end = vma->vm_end;
829 
830 	pgoff = folio_pgoff(folio);
831 
832 	/* if folio start address is not in vma range */
833 	if (!in_range(pgoff, vma->vm_pgoff, vma_pglen))
834 		return false;
835 
836 	addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
837 
838 	return !(addr < start || end - addr < folio_size(folio));
839 }
840 
841 static inline bool
842 folio_within_vma(struct folio *folio, struct vm_area_struct *vma)
843 {
844 	return folio_within_range(folio, vma, vma->vm_start, vma->vm_end);
845 }
846 
847 /*
848  * mlock_vma_folio() and munlock_vma_folio():
849  * should be called with vma's mmap_lock held for read or write,
850  * under page table lock for the pte/pmd being added or removed.
851  *
852  * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at
853  * the end of folio_remove_rmap_*(); but new anon folios are managed by
854  * folio_add_lru_vma() calling mlock_new_folio().
855  */
856 void mlock_folio(struct folio *folio);
857 static inline void mlock_vma_folio(struct folio *folio,
858 				struct vm_area_struct *vma)
859 {
860 	/*
861 	 * The VM_SPECIAL check here serves two purposes.
862 	 * 1) VM_IO check prevents migration from double-counting during mlock.
863 	 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
864 	 *    is never left set on a VM_SPECIAL vma, there is an interval while
865 	 *    file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
866 	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
867 	 */
868 	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED))
869 		mlock_folio(folio);
870 }
871 
872 void munlock_folio(struct folio *folio);
873 static inline void munlock_vma_folio(struct folio *folio,
874 					struct vm_area_struct *vma)
875 {
876 	/*
877 	 * munlock if the function is called. Ideally, we should only
878 	 * do munlock if any page of folio is unmapped from VMA and
879 	 * cause folio not fully mapped to VMA.
880 	 *
881 	 * But it's not easy to confirm that's the situation. So we
882 	 * always munlock the folio and page reclaim will correct it
883 	 * if it's wrong.
884 	 */
885 	if (unlikely(vma->vm_flags & VM_LOCKED))
886 		munlock_folio(folio);
887 }
888 
889 void mlock_new_folio(struct folio *folio);
890 bool need_mlock_drain(int cpu);
891 void mlock_drain_local(void);
892 void mlock_drain_remote(int cpu);
893 
894 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
895 
896 /**
897  * vma_address - Find the virtual address a page range is mapped at
898  * @vma: The vma which maps this object.
899  * @pgoff: The page offset within its object.
900  * @nr_pages: The number of pages to consider.
901  *
902  * If any page in this range is mapped by this VMA, return the first address
903  * where any of these pages appear.  Otherwise, return -EFAULT.
904  */
905 static inline unsigned long vma_address(struct vm_area_struct *vma,
906 		pgoff_t pgoff, unsigned long nr_pages)
907 {
908 	unsigned long address;
909 
910 	if (pgoff >= vma->vm_pgoff) {
911 		address = vma->vm_start +
912 			((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
913 		/* Check for address beyond vma (or wrapped through 0?) */
914 		if (address < vma->vm_start || address >= vma->vm_end)
915 			address = -EFAULT;
916 	} else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
917 		/* Test above avoids possibility of wrap to 0 on 32-bit */
918 		address = vma->vm_start;
919 	} else {
920 		address = -EFAULT;
921 	}
922 	return address;
923 }
924 
925 /*
926  * Then at what user virtual address will none of the range be found in vma?
927  * Assumes that vma_address() already returned a good starting address.
928  */
929 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
930 {
931 	struct vm_area_struct *vma = pvmw->vma;
932 	pgoff_t pgoff;
933 	unsigned long address;
934 
935 	/* Common case, plus ->pgoff is invalid for KSM */
936 	if (pvmw->nr_pages == 1)
937 		return pvmw->address + PAGE_SIZE;
938 
939 	pgoff = pvmw->pgoff + pvmw->nr_pages;
940 	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
941 	/* Check for address beyond vma (or wrapped through 0?) */
942 	if (address < vma->vm_start || address > vma->vm_end)
943 		address = vma->vm_end;
944 	return address;
945 }
946 
947 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
948 						    struct file *fpin)
949 {
950 	int flags = vmf->flags;
951 
952 	if (fpin)
953 		return fpin;
954 
955 	/*
956 	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
957 	 * anything, so we only pin the file and drop the mmap_lock if only
958 	 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
959 	 */
960 	if (fault_flag_allow_retry_first(flags) &&
961 	    !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
962 		fpin = get_file(vmf->vma->vm_file);
963 		release_fault_lock(vmf);
964 	}
965 	return fpin;
966 }
967 #else /* !CONFIG_MMU */
968 static inline void unmap_mapping_folio(struct folio *folio) { }
969 static inline void mlock_new_folio(struct folio *folio) { }
970 static inline bool need_mlock_drain(int cpu) { return false; }
971 static inline void mlock_drain_local(void) { }
972 static inline void mlock_drain_remote(int cpu) { }
973 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
974 {
975 }
976 #endif /* !CONFIG_MMU */
977 
978 /* Memory initialisation debug and verification */
979 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
980 DECLARE_STATIC_KEY_TRUE(deferred_pages);
981 
982 bool __init deferred_grow_zone(struct zone *zone, unsigned int order);
983 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
984 
985 enum mminit_level {
986 	MMINIT_WARNING,
987 	MMINIT_VERIFY,
988 	MMINIT_TRACE
989 };
990 
991 #ifdef CONFIG_DEBUG_MEMORY_INIT
992 
993 extern int mminit_loglevel;
994 
995 #define mminit_dprintk(level, prefix, fmt, arg...) \
996 do { \
997 	if (level < mminit_loglevel) { \
998 		if (level <= MMINIT_WARNING) \
999 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
1000 		else \
1001 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
1002 	} \
1003 } while (0)
1004 
1005 extern void mminit_verify_pageflags_layout(void);
1006 extern void mminit_verify_zonelist(void);
1007 #else
1008 
1009 static inline void mminit_dprintk(enum mminit_level level,
1010 				const char *prefix, const char *fmt, ...)
1011 {
1012 }
1013 
1014 static inline void mminit_verify_pageflags_layout(void)
1015 {
1016 }
1017 
1018 static inline void mminit_verify_zonelist(void)
1019 {
1020 }
1021 #endif /* CONFIG_DEBUG_MEMORY_INIT */
1022 
1023 #define NODE_RECLAIM_NOSCAN	-2
1024 #define NODE_RECLAIM_FULL	-1
1025 #define NODE_RECLAIM_SOME	0
1026 #define NODE_RECLAIM_SUCCESS	1
1027 
1028 #ifdef CONFIG_NUMA
1029 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
1030 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
1031 #else
1032 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
1033 				unsigned int order)
1034 {
1035 	return NODE_RECLAIM_NOSCAN;
1036 }
1037 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
1038 {
1039 	return NUMA_NO_NODE;
1040 }
1041 #endif
1042 
1043 /*
1044  * mm/memory-failure.c
1045  */
1046 void shake_folio(struct folio *folio);
1047 extern int hwpoison_filter(struct page *p);
1048 
1049 extern u32 hwpoison_filter_dev_major;
1050 extern u32 hwpoison_filter_dev_minor;
1051 extern u64 hwpoison_filter_flags_mask;
1052 extern u64 hwpoison_filter_flags_value;
1053 extern u64 hwpoison_filter_memcg;
1054 extern u32 hwpoison_filter_enable;
1055 #define MAGIC_HWPOISON	0x48575053U	/* HWPS */
1056 void SetPageHWPoisonTakenOff(struct page *page);
1057 void ClearPageHWPoisonTakenOff(struct page *page);
1058 bool take_page_off_buddy(struct page *page);
1059 bool put_page_back_buddy(struct page *page);
1060 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
1061 void add_to_kill_ksm(struct task_struct *tsk, struct page *p,
1062 		     struct vm_area_struct *vma, struct list_head *to_kill,
1063 		     unsigned long ksm_addr);
1064 unsigned long page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
1065 
1066 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
1067         unsigned long, unsigned long,
1068         unsigned long, unsigned long);
1069 
1070 extern void set_pageblock_order(void);
1071 struct folio *alloc_migrate_folio(struct folio *src, unsigned long private);
1072 unsigned long reclaim_pages(struct list_head *folio_list);
1073 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
1074 					    struct list_head *folio_list);
1075 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1076 #define ALLOC_WMARK_MIN		WMARK_MIN
1077 #define ALLOC_WMARK_LOW		WMARK_LOW
1078 #define ALLOC_WMARK_HIGH	WMARK_HIGH
1079 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
1080 
1081 /* Mask to get the watermark bits */
1082 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
1083 
1084 /*
1085  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
1086  * cannot assume a reduced access to memory reserves is sufficient for
1087  * !MMU
1088  */
1089 #ifdef CONFIG_MMU
1090 #define ALLOC_OOM		0x08
1091 #else
1092 #define ALLOC_OOM		ALLOC_NO_WATERMARKS
1093 #endif
1094 
1095 #define ALLOC_NON_BLOCK		 0x10 /* Caller cannot block. Allow access
1096 				       * to 25% of the min watermark or
1097 				       * 62.5% if __GFP_HIGH is set.
1098 				       */
1099 #define ALLOC_MIN_RESERVE	 0x20 /* __GFP_HIGH set. Allow access to 50%
1100 				       * of the min watermark.
1101 				       */
1102 #define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
1103 #define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
1104 #ifdef CONFIG_ZONE_DMA32
1105 #define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
1106 #else
1107 #define ALLOC_NOFRAGMENT	  0x0
1108 #endif
1109 #define ALLOC_HIGHATOMIC	0x200 /* Allows access to MIGRATE_HIGHATOMIC */
1110 #define ALLOC_KSWAPD		0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
1111 
1112 /* Flags that allow allocations below the min watermark. */
1113 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
1114 
1115 enum ttu_flags;
1116 struct tlbflush_unmap_batch;
1117 
1118 
1119 /*
1120  * only for MM internal work items which do not depend on
1121  * any allocations or locks which might depend on allocations
1122  */
1123 extern struct workqueue_struct *mm_percpu_wq;
1124 
1125 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1126 void try_to_unmap_flush(void);
1127 void try_to_unmap_flush_dirty(void);
1128 void flush_tlb_batched_pending(struct mm_struct *mm);
1129 #else
1130 static inline void try_to_unmap_flush(void)
1131 {
1132 }
1133 static inline void try_to_unmap_flush_dirty(void)
1134 {
1135 }
1136 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
1137 {
1138 }
1139 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
1140 
1141 extern const struct trace_print_flags pageflag_names[];
1142 extern const struct trace_print_flags pagetype_names[];
1143 extern const struct trace_print_flags vmaflag_names[];
1144 extern const struct trace_print_flags gfpflag_names[];
1145 
1146 static inline bool is_migrate_highatomic(enum migratetype migratetype)
1147 {
1148 	return migratetype == MIGRATE_HIGHATOMIC;
1149 }
1150 
1151 void setup_zone_pageset(struct zone *zone);
1152 
1153 struct migration_target_control {
1154 	int nid;		/* preferred node id */
1155 	nodemask_t *nmask;
1156 	gfp_t gfp_mask;
1157 	enum migrate_reason reason;
1158 };
1159 
1160 /*
1161  * mm/filemap.c
1162  */
1163 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
1164 			      struct folio *folio, loff_t fpos, size_t size);
1165 
1166 /*
1167  * mm/vmalloc.c
1168  */
1169 #ifdef CONFIG_MMU
1170 void __init vmalloc_init(void);
1171 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1172                 pgprot_t prot, struct page **pages, unsigned int page_shift);
1173 #else
1174 static inline void vmalloc_init(void)
1175 {
1176 }
1177 
1178 static inline
1179 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1180                 pgprot_t prot, struct page **pages, unsigned int page_shift)
1181 {
1182 	return -EINVAL;
1183 }
1184 #endif
1185 
1186 int __must_check __vmap_pages_range_noflush(unsigned long addr,
1187 			       unsigned long end, pgprot_t prot,
1188 			       struct page **pages, unsigned int page_shift);
1189 
1190 void vunmap_range_noflush(unsigned long start, unsigned long end);
1191 
1192 void __vunmap_range_noflush(unsigned long start, unsigned long end);
1193 
1194 int numa_migrate_prep(struct folio *folio, struct vm_fault *vmf,
1195 		      unsigned long addr, int page_nid, int *flags);
1196 
1197 void free_zone_device_folio(struct folio *folio);
1198 int migrate_device_coherent_page(struct page *page);
1199 
1200 /*
1201  * mm/gup.c
1202  */
1203 int __must_check try_grab_folio(struct folio *folio, int refs,
1204 				unsigned int flags);
1205 
1206 /*
1207  * mm/huge_memory.c
1208  */
1209 void touch_pud(struct vm_area_struct *vma, unsigned long addr,
1210 	       pud_t *pud, bool write);
1211 void touch_pmd(struct vm_area_struct *vma, unsigned long addr,
1212 	       pmd_t *pmd, bool write);
1213 
1214 enum {
1215 	/* mark page accessed */
1216 	FOLL_TOUCH = 1 << 16,
1217 	/* a retry, previous pass started an IO */
1218 	FOLL_TRIED = 1 << 17,
1219 	/* we are working on non-current tsk/mm */
1220 	FOLL_REMOTE = 1 << 18,
1221 	/* pages must be released via unpin_user_page */
1222 	FOLL_PIN = 1 << 19,
1223 	/* gup_fast: prevent fall-back to slow gup */
1224 	FOLL_FAST_ONLY = 1 << 20,
1225 	/* allow unlocking the mmap lock */
1226 	FOLL_UNLOCKABLE = 1 << 21,
1227 	/* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */
1228 	FOLL_MADV_POPULATE = 1 << 22,
1229 };
1230 
1231 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \
1232 			    FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \
1233 			    FOLL_MADV_POPULATE)
1234 
1235 /*
1236  * Indicates for which pages that are write-protected in the page table,
1237  * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
1238  * GUP pin will remain consistent with the pages mapped into the page tables
1239  * of the MM.
1240  *
1241  * Temporary unmapping of PageAnonExclusive() pages or clearing of
1242  * PageAnonExclusive() has to protect against concurrent GUP:
1243  * * Ordinary GUP: Using the PT lock
1244  * * GUP-fast and fork(): mm->write_protect_seq
1245  * * GUP-fast and KSM or temporary unmapping (swap, migration): see
1246  *    folio_try_share_anon_rmap_*()
1247  *
1248  * Must be called with the (sub)page that's actually referenced via the
1249  * page table entry, which might not necessarily be the head page for a
1250  * PTE-mapped THP.
1251  *
1252  * If the vma is NULL, we're coming from the GUP-fast path and might have
1253  * to fallback to the slow path just to lookup the vma.
1254  */
1255 static inline bool gup_must_unshare(struct vm_area_struct *vma,
1256 				    unsigned int flags, struct page *page)
1257 {
1258 	/*
1259 	 * FOLL_WRITE is implicitly handled correctly as the page table entry
1260 	 * has to be writable -- and if it references (part of) an anonymous
1261 	 * folio, that part is required to be marked exclusive.
1262 	 */
1263 	if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
1264 		return false;
1265 	/*
1266 	 * Note: PageAnon(page) is stable until the page is actually getting
1267 	 * freed.
1268 	 */
1269 	if (!PageAnon(page)) {
1270 		/*
1271 		 * We only care about R/O long-term pining: R/O short-term
1272 		 * pinning does not have the semantics to observe successive
1273 		 * changes through the process page tables.
1274 		 */
1275 		if (!(flags & FOLL_LONGTERM))
1276 			return false;
1277 
1278 		/* We really need the vma ... */
1279 		if (!vma)
1280 			return true;
1281 
1282 		/*
1283 		 * ... because we only care about writable private ("COW")
1284 		 * mappings where we have to break COW early.
1285 		 */
1286 		return is_cow_mapping(vma->vm_flags);
1287 	}
1288 
1289 	/* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */
1290 	if (IS_ENABLED(CONFIG_HAVE_GUP_FAST))
1291 		smp_rmb();
1292 
1293 	/*
1294 	 * Note that PageKsm() pages cannot be exclusive, and consequently,
1295 	 * cannot get pinned.
1296 	 */
1297 	return !PageAnonExclusive(page);
1298 }
1299 
1300 extern bool mirrored_kernelcore;
1301 extern bool memblock_has_mirror(void);
1302 
1303 static __always_inline void vma_set_range(struct vm_area_struct *vma,
1304 					  unsigned long start, unsigned long end,
1305 					  pgoff_t pgoff)
1306 {
1307 	vma->vm_start = start;
1308 	vma->vm_end = end;
1309 	vma->vm_pgoff = pgoff;
1310 }
1311 
1312 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1313 {
1314 	/*
1315 	 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1316 	 * enablements, because when without soft-dirty being compiled in,
1317 	 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1318 	 * will be constantly true.
1319 	 */
1320 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
1321 		return false;
1322 
1323 	/*
1324 	 * Soft-dirty is kind of special: its tracking is enabled when the
1325 	 * vma flags not set.
1326 	 */
1327 	return !(vma->vm_flags & VM_SOFTDIRTY);
1328 }
1329 
1330 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd)
1331 {
1332 	return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd);
1333 }
1334 
1335 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte)
1336 {
1337 	return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte);
1338 }
1339 
1340 void __meminit __init_single_page(struct page *page, unsigned long pfn,
1341 				unsigned long zone, int nid);
1342 
1343 /* shrinker related functions */
1344 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg,
1345 			  int priority);
1346 
1347 #ifdef CONFIG_64BIT
1348 static inline int can_do_mseal(unsigned long flags)
1349 {
1350 	if (flags)
1351 		return -EINVAL;
1352 
1353 	return 0;
1354 }
1355 
1356 bool can_modify_mm(struct mm_struct *mm, unsigned long start,
1357 		unsigned long end);
1358 bool can_modify_mm_madv(struct mm_struct *mm, unsigned long start,
1359 		unsigned long end, int behavior);
1360 #else
1361 static inline int can_do_mseal(unsigned long flags)
1362 {
1363 	return -EPERM;
1364 }
1365 
1366 static inline bool can_modify_mm(struct mm_struct *mm, unsigned long start,
1367 		unsigned long end)
1368 {
1369 	return true;
1370 }
1371 
1372 static inline bool can_modify_mm_madv(struct mm_struct *mm, unsigned long start,
1373 		unsigned long end, int behavior)
1374 {
1375 	return true;
1376 }
1377 #endif
1378 
1379 #ifdef CONFIG_SHRINKER_DEBUG
1380 static inline __printf(2, 0) int shrinker_debugfs_name_alloc(
1381 			struct shrinker *shrinker, const char *fmt, va_list ap)
1382 {
1383 	shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap);
1384 
1385 	return shrinker->name ? 0 : -ENOMEM;
1386 }
1387 
1388 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1389 {
1390 	kfree_const(shrinker->name);
1391 	shrinker->name = NULL;
1392 }
1393 
1394 extern int shrinker_debugfs_add(struct shrinker *shrinker);
1395 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1396 					      int *debugfs_id);
1397 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1398 				    int debugfs_id);
1399 #else /* CONFIG_SHRINKER_DEBUG */
1400 static inline int shrinker_debugfs_add(struct shrinker *shrinker)
1401 {
1402 	return 0;
1403 }
1404 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker,
1405 					      const char *fmt, va_list ap)
1406 {
1407 	return 0;
1408 }
1409 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1410 {
1411 }
1412 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1413 						     int *debugfs_id)
1414 {
1415 	*debugfs_id = -1;
1416 	return NULL;
1417 }
1418 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1419 					   int debugfs_id)
1420 {
1421 }
1422 #endif /* CONFIG_SHRINKER_DEBUG */
1423 
1424 /* Only track the nodes of mappings with shadow entries */
1425 void workingset_update_node(struct xa_node *node);
1426 extern struct list_lru shadow_nodes;
1427 
1428 /* mremap.c */
1429 unsigned long move_page_tables(struct vm_area_struct *vma,
1430 	unsigned long old_addr, struct vm_area_struct *new_vma,
1431 	unsigned long new_addr, unsigned long len,
1432 	bool need_rmap_locks, bool for_stack);
1433 
1434 #endif	/* __MM_INTERNAL_H */
1435