xref: /linux/mm/internal.h (revision 6aacab308a5dfd222b2d23662bbae60c11007cfb)
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/leafops.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  * Maintains state across a page table move. The operation assumes both source
29  * and destination VMAs already exist and are specified by the user.
30  *
31  * Partial moves are permitted, but the old and new ranges must both reside
32  * within a VMA.
33  *
34  * mmap lock must be held in write and VMA write locks must be held on any VMA
35  * that is visible.
36  *
37  * Use the PAGETABLE_MOVE() macro to initialise this struct.
38  *
39  * The old_addr and new_addr fields are updated as the page table move is
40  * executed.
41  *
42  * NOTE: The page table move is affected by reading from [old_addr, old_end),
43  * and old_addr may be updated for better page table alignment, so len_in
44  * represents the length of the range being copied as specified by the user.
45  */
46 struct pagetable_move_control {
47 	struct vm_area_struct *old; /* Source VMA. */
48 	struct vm_area_struct *new; /* Destination VMA. */
49 	unsigned long old_addr; /* Address from which the move begins. */
50 	unsigned long old_end; /* Exclusive address at which old range ends. */
51 	unsigned long new_addr; /* Address to move page tables to. */
52 	unsigned long len_in; /* Bytes to remap specified by user. */
53 
54 	bool need_rmap_locks; /* Do rmap locks need to be taken? */
55 	bool for_stack; /* Is this an early temp stack being moved? */
56 };
57 
58 #define PAGETABLE_MOVE(name, old_, new_, old_addr_, new_addr_, len_)	\
59 	struct pagetable_move_control name = {				\
60 		.old = old_,						\
61 		.new = new_,						\
62 		.old_addr = old_addr_,					\
63 		.old_end = (old_addr_) + (len_),			\
64 		.new_addr = new_addr_,					\
65 		.len_in = len_,						\
66 	}
67 
68 /*
69  * The set of flags that only affect watermark checking and reclaim
70  * behaviour. This is used by the MM to obey the caller constraints
71  * about IO, FS and watermark checking while ignoring placement
72  * hints such as HIGHMEM usage.
73  */
74 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
75 			__GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
76 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
77 			__GFP_NOLOCKDEP)
78 
79 /* The GFP flags allowed during early boot */
80 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
81 
82 /* Control allocation cpuset and node placement constraints */
83 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
84 
85 /* Do not use these with a slab allocator */
86 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
87 
88 /*
89  * Different from WARN_ON_ONCE(), no warning will be issued
90  * when we specify __GFP_NOWARN.
91  */
92 #define WARN_ON_ONCE_GFP(cond, gfp)	({				\
93 	static bool __section(".data..once") __warned;			\
94 	int __ret_warn_once = !!(cond);					\
95 									\
96 	if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
97 		__warned = true;					\
98 		WARN_ON(1);						\
99 	}								\
100 	unlikely(__ret_warn_once);					\
101 })
102 
103 void page_writeback_init(void);
104 
105 /*
106  * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
107  * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit
108  * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE).  Hugetlb currently
109  * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
110  */
111 #define ENTIRELY_MAPPED		0x800000
112 #define FOLIO_PAGES_MAPPED	(ENTIRELY_MAPPED - 1)
113 
114 /*
115  * Flags passed to __show_mem() and show_free_areas() to suppress output in
116  * various contexts.
117  */
118 #define SHOW_MEM_FILTER_NODES		(0x0001u)	/* disallowed nodes */
119 
120 /*
121  * How many individual pages have an elevated _mapcount.  Excludes
122  * the folio's entire_mapcount.
123  *
124  * Don't use this function outside of debugging code.
125  */
126 static inline int folio_nr_pages_mapped(const struct folio *folio)
127 {
128 	if (IS_ENABLED(CONFIG_NO_PAGE_MAPCOUNT))
129 		return -1;
130 	return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
131 }
132 
133 /*
134  * Retrieve the first entry of a folio based on a provided entry within the
135  * folio. We cannot rely on folio->swap as there is no guarantee that it has
136  * been initialized. Used for calling arch_swap_restore()
137  */
138 static inline swp_entry_t folio_swap(swp_entry_t entry,
139 		const struct folio *folio)
140 {
141 	swp_entry_t swap = {
142 		.val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)),
143 	};
144 
145 	return swap;
146 }
147 
148 static inline void *folio_raw_mapping(const struct folio *folio)
149 {
150 	unsigned long mapping = (unsigned long)folio->mapping;
151 
152 	return (void *)(mapping & ~FOLIO_MAPPING_FLAGS);
153 }
154 
155 /*
156  * This is a file-backed mapping, and is about to be memory mapped - invoke its
157  * mmap hook and safely handle error conditions. On error, VMA hooks will be
158  * mutated.
159  *
160  * @file: File which backs the mapping.
161  * @vma:  VMA which we are mapping.
162  *
163  * Returns: 0 if success, error otherwise.
164  */
165 static inline int mmap_file(struct file *file, struct vm_area_struct *vma)
166 {
167 	int err = vfs_mmap(file, vma);
168 
169 	if (likely(!err))
170 		return 0;
171 
172 	/*
173 	 * OK, we tried to call the file hook for mmap(), but an error
174 	 * arose. The mapping is in an inconsistent state and we must not invoke
175 	 * any further hooks on it.
176 	 */
177 	vma->vm_ops = &vma_dummy_vm_ops;
178 
179 	return err;
180 }
181 
182 /*
183  * If the VMA has a close hook then close it, and since closing it might leave
184  * it in an inconsistent state which makes the use of any hooks suspect, clear
185  * them down by installing dummy empty hooks.
186  */
187 static inline void vma_close(struct vm_area_struct *vma)
188 {
189 	if (vma->vm_ops && vma->vm_ops->close) {
190 		vma->vm_ops->close(vma);
191 
192 		/*
193 		 * The mapping is in an inconsistent state, and no further hooks
194 		 * may be invoked upon it.
195 		 */
196 		vma->vm_ops = &vma_dummy_vm_ops;
197 	}
198 }
199 
200 /* unmap_vmas is in mm/memory.c */
201 void unmap_vmas(struct mmu_gather *tlb, struct unmap_desc *unmap);
202 
203 #ifdef CONFIG_MMU
204 
205 static inline void get_anon_vma(struct anon_vma *anon_vma)
206 {
207 	atomic_inc(&anon_vma->refcount);
208 }
209 
210 void __put_anon_vma(struct anon_vma *anon_vma);
211 
212 static inline void put_anon_vma(struct anon_vma *anon_vma)
213 {
214 	if (atomic_dec_and_test(&anon_vma->refcount))
215 		__put_anon_vma(anon_vma);
216 }
217 
218 static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
219 {
220 	down_write(&anon_vma->root->rwsem);
221 }
222 
223 static inline int anon_vma_trylock_write(struct anon_vma *anon_vma)
224 {
225 	return down_write_trylock(&anon_vma->root->rwsem);
226 }
227 
228 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
229 {
230 	up_write(&anon_vma->root->rwsem);
231 }
232 
233 static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
234 {
235 	down_read(&anon_vma->root->rwsem);
236 }
237 
238 static inline int anon_vma_trylock_read(struct anon_vma *anon_vma)
239 {
240 	return down_read_trylock(&anon_vma->root->rwsem);
241 }
242 
243 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
244 {
245 	up_read(&anon_vma->root->rwsem);
246 }
247 
248 struct anon_vma *folio_get_anon_vma(const struct folio *folio);
249 
250 /* Operations which modify VMAs. */
251 enum vma_operation {
252 	VMA_OP_SPLIT,
253 	VMA_OP_MERGE_UNFAULTED,
254 	VMA_OP_REMAP,
255 	VMA_OP_FORK,
256 };
257 
258 int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src,
259 	enum vma_operation operation);
260 int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma);
261 int  __anon_vma_prepare(struct vm_area_struct *vma);
262 void unlink_anon_vmas(struct vm_area_struct *vma);
263 
264 static inline int anon_vma_prepare(struct vm_area_struct *vma)
265 {
266 	if (likely(vma->anon_vma))
267 		return 0;
268 
269 	return __anon_vma_prepare(vma);
270 }
271 
272 /* Flags for folio_pte_batch(). */
273 typedef int __bitwise fpb_t;
274 
275 /* Compare PTEs respecting the dirty bit. */
276 #define FPB_RESPECT_DIRTY		((__force fpb_t)BIT(0))
277 
278 /* Compare PTEs respecting the soft-dirty bit. */
279 #define FPB_RESPECT_SOFT_DIRTY		((__force fpb_t)BIT(1))
280 
281 /* Compare PTEs respecting the writable bit. */
282 #define FPB_RESPECT_WRITE		((__force fpb_t)BIT(2))
283 
284 /*
285  * Merge PTE write bits: if any PTE in the batch is writable, modify the
286  * PTE at @ptentp to be writable.
287  */
288 #define FPB_MERGE_WRITE			((__force fpb_t)BIT(3))
289 
290 /*
291  * Merge PTE young and dirty bits: if any PTE in the batch is young or dirty,
292  * modify the PTE at @ptentp to be young or dirty, respectively.
293  */
294 #define FPB_MERGE_YOUNG_DIRTY		((__force fpb_t)BIT(4))
295 
296 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
297 {
298 	if (!(flags & FPB_RESPECT_DIRTY))
299 		pte = pte_mkclean(pte);
300 	if (likely(!(flags & FPB_RESPECT_SOFT_DIRTY)))
301 		pte = pte_clear_soft_dirty(pte);
302 	if (likely(!(flags & FPB_RESPECT_WRITE)))
303 		pte = pte_wrprotect(pte);
304 	return pte_mkold(pte);
305 }
306 
307 /**
308  * folio_pte_batch_flags - detect a PTE batch for a large folio
309  * @folio: The large folio to detect a PTE batch for.
310  * @vma: The VMA. Only relevant with FPB_MERGE_WRITE, otherwise can be NULL.
311  * @ptep: Page table pointer for the first entry.
312  * @ptentp: Pointer to a COPY of the first page table entry whose flags this
313  *	    function updates based on @flags if appropriate.
314  * @max_nr: The maximum number of table entries to consider.
315  * @flags: Flags to modify the PTE batch semantics.
316  *
317  * Detect a PTE batch: consecutive (present) PTEs that map consecutive
318  * pages of the same large folio in a single VMA and a single page table.
319  *
320  * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN,
321  * the accessed bit, writable bit, dirty bit (unless FPB_RESPECT_DIRTY is set)
322  * and soft-dirty bit (unless FPB_RESPECT_SOFT_DIRTY is set).
323  *
324  * @ptep must map any page of the folio. max_nr must be at least one and
325  * must be limited by the caller so scanning cannot exceed a single VMA and
326  * a single page table.
327  *
328  * Depending on the FPB_MERGE_* flags, the pte stored at @ptentp will
329  * be updated: it's crucial that a pointer to a COPY of the first
330  * page table entry, obtained through ptep_get(), is provided as @ptentp.
331  *
332  * This function will be inlined to optimize based on the input parameters;
333  * consider using folio_pte_batch() instead if applicable.
334  *
335  * Return: the number of table entries in the batch.
336  */
337 static inline unsigned int folio_pte_batch_flags(struct folio *folio,
338 		struct vm_area_struct *vma, pte_t *ptep, pte_t *ptentp,
339 		unsigned int max_nr, fpb_t flags)
340 {
341 	bool any_writable = false, any_young = false, any_dirty = false;
342 	pte_t expected_pte, pte = *ptentp;
343 	unsigned int nr, cur_nr;
344 
345 	VM_WARN_ON_FOLIO(!pte_present(pte), folio);
346 	VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio);
347 	VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio);
348 	/*
349 	 * Ensure this is a pointer to a copy not a pointer into a page table.
350 	 * If this is a stack value, it won't be a valid virtual address, but
351 	 * that's fine because it also cannot be pointing into the page table.
352 	 */
353 	VM_WARN_ON(virt_addr_valid(ptentp) && PageTable(virt_to_page(ptentp)));
354 
355 	/* Limit max_nr to the actual remaining PFNs in the folio we could batch. */
356 	max_nr = min_t(unsigned long, max_nr,
357 		       folio_pfn(folio) + folio_nr_pages(folio) - pte_pfn(pte));
358 
359 	nr = pte_batch_hint(ptep, pte);
360 	expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags);
361 	ptep = ptep + nr;
362 
363 	while (nr < max_nr) {
364 		pte = ptep_get(ptep);
365 
366 		if (!pte_same(__pte_batch_clear_ignored(pte, flags), expected_pte))
367 			break;
368 
369 		if (flags & FPB_MERGE_WRITE)
370 			any_writable |= pte_write(pte);
371 		if (flags & FPB_MERGE_YOUNG_DIRTY) {
372 			any_young |= pte_young(pte);
373 			any_dirty |= pte_dirty(pte);
374 		}
375 
376 		cur_nr = pte_batch_hint(ptep, pte);
377 		expected_pte = pte_advance_pfn(expected_pte, cur_nr);
378 		ptep += cur_nr;
379 		nr += cur_nr;
380 	}
381 
382 	if (any_writable)
383 		*ptentp = pte_mkwrite(*ptentp, vma);
384 	if (any_young)
385 		*ptentp = pte_mkyoung(*ptentp);
386 	if (any_dirty)
387 		*ptentp = pte_mkdirty(*ptentp);
388 
389 	return min(nr, max_nr);
390 }
391 
392 unsigned int folio_pte_batch(struct folio *folio, pte_t *ptep, pte_t pte,
393 		unsigned int max_nr);
394 
395 /**
396  * pte_move_swp_offset - Move the swap entry offset field of a swap pte
397  *	 forward or backward by delta
398  * @pte: The initial pte state; must be a swap entry
399  * @delta: The direction and the offset we are moving; forward if delta
400  *	 is positive; backward if delta is negative
401  *
402  * Moves the swap offset, while maintaining all other fields, including
403  * swap type, and any swp pte bits. The resulting pte is returned.
404  */
405 static inline pte_t pte_move_swp_offset(pte_t pte, long delta)
406 {
407 	const softleaf_t entry = softleaf_from_pte(pte);
408 	pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry),
409 						   (swp_offset(entry) + delta)));
410 
411 	if (pte_swp_soft_dirty(pte))
412 		new = pte_swp_mksoft_dirty(new);
413 	if (pte_swp_exclusive(pte))
414 		new = pte_swp_mkexclusive(new);
415 	if (pte_swp_uffd_wp(pte))
416 		new = pte_swp_mkuffd_wp(new);
417 
418 	return new;
419 }
420 
421 
422 /**
423  * pte_next_swp_offset - Increment the swap entry offset field of a swap pte.
424  * @pte: The initial pte state; must be a swap entry.
425  *
426  * Increments the swap offset, while maintaining all other fields, including
427  * swap type, and any swp pte bits. The resulting pte is returned.
428  */
429 static inline pte_t pte_next_swp_offset(pte_t pte)
430 {
431 	return pte_move_swp_offset(pte, 1);
432 }
433 
434 /**
435  * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries
436  * @start_ptep: Page table pointer for the first entry.
437  * @max_nr: The maximum number of table entries to consider.
438  * @pte: Page table entry for the first entry.
439  *
440  * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs
441  * containing swap entries all with consecutive offsets and targeting the same
442  * swap type, all with matching swp pte bits.
443  *
444  * max_nr must be at least one and must be limited by the caller so scanning
445  * cannot exceed a single page table.
446  *
447  * Return: the number of table entries in the batch.
448  */
449 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte)
450 {
451 	pte_t expected_pte = pte_next_swp_offset(pte);
452 	const pte_t *end_ptep = start_ptep + max_nr;
453 	const softleaf_t entry = softleaf_from_pte(pte);
454 	pte_t *ptep = start_ptep + 1;
455 	unsigned short cgroup_id;
456 
457 	VM_WARN_ON(max_nr < 1);
458 	VM_WARN_ON(!softleaf_is_swap(entry));
459 
460 	cgroup_id = lookup_swap_cgroup_id(entry);
461 	while (ptep < end_ptep) {
462 		softleaf_t entry;
463 
464 		pte = ptep_get(ptep);
465 
466 		if (!pte_same(pte, expected_pte))
467 			break;
468 		entry = softleaf_from_pte(pte);
469 		if (lookup_swap_cgroup_id(entry) != cgroup_id)
470 			break;
471 		expected_pte = pte_next_swp_offset(expected_pte);
472 		ptep++;
473 	}
474 
475 	return ptep - start_ptep;
476 }
477 #endif /* CONFIG_MMU */
478 
479 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
480 						int nr_throttled);
481 static inline void acct_reclaim_writeback(struct folio *folio)
482 {
483 	pg_data_t *pgdat = folio_pgdat(folio);
484 	int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
485 
486 	if (nr_throttled)
487 		__acct_reclaim_writeback(pgdat, folio, nr_throttled);
488 }
489 
490 static inline void wake_throttle_isolated(pg_data_t *pgdat)
491 {
492 	wait_queue_head_t *wqh;
493 
494 	wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
495 	if (waitqueue_active(wqh))
496 		wake_up(wqh);
497 }
498 
499 vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf);
500 static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf)
501 {
502 	vm_fault_t ret = __vmf_anon_prepare(vmf);
503 
504 	if (unlikely(ret & VM_FAULT_RETRY))
505 		vma_end_read(vmf->vma);
506 	return ret;
507 }
508 
509 vm_fault_t do_swap_page(struct vm_fault *vmf);
510 void folio_rotate_reclaimable(struct folio *folio);
511 bool __folio_end_writeback(struct folio *folio);
512 void deactivate_file_folio(struct folio *folio);
513 void folio_activate(struct folio *folio);
514 
515 void free_pgtables(struct mmu_gather *tlb, struct unmap_desc *desc);
516 
517 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
518 
519 struct zap_details;
520 void unmap_page_range(struct mmu_gather *tlb,
521 			     struct vm_area_struct *vma,
522 			     unsigned long addr, unsigned long end,
523 			     struct zap_details *details);
524 void zap_page_range_single_batched(struct mmu_gather *tlb,
525 		struct vm_area_struct *vma, unsigned long addr,
526 		unsigned long size, struct zap_details *details);
527 int folio_unmap_invalidate(struct address_space *mapping, struct folio *folio,
528 			   gfp_t gfp);
529 
530 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *);
531 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
532 static inline void force_page_cache_readahead(struct address_space *mapping,
533 		struct file *file, pgoff_t index, unsigned long nr_to_read)
534 {
535 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
536 	force_page_cache_ra(&ractl, nr_to_read);
537 }
538 
539 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
540 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
541 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
542 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
543 void filemap_free_folio(struct address_space *mapping, struct folio *folio);
544 int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
545 bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
546 		loff_t end);
547 long mapping_evict_folio(struct address_space *mapping, struct folio *folio);
548 unsigned long mapping_try_invalidate(struct address_space *mapping,
549 		pgoff_t start, pgoff_t end, unsigned long *nr_failed);
550 
551 /**
552  * folio_evictable - Test whether a folio is evictable.
553  * @folio: The folio to test.
554  *
555  * Test whether @folio is evictable -- i.e., should be placed on
556  * active/inactive lists vs unevictable list.
557  *
558  * Reasons folio might not be evictable:
559  * 1. folio's mapping marked unevictable
560  * 2. One of the pages in the folio is part of an mlocked VMA
561  */
562 static inline bool folio_evictable(struct folio *folio)
563 {
564 	bool ret;
565 
566 	/* Prevent address_space of inode and swap cache from being freed */
567 	rcu_read_lock();
568 	ret = !mapping_unevictable(folio_mapping(folio)) &&
569 			!folio_test_mlocked(folio);
570 	rcu_read_unlock();
571 	return ret;
572 }
573 
574 /*
575  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
576  * a count of one.
577  */
578 static inline void set_page_refcounted(struct page *page)
579 {
580 	VM_BUG_ON_PAGE(PageTail(page), page);
581 	VM_BUG_ON_PAGE(page_ref_count(page), page);
582 	set_page_count(page, 1);
583 }
584 
585 static inline void set_pages_refcounted(struct page *page, unsigned long nr_pages)
586 {
587 	unsigned long pfn = page_to_pfn(page);
588 
589 	for (; nr_pages--; pfn++)
590 		set_page_refcounted(pfn_to_page(pfn));
591 }
592 
593 /*
594  * Return true if a folio needs ->release_folio() calling upon it.
595  */
596 static inline bool folio_needs_release(struct folio *folio)
597 {
598 	struct address_space *mapping = folio_mapping(folio);
599 
600 	return folio_has_private(folio) ||
601 		(mapping && mapping_release_always(mapping));
602 }
603 
604 extern unsigned long highest_memmap_pfn;
605 
606 /*
607  * Maximum number of reclaim retries without progress before the OOM
608  * killer is consider the only way forward.
609  */
610 #define MAX_RECLAIM_RETRIES 16
611 
612 /*
613  * in mm/vmscan.c:
614  */
615 bool folio_isolate_lru(struct folio *folio);
616 void folio_putback_lru(struct folio *folio);
617 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
618 int user_proactive_reclaim(char *buf,
619 			   struct mem_cgroup *memcg, pg_data_t *pgdat);
620 
621 /*
622  * in mm/rmap.c:
623  */
624 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
625 
626 /*
627  * in mm/page_alloc.c
628  */
629 #define K(x) ((x) << (PAGE_SHIFT-10))
630 
631 extern char * const zone_names[MAX_NR_ZONES];
632 
633 /* perform sanity checks on struct pages being allocated or freed */
634 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
635 
636 extern int min_free_kbytes;
637 extern int defrag_mode;
638 
639 void setup_per_zone_wmarks(void);
640 void calculate_min_free_kbytes(void);
641 int __meminit init_per_zone_wmark_min(void);
642 void page_alloc_sysctl_init(void);
643 
644 /*
645  * Structure for holding the mostly immutable allocation parameters passed
646  * between functions involved in allocations, including the alloc_pages*
647  * family of functions.
648  *
649  * nodemask, migratetype and highest_zoneidx are initialized only once in
650  * __alloc_pages() and then never change.
651  *
652  * zonelist, preferred_zone and highest_zoneidx are set first in
653  * __alloc_pages() for the fast path, and might be later changed
654  * in __alloc_pages_slowpath(). All other functions pass the whole structure
655  * by a const pointer.
656  */
657 struct alloc_context {
658 	struct zonelist *zonelist;
659 	nodemask_t *nodemask;
660 	struct zoneref *preferred_zoneref;
661 	int migratetype;
662 
663 	/*
664 	 * highest_zoneidx represents highest usable zone index of
665 	 * the allocation request. Due to the nature of the zone,
666 	 * memory on lower zone than the highest_zoneidx will be
667 	 * protected by lowmem_reserve[highest_zoneidx].
668 	 *
669 	 * highest_zoneidx is also used by reclaim/compaction to limit
670 	 * the target zone since higher zone than this index cannot be
671 	 * usable for this allocation request.
672 	 */
673 	enum zone_type highest_zoneidx;
674 	bool spread_dirty_pages;
675 };
676 
677 /*
678  * This function returns the order of a free page in the buddy system. In
679  * general, page_zone(page)->lock must be held by the caller to prevent the
680  * page from being allocated in parallel and returning garbage as the order.
681  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
682  * page cannot be allocated or merged in parallel. Alternatively, it must
683  * handle invalid values gracefully, and use buddy_order_unsafe() below.
684  */
685 static inline unsigned int buddy_order(struct page *page)
686 {
687 	/* PageBuddy() must be checked by the caller */
688 	return page_private(page);
689 }
690 
691 /*
692  * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
693  * PageBuddy() should be checked first by the caller to minimize race window,
694  * and invalid values must be handled gracefully.
695  *
696  * READ_ONCE is used so that if the caller assigns the result into a local
697  * variable and e.g. tests it for valid range before using, the compiler cannot
698  * decide to remove the variable and inline the page_private(page) multiple
699  * times, potentially observing different values in the tests and the actual
700  * use of the result.
701  */
702 #define buddy_order_unsafe(page)	READ_ONCE(page_private(page))
703 
704 /*
705  * This function checks whether a page is free && is the buddy
706  * we can coalesce a page and its buddy if
707  * (a) the buddy is not in a hole (check before calling!) &&
708  * (b) the buddy is in the buddy system &&
709  * (c) a page and its buddy have the same order &&
710  * (d) a page and its buddy are in the same zone.
711  *
712  * For recording whether a page is in the buddy system, we set PageBuddy.
713  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
714  *
715  * For recording page's order, we use page_private(page).
716  */
717 static inline bool page_is_buddy(struct page *page, struct page *buddy,
718 				 unsigned int order)
719 {
720 	if (!page_is_guard(buddy) && !PageBuddy(buddy))
721 		return false;
722 
723 	if (buddy_order(buddy) != order)
724 		return false;
725 
726 	/*
727 	 * zone check is done late to avoid uselessly calculating
728 	 * zone/node ids for pages that could never merge.
729 	 */
730 	if (page_zone_id(page) != page_zone_id(buddy))
731 		return false;
732 
733 	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
734 
735 	return true;
736 }
737 
738 /*
739  * Locate the struct page for both the matching buddy in our
740  * pair (buddy1) and the combined O(n+1) page they form (page).
741  *
742  * 1) Any buddy B1 will have an order O twin B2 which satisfies
743  * the following equation:
744  *     B2 = B1 ^ (1 << O)
745  * For example, if the starting buddy (buddy2) is #8 its order
746  * 1 buddy is #10:
747  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
748  *
749  * 2) Any buddy B will have an order O+1 parent P which
750  * satisfies the following equation:
751  *     P = B & ~(1 << O)
752  *
753  * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER
754  */
755 static inline unsigned long
756 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
757 {
758 	return page_pfn ^ (1 << order);
759 }
760 
761 /*
762  * Find the buddy of @page and validate it.
763  * @page: The input page
764  * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
765  *       function is used in the performance-critical __free_one_page().
766  * @order: The order of the page
767  * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
768  *             page_to_pfn().
769  *
770  * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
771  * not the same as @page. The validation is necessary before use it.
772  *
773  * Return: the found buddy page or NULL if not found.
774  */
775 static inline struct page *find_buddy_page_pfn(struct page *page,
776 			unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
777 {
778 	unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
779 	struct page *buddy;
780 
781 	buddy = page + (__buddy_pfn - pfn);
782 	if (buddy_pfn)
783 		*buddy_pfn = __buddy_pfn;
784 
785 	if (page_is_buddy(page, buddy, order))
786 		return buddy;
787 	return NULL;
788 }
789 
790 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
791 				unsigned long end_pfn, struct zone *zone);
792 
793 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
794 				unsigned long end_pfn, struct zone *zone)
795 {
796 	if (zone->contiguous)
797 		return pfn_to_page(start_pfn);
798 
799 	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
800 }
801 
802 void set_zone_contiguous(struct zone *zone);
803 bool pfn_range_intersects_zones(int nid, unsigned long start_pfn,
804 			   unsigned long nr_pages);
805 
806 static inline void clear_zone_contiguous(struct zone *zone)
807 {
808 	zone->contiguous = false;
809 }
810 
811 extern int __isolate_free_page(struct page *page, unsigned int order);
812 extern void __putback_isolated_page(struct page *page, unsigned int order,
813 				    int mt);
814 extern void memblock_free_pages(struct page *page, unsigned long pfn,
815 					unsigned int order);
816 extern void __free_pages_core(struct page *page, unsigned int order,
817 		enum meminit_context context);
818 
819 /*
820  * This will have no effect, other than possibly generating a warning, if the
821  * caller passes in a non-large folio.
822  */
823 static inline void folio_set_order(struct folio *folio, unsigned int order)
824 {
825 	if (WARN_ON_ONCE(!order || !folio_test_large(folio)))
826 		return;
827 	VM_WARN_ON_ONCE(order > MAX_FOLIO_ORDER);
828 
829 	folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order;
830 #ifdef NR_PAGES_IN_LARGE_FOLIO
831 	folio->_nr_pages = 1U << order;
832 #endif
833 }
834 
835 bool __folio_unqueue_deferred_split(struct folio *folio);
836 static inline bool folio_unqueue_deferred_split(struct folio *folio)
837 {
838 	if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio))
839 		return false;
840 
841 	/*
842 	 * At this point, there is no one trying to add the folio to
843 	 * deferred_list. If folio is not in deferred_list, it's safe
844 	 * to check without acquiring the split_queue_lock.
845 	 */
846 	if (data_race(list_empty(&folio->_deferred_list)))
847 		return false;
848 
849 	return __folio_unqueue_deferred_split(folio);
850 }
851 
852 static inline struct folio *page_rmappable_folio(struct page *page)
853 {
854 	struct folio *folio = (struct folio *)page;
855 
856 	if (folio && folio_test_large(folio))
857 		folio_set_large_rmappable(folio);
858 	return folio;
859 }
860 
861 static inline void prep_compound_head(struct page *page, unsigned int order)
862 {
863 	struct folio *folio = (struct folio *)page;
864 
865 	folio_set_order(folio, order);
866 	atomic_set(&folio->_large_mapcount, -1);
867 	if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
868 		atomic_set(&folio->_nr_pages_mapped, 0);
869 	if (IS_ENABLED(CONFIG_MM_ID)) {
870 		folio->_mm_ids = 0;
871 		folio->_mm_id_mapcount[0] = -1;
872 		folio->_mm_id_mapcount[1] = -1;
873 	}
874 	if (IS_ENABLED(CONFIG_64BIT) || order > 1) {
875 		atomic_set(&folio->_pincount, 0);
876 		atomic_set(&folio->_entire_mapcount, -1);
877 	}
878 	if (order > 1)
879 		INIT_LIST_HEAD(&folio->_deferred_list);
880 }
881 
882 static inline void prep_compound_tail(struct page *head, int tail_idx)
883 {
884 	struct page *p = head + tail_idx;
885 
886 	p->mapping = TAIL_MAPPING;
887 	set_compound_head(p, head);
888 	set_page_private(p, 0);
889 }
890 
891 void post_alloc_hook(struct page *page, unsigned int order, gfp_t gfp_flags);
892 extern bool free_pages_prepare(struct page *page, unsigned int order);
893 
894 extern int user_min_free_kbytes;
895 
896 struct page *__alloc_frozen_pages_noprof(gfp_t, unsigned int order, int nid,
897 		nodemask_t *);
898 #define __alloc_frozen_pages(...) \
899 	alloc_hooks(__alloc_frozen_pages_noprof(__VA_ARGS__))
900 void free_frozen_pages(struct page *page, unsigned int order);
901 void free_unref_folios(struct folio_batch *fbatch);
902 
903 #ifdef CONFIG_NUMA
904 struct page *alloc_frozen_pages_noprof(gfp_t, unsigned int order);
905 #else
906 static inline struct page *alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order)
907 {
908 	return __alloc_frozen_pages_noprof(gfp, order, numa_node_id(), NULL);
909 }
910 #endif
911 
912 #define alloc_frozen_pages(...) \
913 	alloc_hooks(alloc_frozen_pages_noprof(__VA_ARGS__))
914 
915 struct page *alloc_frozen_pages_nolock_noprof(gfp_t gfp_flags, int nid, unsigned int order);
916 #define alloc_frozen_pages_nolock(...) \
917 	alloc_hooks(alloc_frozen_pages_nolock_noprof(__VA_ARGS__))
918 
919 extern void zone_pcp_reset(struct zone *zone);
920 extern void zone_pcp_disable(struct zone *zone);
921 extern void zone_pcp_enable(struct zone *zone);
922 extern void zone_pcp_init(struct zone *zone);
923 
924 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
925 			  phys_addr_t min_addr,
926 			  int nid, bool exact_nid);
927 
928 void memmap_init_range(unsigned long, int, unsigned long, unsigned long,
929 		unsigned long, enum meminit_context, struct vmem_altmap *, int,
930 		bool);
931 
932 #ifdef CONFIG_SPARSEMEM
933 void sparse_init(void);
934 #else
935 static inline void sparse_init(void) {}
936 #endif /* CONFIG_SPARSEMEM */
937 
938 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
939 
940 /*
941  * in mm/compaction.c
942  */
943 /*
944  * compact_control is used to track pages being migrated and the free pages
945  * they are being migrated to during memory compaction. The free_pfn starts
946  * at the end of a zone and migrate_pfn begins at the start. Movable pages
947  * are moved to the end of a zone during a compaction run and the run
948  * completes when free_pfn <= migrate_pfn
949  */
950 struct compact_control {
951 	struct list_head freepages[NR_PAGE_ORDERS];	/* List of free pages to migrate to */
952 	struct list_head migratepages;	/* List of pages being migrated */
953 	unsigned int nr_freepages;	/* Number of isolated free pages */
954 	unsigned int nr_migratepages;	/* Number of pages to migrate */
955 	unsigned long free_pfn;		/* isolate_freepages search base */
956 	/*
957 	 * Acts as an in/out parameter to page isolation for migration.
958 	 * isolate_migratepages uses it as a search base.
959 	 * isolate_migratepages_block will update the value to the next pfn
960 	 * after the last isolated one.
961 	 */
962 	unsigned long migrate_pfn;
963 	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
964 	struct zone *zone;
965 	unsigned long total_migrate_scanned;
966 	unsigned long total_free_scanned;
967 	unsigned short fast_search_fail;/* failures to use free list searches */
968 	short search_order;		/* order to start a fast search at */
969 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
970 	int order;			/* order a direct compactor needs */
971 	int migratetype;		/* migratetype of direct compactor */
972 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
973 	const int highest_zoneidx;	/* zone index of a direct compactor */
974 	enum migrate_mode mode;		/* Async or sync migration mode */
975 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
976 	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
977 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
978 	bool direct_compaction;		/* False from kcompactd or /proc/... */
979 	bool proactive_compaction;	/* kcompactd proactive compaction */
980 	bool whole_zone;		/* Whole zone should/has been scanned */
981 	bool contended;			/* Signal lock contention */
982 	bool finish_pageblock;		/* Scan the remainder of a pageblock. Used
983 					 * when there are potentially transient
984 					 * isolation or migration failures to
985 					 * ensure forward progress.
986 					 */
987 	bool alloc_contig;		/* alloc_contig_range allocation */
988 };
989 
990 /*
991  * Used in direct compaction when a page should be taken from the freelists
992  * immediately when one is created during the free path.
993  */
994 struct capture_control {
995 	struct compact_control *cc;
996 	struct page *page;
997 };
998 
999 unsigned long
1000 isolate_freepages_range(struct compact_control *cc,
1001 			unsigned long start_pfn, unsigned long end_pfn);
1002 int
1003 isolate_migratepages_range(struct compact_control *cc,
1004 			   unsigned long low_pfn, unsigned long end_pfn);
1005 
1006 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1007 void init_cma_reserved_pageblock(struct page *page);
1008 
1009 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
1010 
1011 struct cma;
1012 
1013 #ifdef CONFIG_CMA
1014 bool cma_validate_zones(struct cma *cma);
1015 void *cma_reserve_early(struct cma *cma, unsigned long size);
1016 void init_cma_pageblock(struct page *page);
1017 #else
1018 static inline bool cma_validate_zones(struct cma *cma)
1019 {
1020 	return false;
1021 }
1022 static inline void *cma_reserve_early(struct cma *cma, unsigned long size)
1023 {
1024 	return NULL;
1025 }
1026 static inline void init_cma_pageblock(struct page *page)
1027 {
1028 }
1029 #endif
1030 
1031 
1032 int find_suitable_fallback(struct free_area *area, unsigned int order,
1033 			   int migratetype, bool claimable);
1034 
1035 static inline bool free_area_empty(struct free_area *area, int migratetype)
1036 {
1037 	return list_empty(&area->free_list[migratetype]);
1038 }
1039 
1040 /* mm/util.c */
1041 struct anon_vma *folio_anon_vma(const struct folio *folio);
1042 
1043 #ifdef CONFIG_MMU
1044 void unmap_mapping_folio(struct folio *folio);
1045 extern long populate_vma_page_range(struct vm_area_struct *vma,
1046 		unsigned long start, unsigned long end, int *locked);
1047 extern long faultin_page_range(struct mm_struct *mm, unsigned long start,
1048 		unsigned long end, bool write, int *locked);
1049 bool mlock_future_ok(const struct mm_struct *mm, bool is_vma_locked,
1050 		unsigned long bytes);
1051 
1052 /*
1053  * NOTE: This function can't tell whether the folio is "fully mapped" in the
1054  * range.
1055  * "fully mapped" means all the pages of folio is associated with the page
1056  * table of range while this function just check whether the folio range is
1057  * within the range [start, end). Function caller needs to do page table
1058  * check if it cares about the page table association.
1059  *
1060  * Typical usage (like mlock or madvise) is:
1061  * Caller knows at least 1 page of folio is associated with page table of VMA
1062  * and the range [start, end) is intersect with the VMA range. Caller wants
1063  * to know whether the folio is fully associated with the range. It calls
1064  * this function to check whether the folio is in the range first. Then checks
1065  * the page table to know whether the folio is fully mapped to the range.
1066  */
1067 static inline bool
1068 folio_within_range(struct folio *folio, struct vm_area_struct *vma,
1069 		unsigned long start, unsigned long end)
1070 {
1071 	pgoff_t pgoff, addr;
1072 	unsigned long vma_pglen = vma_pages(vma);
1073 
1074 	VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio);
1075 	if (start > end)
1076 		return false;
1077 
1078 	if (start < vma->vm_start)
1079 		start = vma->vm_start;
1080 
1081 	if (end > vma->vm_end)
1082 		end = vma->vm_end;
1083 
1084 	pgoff = folio_pgoff(folio);
1085 
1086 	/* if folio start address is not in vma range */
1087 	if (!in_range(pgoff, vma->vm_pgoff, vma_pglen))
1088 		return false;
1089 
1090 	addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1091 
1092 	return !(addr < start || end - addr < folio_size(folio));
1093 }
1094 
1095 static inline bool
1096 folio_within_vma(struct folio *folio, struct vm_area_struct *vma)
1097 {
1098 	return folio_within_range(folio, vma, vma->vm_start, vma->vm_end);
1099 }
1100 
1101 /*
1102  * mlock_vma_folio() and munlock_vma_folio():
1103  * should be called with vma's mmap_lock held for read or write,
1104  * under page table lock for the pte/pmd being added or removed.
1105  *
1106  * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at
1107  * the end of folio_remove_rmap_*(); but new anon folios are managed by
1108  * folio_add_lru_vma() calling mlock_new_folio().
1109  */
1110 void mlock_folio(struct folio *folio);
1111 static inline void mlock_vma_folio(struct folio *folio,
1112 				struct vm_area_struct *vma)
1113 {
1114 	/*
1115 	 * The VM_SPECIAL check here serves two purposes.
1116 	 * 1) VM_IO check prevents migration from double-counting during mlock.
1117 	 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
1118 	 *    is never left set on a VM_SPECIAL vma, there is an interval while
1119 	 *    file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
1120 	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
1121 	 */
1122 	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED))
1123 		mlock_folio(folio);
1124 }
1125 
1126 void munlock_folio(struct folio *folio);
1127 static inline void munlock_vma_folio(struct folio *folio,
1128 					struct vm_area_struct *vma)
1129 {
1130 	/*
1131 	 * munlock if the function is called. Ideally, we should only
1132 	 * do munlock if any page of folio is unmapped from VMA and
1133 	 * cause folio not fully mapped to VMA.
1134 	 *
1135 	 * But it's not easy to confirm that's the situation. So we
1136 	 * always munlock the folio and page reclaim will correct it
1137 	 * if it's wrong.
1138 	 */
1139 	if (unlikely(vma->vm_flags & VM_LOCKED))
1140 		munlock_folio(folio);
1141 }
1142 
1143 void mlock_new_folio(struct folio *folio);
1144 bool need_mlock_drain(int cpu);
1145 void mlock_drain_local(void);
1146 void mlock_drain_remote(int cpu);
1147 
1148 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
1149 
1150 /**
1151  * vma_address - Find the virtual address a page range is mapped at
1152  * @vma: The vma which maps this object.
1153  * @pgoff: The page offset within its object.
1154  * @nr_pages: The number of pages to consider.
1155  *
1156  * If any page in this range is mapped by this VMA, return the first address
1157  * where any of these pages appear.  Otherwise, return -EFAULT.
1158  */
1159 static inline unsigned long vma_address(const struct vm_area_struct *vma,
1160 		pgoff_t pgoff, unsigned long nr_pages)
1161 {
1162 	unsigned long address;
1163 
1164 	if (pgoff >= vma->vm_pgoff) {
1165 		address = vma->vm_start +
1166 			((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1167 		/* Check for address beyond vma (or wrapped through 0?) */
1168 		if (address < vma->vm_start || address >= vma->vm_end)
1169 			address = -EFAULT;
1170 	} else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
1171 		/* Test above avoids possibility of wrap to 0 on 32-bit */
1172 		address = vma->vm_start;
1173 	} else {
1174 		address = -EFAULT;
1175 	}
1176 	return address;
1177 }
1178 
1179 /*
1180  * Then at what user virtual address will none of the range be found in vma?
1181  * Assumes that vma_address() already returned a good starting address.
1182  */
1183 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
1184 {
1185 	struct vm_area_struct *vma = pvmw->vma;
1186 	pgoff_t pgoff;
1187 	unsigned long address;
1188 
1189 	/* Common case, plus ->pgoff is invalid for KSM */
1190 	if (pvmw->nr_pages == 1)
1191 		return pvmw->address + PAGE_SIZE;
1192 
1193 	pgoff = pvmw->pgoff + pvmw->nr_pages;
1194 	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1195 	/* Check for address beyond vma (or wrapped through 0?) */
1196 	if (address < vma->vm_start || address > vma->vm_end)
1197 		address = vma->vm_end;
1198 	return address;
1199 }
1200 
1201 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
1202 						    struct file *fpin)
1203 {
1204 	int flags = vmf->flags;
1205 
1206 	if (fpin)
1207 		return fpin;
1208 
1209 	/*
1210 	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
1211 	 * anything, so we only pin the file and drop the mmap_lock if only
1212 	 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
1213 	 */
1214 	if (fault_flag_allow_retry_first(flags) &&
1215 	    !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
1216 		fpin = get_file(vmf->vma->vm_file);
1217 		release_fault_lock(vmf);
1218 	}
1219 	return fpin;
1220 }
1221 #else /* !CONFIG_MMU */
1222 static inline void unmap_mapping_folio(struct folio *folio) { }
1223 static inline void mlock_new_folio(struct folio *folio) { }
1224 static inline bool need_mlock_drain(int cpu) { return false; }
1225 static inline void mlock_drain_local(void) { }
1226 static inline void mlock_drain_remote(int cpu) { }
1227 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
1228 {
1229 }
1230 #endif /* !CONFIG_MMU */
1231 
1232 /* Memory initialisation debug and verification */
1233 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1234 DECLARE_STATIC_KEY_TRUE(deferred_pages);
1235 
1236 bool __init deferred_grow_zone(struct zone *zone, unsigned int order);
1237 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1238 
1239 void init_deferred_page(unsigned long pfn, int nid);
1240 
1241 enum mminit_level {
1242 	MMINIT_WARNING,
1243 	MMINIT_VERIFY,
1244 	MMINIT_TRACE
1245 };
1246 
1247 #ifdef CONFIG_DEBUG_MEMORY_INIT
1248 
1249 extern int mminit_loglevel;
1250 
1251 #define mminit_dprintk(level, prefix, fmt, arg...) \
1252 do { \
1253 	if (level < mminit_loglevel) { \
1254 		if (level <= MMINIT_WARNING) \
1255 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
1256 		else \
1257 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
1258 	} \
1259 } while (0)
1260 
1261 extern void mminit_verify_pageflags_layout(void);
1262 extern void mminit_verify_zonelist(void);
1263 #else
1264 
1265 static inline void mminit_dprintk(enum mminit_level level,
1266 				const char *prefix, const char *fmt, ...)
1267 {
1268 }
1269 
1270 static inline void mminit_verify_pageflags_layout(void)
1271 {
1272 }
1273 
1274 static inline void mminit_verify_zonelist(void)
1275 {
1276 }
1277 #endif /* CONFIG_DEBUG_MEMORY_INIT */
1278 
1279 #define NODE_RECLAIM_NOSCAN	-2
1280 #define NODE_RECLAIM_FULL	-1
1281 #define NODE_RECLAIM_SOME	0
1282 #define NODE_RECLAIM_SUCCESS	1
1283 
1284 #ifdef CONFIG_NUMA
1285 extern int node_reclaim_mode;
1286 
1287 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
1288 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
1289 #else
1290 #define node_reclaim_mode 0
1291 
1292 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
1293 				unsigned int order)
1294 {
1295 	return NODE_RECLAIM_NOSCAN;
1296 }
1297 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
1298 {
1299 	return NUMA_NO_NODE;
1300 }
1301 #endif
1302 
1303 static inline bool node_reclaim_enabled(void)
1304 {
1305 	/* Is any node_reclaim_mode bit set? */
1306 	return node_reclaim_mode & (RECLAIM_ZONE|RECLAIM_WRITE|RECLAIM_UNMAP);
1307 }
1308 
1309 /*
1310  * mm/memory-failure.c
1311  */
1312 #ifdef CONFIG_MEMORY_FAILURE
1313 int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill);
1314 void shake_folio(struct folio *folio);
1315 typedef int hwpoison_filter_func_t(struct page *p);
1316 void hwpoison_filter_register(hwpoison_filter_func_t *filter);
1317 void hwpoison_filter_unregister(void);
1318 
1319 #define MAGIC_HWPOISON	0x48575053U	/* HWPS */
1320 void SetPageHWPoisonTakenOff(struct page *page);
1321 void ClearPageHWPoisonTakenOff(struct page *page);
1322 bool take_page_off_buddy(struct page *page);
1323 bool put_page_back_buddy(struct page *page);
1324 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
1325 void add_to_kill_ksm(struct task_struct *tsk, const struct page *p,
1326 		     struct vm_area_struct *vma, struct list_head *to_kill,
1327 		     unsigned long ksm_addr);
1328 unsigned long page_mapped_in_vma(const struct page *page,
1329 		struct vm_area_struct *vma);
1330 
1331 #else
1332 static inline int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill)
1333 {
1334 	return -EBUSY;
1335 }
1336 #endif
1337 
1338 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
1339         unsigned long, unsigned long,
1340         unsigned long, unsigned long);
1341 
1342 extern void set_pageblock_order(void);
1343 unsigned long reclaim_pages(struct list_head *folio_list);
1344 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
1345 					    struct list_head *folio_list);
1346 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1347 #define ALLOC_WMARK_MIN		WMARK_MIN
1348 #define ALLOC_WMARK_LOW		WMARK_LOW
1349 #define ALLOC_WMARK_HIGH	WMARK_HIGH
1350 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
1351 
1352 /* Mask to get the watermark bits */
1353 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
1354 
1355 /*
1356  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
1357  * cannot assume a reduced access to memory reserves is sufficient for
1358  * !MMU
1359  */
1360 #ifdef CONFIG_MMU
1361 #define ALLOC_OOM		0x08
1362 #else
1363 #define ALLOC_OOM		ALLOC_NO_WATERMARKS
1364 #endif
1365 
1366 #define ALLOC_NON_BLOCK		 0x10 /* Caller cannot block. Allow access
1367 				       * to 25% of the min watermark or
1368 				       * 62.5% if __GFP_HIGH is set.
1369 				       */
1370 #define ALLOC_MIN_RESERVE	 0x20 /* __GFP_HIGH set. Allow access to 50%
1371 				       * of the min watermark.
1372 				       */
1373 #define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
1374 #define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
1375 #ifdef CONFIG_ZONE_DMA32
1376 #define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
1377 #else
1378 #define ALLOC_NOFRAGMENT	  0x0
1379 #endif
1380 #define ALLOC_HIGHATOMIC	0x200 /* Allows access to MIGRATE_HIGHATOMIC */
1381 #define ALLOC_TRYLOCK		0x400 /* Only use spin_trylock in allocation path */
1382 #define ALLOC_KSWAPD		0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
1383 
1384 /* Flags that allow allocations below the min watermark. */
1385 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
1386 
1387 enum ttu_flags;
1388 struct tlbflush_unmap_batch;
1389 
1390 
1391 /*
1392  * only for MM internal work items which do not depend on
1393  * any allocations or locks which might depend on allocations
1394  */
1395 extern struct workqueue_struct *mm_percpu_wq;
1396 
1397 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1398 void try_to_unmap_flush(void);
1399 void try_to_unmap_flush_dirty(void);
1400 void flush_tlb_batched_pending(struct mm_struct *mm);
1401 #else
1402 static inline void try_to_unmap_flush(void)
1403 {
1404 }
1405 static inline void try_to_unmap_flush_dirty(void)
1406 {
1407 }
1408 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
1409 {
1410 }
1411 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
1412 
1413 extern const struct trace_print_flags pageflag_names[];
1414 extern const struct trace_print_flags vmaflag_names[];
1415 extern const struct trace_print_flags gfpflag_names[];
1416 
1417 void setup_zone_pageset(struct zone *zone);
1418 
1419 struct migration_target_control {
1420 	int nid;		/* preferred node id */
1421 	nodemask_t *nmask;
1422 	gfp_t gfp_mask;
1423 	enum migrate_reason reason;
1424 };
1425 
1426 /*
1427  * mm/filemap.c
1428  */
1429 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
1430 			      struct folio *folio, loff_t fpos, size_t size);
1431 
1432 /*
1433  * mm/vmalloc.c
1434  */
1435 #ifdef CONFIG_MMU
1436 void __init vmalloc_init(void);
1437 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1438 	pgprot_t prot, struct page **pages, unsigned int page_shift, gfp_t gfp_mask);
1439 unsigned int get_vm_area_page_order(struct vm_struct *vm);
1440 #else
1441 static inline void vmalloc_init(void)
1442 {
1443 }
1444 
1445 static inline
1446 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1447 	pgprot_t prot, struct page **pages, unsigned int page_shift, gfp_t gfp_mask)
1448 {
1449 	return -EINVAL;
1450 }
1451 #endif
1452 
1453 int __must_check __vmap_pages_range_noflush(unsigned long addr,
1454 			       unsigned long end, pgprot_t prot,
1455 			       struct page **pages, unsigned int page_shift);
1456 
1457 void vunmap_range_noflush(unsigned long start, unsigned long end);
1458 
1459 void __vunmap_range_noflush(unsigned long start, unsigned long end);
1460 
1461 static inline bool vma_is_single_threaded_private(struct vm_area_struct *vma)
1462 {
1463 	if (vma->vm_flags & VM_SHARED)
1464 		return false;
1465 
1466 	return atomic_read(&vma->vm_mm->mm_users) == 1;
1467 }
1468 
1469 #ifdef CONFIG_NUMA_BALANCING
1470 bool folio_can_map_prot_numa(struct folio *folio, struct vm_area_struct *vma,
1471 		bool is_private_single_threaded);
1472 
1473 #else
1474 static inline bool folio_can_map_prot_numa(struct folio *folio,
1475 		struct vm_area_struct *vma, bool is_private_single_threaded)
1476 {
1477 	return false;
1478 }
1479 #endif
1480 
1481 int numa_migrate_check(struct folio *folio, struct vm_fault *vmf,
1482 		      unsigned long addr, int *flags, bool writable,
1483 		      int *last_cpupid);
1484 
1485 void free_zone_device_folio(struct folio *folio);
1486 int migrate_device_coherent_folio(struct folio *folio);
1487 
1488 struct vm_struct *__get_vm_area_node(unsigned long size,
1489 				     unsigned long align, unsigned long shift,
1490 				     unsigned long vm_flags, unsigned long start,
1491 				     unsigned long end, int node, gfp_t gfp_mask,
1492 				     const void *caller);
1493 
1494 /*
1495  * mm/gup.c
1496  */
1497 int __must_check try_grab_folio(struct folio *folio, int refs,
1498 				unsigned int flags);
1499 
1500 /*
1501  * mm/huge_memory.c
1502  */
1503 void touch_pud(struct vm_area_struct *vma, unsigned long addr,
1504 	       pud_t *pud, bool write);
1505 bool touch_pmd(struct vm_area_struct *vma, unsigned long addr,
1506 	       pmd_t *pmd, bool write);
1507 
1508 /*
1509  * Parses a string with mem suffixes into its order. Useful to parse kernel
1510  * parameters.
1511  */
1512 static inline int get_order_from_str(const char *size_str,
1513 				     unsigned long valid_orders)
1514 {
1515 	unsigned long size;
1516 	char *endptr;
1517 	int order;
1518 
1519 	size = memparse(size_str, &endptr);
1520 
1521 	if (!is_power_of_2(size))
1522 		return -EINVAL;
1523 	order = get_order(size);
1524 	if (BIT(order) & ~valid_orders)
1525 		return -EINVAL;
1526 
1527 	return order;
1528 }
1529 
1530 enum {
1531 	/* mark page accessed */
1532 	FOLL_TOUCH = 1 << 16,
1533 	/* a retry, previous pass started an IO */
1534 	FOLL_TRIED = 1 << 17,
1535 	/* we are working on non-current tsk/mm */
1536 	FOLL_REMOTE = 1 << 18,
1537 	/* pages must be released via unpin_user_page */
1538 	FOLL_PIN = 1 << 19,
1539 	/* gup_fast: prevent fall-back to slow gup */
1540 	FOLL_FAST_ONLY = 1 << 20,
1541 	/* allow unlocking the mmap lock */
1542 	FOLL_UNLOCKABLE = 1 << 21,
1543 	/* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */
1544 	FOLL_MADV_POPULATE = 1 << 22,
1545 };
1546 
1547 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \
1548 			    FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \
1549 			    FOLL_MADV_POPULATE)
1550 
1551 /*
1552  * Indicates for which pages that are write-protected in the page table,
1553  * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
1554  * GUP pin will remain consistent with the pages mapped into the page tables
1555  * of the MM.
1556  *
1557  * Temporary unmapping of PageAnonExclusive() pages or clearing of
1558  * PageAnonExclusive() has to protect against concurrent GUP:
1559  * * Ordinary GUP: Using the PT lock
1560  * * GUP-fast and fork(): mm->write_protect_seq
1561  * * GUP-fast and KSM or temporary unmapping (swap, migration): see
1562  *    folio_try_share_anon_rmap_*()
1563  *
1564  * Must be called with the (sub)page that's actually referenced via the
1565  * page table entry, which might not necessarily be the head page for a
1566  * PTE-mapped THP.
1567  *
1568  * If the vma is NULL, we're coming from the GUP-fast path and might have
1569  * to fallback to the slow path just to lookup the vma.
1570  */
1571 static inline bool gup_must_unshare(struct vm_area_struct *vma,
1572 				    unsigned int flags, struct page *page)
1573 {
1574 	/*
1575 	 * FOLL_WRITE is implicitly handled correctly as the page table entry
1576 	 * has to be writable -- and if it references (part of) an anonymous
1577 	 * folio, that part is required to be marked exclusive.
1578 	 */
1579 	if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
1580 		return false;
1581 	/*
1582 	 * Note: PageAnon(page) is stable until the page is actually getting
1583 	 * freed.
1584 	 */
1585 	if (!PageAnon(page)) {
1586 		/*
1587 		 * We only care about R/O long-term pining: R/O short-term
1588 		 * pinning does not have the semantics to observe successive
1589 		 * changes through the process page tables.
1590 		 */
1591 		if (!(flags & FOLL_LONGTERM))
1592 			return false;
1593 
1594 		/* We really need the vma ... */
1595 		if (!vma)
1596 			return true;
1597 
1598 		/*
1599 		 * ... because we only care about writable private ("COW")
1600 		 * mappings where we have to break COW early.
1601 		 */
1602 		return is_cow_mapping(vma->vm_flags);
1603 	}
1604 
1605 	/* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */
1606 	if (IS_ENABLED(CONFIG_HAVE_GUP_FAST))
1607 		smp_rmb();
1608 
1609 	/*
1610 	 * Note that KSM pages cannot be exclusive, and consequently,
1611 	 * cannot get pinned.
1612 	 */
1613 	return !PageAnonExclusive(page);
1614 }
1615 
1616 extern bool mirrored_kernelcore;
1617 bool memblock_has_mirror(void);
1618 void memblock_free_all(void);
1619 
1620 static __always_inline void vma_set_range(struct vm_area_struct *vma,
1621 					  unsigned long start, unsigned long end,
1622 					  pgoff_t pgoff)
1623 {
1624 	vma->vm_start = start;
1625 	vma->vm_end = end;
1626 	vma->vm_pgoff = pgoff;
1627 }
1628 
1629 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1630 {
1631 	/*
1632 	 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1633 	 * enablements, because when without soft-dirty being compiled in,
1634 	 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1635 	 * will be constantly true.
1636 	 */
1637 	if (!pgtable_supports_soft_dirty())
1638 		return false;
1639 
1640 	/*
1641 	 * Soft-dirty is kind of special: its tracking is enabled when the
1642 	 * vma flags not set.
1643 	 */
1644 	return !(vma->vm_flags & VM_SOFTDIRTY);
1645 }
1646 
1647 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd)
1648 {
1649 	return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd);
1650 }
1651 
1652 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte)
1653 {
1654 	return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte);
1655 }
1656 
1657 void __meminit __init_single_page(struct page *page, unsigned long pfn,
1658 				unsigned long zone, int nid);
1659 void __meminit __init_page_from_nid(unsigned long pfn, int nid);
1660 
1661 /* shrinker related functions */
1662 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg,
1663 			  int priority);
1664 
1665 int shmem_add_to_page_cache(struct folio *folio,
1666 			    struct address_space *mapping,
1667 			    pgoff_t index, void *expected, gfp_t gfp);
1668 int shmem_inode_acct_blocks(struct inode *inode, long pages);
1669 bool shmem_recalc_inode(struct inode *inode, long alloced, long swapped);
1670 
1671 #ifdef CONFIG_SHRINKER_DEBUG
1672 static inline __printf(2, 0) int shrinker_debugfs_name_alloc(
1673 			struct shrinker *shrinker, const char *fmt, va_list ap)
1674 {
1675 	shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap);
1676 
1677 	return shrinker->name ? 0 : -ENOMEM;
1678 }
1679 
1680 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1681 {
1682 	kfree_const(shrinker->name);
1683 	shrinker->name = NULL;
1684 }
1685 
1686 extern int shrinker_debugfs_add(struct shrinker *shrinker);
1687 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1688 					      int *debugfs_id);
1689 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1690 				    int debugfs_id);
1691 #else /* CONFIG_SHRINKER_DEBUG */
1692 static inline int shrinker_debugfs_add(struct shrinker *shrinker)
1693 {
1694 	return 0;
1695 }
1696 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker,
1697 					      const char *fmt, va_list ap)
1698 {
1699 	return 0;
1700 }
1701 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1702 {
1703 }
1704 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1705 						     int *debugfs_id)
1706 {
1707 	*debugfs_id = -1;
1708 	return NULL;
1709 }
1710 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1711 					   int debugfs_id)
1712 {
1713 }
1714 #endif /* CONFIG_SHRINKER_DEBUG */
1715 
1716 /* Only track the nodes of mappings with shadow entries */
1717 void workingset_update_node(struct xa_node *node);
1718 extern struct list_lru shadow_nodes;
1719 #define mapping_set_update(xas, mapping) do {			\
1720 	if (!dax_mapping(mapping) && !shmem_mapping(mapping)) {	\
1721 		xas_set_update(xas, workingset_update_node);	\
1722 		xas_set_lru(xas, &shadow_nodes);		\
1723 	}							\
1724 } while (0)
1725 
1726 /* mremap.c */
1727 unsigned long move_page_tables(struct pagetable_move_control *pmc);
1728 
1729 #ifdef CONFIG_UNACCEPTED_MEMORY
1730 void accept_page(struct page *page);
1731 #else /* CONFIG_UNACCEPTED_MEMORY */
1732 static inline void accept_page(struct page *page)
1733 {
1734 }
1735 #endif /* CONFIG_UNACCEPTED_MEMORY */
1736 
1737 /* pagewalk.c */
1738 int walk_page_range_mm_unsafe(struct mm_struct *mm, unsigned long start,
1739 		unsigned long end, const struct mm_walk_ops *ops,
1740 		void *private);
1741 int walk_page_range_vma_unsafe(struct vm_area_struct *vma, unsigned long start,
1742 		unsigned long end, const struct mm_walk_ops *ops,
1743 		void *private);
1744 int walk_page_range_debug(struct mm_struct *mm, unsigned long start,
1745 			  unsigned long end, const struct mm_walk_ops *ops,
1746 			  pgd_t *pgd, void *private);
1747 
1748 void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm);
1749 int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm);
1750 
1751 void remap_pfn_range_prepare(struct vm_area_desc *desc, unsigned long pfn);
1752 int remap_pfn_range_complete(struct vm_area_struct *vma, unsigned long addr,
1753 		unsigned long pfn, unsigned long size, pgprot_t pgprot);
1754 
1755 static inline void io_remap_pfn_range_prepare(struct vm_area_desc *desc,
1756 		unsigned long orig_pfn, unsigned long size)
1757 {
1758 	const unsigned long pfn = io_remap_pfn_range_pfn(orig_pfn, size);
1759 
1760 	return remap_pfn_range_prepare(desc, pfn);
1761 }
1762 
1763 static inline int io_remap_pfn_range_complete(struct vm_area_struct *vma,
1764 		unsigned long addr, unsigned long orig_pfn, unsigned long size,
1765 		pgprot_t orig_prot)
1766 {
1767 	const unsigned long pfn = io_remap_pfn_range_pfn(orig_pfn, size);
1768 	const pgprot_t prot = pgprot_decrypted(orig_prot);
1769 
1770 	return remap_pfn_range_complete(vma, addr, pfn, size, prot);
1771 }
1772 
1773 #endif	/* __MM_INTERNAL_H */
1774