xref: /openbmc/linux/mm/internal.h (revision 2bad466c)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* internal.h: mm/ internal definitions
3  *
4  * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 #ifndef __MM_INTERNAL_H
8 #define __MM_INTERNAL_H
9 
10 #include <linux/fs.h>
11 #include <linux/mm.h>
12 #include <linux/pagemap.h>
13 #include <linux/rmap.h>
14 #include <linux/tracepoint-defs.h>
15 
16 struct folio_batch;
17 
18 /*
19  * The set of flags that only affect watermark checking and reclaim
20  * behaviour. This is used by the MM to obey the caller constraints
21  * about IO, FS and watermark checking while ignoring placement
22  * hints such as HIGHMEM usage.
23  */
24 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
25 			__GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
26 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
27 			__GFP_NOLOCKDEP)
28 
29 /* The GFP flags allowed during early boot */
30 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
31 
32 /* Control allocation cpuset and node placement constraints */
33 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
34 
35 /* Do not use these with a slab allocator */
36 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
37 
38 /*
39  * Different from WARN_ON_ONCE(), no warning will be issued
40  * when we specify __GFP_NOWARN.
41  */
42 #define WARN_ON_ONCE_GFP(cond, gfp)	({				\
43 	static bool __section(".data.once") __warned;			\
44 	int __ret_warn_once = !!(cond);					\
45 									\
46 	if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
47 		__warned = true;					\
48 		WARN_ON(1);						\
49 	}								\
50 	unlikely(__ret_warn_once);					\
51 })
52 
53 void page_writeback_init(void);
54 
55 /*
56  * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
57  * its nr_pages_mapped would be 0x400000: choose the COMPOUND_MAPPED bit
58  * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE).  Hugetlb currently
59  * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
60  */
61 #define COMPOUND_MAPPED		0x800000
62 #define FOLIO_PAGES_MAPPED	(COMPOUND_MAPPED - 1)
63 
64 /*
65  * How many individual pages have an elevated _mapcount.  Excludes
66  * the folio's entire_mapcount.
67  */
68 static inline int folio_nr_pages_mapped(struct folio *folio)
69 {
70 	return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
71 }
72 
73 static inline void *folio_raw_mapping(struct folio *folio)
74 {
75 	unsigned long mapping = (unsigned long)folio->mapping;
76 
77 	return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
78 }
79 
80 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
81 						int nr_throttled);
82 static inline void acct_reclaim_writeback(struct folio *folio)
83 {
84 	pg_data_t *pgdat = folio_pgdat(folio);
85 	int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
86 
87 	if (nr_throttled)
88 		__acct_reclaim_writeback(pgdat, folio, nr_throttled);
89 }
90 
91 static inline void wake_throttle_isolated(pg_data_t *pgdat)
92 {
93 	wait_queue_head_t *wqh;
94 
95 	wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
96 	if (waitqueue_active(wqh))
97 		wake_up(wqh);
98 }
99 
100 vm_fault_t do_swap_page(struct vm_fault *vmf);
101 void folio_rotate_reclaimable(struct folio *folio);
102 bool __folio_end_writeback(struct folio *folio);
103 void deactivate_file_folio(struct folio *folio);
104 void folio_activate(struct folio *folio);
105 
106 void free_pgtables(struct mmu_gather *tlb, struct maple_tree *mt,
107 		   struct vm_area_struct *start_vma, unsigned long floor,
108 		   unsigned long ceiling);
109 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
110 
111 struct zap_details;
112 void unmap_page_range(struct mmu_gather *tlb,
113 			     struct vm_area_struct *vma,
114 			     unsigned long addr, unsigned long end,
115 			     struct zap_details *details);
116 
117 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *,
118 		unsigned int order);
119 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
120 static inline void force_page_cache_readahead(struct address_space *mapping,
121 		struct file *file, pgoff_t index, unsigned long nr_to_read)
122 {
123 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
124 	force_page_cache_ra(&ractl, nr_to_read);
125 }
126 
127 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
128 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
129 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
130 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
131 void filemap_free_folio(struct address_space *mapping, struct folio *folio);
132 int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
133 bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
134 		loff_t end);
135 long invalidate_inode_page(struct page *page);
136 unsigned long invalidate_mapping_pagevec(struct address_space *mapping,
137 		pgoff_t start, pgoff_t end, unsigned long *nr_pagevec);
138 
139 /**
140  * folio_evictable - Test whether a folio is evictable.
141  * @folio: The folio to test.
142  *
143  * Test whether @folio is evictable -- i.e., should be placed on
144  * active/inactive lists vs unevictable list.
145  *
146  * Reasons folio might not be evictable:
147  * 1. folio's mapping marked unevictable
148  * 2. One of the pages in the folio is part of an mlocked VMA
149  */
150 static inline bool folio_evictable(struct folio *folio)
151 {
152 	bool ret;
153 
154 	/* Prevent address_space of inode and swap cache from being freed */
155 	rcu_read_lock();
156 	ret = !mapping_unevictable(folio_mapping(folio)) &&
157 			!folio_test_mlocked(folio);
158 	rcu_read_unlock();
159 	return ret;
160 }
161 
162 /*
163  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
164  * a count of one.
165  */
166 static inline void set_page_refcounted(struct page *page)
167 {
168 	VM_BUG_ON_PAGE(PageTail(page), page);
169 	VM_BUG_ON_PAGE(page_ref_count(page), page);
170 	set_page_count(page, 1);
171 }
172 
173 extern unsigned long highest_memmap_pfn;
174 
175 /*
176  * Maximum number of reclaim retries without progress before the OOM
177  * killer is consider the only way forward.
178  */
179 #define MAX_RECLAIM_RETRIES 16
180 
181 /*
182  * in mm/early_ioremap.c
183  */
184 pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr,
185 					unsigned long size, pgprot_t prot);
186 
187 /*
188  * in mm/vmscan.c:
189  */
190 bool isolate_lru_page(struct page *page);
191 bool folio_isolate_lru(struct folio *folio);
192 void putback_lru_page(struct page *page);
193 void folio_putback_lru(struct folio *folio);
194 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
195 
196 /*
197  * in mm/rmap.c:
198  */
199 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
200 
201 /*
202  * in mm/page_alloc.c
203  */
204 
205 /*
206  * Structure for holding the mostly immutable allocation parameters passed
207  * between functions involved in allocations, including the alloc_pages*
208  * family of functions.
209  *
210  * nodemask, migratetype and highest_zoneidx are initialized only once in
211  * __alloc_pages() and then never change.
212  *
213  * zonelist, preferred_zone and highest_zoneidx are set first in
214  * __alloc_pages() for the fast path, and might be later changed
215  * in __alloc_pages_slowpath(). All other functions pass the whole structure
216  * by a const pointer.
217  */
218 struct alloc_context {
219 	struct zonelist *zonelist;
220 	nodemask_t *nodemask;
221 	struct zoneref *preferred_zoneref;
222 	int migratetype;
223 
224 	/*
225 	 * highest_zoneidx represents highest usable zone index of
226 	 * the allocation request. Due to the nature of the zone,
227 	 * memory on lower zone than the highest_zoneidx will be
228 	 * protected by lowmem_reserve[highest_zoneidx].
229 	 *
230 	 * highest_zoneidx is also used by reclaim/compaction to limit
231 	 * the target zone since higher zone than this index cannot be
232 	 * usable for this allocation request.
233 	 */
234 	enum zone_type highest_zoneidx;
235 	bool spread_dirty_pages;
236 };
237 
238 /*
239  * This function returns the order of a free page in the buddy system. In
240  * general, page_zone(page)->lock must be held by the caller to prevent the
241  * page from being allocated in parallel and returning garbage as the order.
242  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
243  * page cannot be allocated or merged in parallel. Alternatively, it must
244  * handle invalid values gracefully, and use buddy_order_unsafe() below.
245  */
246 static inline unsigned int buddy_order(struct page *page)
247 {
248 	/* PageBuddy() must be checked by the caller */
249 	return page_private(page);
250 }
251 
252 /*
253  * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
254  * PageBuddy() should be checked first by the caller to minimize race window,
255  * and invalid values must be handled gracefully.
256  *
257  * READ_ONCE is used so that if the caller assigns the result into a local
258  * variable and e.g. tests it for valid range before using, the compiler cannot
259  * decide to remove the variable and inline the page_private(page) multiple
260  * times, potentially observing different values in the tests and the actual
261  * use of the result.
262  */
263 #define buddy_order_unsafe(page)	READ_ONCE(page_private(page))
264 
265 /*
266  * This function checks whether a page is free && is the buddy
267  * we can coalesce a page and its buddy if
268  * (a) the buddy is not in a hole (check before calling!) &&
269  * (b) the buddy is in the buddy system &&
270  * (c) a page and its buddy have the same order &&
271  * (d) a page and its buddy are in the same zone.
272  *
273  * For recording whether a page is in the buddy system, we set PageBuddy.
274  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
275  *
276  * For recording page's order, we use page_private(page).
277  */
278 static inline bool page_is_buddy(struct page *page, struct page *buddy,
279 				 unsigned int order)
280 {
281 	if (!page_is_guard(buddy) && !PageBuddy(buddy))
282 		return false;
283 
284 	if (buddy_order(buddy) != order)
285 		return false;
286 
287 	/*
288 	 * zone check is done late to avoid uselessly calculating
289 	 * zone/node ids for pages that could never merge.
290 	 */
291 	if (page_zone_id(page) != page_zone_id(buddy))
292 		return false;
293 
294 	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
295 
296 	return true;
297 }
298 
299 /*
300  * Locate the struct page for both the matching buddy in our
301  * pair (buddy1) and the combined O(n+1) page they form (page).
302  *
303  * 1) Any buddy B1 will have an order O twin B2 which satisfies
304  * the following equation:
305  *     B2 = B1 ^ (1 << O)
306  * For example, if the starting buddy (buddy2) is #8 its order
307  * 1 buddy is #10:
308  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
309  *
310  * 2) Any buddy B will have an order O+1 parent P which
311  * satisfies the following equation:
312  *     P = B & ~(1 << O)
313  *
314  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
315  */
316 static inline unsigned long
317 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
318 {
319 	return page_pfn ^ (1 << order);
320 }
321 
322 /*
323  * Find the buddy of @page and validate it.
324  * @page: The input page
325  * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
326  *       function is used in the performance-critical __free_one_page().
327  * @order: The order of the page
328  * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
329  *             page_to_pfn().
330  *
331  * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
332  * not the same as @page. The validation is necessary before use it.
333  *
334  * Return: the found buddy page or NULL if not found.
335  */
336 static inline struct page *find_buddy_page_pfn(struct page *page,
337 			unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
338 {
339 	unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
340 	struct page *buddy;
341 
342 	buddy = page + (__buddy_pfn - pfn);
343 	if (buddy_pfn)
344 		*buddy_pfn = __buddy_pfn;
345 
346 	if (page_is_buddy(page, buddy, order))
347 		return buddy;
348 	return NULL;
349 }
350 
351 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
352 				unsigned long end_pfn, struct zone *zone);
353 
354 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
355 				unsigned long end_pfn, struct zone *zone)
356 {
357 	if (zone->contiguous)
358 		return pfn_to_page(start_pfn);
359 
360 	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
361 }
362 
363 extern int __isolate_free_page(struct page *page, unsigned int order);
364 extern void __putback_isolated_page(struct page *page, unsigned int order,
365 				    int mt);
366 extern void memblock_free_pages(struct page *page, unsigned long pfn,
367 					unsigned int order);
368 extern void __free_pages_core(struct page *page, unsigned int order);
369 extern void prep_compound_page(struct page *page, unsigned int order);
370 extern void post_alloc_hook(struct page *page, unsigned int order,
371 					gfp_t gfp_flags);
372 extern int user_min_free_kbytes;
373 
374 extern void free_unref_page(struct page *page, unsigned int order);
375 extern void free_unref_page_list(struct list_head *list);
376 
377 extern void zone_pcp_reset(struct zone *zone);
378 extern void zone_pcp_disable(struct zone *zone);
379 extern void zone_pcp_enable(struct zone *zone);
380 
381 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
382 			  phys_addr_t min_addr,
383 			  int nid, bool exact_nid);
384 
385 int split_free_page(struct page *free_page,
386 			unsigned int order, unsigned long split_pfn_offset);
387 
388 /*
389  * This will have no effect, other than possibly generating a warning, if the
390  * caller passes in a non-large folio.
391  */
392 static inline void folio_set_order(struct folio *folio, unsigned int order)
393 {
394 	if (WARN_ON_ONCE(!folio_test_large(folio)))
395 		return;
396 
397 	folio->_folio_order = order;
398 #ifdef CONFIG_64BIT
399 	/*
400 	 * When hugetlb dissolves a folio, we need to clear the tail
401 	 * page, rather than setting nr_pages to 1.
402 	 */
403 	folio->_folio_nr_pages = order ? 1U << order : 0;
404 #endif
405 }
406 
407 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
408 
409 /*
410  * in mm/compaction.c
411  */
412 /*
413  * compact_control is used to track pages being migrated and the free pages
414  * they are being migrated to during memory compaction. The free_pfn starts
415  * at the end of a zone and migrate_pfn begins at the start. Movable pages
416  * are moved to the end of a zone during a compaction run and the run
417  * completes when free_pfn <= migrate_pfn
418  */
419 struct compact_control {
420 	struct list_head freepages;	/* List of free pages to migrate to */
421 	struct list_head migratepages;	/* List of pages being migrated */
422 	unsigned int nr_freepages;	/* Number of isolated free pages */
423 	unsigned int nr_migratepages;	/* Number of pages to migrate */
424 	unsigned long free_pfn;		/* isolate_freepages search base */
425 	/*
426 	 * Acts as an in/out parameter to page isolation for migration.
427 	 * isolate_migratepages uses it as a search base.
428 	 * isolate_migratepages_block will update the value to the next pfn
429 	 * after the last isolated one.
430 	 */
431 	unsigned long migrate_pfn;
432 	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
433 	struct zone *zone;
434 	unsigned long total_migrate_scanned;
435 	unsigned long total_free_scanned;
436 	unsigned short fast_search_fail;/* failures to use free list searches */
437 	short search_order;		/* order to start a fast search at */
438 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
439 	int order;			/* order a direct compactor needs */
440 	int migratetype;		/* migratetype of direct compactor */
441 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
442 	const int highest_zoneidx;	/* zone index of a direct compactor */
443 	enum migrate_mode mode;		/* Async or sync migration mode */
444 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
445 	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
446 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
447 	bool direct_compaction;		/* False from kcompactd or /proc/... */
448 	bool proactive_compaction;	/* kcompactd proactive compaction */
449 	bool whole_zone;		/* Whole zone should/has been scanned */
450 	bool contended;			/* Signal lock contention */
451 	bool finish_pageblock;		/* Scan the remainder of a pageblock. Used
452 					 * when there are potentially transient
453 					 * isolation or migration failures to
454 					 * ensure forward progress.
455 					 */
456 	bool alloc_contig;		/* alloc_contig_range allocation */
457 };
458 
459 /*
460  * Used in direct compaction when a page should be taken from the freelists
461  * immediately when one is created during the free path.
462  */
463 struct capture_control {
464 	struct compact_control *cc;
465 	struct page *page;
466 };
467 
468 unsigned long
469 isolate_freepages_range(struct compact_control *cc,
470 			unsigned long start_pfn, unsigned long end_pfn);
471 int
472 isolate_migratepages_range(struct compact_control *cc,
473 			   unsigned long low_pfn, unsigned long end_pfn);
474 
475 int __alloc_contig_migrate_range(struct compact_control *cc,
476 					unsigned long start, unsigned long end);
477 #endif
478 int find_suitable_fallback(struct free_area *area, unsigned int order,
479 			int migratetype, bool only_stealable, bool *can_steal);
480 
481 /*
482  * These three helpers classifies VMAs for virtual memory accounting.
483  */
484 
485 /*
486  * Executable code area - executable, not writable, not stack
487  */
488 static inline bool is_exec_mapping(vm_flags_t flags)
489 {
490 	return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
491 }
492 
493 /*
494  * Stack area - automatically grows in one direction
495  *
496  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
497  * do_mmap() forbids all other combinations.
498  */
499 static inline bool is_stack_mapping(vm_flags_t flags)
500 {
501 	return (flags & VM_STACK) == VM_STACK;
502 }
503 
504 /*
505  * Data area - private, writable, not stack
506  */
507 static inline bool is_data_mapping(vm_flags_t flags)
508 {
509 	return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
510 }
511 
512 /* mm/util.c */
513 struct anon_vma *folio_anon_vma(struct folio *folio);
514 
515 #ifdef CONFIG_MMU
516 void unmap_mapping_folio(struct folio *folio);
517 extern long populate_vma_page_range(struct vm_area_struct *vma,
518 		unsigned long start, unsigned long end, int *locked);
519 extern long faultin_vma_page_range(struct vm_area_struct *vma,
520 				   unsigned long start, unsigned long end,
521 				   bool write, int *locked);
522 extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
523 			      unsigned long len);
524 /*
525  * mlock_vma_folio() and munlock_vma_folio():
526  * should be called with vma's mmap_lock held for read or write,
527  * under page table lock for the pte/pmd being added or removed.
528  *
529  * mlock is usually called at the end of page_add_*_rmap(), munlock at
530  * the end of page_remove_rmap(); but new anon folios are managed by
531  * folio_add_lru_vma() calling mlock_new_folio().
532  *
533  * @compound is used to include pmd mappings of THPs, but filter out
534  * pte mappings of THPs, which cannot be consistently counted: a pte
535  * mapping of the THP head cannot be distinguished by the page alone.
536  */
537 void mlock_folio(struct folio *folio);
538 static inline void mlock_vma_folio(struct folio *folio,
539 			struct vm_area_struct *vma, bool compound)
540 {
541 	/*
542 	 * The VM_SPECIAL check here serves two purposes.
543 	 * 1) VM_IO check prevents migration from double-counting during mlock.
544 	 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
545 	 *    is never left set on a VM_SPECIAL vma, there is an interval while
546 	 *    file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
547 	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
548 	 */
549 	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) &&
550 	    (compound || !folio_test_large(folio)))
551 		mlock_folio(folio);
552 }
553 
554 void munlock_folio(struct folio *folio);
555 static inline void munlock_vma_folio(struct folio *folio,
556 			struct vm_area_struct *vma, bool compound)
557 {
558 	if (unlikely(vma->vm_flags & VM_LOCKED) &&
559 	    (compound || !folio_test_large(folio)))
560 		munlock_folio(folio);
561 }
562 
563 void mlock_new_folio(struct folio *folio);
564 bool need_mlock_drain(int cpu);
565 void mlock_drain_local(void);
566 void mlock_drain_remote(int cpu);
567 
568 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
569 
570 /*
571  * Return the start of user virtual address at the specific offset within
572  * a vma.
573  */
574 static inline unsigned long
575 vma_pgoff_address(pgoff_t pgoff, unsigned long nr_pages,
576 		  struct vm_area_struct *vma)
577 {
578 	unsigned long address;
579 
580 	if (pgoff >= vma->vm_pgoff) {
581 		address = vma->vm_start +
582 			((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
583 		/* Check for address beyond vma (or wrapped through 0?) */
584 		if (address < vma->vm_start || address >= vma->vm_end)
585 			address = -EFAULT;
586 	} else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
587 		/* Test above avoids possibility of wrap to 0 on 32-bit */
588 		address = vma->vm_start;
589 	} else {
590 		address = -EFAULT;
591 	}
592 	return address;
593 }
594 
595 /*
596  * Return the start of user virtual address of a page within a vma.
597  * Returns -EFAULT if all of the page is outside the range of vma.
598  * If page is a compound head, the entire compound page is considered.
599  */
600 static inline unsigned long
601 vma_address(struct page *page, struct vm_area_struct *vma)
602 {
603 	VM_BUG_ON_PAGE(PageKsm(page), page);	/* KSM page->index unusable */
604 	return vma_pgoff_address(page_to_pgoff(page), compound_nr(page), vma);
605 }
606 
607 /*
608  * Then at what user virtual address will none of the range be found in vma?
609  * Assumes that vma_address() already returned a good starting address.
610  */
611 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
612 {
613 	struct vm_area_struct *vma = pvmw->vma;
614 	pgoff_t pgoff;
615 	unsigned long address;
616 
617 	/* Common case, plus ->pgoff is invalid for KSM */
618 	if (pvmw->nr_pages == 1)
619 		return pvmw->address + PAGE_SIZE;
620 
621 	pgoff = pvmw->pgoff + pvmw->nr_pages;
622 	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
623 	/* Check for address beyond vma (or wrapped through 0?) */
624 	if (address < vma->vm_start || address > vma->vm_end)
625 		address = vma->vm_end;
626 	return address;
627 }
628 
629 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
630 						    struct file *fpin)
631 {
632 	int flags = vmf->flags;
633 
634 	if (fpin)
635 		return fpin;
636 
637 	/*
638 	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
639 	 * anything, so we only pin the file and drop the mmap_lock if only
640 	 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
641 	 */
642 	if (fault_flag_allow_retry_first(flags) &&
643 	    !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
644 		fpin = get_file(vmf->vma->vm_file);
645 		mmap_read_unlock(vmf->vma->vm_mm);
646 	}
647 	return fpin;
648 }
649 #else /* !CONFIG_MMU */
650 static inline void unmap_mapping_folio(struct folio *folio) { }
651 static inline void mlock_new_folio(struct folio *folio) { }
652 static inline bool need_mlock_drain(int cpu) { return false; }
653 static inline void mlock_drain_local(void) { }
654 static inline void mlock_drain_remote(int cpu) { }
655 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
656 {
657 }
658 #endif /* !CONFIG_MMU */
659 
660 /* Memory initialisation debug and verification */
661 enum mminit_level {
662 	MMINIT_WARNING,
663 	MMINIT_VERIFY,
664 	MMINIT_TRACE
665 };
666 
667 #ifdef CONFIG_DEBUG_MEMORY_INIT
668 
669 extern int mminit_loglevel;
670 
671 #define mminit_dprintk(level, prefix, fmt, arg...) \
672 do { \
673 	if (level < mminit_loglevel) { \
674 		if (level <= MMINIT_WARNING) \
675 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
676 		else \
677 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
678 	} \
679 } while (0)
680 
681 extern void mminit_verify_pageflags_layout(void);
682 extern void mminit_verify_zonelist(void);
683 #else
684 
685 static inline void mminit_dprintk(enum mminit_level level,
686 				const char *prefix, const char *fmt, ...)
687 {
688 }
689 
690 static inline void mminit_verify_pageflags_layout(void)
691 {
692 }
693 
694 static inline void mminit_verify_zonelist(void)
695 {
696 }
697 #endif /* CONFIG_DEBUG_MEMORY_INIT */
698 
699 #define NODE_RECLAIM_NOSCAN	-2
700 #define NODE_RECLAIM_FULL	-1
701 #define NODE_RECLAIM_SOME	0
702 #define NODE_RECLAIM_SUCCESS	1
703 
704 #ifdef CONFIG_NUMA
705 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
706 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
707 #else
708 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
709 				unsigned int order)
710 {
711 	return NODE_RECLAIM_NOSCAN;
712 }
713 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
714 {
715 	return NUMA_NO_NODE;
716 }
717 #endif
718 
719 /*
720  * mm/memory-failure.c
721  */
722 extern int hwpoison_filter(struct page *p);
723 
724 extern u32 hwpoison_filter_dev_major;
725 extern u32 hwpoison_filter_dev_minor;
726 extern u64 hwpoison_filter_flags_mask;
727 extern u64 hwpoison_filter_flags_value;
728 extern u64 hwpoison_filter_memcg;
729 extern u32 hwpoison_filter_enable;
730 
731 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
732         unsigned long, unsigned long,
733         unsigned long, unsigned long);
734 
735 extern void set_pageblock_order(void);
736 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
737 					    struct list_head *page_list);
738 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
739 #define ALLOC_WMARK_MIN		WMARK_MIN
740 #define ALLOC_WMARK_LOW		WMARK_LOW
741 #define ALLOC_WMARK_HIGH	WMARK_HIGH
742 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
743 
744 /* Mask to get the watermark bits */
745 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
746 
747 /*
748  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
749  * cannot assume a reduced access to memory reserves is sufficient for
750  * !MMU
751  */
752 #ifdef CONFIG_MMU
753 #define ALLOC_OOM		0x08
754 #else
755 #define ALLOC_OOM		ALLOC_NO_WATERMARKS
756 #endif
757 
758 #define ALLOC_NON_BLOCK		 0x10 /* Caller cannot block. Allow access
759 				       * to 25% of the min watermark or
760 				       * 62.5% if __GFP_HIGH is set.
761 				       */
762 #define ALLOC_MIN_RESERVE	 0x20 /* __GFP_HIGH set. Allow access to 50%
763 				       * of the min watermark.
764 				       */
765 #define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
766 #define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
767 #ifdef CONFIG_ZONE_DMA32
768 #define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
769 #else
770 #define ALLOC_NOFRAGMENT	  0x0
771 #endif
772 #define ALLOC_HIGHATOMIC	0x200 /* Allows access to MIGRATE_HIGHATOMIC */
773 #define ALLOC_KSWAPD		0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
774 
775 /* Flags that allow allocations below the min watermark. */
776 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
777 
778 enum ttu_flags;
779 struct tlbflush_unmap_batch;
780 
781 
782 /*
783  * only for MM internal work items which do not depend on
784  * any allocations or locks which might depend on allocations
785  */
786 extern struct workqueue_struct *mm_percpu_wq;
787 
788 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
789 void try_to_unmap_flush(void);
790 void try_to_unmap_flush_dirty(void);
791 void flush_tlb_batched_pending(struct mm_struct *mm);
792 #else
793 static inline void try_to_unmap_flush(void)
794 {
795 }
796 static inline void try_to_unmap_flush_dirty(void)
797 {
798 }
799 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
800 {
801 }
802 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
803 
804 extern const struct trace_print_flags pageflag_names[];
805 extern const struct trace_print_flags pagetype_names[];
806 extern const struct trace_print_flags vmaflag_names[];
807 extern const struct trace_print_flags gfpflag_names[];
808 
809 static inline bool is_migrate_highatomic(enum migratetype migratetype)
810 {
811 	return migratetype == MIGRATE_HIGHATOMIC;
812 }
813 
814 static inline bool is_migrate_highatomic_page(struct page *page)
815 {
816 	return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
817 }
818 
819 void setup_zone_pageset(struct zone *zone);
820 
821 struct migration_target_control {
822 	int nid;		/* preferred node id */
823 	nodemask_t *nmask;
824 	gfp_t gfp_mask;
825 };
826 
827 /*
828  * mm/filemap.c
829  */
830 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
831 			      struct folio *folio, loff_t fpos, size_t size);
832 
833 /*
834  * mm/vmalloc.c
835  */
836 #ifdef CONFIG_MMU
837 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
838                 pgprot_t prot, struct page **pages, unsigned int page_shift);
839 #else
840 static inline
841 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
842                 pgprot_t prot, struct page **pages, unsigned int page_shift)
843 {
844 	return -EINVAL;
845 }
846 #endif
847 
848 int __vmap_pages_range_noflush(unsigned long addr, unsigned long end,
849 			       pgprot_t prot, struct page **pages,
850 			       unsigned int page_shift);
851 
852 void vunmap_range_noflush(unsigned long start, unsigned long end);
853 
854 void __vunmap_range_noflush(unsigned long start, unsigned long end);
855 
856 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
857 		      unsigned long addr, int page_nid, int *flags);
858 
859 void free_zone_device_page(struct page *page);
860 int migrate_device_coherent_page(struct page *page);
861 
862 /*
863  * mm/gup.c
864  */
865 struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags);
866 int __must_check try_grab_page(struct page *page, unsigned int flags);
867 
868 enum {
869 	/* mark page accessed */
870 	FOLL_TOUCH = 1 << 16,
871 	/* a retry, previous pass started an IO */
872 	FOLL_TRIED = 1 << 17,
873 	/* we are working on non-current tsk/mm */
874 	FOLL_REMOTE = 1 << 18,
875 	/* pages must be released via unpin_user_page */
876 	FOLL_PIN = 1 << 19,
877 	/* gup_fast: prevent fall-back to slow gup */
878 	FOLL_FAST_ONLY = 1 << 20,
879 	/* allow unlocking the mmap lock */
880 	FOLL_UNLOCKABLE = 1 << 21,
881 };
882 
883 /*
884  * Indicates for which pages that are write-protected in the page table,
885  * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
886  * GUP pin will remain consistent with the pages mapped into the page tables
887  * of the MM.
888  *
889  * Temporary unmapping of PageAnonExclusive() pages or clearing of
890  * PageAnonExclusive() has to protect against concurrent GUP:
891  * * Ordinary GUP: Using the PT lock
892  * * GUP-fast and fork(): mm->write_protect_seq
893  * * GUP-fast and KSM or temporary unmapping (swap, migration): see
894  *    page_try_share_anon_rmap()
895  *
896  * Must be called with the (sub)page that's actually referenced via the
897  * page table entry, which might not necessarily be the head page for a
898  * PTE-mapped THP.
899  *
900  * If the vma is NULL, we're coming from the GUP-fast path and might have
901  * to fallback to the slow path just to lookup the vma.
902  */
903 static inline bool gup_must_unshare(struct vm_area_struct *vma,
904 				    unsigned int flags, struct page *page)
905 {
906 	/*
907 	 * FOLL_WRITE is implicitly handled correctly as the page table entry
908 	 * has to be writable -- and if it references (part of) an anonymous
909 	 * folio, that part is required to be marked exclusive.
910 	 */
911 	if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
912 		return false;
913 	/*
914 	 * Note: PageAnon(page) is stable until the page is actually getting
915 	 * freed.
916 	 */
917 	if (!PageAnon(page)) {
918 		/*
919 		 * We only care about R/O long-term pining: R/O short-term
920 		 * pinning does not have the semantics to observe successive
921 		 * changes through the process page tables.
922 		 */
923 		if (!(flags & FOLL_LONGTERM))
924 			return false;
925 
926 		/* We really need the vma ... */
927 		if (!vma)
928 			return true;
929 
930 		/*
931 		 * ... because we only care about writable private ("COW")
932 		 * mappings where we have to break COW early.
933 		 */
934 		return is_cow_mapping(vma->vm_flags);
935 	}
936 
937 	/* Paired with a memory barrier in page_try_share_anon_rmap(). */
938 	if (IS_ENABLED(CONFIG_HAVE_FAST_GUP))
939 		smp_rmb();
940 
941 	/*
942 	 * Note that PageKsm() pages cannot be exclusive, and consequently,
943 	 * cannot get pinned.
944 	 */
945 	return !PageAnonExclusive(page);
946 }
947 
948 extern bool mirrored_kernelcore;
949 
950 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
951 {
952 	/*
953 	 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
954 	 * enablements, because when without soft-dirty being compiled in,
955 	 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
956 	 * will be constantly true.
957 	 */
958 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
959 		return false;
960 
961 	/*
962 	 * Soft-dirty is kind of special: its tracking is enabled when the
963 	 * vma flags not set.
964 	 */
965 	return !(vma->vm_flags & VM_SOFTDIRTY);
966 }
967 
968 /*
969  * VMA Iterator functions shared between nommu and mmap
970  */
971 static inline int vma_iter_prealloc(struct vma_iterator *vmi)
972 {
973 	return mas_preallocate(&vmi->mas, GFP_KERNEL);
974 }
975 
976 static inline void vma_iter_clear(struct vma_iterator *vmi,
977 				  unsigned long start, unsigned long end)
978 {
979 	mas_set_range(&vmi->mas, start, end - 1);
980 	mas_store_prealloc(&vmi->mas, NULL);
981 }
982 
983 static inline struct vm_area_struct *vma_iter_load(struct vma_iterator *vmi)
984 {
985 	return mas_walk(&vmi->mas);
986 }
987 
988 /* Store a VMA with preallocated memory */
989 static inline void vma_iter_store(struct vma_iterator *vmi,
990 				  struct vm_area_struct *vma)
991 {
992 
993 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
994 	if (WARN_ON(vmi->mas.node != MAS_START && vmi->mas.index > vma->vm_start)) {
995 		printk("%lu > %lu\n", vmi->mas.index, vma->vm_start);
996 		printk("store of vma %lu-%lu", vma->vm_start, vma->vm_end);
997 		printk("into slot    %lu-%lu", vmi->mas.index, vmi->mas.last);
998 		mt_dump(vmi->mas.tree);
999 	}
1000 	if (WARN_ON(vmi->mas.node != MAS_START && vmi->mas.last <  vma->vm_start)) {
1001 		printk("%lu < %lu\n", vmi->mas.last, vma->vm_start);
1002 		printk("store of vma %lu-%lu", vma->vm_start, vma->vm_end);
1003 		printk("into slot    %lu-%lu", vmi->mas.index, vmi->mas.last);
1004 		mt_dump(vmi->mas.tree);
1005 	}
1006 #endif
1007 
1008 	if (vmi->mas.node != MAS_START &&
1009 	    ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
1010 		vma_iter_invalidate(vmi);
1011 
1012 	vmi->mas.index = vma->vm_start;
1013 	vmi->mas.last = vma->vm_end - 1;
1014 	mas_store_prealloc(&vmi->mas, vma);
1015 }
1016 
1017 static inline int vma_iter_store_gfp(struct vma_iterator *vmi,
1018 			struct vm_area_struct *vma, gfp_t gfp)
1019 {
1020 	if (vmi->mas.node != MAS_START &&
1021 	    ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
1022 		vma_iter_invalidate(vmi);
1023 
1024 	vmi->mas.index = vma->vm_start;
1025 	vmi->mas.last = vma->vm_end - 1;
1026 	mas_store_gfp(&vmi->mas, vma, gfp);
1027 	if (unlikely(mas_is_err(&vmi->mas)))
1028 		return -ENOMEM;
1029 
1030 	return 0;
1031 }
1032 
1033 /*
1034  * VMA lock generalization
1035  */
1036 struct vma_prepare {
1037 	struct vm_area_struct *vma;
1038 	struct vm_area_struct *adj_next;
1039 	struct file *file;
1040 	struct address_space *mapping;
1041 	struct anon_vma *anon_vma;
1042 	struct vm_area_struct *insert;
1043 	struct vm_area_struct *remove;
1044 	struct vm_area_struct *remove2;
1045 };
1046 #endif	/* __MM_INTERNAL_H */
1047