xref: /linux/mm/internal.h (revision d8a142058f39a32ae6c7cd5a786c656133c717fb)
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/pagewalk.h>
16 #include <linux/rmap.h>
17 #include <linux/swap.h>
18 #include <linux/swapops.h>
19 #include <linux/swap_cgroup.h>
20 #include <linux/tracepoint-defs.h>
21 
22 /* Internal core VMA manipulation functions. */
23 #include "vma.h"
24 
25 struct folio_batch;
26 
27 /*
28  * The set of flags that only affect watermark checking and reclaim
29  * behaviour. This is used by the MM to obey the caller constraints
30  * about IO, FS and watermark checking while ignoring placement
31  * hints such as HIGHMEM usage.
32  */
33 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
34 			__GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
35 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
36 			__GFP_NOLOCKDEP)
37 
38 /* The GFP flags allowed during early boot */
39 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
40 
41 /* Control allocation cpuset and node placement constraints */
42 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
43 
44 /* Do not use these with a slab allocator */
45 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
46 
47 /*
48  * Different from WARN_ON_ONCE(), no warning will be issued
49  * when we specify __GFP_NOWARN.
50  */
51 #define WARN_ON_ONCE_GFP(cond, gfp)	({				\
52 	static bool __section(".data..once") __warned;			\
53 	int __ret_warn_once = !!(cond);					\
54 									\
55 	if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
56 		__warned = true;					\
57 		WARN_ON(1);						\
58 	}								\
59 	unlikely(__ret_warn_once);					\
60 })
61 
62 void page_writeback_init(void);
63 
64 /*
65  * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
66  * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit
67  * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE).  Hugetlb currently
68  * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
69  */
70 #define ENTIRELY_MAPPED		0x800000
71 #define FOLIO_PAGES_MAPPED	(ENTIRELY_MAPPED - 1)
72 
73 /*
74  * Flags passed to __show_mem() and show_free_areas() to suppress output in
75  * various contexts.
76  */
77 #define SHOW_MEM_FILTER_NODES		(0x0001u)	/* disallowed nodes */
78 
79 /*
80  * How many individual pages have an elevated _mapcount.  Excludes
81  * the folio's entire_mapcount.
82  *
83  * Don't use this function outside of debugging code.
84  */
85 static inline int folio_nr_pages_mapped(const struct folio *folio)
86 {
87 	return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
88 }
89 
90 /*
91  * Retrieve the first entry of a folio based on a provided entry within the
92  * folio. We cannot rely on folio->swap as there is no guarantee that it has
93  * been initialized. Used for calling arch_swap_restore()
94  */
95 static inline swp_entry_t folio_swap(swp_entry_t entry,
96 		const struct folio *folio)
97 {
98 	swp_entry_t swap = {
99 		.val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)),
100 	};
101 
102 	return swap;
103 }
104 
105 static inline void *folio_raw_mapping(const struct folio *folio)
106 {
107 	unsigned long mapping = (unsigned long)folio->mapping;
108 
109 	return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
110 }
111 
112 /*
113  * This is a file-backed mapping, and is about to be memory mapped - invoke its
114  * mmap hook and safely handle error conditions. On error, VMA hooks will be
115  * mutated.
116  *
117  * @file: File which backs the mapping.
118  * @vma:  VMA which we are mapping.
119  *
120  * Returns: 0 if success, error otherwise.
121  */
122 static inline int mmap_file(struct file *file, struct vm_area_struct *vma)
123 {
124 	int err = call_mmap(file, vma);
125 
126 	if (likely(!err))
127 		return 0;
128 
129 	/*
130 	 * OK, we tried to call the file hook for mmap(), but an error
131 	 * arose. The mapping is in an inconsistent state and we most not invoke
132 	 * any further hooks on it.
133 	 */
134 	vma->vm_ops = &vma_dummy_vm_ops;
135 
136 	return err;
137 }
138 
139 /*
140  * If the VMA has a close hook then close it, and since closing it might leave
141  * it in an inconsistent state which makes the use of any hooks suspect, clear
142  * them down by installing dummy empty hooks.
143  */
144 static inline void vma_close(struct vm_area_struct *vma)
145 {
146 	if (vma->vm_ops && vma->vm_ops->close) {
147 		vma->vm_ops->close(vma);
148 
149 		/*
150 		 * The mapping is in an inconsistent state, and no further hooks
151 		 * may be invoked upon it.
152 		 */
153 		vma->vm_ops = &vma_dummy_vm_ops;
154 	}
155 }
156 
157 #ifdef CONFIG_MMU
158 
159 /* Flags for folio_pte_batch(). */
160 typedef int __bitwise fpb_t;
161 
162 /* Compare PTEs after pte_mkclean(), ignoring the dirty bit. */
163 #define FPB_IGNORE_DIRTY		((__force fpb_t)BIT(0))
164 
165 /* Compare PTEs after pte_clear_soft_dirty(), ignoring the soft-dirty bit. */
166 #define FPB_IGNORE_SOFT_DIRTY		((__force fpb_t)BIT(1))
167 
168 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
169 {
170 	if (flags & FPB_IGNORE_DIRTY)
171 		pte = pte_mkclean(pte);
172 	if (likely(flags & FPB_IGNORE_SOFT_DIRTY))
173 		pte = pte_clear_soft_dirty(pte);
174 	return pte_wrprotect(pte_mkold(pte));
175 }
176 
177 /**
178  * folio_pte_batch - detect a PTE batch for a large folio
179  * @folio: The large folio to detect a PTE batch for.
180  * @addr: The user virtual address the first page is mapped at.
181  * @start_ptep: Page table pointer for the first entry.
182  * @pte: Page table entry for the first page.
183  * @max_nr: The maximum number of table entries to consider.
184  * @flags: Flags to modify the PTE batch semantics.
185  * @any_writable: Optional pointer to indicate whether any entry except the
186  *		  first one is writable.
187  * @any_young: Optional pointer to indicate whether any entry except the
188  *		  first one is young.
189  * @any_dirty: Optional pointer to indicate whether any entry except the
190  *		  first one is dirty.
191  *
192  * Detect a PTE batch: consecutive (present) PTEs that map consecutive
193  * pages of the same large folio.
194  *
195  * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN,
196  * the accessed bit, writable bit, dirty bit (with FPB_IGNORE_DIRTY) and
197  * soft-dirty bit (with FPB_IGNORE_SOFT_DIRTY).
198  *
199  * start_ptep must map any page of the folio. max_nr must be at least one and
200  * must be limited by the caller so scanning cannot exceed a single page table.
201  *
202  * Return: the number of table entries in the batch.
203  */
204 static inline int folio_pte_batch(struct folio *folio, unsigned long addr,
205 		pte_t *start_ptep, pte_t pte, int max_nr, fpb_t flags,
206 		bool *any_writable, bool *any_young, bool *any_dirty)
207 {
208 	unsigned long folio_end_pfn = folio_pfn(folio) + folio_nr_pages(folio);
209 	const pte_t *end_ptep = start_ptep + max_nr;
210 	pte_t expected_pte, *ptep;
211 	bool writable, young, dirty;
212 	int nr;
213 
214 	if (any_writable)
215 		*any_writable = false;
216 	if (any_young)
217 		*any_young = false;
218 	if (any_dirty)
219 		*any_dirty = false;
220 
221 	VM_WARN_ON_FOLIO(!pte_present(pte), folio);
222 	VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio);
223 	VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio);
224 
225 	nr = pte_batch_hint(start_ptep, pte);
226 	expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags);
227 	ptep = start_ptep + nr;
228 
229 	while (ptep < end_ptep) {
230 		pte = ptep_get(ptep);
231 		if (any_writable)
232 			writable = !!pte_write(pte);
233 		if (any_young)
234 			young = !!pte_young(pte);
235 		if (any_dirty)
236 			dirty = !!pte_dirty(pte);
237 		pte = __pte_batch_clear_ignored(pte, flags);
238 
239 		if (!pte_same(pte, expected_pte))
240 			break;
241 
242 		/*
243 		 * Stop immediately once we reached the end of the folio. In
244 		 * corner cases the next PFN might fall into a different
245 		 * folio.
246 		 */
247 		if (pte_pfn(pte) >= folio_end_pfn)
248 			break;
249 
250 		if (any_writable)
251 			*any_writable |= writable;
252 		if (any_young)
253 			*any_young |= young;
254 		if (any_dirty)
255 			*any_dirty |= dirty;
256 
257 		nr = pte_batch_hint(ptep, pte);
258 		expected_pte = pte_advance_pfn(expected_pte, nr);
259 		ptep += nr;
260 	}
261 
262 	return min(ptep - start_ptep, max_nr);
263 }
264 
265 /**
266  * pte_move_swp_offset - Move the swap entry offset field of a swap pte
267  *	 forward or backward by delta
268  * @pte: The initial pte state; is_swap_pte(pte) must be true and
269  *	 non_swap_entry() must be false.
270  * @delta: The direction and the offset we are moving; forward if delta
271  *	 is positive; backward if delta is negative
272  *
273  * Moves the swap offset, while maintaining all other fields, including
274  * swap type, and any swp pte bits. The resulting pte is returned.
275  */
276 static inline pte_t pte_move_swp_offset(pte_t pte, long delta)
277 {
278 	swp_entry_t entry = pte_to_swp_entry(pte);
279 	pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry),
280 						   (swp_offset(entry) + delta)));
281 
282 	if (pte_swp_soft_dirty(pte))
283 		new = pte_swp_mksoft_dirty(new);
284 	if (pte_swp_exclusive(pte))
285 		new = pte_swp_mkexclusive(new);
286 	if (pte_swp_uffd_wp(pte))
287 		new = pte_swp_mkuffd_wp(new);
288 
289 	return new;
290 }
291 
292 
293 /**
294  * pte_next_swp_offset - Increment the swap entry offset field of a swap pte.
295  * @pte: The initial pte state; is_swap_pte(pte) must be true and
296  *	 non_swap_entry() must be false.
297  *
298  * Increments the swap offset, while maintaining all other fields, including
299  * swap type, and any swp pte bits. The resulting pte is returned.
300  */
301 static inline pte_t pte_next_swp_offset(pte_t pte)
302 {
303 	return pte_move_swp_offset(pte, 1);
304 }
305 
306 /**
307  * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries
308  * @start_ptep: Page table pointer for the first entry.
309  * @max_nr: The maximum number of table entries to consider.
310  * @pte: Page table entry for the first entry.
311  *
312  * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs
313  * containing swap entries all with consecutive offsets and targeting the same
314  * swap type, all with matching swp pte bits.
315  *
316  * max_nr must be at least one and must be limited by the caller so scanning
317  * cannot exceed a single page table.
318  *
319  * Return: the number of table entries in the batch.
320  */
321 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte)
322 {
323 	pte_t expected_pte = pte_next_swp_offset(pte);
324 	const pte_t *end_ptep = start_ptep + max_nr;
325 	swp_entry_t entry = pte_to_swp_entry(pte);
326 	pte_t *ptep = start_ptep + 1;
327 	unsigned short cgroup_id;
328 
329 	VM_WARN_ON(max_nr < 1);
330 	VM_WARN_ON(!is_swap_pte(pte));
331 	VM_WARN_ON(non_swap_entry(entry));
332 
333 	cgroup_id = lookup_swap_cgroup_id(entry);
334 	while (ptep < end_ptep) {
335 		pte = ptep_get(ptep);
336 
337 		if (!pte_same(pte, expected_pte))
338 			break;
339 		if (lookup_swap_cgroup_id(pte_to_swp_entry(pte)) != cgroup_id)
340 			break;
341 		expected_pte = pte_next_swp_offset(expected_pte);
342 		ptep++;
343 	}
344 
345 	return ptep - start_ptep;
346 }
347 #endif /* CONFIG_MMU */
348 
349 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
350 						int nr_throttled);
351 static inline void acct_reclaim_writeback(struct folio *folio)
352 {
353 	pg_data_t *pgdat = folio_pgdat(folio);
354 	int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
355 
356 	if (nr_throttled)
357 		__acct_reclaim_writeback(pgdat, folio, nr_throttled);
358 }
359 
360 static inline void wake_throttle_isolated(pg_data_t *pgdat)
361 {
362 	wait_queue_head_t *wqh;
363 
364 	wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
365 	if (waitqueue_active(wqh))
366 		wake_up(wqh);
367 }
368 
369 vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf);
370 static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf)
371 {
372 	vm_fault_t ret = __vmf_anon_prepare(vmf);
373 
374 	if (unlikely(ret & VM_FAULT_RETRY))
375 		vma_end_read(vmf->vma);
376 	return ret;
377 }
378 
379 vm_fault_t do_swap_page(struct vm_fault *vmf);
380 void folio_rotate_reclaimable(struct folio *folio);
381 bool __folio_end_writeback(struct folio *folio);
382 void deactivate_file_folio(struct folio *folio);
383 void folio_activate(struct folio *folio);
384 
385 void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas,
386 		   struct vm_area_struct *start_vma, unsigned long floor,
387 		   unsigned long ceiling, bool mm_wr_locked);
388 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
389 
390 struct zap_details;
391 void unmap_page_range(struct mmu_gather *tlb,
392 			     struct vm_area_struct *vma,
393 			     unsigned long addr, unsigned long end,
394 			     struct zap_details *details);
395 
396 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *,
397 		unsigned int order);
398 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
399 static inline void force_page_cache_readahead(struct address_space *mapping,
400 		struct file *file, pgoff_t index, unsigned long nr_to_read)
401 {
402 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
403 	force_page_cache_ra(&ractl, nr_to_read);
404 }
405 
406 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
407 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
408 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
409 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
410 void filemap_free_folio(struct address_space *mapping, struct folio *folio);
411 int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
412 bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
413 		loff_t end);
414 long mapping_evict_folio(struct address_space *mapping, struct folio *folio);
415 unsigned long mapping_try_invalidate(struct address_space *mapping,
416 		pgoff_t start, pgoff_t end, unsigned long *nr_failed);
417 
418 /**
419  * folio_evictable - Test whether a folio is evictable.
420  * @folio: The folio to test.
421  *
422  * Test whether @folio is evictable -- i.e., should be placed on
423  * active/inactive lists vs unevictable list.
424  *
425  * Reasons folio might not be evictable:
426  * 1. folio's mapping marked unevictable
427  * 2. One of the pages in the folio is part of an mlocked VMA
428  */
429 static inline bool folio_evictable(struct folio *folio)
430 {
431 	bool ret;
432 
433 	/* Prevent address_space of inode and swap cache from being freed */
434 	rcu_read_lock();
435 	ret = !mapping_unevictable(folio_mapping(folio)) &&
436 			!folio_test_mlocked(folio);
437 	rcu_read_unlock();
438 	return ret;
439 }
440 
441 /*
442  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
443  * a count of one.
444  */
445 static inline void set_page_refcounted(struct page *page)
446 {
447 	VM_BUG_ON_PAGE(PageTail(page), page);
448 	VM_BUG_ON_PAGE(page_ref_count(page), page);
449 	set_page_count(page, 1);
450 }
451 
452 /*
453  * Return true if a folio needs ->release_folio() calling upon it.
454  */
455 static inline bool folio_needs_release(struct folio *folio)
456 {
457 	struct address_space *mapping = folio_mapping(folio);
458 
459 	return folio_has_private(folio) ||
460 		(mapping && mapping_release_always(mapping));
461 }
462 
463 extern unsigned long highest_memmap_pfn;
464 
465 /*
466  * Maximum number of reclaim retries without progress before the OOM
467  * killer is consider the only way forward.
468  */
469 #define MAX_RECLAIM_RETRIES 16
470 
471 /*
472  * in mm/vmscan.c:
473  */
474 bool folio_isolate_lru(struct folio *folio);
475 void folio_putback_lru(struct folio *folio);
476 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
477 
478 /*
479  * in mm/rmap.c:
480  */
481 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
482 
483 /*
484  * in mm/page_alloc.c
485  */
486 #define K(x) ((x) << (PAGE_SHIFT-10))
487 
488 extern char * const zone_names[MAX_NR_ZONES];
489 
490 /* perform sanity checks on struct pages being allocated or freed */
491 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
492 
493 extern int min_free_kbytes;
494 
495 void setup_per_zone_wmarks(void);
496 void calculate_min_free_kbytes(void);
497 int __meminit init_per_zone_wmark_min(void);
498 void page_alloc_sysctl_init(void);
499 
500 /*
501  * Structure for holding the mostly immutable allocation parameters passed
502  * between functions involved in allocations, including the alloc_pages*
503  * family of functions.
504  *
505  * nodemask, migratetype and highest_zoneidx are initialized only once in
506  * __alloc_pages() and then never change.
507  *
508  * zonelist, preferred_zone and highest_zoneidx are set first in
509  * __alloc_pages() for the fast path, and might be later changed
510  * in __alloc_pages_slowpath(). All other functions pass the whole structure
511  * by a const pointer.
512  */
513 struct alloc_context {
514 	struct zonelist *zonelist;
515 	nodemask_t *nodemask;
516 	struct zoneref *preferred_zoneref;
517 	int migratetype;
518 
519 	/*
520 	 * highest_zoneidx represents highest usable zone index of
521 	 * the allocation request. Due to the nature of the zone,
522 	 * memory on lower zone than the highest_zoneidx will be
523 	 * protected by lowmem_reserve[highest_zoneidx].
524 	 *
525 	 * highest_zoneidx is also used by reclaim/compaction to limit
526 	 * the target zone since higher zone than this index cannot be
527 	 * usable for this allocation request.
528 	 */
529 	enum zone_type highest_zoneidx;
530 	bool spread_dirty_pages;
531 };
532 
533 /*
534  * This function returns the order of a free page in the buddy system. In
535  * general, page_zone(page)->lock must be held by the caller to prevent the
536  * page from being allocated in parallel and returning garbage as the order.
537  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
538  * page cannot be allocated or merged in parallel. Alternatively, it must
539  * handle invalid values gracefully, and use buddy_order_unsafe() below.
540  */
541 static inline unsigned int buddy_order(struct page *page)
542 {
543 	/* PageBuddy() must be checked by the caller */
544 	return page_private(page);
545 }
546 
547 /*
548  * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
549  * PageBuddy() should be checked first by the caller to minimize race window,
550  * and invalid values must be handled gracefully.
551  *
552  * READ_ONCE is used so that if the caller assigns the result into a local
553  * variable and e.g. tests it for valid range before using, the compiler cannot
554  * decide to remove the variable and inline the page_private(page) multiple
555  * times, potentially observing different values in the tests and the actual
556  * use of the result.
557  */
558 #define buddy_order_unsafe(page)	READ_ONCE(page_private(page))
559 
560 /*
561  * This function checks whether a page is free && is the buddy
562  * we can coalesce a page and its buddy if
563  * (a) the buddy is not in a hole (check before calling!) &&
564  * (b) the buddy is in the buddy system &&
565  * (c) a page and its buddy have the same order &&
566  * (d) a page and its buddy are in the same zone.
567  *
568  * For recording whether a page is in the buddy system, we set PageBuddy.
569  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
570  *
571  * For recording page's order, we use page_private(page).
572  */
573 static inline bool page_is_buddy(struct page *page, struct page *buddy,
574 				 unsigned int order)
575 {
576 	if (!page_is_guard(buddy) && !PageBuddy(buddy))
577 		return false;
578 
579 	if (buddy_order(buddy) != order)
580 		return false;
581 
582 	/*
583 	 * zone check is done late to avoid uselessly calculating
584 	 * zone/node ids for pages that could never merge.
585 	 */
586 	if (page_zone_id(page) != page_zone_id(buddy))
587 		return false;
588 
589 	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
590 
591 	return true;
592 }
593 
594 /*
595  * Locate the struct page for both the matching buddy in our
596  * pair (buddy1) and the combined O(n+1) page they form (page).
597  *
598  * 1) Any buddy B1 will have an order O twin B2 which satisfies
599  * the following equation:
600  *     B2 = B1 ^ (1 << O)
601  * For example, if the starting buddy (buddy2) is #8 its order
602  * 1 buddy is #10:
603  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
604  *
605  * 2) Any buddy B will have an order O+1 parent P which
606  * satisfies the following equation:
607  *     P = B & ~(1 << O)
608  *
609  * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER
610  */
611 static inline unsigned long
612 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
613 {
614 	return page_pfn ^ (1 << order);
615 }
616 
617 /*
618  * Find the buddy of @page and validate it.
619  * @page: The input page
620  * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
621  *       function is used in the performance-critical __free_one_page().
622  * @order: The order of the page
623  * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
624  *             page_to_pfn().
625  *
626  * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
627  * not the same as @page. The validation is necessary before use it.
628  *
629  * Return: the found buddy page or NULL if not found.
630  */
631 static inline struct page *find_buddy_page_pfn(struct page *page,
632 			unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
633 {
634 	unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
635 	struct page *buddy;
636 
637 	buddy = page + (__buddy_pfn - pfn);
638 	if (buddy_pfn)
639 		*buddy_pfn = __buddy_pfn;
640 
641 	if (page_is_buddy(page, buddy, order))
642 		return buddy;
643 	return NULL;
644 }
645 
646 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
647 				unsigned long end_pfn, struct zone *zone);
648 
649 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
650 				unsigned long end_pfn, struct zone *zone)
651 {
652 	if (zone->contiguous)
653 		return pfn_to_page(start_pfn);
654 
655 	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
656 }
657 
658 void set_zone_contiguous(struct zone *zone);
659 
660 static inline void clear_zone_contiguous(struct zone *zone)
661 {
662 	zone->contiguous = false;
663 }
664 
665 extern int __isolate_free_page(struct page *page, unsigned int order);
666 extern void __putback_isolated_page(struct page *page, unsigned int order,
667 				    int mt);
668 extern void memblock_free_pages(struct page *page, unsigned long pfn,
669 					unsigned int order);
670 extern void __free_pages_core(struct page *page, unsigned int order,
671 		enum meminit_context context);
672 
673 /*
674  * This will have no effect, other than possibly generating a warning, if the
675  * caller passes in a non-large folio.
676  */
677 static inline void folio_set_order(struct folio *folio, unsigned int order)
678 {
679 	if (WARN_ON_ONCE(!order || !folio_test_large(folio)))
680 		return;
681 
682 	folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order;
683 #ifdef CONFIG_64BIT
684 	folio->_folio_nr_pages = 1U << order;
685 #endif
686 }
687 
688 bool __folio_unqueue_deferred_split(struct folio *folio);
689 static inline bool folio_unqueue_deferred_split(struct folio *folio)
690 {
691 	if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio))
692 		return false;
693 
694 	/*
695 	 * At this point, there is no one trying to add the folio to
696 	 * deferred_list. If folio is not in deferred_list, it's safe
697 	 * to check without acquiring the split_queue_lock.
698 	 */
699 	if (data_race(list_empty(&folio->_deferred_list)))
700 		return false;
701 
702 	return __folio_unqueue_deferred_split(folio);
703 }
704 
705 static inline struct folio *page_rmappable_folio(struct page *page)
706 {
707 	struct folio *folio = (struct folio *)page;
708 
709 	if (folio && folio_test_large(folio))
710 		folio_set_large_rmappable(folio);
711 	return folio;
712 }
713 
714 static inline void prep_compound_head(struct page *page, unsigned int order)
715 {
716 	struct folio *folio = (struct folio *)page;
717 
718 	folio_set_order(folio, order);
719 	atomic_set(&folio->_large_mapcount, -1);
720 	atomic_set(&folio->_entire_mapcount, -1);
721 	atomic_set(&folio->_nr_pages_mapped, 0);
722 	atomic_set(&folio->_pincount, 0);
723 	if (order > 1)
724 		INIT_LIST_HEAD(&folio->_deferred_list);
725 }
726 
727 static inline void prep_compound_tail(struct page *head, int tail_idx)
728 {
729 	struct page *p = head + tail_idx;
730 
731 	p->mapping = TAIL_MAPPING;
732 	set_compound_head(p, head);
733 	set_page_private(p, 0);
734 }
735 
736 extern void prep_compound_page(struct page *page, unsigned int order);
737 
738 void post_alloc_hook(struct page *page, unsigned int order, gfp_t gfp_flags);
739 extern bool free_pages_prepare(struct page *page, unsigned int order);
740 
741 extern int user_min_free_kbytes;
742 
743 struct page *__alloc_frozen_pages_noprof(gfp_t, unsigned int order, int nid,
744 		nodemask_t *);
745 #define __alloc_frozen_pages(...) \
746 	alloc_hooks(__alloc_frozen_pages_noprof(__VA_ARGS__))
747 void free_frozen_pages(struct page *page, unsigned int order);
748 void free_unref_folios(struct folio_batch *fbatch);
749 
750 #ifdef CONFIG_NUMA
751 struct page *alloc_frozen_pages_noprof(gfp_t, unsigned int order);
752 #else
753 static inline struct page *alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order)
754 {
755 	return __alloc_frozen_pages_noprof(gfp, order, numa_node_id(), NULL);
756 }
757 #endif
758 
759 #define alloc_frozen_pages(...) \
760 	alloc_hooks(alloc_frozen_pages_noprof(__VA_ARGS__))
761 
762 extern void zone_pcp_reset(struct zone *zone);
763 extern void zone_pcp_disable(struct zone *zone);
764 extern void zone_pcp_enable(struct zone *zone);
765 extern void zone_pcp_init(struct zone *zone);
766 
767 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
768 			  phys_addr_t min_addr,
769 			  int nid, bool exact_nid);
770 
771 void memmap_init_range(unsigned long, int, unsigned long, unsigned long,
772 		unsigned long, enum meminit_context, struct vmem_altmap *, int);
773 
774 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
775 
776 /*
777  * in mm/compaction.c
778  */
779 /*
780  * compact_control is used to track pages being migrated and the free pages
781  * they are being migrated to during memory compaction. The free_pfn starts
782  * at the end of a zone and migrate_pfn begins at the start. Movable pages
783  * are moved to the end of a zone during a compaction run and the run
784  * completes when free_pfn <= migrate_pfn
785  */
786 struct compact_control {
787 	struct list_head freepages[NR_PAGE_ORDERS];	/* List of free pages to migrate to */
788 	struct list_head migratepages;	/* List of pages being migrated */
789 	unsigned int nr_freepages;	/* Number of isolated free pages */
790 	unsigned int nr_migratepages;	/* Number of pages to migrate */
791 	unsigned long free_pfn;		/* isolate_freepages search base */
792 	/*
793 	 * Acts as an in/out parameter to page isolation for migration.
794 	 * isolate_migratepages uses it as a search base.
795 	 * isolate_migratepages_block will update the value to the next pfn
796 	 * after the last isolated one.
797 	 */
798 	unsigned long migrate_pfn;
799 	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
800 	struct zone *zone;
801 	unsigned long total_migrate_scanned;
802 	unsigned long total_free_scanned;
803 	unsigned short fast_search_fail;/* failures to use free list searches */
804 	short search_order;		/* order to start a fast search at */
805 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
806 	int order;			/* order a direct compactor needs */
807 	int migratetype;		/* migratetype of direct compactor */
808 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
809 	const int highest_zoneidx;	/* zone index of a direct compactor */
810 	enum migrate_mode mode;		/* Async or sync migration mode */
811 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
812 	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
813 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
814 	bool direct_compaction;		/* False from kcompactd or /proc/... */
815 	bool proactive_compaction;	/* kcompactd proactive compaction */
816 	bool whole_zone;		/* Whole zone should/has been scanned */
817 	bool contended;			/* Signal lock contention */
818 	bool finish_pageblock;		/* Scan the remainder of a pageblock. Used
819 					 * when there are potentially transient
820 					 * isolation or migration failures to
821 					 * ensure forward progress.
822 					 */
823 	bool alloc_contig;		/* alloc_contig_range allocation */
824 };
825 
826 /*
827  * Used in direct compaction when a page should be taken from the freelists
828  * immediately when one is created during the free path.
829  */
830 struct capture_control {
831 	struct compact_control *cc;
832 	struct page *page;
833 };
834 
835 unsigned long
836 isolate_freepages_range(struct compact_control *cc,
837 			unsigned long start_pfn, unsigned long end_pfn);
838 int
839 isolate_migratepages_range(struct compact_control *cc,
840 			   unsigned long low_pfn, unsigned long end_pfn);
841 
842 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
843 void init_cma_reserved_pageblock(struct page *page);
844 
845 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
846 
847 int find_suitable_fallback(struct free_area *area, unsigned int order,
848 			int migratetype, bool only_stealable, bool *can_steal);
849 
850 static inline bool free_area_empty(struct free_area *area, int migratetype)
851 {
852 	return list_empty(&area->free_list[migratetype]);
853 }
854 
855 /* mm/util.c */
856 struct anon_vma *folio_anon_vma(const struct folio *folio);
857 
858 #ifdef CONFIG_MMU
859 void unmap_mapping_folio(struct folio *folio);
860 extern long populate_vma_page_range(struct vm_area_struct *vma,
861 		unsigned long start, unsigned long end, int *locked);
862 extern long faultin_page_range(struct mm_struct *mm, unsigned long start,
863 		unsigned long end, bool write, int *locked);
864 extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags,
865 			       unsigned long bytes);
866 
867 /*
868  * NOTE: This function can't tell whether the folio is "fully mapped" in the
869  * range.
870  * "fully mapped" means all the pages of folio is associated with the page
871  * table of range while this function just check whether the folio range is
872  * within the range [start, end). Function caller needs to do page table
873  * check if it cares about the page table association.
874  *
875  * Typical usage (like mlock or madvise) is:
876  * Caller knows at least 1 page of folio is associated with page table of VMA
877  * and the range [start, end) is intersect with the VMA range. Caller wants
878  * to know whether the folio is fully associated with the range. It calls
879  * this function to check whether the folio is in the range first. Then checks
880  * the page table to know whether the folio is fully mapped to the range.
881  */
882 static inline bool
883 folio_within_range(struct folio *folio, struct vm_area_struct *vma,
884 		unsigned long start, unsigned long end)
885 {
886 	pgoff_t pgoff, addr;
887 	unsigned long vma_pglen = vma_pages(vma);
888 
889 	VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio);
890 	if (start > end)
891 		return false;
892 
893 	if (start < vma->vm_start)
894 		start = vma->vm_start;
895 
896 	if (end > vma->vm_end)
897 		end = vma->vm_end;
898 
899 	pgoff = folio_pgoff(folio);
900 
901 	/* if folio start address is not in vma range */
902 	if (!in_range(pgoff, vma->vm_pgoff, vma_pglen))
903 		return false;
904 
905 	addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
906 
907 	return !(addr < start || end - addr < folio_size(folio));
908 }
909 
910 static inline bool
911 folio_within_vma(struct folio *folio, struct vm_area_struct *vma)
912 {
913 	return folio_within_range(folio, vma, vma->vm_start, vma->vm_end);
914 }
915 
916 /*
917  * mlock_vma_folio() and munlock_vma_folio():
918  * should be called with vma's mmap_lock held for read or write,
919  * under page table lock for the pte/pmd being added or removed.
920  *
921  * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at
922  * the end of folio_remove_rmap_*(); but new anon folios are managed by
923  * folio_add_lru_vma() calling mlock_new_folio().
924  */
925 void mlock_folio(struct folio *folio);
926 static inline void mlock_vma_folio(struct folio *folio,
927 				struct vm_area_struct *vma)
928 {
929 	/*
930 	 * The VM_SPECIAL check here serves two purposes.
931 	 * 1) VM_IO check prevents migration from double-counting during mlock.
932 	 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
933 	 *    is never left set on a VM_SPECIAL vma, there is an interval while
934 	 *    file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
935 	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
936 	 */
937 	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED))
938 		mlock_folio(folio);
939 }
940 
941 void munlock_folio(struct folio *folio);
942 static inline void munlock_vma_folio(struct folio *folio,
943 					struct vm_area_struct *vma)
944 {
945 	/*
946 	 * munlock if the function is called. Ideally, we should only
947 	 * do munlock if any page of folio is unmapped from VMA and
948 	 * cause folio not fully mapped to VMA.
949 	 *
950 	 * But it's not easy to confirm that's the situation. So we
951 	 * always munlock the folio and page reclaim will correct it
952 	 * if it's wrong.
953 	 */
954 	if (unlikely(vma->vm_flags & VM_LOCKED))
955 		munlock_folio(folio);
956 }
957 
958 void mlock_new_folio(struct folio *folio);
959 bool need_mlock_drain(int cpu);
960 void mlock_drain_local(void);
961 void mlock_drain_remote(int cpu);
962 
963 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
964 
965 /**
966  * vma_address - Find the virtual address a page range is mapped at
967  * @vma: The vma which maps this object.
968  * @pgoff: The page offset within its object.
969  * @nr_pages: The number of pages to consider.
970  *
971  * If any page in this range is mapped by this VMA, return the first address
972  * where any of these pages appear.  Otherwise, return -EFAULT.
973  */
974 static inline unsigned long vma_address(const struct vm_area_struct *vma,
975 		pgoff_t pgoff, unsigned long nr_pages)
976 {
977 	unsigned long address;
978 
979 	if (pgoff >= vma->vm_pgoff) {
980 		address = vma->vm_start +
981 			((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
982 		/* Check for address beyond vma (or wrapped through 0?) */
983 		if (address < vma->vm_start || address >= vma->vm_end)
984 			address = -EFAULT;
985 	} else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
986 		/* Test above avoids possibility of wrap to 0 on 32-bit */
987 		address = vma->vm_start;
988 	} else {
989 		address = -EFAULT;
990 	}
991 	return address;
992 }
993 
994 /*
995  * Then at what user virtual address will none of the range be found in vma?
996  * Assumes that vma_address() already returned a good starting address.
997  */
998 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
999 {
1000 	struct vm_area_struct *vma = pvmw->vma;
1001 	pgoff_t pgoff;
1002 	unsigned long address;
1003 
1004 	/* Common case, plus ->pgoff is invalid for KSM */
1005 	if (pvmw->nr_pages == 1)
1006 		return pvmw->address + PAGE_SIZE;
1007 
1008 	pgoff = pvmw->pgoff + pvmw->nr_pages;
1009 	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1010 	/* Check for address beyond vma (or wrapped through 0?) */
1011 	if (address < vma->vm_start || address > vma->vm_end)
1012 		address = vma->vm_end;
1013 	return address;
1014 }
1015 
1016 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
1017 						    struct file *fpin)
1018 {
1019 	int flags = vmf->flags;
1020 
1021 	if (fpin)
1022 		return fpin;
1023 
1024 	/*
1025 	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
1026 	 * anything, so we only pin the file and drop the mmap_lock if only
1027 	 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
1028 	 */
1029 	if (fault_flag_allow_retry_first(flags) &&
1030 	    !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
1031 		fpin = get_file(vmf->vma->vm_file);
1032 		release_fault_lock(vmf);
1033 	}
1034 	return fpin;
1035 }
1036 #else /* !CONFIG_MMU */
1037 static inline void unmap_mapping_folio(struct folio *folio) { }
1038 static inline void mlock_new_folio(struct folio *folio) { }
1039 static inline bool need_mlock_drain(int cpu) { return false; }
1040 static inline void mlock_drain_local(void) { }
1041 static inline void mlock_drain_remote(int cpu) { }
1042 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
1043 {
1044 }
1045 #endif /* !CONFIG_MMU */
1046 
1047 /* Memory initialisation debug and verification */
1048 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1049 DECLARE_STATIC_KEY_TRUE(deferred_pages);
1050 
1051 bool __init deferred_grow_zone(struct zone *zone, unsigned int order);
1052 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1053 
1054 enum mminit_level {
1055 	MMINIT_WARNING,
1056 	MMINIT_VERIFY,
1057 	MMINIT_TRACE
1058 };
1059 
1060 #ifdef CONFIG_DEBUG_MEMORY_INIT
1061 
1062 extern int mminit_loglevel;
1063 
1064 #define mminit_dprintk(level, prefix, fmt, arg...) \
1065 do { \
1066 	if (level < mminit_loglevel) { \
1067 		if (level <= MMINIT_WARNING) \
1068 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
1069 		else \
1070 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
1071 	} \
1072 } while (0)
1073 
1074 extern void mminit_verify_pageflags_layout(void);
1075 extern void mminit_verify_zonelist(void);
1076 #else
1077 
1078 static inline void mminit_dprintk(enum mminit_level level,
1079 				const char *prefix, const char *fmt, ...)
1080 {
1081 }
1082 
1083 static inline void mminit_verify_pageflags_layout(void)
1084 {
1085 }
1086 
1087 static inline void mminit_verify_zonelist(void)
1088 {
1089 }
1090 #endif /* CONFIG_DEBUG_MEMORY_INIT */
1091 
1092 #define NODE_RECLAIM_NOSCAN	-2
1093 #define NODE_RECLAIM_FULL	-1
1094 #define NODE_RECLAIM_SOME	0
1095 #define NODE_RECLAIM_SUCCESS	1
1096 
1097 #ifdef CONFIG_NUMA
1098 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
1099 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
1100 #else
1101 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
1102 				unsigned int order)
1103 {
1104 	return NODE_RECLAIM_NOSCAN;
1105 }
1106 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
1107 {
1108 	return NUMA_NO_NODE;
1109 }
1110 #endif
1111 
1112 /*
1113  * mm/memory-failure.c
1114  */
1115 #ifdef CONFIG_MEMORY_FAILURE
1116 void unmap_poisoned_folio(struct folio *folio, enum ttu_flags ttu);
1117 void shake_folio(struct folio *folio);
1118 extern int hwpoison_filter(struct page *p);
1119 
1120 extern u32 hwpoison_filter_dev_major;
1121 extern u32 hwpoison_filter_dev_minor;
1122 extern u64 hwpoison_filter_flags_mask;
1123 extern u64 hwpoison_filter_flags_value;
1124 extern u64 hwpoison_filter_memcg;
1125 extern u32 hwpoison_filter_enable;
1126 #define MAGIC_HWPOISON	0x48575053U	/* HWPS */
1127 void SetPageHWPoisonTakenOff(struct page *page);
1128 void ClearPageHWPoisonTakenOff(struct page *page);
1129 bool take_page_off_buddy(struct page *page);
1130 bool put_page_back_buddy(struct page *page);
1131 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
1132 void add_to_kill_ksm(struct task_struct *tsk, const struct page *p,
1133 		     struct vm_area_struct *vma, struct list_head *to_kill,
1134 		     unsigned long ksm_addr);
1135 unsigned long page_mapped_in_vma(const struct page *page,
1136 		struct vm_area_struct *vma);
1137 
1138 #else
1139 static inline void unmap_poisoned_folio(struct folio *folio, enum ttu_flags ttu)
1140 {
1141 }
1142 #endif
1143 
1144 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
1145         unsigned long, unsigned long,
1146         unsigned long, unsigned long);
1147 
1148 extern void set_pageblock_order(void);
1149 struct folio *alloc_migrate_folio(struct folio *src, unsigned long private);
1150 unsigned long reclaim_pages(struct list_head *folio_list);
1151 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
1152 					    struct list_head *folio_list);
1153 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1154 #define ALLOC_WMARK_MIN		WMARK_MIN
1155 #define ALLOC_WMARK_LOW		WMARK_LOW
1156 #define ALLOC_WMARK_HIGH	WMARK_HIGH
1157 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
1158 
1159 /* Mask to get the watermark bits */
1160 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
1161 
1162 /*
1163  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
1164  * cannot assume a reduced access to memory reserves is sufficient for
1165  * !MMU
1166  */
1167 #ifdef CONFIG_MMU
1168 #define ALLOC_OOM		0x08
1169 #else
1170 #define ALLOC_OOM		ALLOC_NO_WATERMARKS
1171 #endif
1172 
1173 #define ALLOC_NON_BLOCK		 0x10 /* Caller cannot block. Allow access
1174 				       * to 25% of the min watermark or
1175 				       * 62.5% if __GFP_HIGH is set.
1176 				       */
1177 #define ALLOC_MIN_RESERVE	 0x20 /* __GFP_HIGH set. Allow access to 50%
1178 				       * of the min watermark.
1179 				       */
1180 #define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
1181 #define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
1182 #ifdef CONFIG_ZONE_DMA32
1183 #define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
1184 #else
1185 #define ALLOC_NOFRAGMENT	  0x0
1186 #endif
1187 #define ALLOC_HIGHATOMIC	0x200 /* Allows access to MIGRATE_HIGHATOMIC */
1188 #define ALLOC_KSWAPD		0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
1189 
1190 /* Flags that allow allocations below the min watermark. */
1191 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
1192 
1193 enum ttu_flags;
1194 struct tlbflush_unmap_batch;
1195 
1196 
1197 /*
1198  * only for MM internal work items which do not depend on
1199  * any allocations or locks which might depend on allocations
1200  */
1201 extern struct workqueue_struct *mm_percpu_wq;
1202 
1203 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1204 void try_to_unmap_flush(void);
1205 void try_to_unmap_flush_dirty(void);
1206 void flush_tlb_batched_pending(struct mm_struct *mm);
1207 #else
1208 static inline void try_to_unmap_flush(void)
1209 {
1210 }
1211 static inline void try_to_unmap_flush_dirty(void)
1212 {
1213 }
1214 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
1215 {
1216 }
1217 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
1218 
1219 extern const struct trace_print_flags pageflag_names[];
1220 extern const struct trace_print_flags vmaflag_names[];
1221 extern const struct trace_print_flags gfpflag_names[];
1222 
1223 static inline bool is_migrate_highatomic(enum migratetype migratetype)
1224 {
1225 	return migratetype == MIGRATE_HIGHATOMIC;
1226 }
1227 
1228 void setup_zone_pageset(struct zone *zone);
1229 
1230 struct migration_target_control {
1231 	int nid;		/* preferred node id */
1232 	nodemask_t *nmask;
1233 	gfp_t gfp_mask;
1234 	enum migrate_reason reason;
1235 };
1236 
1237 /*
1238  * mm/filemap.c
1239  */
1240 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
1241 			      struct folio *folio, loff_t fpos, size_t size);
1242 
1243 /*
1244  * mm/vmalloc.c
1245  */
1246 #ifdef CONFIG_MMU
1247 void __init vmalloc_init(void);
1248 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1249                 pgprot_t prot, struct page **pages, unsigned int page_shift);
1250 unsigned int get_vm_area_page_order(struct vm_struct *vm);
1251 #else
1252 static inline void vmalloc_init(void)
1253 {
1254 }
1255 
1256 static inline
1257 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1258                 pgprot_t prot, struct page **pages, unsigned int page_shift)
1259 {
1260 	return -EINVAL;
1261 }
1262 #endif
1263 
1264 int __must_check __vmap_pages_range_noflush(unsigned long addr,
1265 			       unsigned long end, pgprot_t prot,
1266 			       struct page **pages, unsigned int page_shift);
1267 
1268 void vunmap_range_noflush(unsigned long start, unsigned long end);
1269 
1270 void __vunmap_range_noflush(unsigned long start, unsigned long end);
1271 
1272 int numa_migrate_check(struct folio *folio, struct vm_fault *vmf,
1273 		      unsigned long addr, int *flags, bool writable,
1274 		      int *last_cpupid);
1275 
1276 void free_zone_device_folio(struct folio *folio);
1277 int migrate_device_coherent_folio(struct folio *folio);
1278 
1279 struct vm_struct *__get_vm_area_node(unsigned long size,
1280 				     unsigned long align, unsigned long shift,
1281 				     unsigned long flags, unsigned long start,
1282 				     unsigned long end, int node, gfp_t gfp_mask,
1283 				     const void *caller);
1284 
1285 /*
1286  * mm/gup.c
1287  */
1288 int __must_check try_grab_folio(struct folio *folio, int refs,
1289 				unsigned int flags);
1290 
1291 /*
1292  * mm/huge_memory.c
1293  */
1294 void touch_pud(struct vm_area_struct *vma, unsigned long addr,
1295 	       pud_t *pud, bool write);
1296 void touch_pmd(struct vm_area_struct *vma, unsigned long addr,
1297 	       pmd_t *pmd, bool write);
1298 
1299 /*
1300  * Parses a string with mem suffixes into its order. Useful to parse kernel
1301  * parameters.
1302  */
1303 static inline int get_order_from_str(const char *size_str,
1304 				     unsigned long valid_orders)
1305 {
1306 	unsigned long size;
1307 	char *endptr;
1308 	int order;
1309 
1310 	size = memparse(size_str, &endptr);
1311 
1312 	if (!is_power_of_2(size))
1313 		return -EINVAL;
1314 	order = get_order(size);
1315 	if (BIT(order) & ~valid_orders)
1316 		return -EINVAL;
1317 
1318 	return order;
1319 }
1320 
1321 enum {
1322 	/* mark page accessed */
1323 	FOLL_TOUCH = 1 << 16,
1324 	/* a retry, previous pass started an IO */
1325 	FOLL_TRIED = 1 << 17,
1326 	/* we are working on non-current tsk/mm */
1327 	FOLL_REMOTE = 1 << 18,
1328 	/* pages must be released via unpin_user_page */
1329 	FOLL_PIN = 1 << 19,
1330 	/* gup_fast: prevent fall-back to slow gup */
1331 	FOLL_FAST_ONLY = 1 << 20,
1332 	/* allow unlocking the mmap lock */
1333 	FOLL_UNLOCKABLE = 1 << 21,
1334 	/* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */
1335 	FOLL_MADV_POPULATE = 1 << 22,
1336 };
1337 
1338 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \
1339 			    FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \
1340 			    FOLL_MADV_POPULATE)
1341 
1342 /*
1343  * Indicates for which pages that are write-protected in the page table,
1344  * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
1345  * GUP pin will remain consistent with the pages mapped into the page tables
1346  * of the MM.
1347  *
1348  * Temporary unmapping of PageAnonExclusive() pages or clearing of
1349  * PageAnonExclusive() has to protect against concurrent GUP:
1350  * * Ordinary GUP: Using the PT lock
1351  * * GUP-fast and fork(): mm->write_protect_seq
1352  * * GUP-fast and KSM or temporary unmapping (swap, migration): see
1353  *    folio_try_share_anon_rmap_*()
1354  *
1355  * Must be called with the (sub)page that's actually referenced via the
1356  * page table entry, which might not necessarily be the head page for a
1357  * PTE-mapped THP.
1358  *
1359  * If the vma is NULL, we're coming from the GUP-fast path and might have
1360  * to fallback to the slow path just to lookup the vma.
1361  */
1362 static inline bool gup_must_unshare(struct vm_area_struct *vma,
1363 				    unsigned int flags, struct page *page)
1364 {
1365 	/*
1366 	 * FOLL_WRITE is implicitly handled correctly as the page table entry
1367 	 * has to be writable -- and if it references (part of) an anonymous
1368 	 * folio, that part is required to be marked exclusive.
1369 	 */
1370 	if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
1371 		return false;
1372 	/*
1373 	 * Note: PageAnon(page) is stable until the page is actually getting
1374 	 * freed.
1375 	 */
1376 	if (!PageAnon(page)) {
1377 		/*
1378 		 * We only care about R/O long-term pining: R/O short-term
1379 		 * pinning does not have the semantics to observe successive
1380 		 * changes through the process page tables.
1381 		 */
1382 		if (!(flags & FOLL_LONGTERM))
1383 			return false;
1384 
1385 		/* We really need the vma ... */
1386 		if (!vma)
1387 			return true;
1388 
1389 		/*
1390 		 * ... because we only care about writable private ("COW")
1391 		 * mappings where we have to break COW early.
1392 		 */
1393 		return is_cow_mapping(vma->vm_flags);
1394 	}
1395 
1396 	/* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */
1397 	if (IS_ENABLED(CONFIG_HAVE_GUP_FAST))
1398 		smp_rmb();
1399 
1400 	/*
1401 	 * Note that KSM pages cannot be exclusive, and consequently,
1402 	 * cannot get pinned.
1403 	 */
1404 	return !PageAnonExclusive(page);
1405 }
1406 
1407 extern bool mirrored_kernelcore;
1408 extern bool memblock_has_mirror(void);
1409 
1410 static __always_inline void vma_set_range(struct vm_area_struct *vma,
1411 					  unsigned long start, unsigned long end,
1412 					  pgoff_t pgoff)
1413 {
1414 	vma->vm_start = start;
1415 	vma->vm_end = end;
1416 	vma->vm_pgoff = pgoff;
1417 }
1418 
1419 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1420 {
1421 	/*
1422 	 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1423 	 * enablements, because when without soft-dirty being compiled in,
1424 	 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1425 	 * will be constantly true.
1426 	 */
1427 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
1428 		return false;
1429 
1430 	/*
1431 	 * Soft-dirty is kind of special: its tracking is enabled when the
1432 	 * vma flags not set.
1433 	 */
1434 	return !(vma->vm_flags & VM_SOFTDIRTY);
1435 }
1436 
1437 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd)
1438 {
1439 	return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd);
1440 }
1441 
1442 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte)
1443 {
1444 	return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte);
1445 }
1446 
1447 void __meminit __init_single_page(struct page *page, unsigned long pfn,
1448 				unsigned long zone, int nid);
1449 
1450 /* shrinker related functions */
1451 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg,
1452 			  int priority);
1453 
1454 #ifdef CONFIG_SHRINKER_DEBUG
1455 static inline __printf(2, 0) int shrinker_debugfs_name_alloc(
1456 			struct shrinker *shrinker, const char *fmt, va_list ap)
1457 {
1458 	shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap);
1459 
1460 	return shrinker->name ? 0 : -ENOMEM;
1461 }
1462 
1463 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1464 {
1465 	kfree_const(shrinker->name);
1466 	shrinker->name = NULL;
1467 }
1468 
1469 extern int shrinker_debugfs_add(struct shrinker *shrinker);
1470 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1471 					      int *debugfs_id);
1472 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1473 				    int debugfs_id);
1474 #else /* CONFIG_SHRINKER_DEBUG */
1475 static inline int shrinker_debugfs_add(struct shrinker *shrinker)
1476 {
1477 	return 0;
1478 }
1479 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker,
1480 					      const char *fmt, va_list ap)
1481 {
1482 	return 0;
1483 }
1484 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1485 {
1486 }
1487 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1488 						     int *debugfs_id)
1489 {
1490 	*debugfs_id = -1;
1491 	return NULL;
1492 }
1493 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1494 					   int debugfs_id)
1495 {
1496 }
1497 #endif /* CONFIG_SHRINKER_DEBUG */
1498 
1499 /* Only track the nodes of mappings with shadow entries */
1500 void workingset_update_node(struct xa_node *node);
1501 extern struct list_lru shadow_nodes;
1502 #define mapping_set_update(xas, mapping) do {			\
1503 	if (!dax_mapping(mapping) && !shmem_mapping(mapping)) {	\
1504 		xas_set_update(xas, workingset_update_node);	\
1505 		xas_set_lru(xas, &shadow_nodes);		\
1506 	}							\
1507 } while (0)
1508 
1509 /* mremap.c */
1510 unsigned long move_page_tables(struct vm_area_struct *vma,
1511 	unsigned long old_addr, struct vm_area_struct *new_vma,
1512 	unsigned long new_addr, unsigned long len,
1513 	bool need_rmap_locks, bool for_stack);
1514 
1515 #ifdef CONFIG_UNACCEPTED_MEMORY
1516 void accept_page(struct page *page);
1517 #else /* CONFIG_UNACCEPTED_MEMORY */
1518 static inline void accept_page(struct page *page)
1519 {
1520 }
1521 #endif /* CONFIG_UNACCEPTED_MEMORY */
1522 
1523 /* pagewalk.c */
1524 int walk_page_range_mm(struct mm_struct *mm, unsigned long start,
1525 		unsigned long end, const struct mm_walk_ops *ops,
1526 		void *private);
1527 
1528 /* pt_reclaim.c */
1529 bool try_get_and_clear_pmd(struct mm_struct *mm, pmd_t *pmd, pmd_t *pmdval);
1530 void free_pte(struct mm_struct *mm, unsigned long addr, struct mmu_gather *tlb,
1531 	      pmd_t pmdval);
1532 void try_to_free_pte(struct mm_struct *mm, pmd_t *pmd, unsigned long addr,
1533 		     struct mmu_gather *tlb);
1534 
1535 #ifdef CONFIG_PT_RECLAIM
1536 bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
1537 			   struct zap_details *details);
1538 #else
1539 static inline bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
1540 					 struct zap_details *details)
1541 {
1542 	return false;
1543 }
1544 #endif /* CONFIG_PT_RECLAIM */
1545 
1546 
1547 #endif	/* __MM_INTERNAL_H */
1548