xref: /linux/mm/internal.h (revision beace86e61e465dba204a268ab3f3377153a4973)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* internal.h: mm/ internal definitions
3  *
4  * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 #ifndef __MM_INTERNAL_H
8 #define __MM_INTERNAL_H
9 
10 #include <linux/fs.h>
11 #include <linux/khugepaged.h>
12 #include <linux/mm.h>
13 #include <linux/mm_inline.h>
14 #include <linux/pagemap.h>
15 #include <linux/pagewalk.h>
16 #include <linux/rmap.h>
17 #include <linux/swap.h>
18 #include <linux/swapops.h>
19 #include <linux/swap_cgroup.h>
20 #include <linux/tracepoint-defs.h>
21 
22 /* Internal core VMA manipulation functions. */
23 #include "vma.h"
24 
25 struct folio_batch;
26 
27 /*
28  * 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 most 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 #ifdef CONFIG_MMU
201 
202 /* Flags for folio_pte_batch(). */
203 typedef int __bitwise fpb_t;
204 
205 /* Compare PTEs respecting the dirty bit. */
206 #define FPB_RESPECT_DIRTY		((__force fpb_t)BIT(0))
207 
208 /* Compare PTEs respecting the soft-dirty bit. */
209 #define FPB_RESPECT_SOFT_DIRTY		((__force fpb_t)BIT(1))
210 
211 /* Compare PTEs respecting the writable bit. */
212 #define FPB_RESPECT_WRITE		((__force fpb_t)BIT(2))
213 
214 /*
215  * Merge PTE write bits: if any PTE in the batch is writable, modify the
216  * PTE at @ptentp to be writable.
217  */
218 #define FPB_MERGE_WRITE			((__force fpb_t)BIT(3))
219 
220 /*
221  * Merge PTE young and dirty bits: if any PTE in the batch is young or dirty,
222  * modify the PTE at @ptentp to be young or dirty, respectively.
223  */
224 #define FPB_MERGE_YOUNG_DIRTY		((__force fpb_t)BIT(4))
225 
226 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
227 {
228 	if (!(flags & FPB_RESPECT_DIRTY))
229 		pte = pte_mkclean(pte);
230 	if (likely(!(flags & FPB_RESPECT_SOFT_DIRTY)))
231 		pte = pte_clear_soft_dirty(pte);
232 	if (likely(!(flags & FPB_RESPECT_WRITE)))
233 		pte = pte_wrprotect(pte);
234 	return pte_mkold(pte);
235 }
236 
237 /**
238  * folio_pte_batch_flags - detect a PTE batch for a large folio
239  * @folio: The large folio to detect a PTE batch for.
240  * @vma: The VMA. Only relevant with FPB_MERGE_WRITE, otherwise can be NULL.
241  * @ptep: Page table pointer for the first entry.
242  * @ptentp: Pointer to a COPY of the first page table entry whose flags this
243  *	    function updates based on @flags if appropriate.
244  * @max_nr: The maximum number of table entries to consider.
245  * @flags: Flags to modify the PTE batch semantics.
246  *
247  * Detect a PTE batch: consecutive (present) PTEs that map consecutive
248  * pages of the same large folio in a single VMA and a single page table.
249  *
250  * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN,
251  * the accessed bit, writable bit, dirty bit (unless FPB_RESPECT_DIRTY is set)
252  * and soft-dirty bit (unless FPB_RESPECT_SOFT_DIRTY is set).
253  *
254  * @ptep must map any page of the folio. max_nr must be at least one and
255  * must be limited by the caller so scanning cannot exceed a single VMA and
256  * a single page table.
257  *
258  * Depending on the FPB_MERGE_* flags, the pte stored at @ptentp will
259  * be updated: it's crucial that a pointer to a COPY of the first
260  * page table entry, obtained through ptep_get(), is provided as @ptentp.
261  *
262  * This function will be inlined to optimize based on the input parameters;
263  * consider using folio_pte_batch() instead if applicable.
264  *
265  * Return: the number of table entries in the batch.
266  */
267 static inline unsigned int folio_pte_batch_flags(struct folio *folio,
268 		struct vm_area_struct *vma, pte_t *ptep, pte_t *ptentp,
269 		unsigned int max_nr, fpb_t flags)
270 {
271 	bool any_writable = false, any_young = false, any_dirty = false;
272 	pte_t expected_pte, pte = *ptentp;
273 	unsigned int nr, cur_nr;
274 
275 	VM_WARN_ON_FOLIO(!pte_present(pte), folio);
276 	VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio);
277 	VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio);
278 	/*
279 	 * Ensure this is a pointer to a copy not a pointer into a page table.
280 	 * If this is a stack value, it won't be a valid virtual address, but
281 	 * that's fine because it also cannot be pointing into the page table.
282 	 */
283 	VM_WARN_ON(virt_addr_valid(ptentp) && PageTable(virt_to_page(ptentp)));
284 
285 	/* Limit max_nr to the actual remaining PFNs in the folio we could batch. */
286 	max_nr = min_t(unsigned long, max_nr,
287 		       folio_pfn(folio) + folio_nr_pages(folio) - pte_pfn(pte));
288 
289 	nr = pte_batch_hint(ptep, pte);
290 	expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags);
291 	ptep = ptep + nr;
292 
293 	while (nr < max_nr) {
294 		pte = ptep_get(ptep);
295 
296 		if (!pte_same(__pte_batch_clear_ignored(pte, flags), expected_pte))
297 			break;
298 
299 		if (flags & FPB_MERGE_WRITE)
300 			any_writable |= pte_write(pte);
301 		if (flags & FPB_MERGE_YOUNG_DIRTY) {
302 			any_young |= pte_young(pte);
303 			any_dirty |= pte_dirty(pte);
304 		}
305 
306 		cur_nr = pte_batch_hint(ptep, pte);
307 		expected_pte = pte_advance_pfn(expected_pte, cur_nr);
308 		ptep += cur_nr;
309 		nr += cur_nr;
310 	}
311 
312 	if (any_writable)
313 		*ptentp = pte_mkwrite(*ptentp, vma);
314 	if (any_young)
315 		*ptentp = pte_mkyoung(*ptentp);
316 	if (any_dirty)
317 		*ptentp = pte_mkdirty(*ptentp);
318 
319 	return min(nr, max_nr);
320 }
321 
322 unsigned int folio_pte_batch(struct folio *folio, pte_t *ptep, pte_t pte,
323 		unsigned int max_nr);
324 
325 /**
326  * pte_move_swp_offset - Move the swap entry offset field of a swap pte
327  *	 forward or backward by delta
328  * @pte: The initial pte state; is_swap_pte(pte) must be true and
329  *	 non_swap_entry() must be false.
330  * @delta: The direction and the offset we are moving; forward if delta
331  *	 is positive; backward if delta is negative
332  *
333  * Moves the swap offset, while maintaining all other fields, including
334  * swap type, and any swp pte bits. The resulting pte is returned.
335  */
336 static inline pte_t pte_move_swp_offset(pte_t pte, long delta)
337 {
338 	swp_entry_t entry = pte_to_swp_entry(pte);
339 	pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry),
340 						   (swp_offset(entry) + delta)));
341 
342 	if (pte_swp_soft_dirty(pte))
343 		new = pte_swp_mksoft_dirty(new);
344 	if (pte_swp_exclusive(pte))
345 		new = pte_swp_mkexclusive(new);
346 	if (pte_swp_uffd_wp(pte))
347 		new = pte_swp_mkuffd_wp(new);
348 
349 	return new;
350 }
351 
352 
353 /**
354  * pte_next_swp_offset - Increment the swap entry offset field of a swap pte.
355  * @pte: The initial pte state; is_swap_pte(pte) must be true and
356  *	 non_swap_entry() must be false.
357  *
358  * Increments the swap offset, while maintaining all other fields, including
359  * swap type, and any swp pte bits. The resulting pte is returned.
360  */
361 static inline pte_t pte_next_swp_offset(pte_t pte)
362 {
363 	return pte_move_swp_offset(pte, 1);
364 }
365 
366 /**
367  * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries
368  * @start_ptep: Page table pointer for the first entry.
369  * @max_nr: The maximum number of table entries to consider.
370  * @pte: Page table entry for the first entry.
371  *
372  * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs
373  * containing swap entries all with consecutive offsets and targeting the same
374  * swap type, all with matching swp pte bits.
375  *
376  * max_nr must be at least one and must be limited by the caller so scanning
377  * cannot exceed a single page table.
378  *
379  * Return: the number of table entries in the batch.
380  */
381 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte)
382 {
383 	pte_t expected_pte = pte_next_swp_offset(pte);
384 	const pte_t *end_ptep = start_ptep + max_nr;
385 	swp_entry_t entry = pte_to_swp_entry(pte);
386 	pte_t *ptep = start_ptep + 1;
387 	unsigned short cgroup_id;
388 
389 	VM_WARN_ON(max_nr < 1);
390 	VM_WARN_ON(!is_swap_pte(pte));
391 	VM_WARN_ON(non_swap_entry(entry));
392 
393 	cgroup_id = lookup_swap_cgroup_id(entry);
394 	while (ptep < end_ptep) {
395 		pte = ptep_get(ptep);
396 
397 		if (!pte_same(pte, expected_pte))
398 			break;
399 		if (lookup_swap_cgroup_id(pte_to_swp_entry(pte)) != cgroup_id)
400 			break;
401 		expected_pte = pte_next_swp_offset(expected_pte);
402 		ptep++;
403 	}
404 
405 	return ptep - start_ptep;
406 }
407 #endif /* CONFIG_MMU */
408 
409 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
410 						int nr_throttled);
411 static inline void acct_reclaim_writeback(struct folio *folio)
412 {
413 	pg_data_t *pgdat = folio_pgdat(folio);
414 	int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
415 
416 	if (nr_throttled)
417 		__acct_reclaim_writeback(pgdat, folio, nr_throttled);
418 }
419 
420 static inline void wake_throttle_isolated(pg_data_t *pgdat)
421 {
422 	wait_queue_head_t *wqh;
423 
424 	wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
425 	if (waitqueue_active(wqh))
426 		wake_up(wqh);
427 }
428 
429 vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf);
430 static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf)
431 {
432 	vm_fault_t ret = __vmf_anon_prepare(vmf);
433 
434 	if (unlikely(ret & VM_FAULT_RETRY))
435 		vma_end_read(vmf->vma);
436 	return ret;
437 }
438 
439 vm_fault_t do_swap_page(struct vm_fault *vmf);
440 void folio_rotate_reclaimable(struct folio *folio);
441 bool __folio_end_writeback(struct folio *folio);
442 void deactivate_file_folio(struct folio *folio);
443 void folio_activate(struct folio *folio);
444 
445 void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas,
446 		   struct vm_area_struct *start_vma, unsigned long floor,
447 		   unsigned long ceiling, bool mm_wr_locked);
448 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
449 
450 struct zap_details;
451 void unmap_page_range(struct mmu_gather *tlb,
452 			     struct vm_area_struct *vma,
453 			     unsigned long addr, unsigned long end,
454 			     struct zap_details *details);
455 void zap_page_range_single_batched(struct mmu_gather *tlb,
456 		struct vm_area_struct *vma, unsigned long addr,
457 		unsigned long size, struct zap_details *details);
458 int folio_unmap_invalidate(struct address_space *mapping, struct folio *folio,
459 			   gfp_t gfp);
460 
461 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *);
462 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
463 static inline void force_page_cache_readahead(struct address_space *mapping,
464 		struct file *file, pgoff_t index, unsigned long nr_to_read)
465 {
466 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
467 	force_page_cache_ra(&ractl, nr_to_read);
468 }
469 
470 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
471 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
472 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
473 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
474 void filemap_free_folio(struct address_space *mapping, struct folio *folio);
475 int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
476 bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
477 		loff_t end);
478 long mapping_evict_folio(struct address_space *mapping, struct folio *folio);
479 unsigned long mapping_try_invalidate(struct address_space *mapping,
480 		pgoff_t start, pgoff_t end, unsigned long *nr_failed);
481 
482 /**
483  * folio_evictable - Test whether a folio is evictable.
484  * @folio: The folio to test.
485  *
486  * Test whether @folio is evictable -- i.e., should be placed on
487  * active/inactive lists vs unevictable list.
488  *
489  * Reasons folio might not be evictable:
490  * 1. folio's mapping marked unevictable
491  * 2. One of the pages in the folio is part of an mlocked VMA
492  */
493 static inline bool folio_evictable(struct folio *folio)
494 {
495 	bool ret;
496 
497 	/* Prevent address_space of inode and swap cache from being freed */
498 	rcu_read_lock();
499 	ret = !mapping_unevictable(folio_mapping(folio)) &&
500 			!folio_test_mlocked(folio);
501 	rcu_read_unlock();
502 	return ret;
503 }
504 
505 /*
506  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
507  * a count of one.
508  */
509 static inline void set_page_refcounted(struct page *page)
510 {
511 	VM_BUG_ON_PAGE(PageTail(page), page);
512 	VM_BUG_ON_PAGE(page_ref_count(page), page);
513 	set_page_count(page, 1);
514 }
515 
516 /*
517  * Return true if a folio needs ->release_folio() calling upon it.
518  */
519 static inline bool folio_needs_release(struct folio *folio)
520 {
521 	struct address_space *mapping = folio_mapping(folio);
522 
523 	return folio_has_private(folio) ||
524 		(mapping && mapping_release_always(mapping));
525 }
526 
527 extern unsigned long highest_memmap_pfn;
528 
529 /*
530  * Maximum number of reclaim retries without progress before the OOM
531  * killer is consider the only way forward.
532  */
533 #define MAX_RECLAIM_RETRIES 16
534 
535 /*
536  * in mm/vmscan.c:
537  */
538 bool folio_isolate_lru(struct folio *folio);
539 void folio_putback_lru(struct folio *folio);
540 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
541 #ifdef CONFIG_NUMA
542 int user_proactive_reclaim(char *buf,
543 			   struct mem_cgroup *memcg, pg_data_t *pgdat);
544 #else
545 static inline int user_proactive_reclaim(char *buf,
546 			   struct mem_cgroup *memcg, pg_data_t *pgdat)
547 {
548 	return 0;
549 }
550 #endif
551 
552 /*
553  * in mm/rmap.c:
554  */
555 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
556 
557 /*
558  * in mm/page_alloc.c
559  */
560 #define K(x) ((x) << (PAGE_SHIFT-10))
561 
562 extern char * const zone_names[MAX_NR_ZONES];
563 
564 /* perform sanity checks on struct pages being allocated or freed */
565 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
566 
567 extern int min_free_kbytes;
568 extern int defrag_mode;
569 
570 void setup_per_zone_wmarks(void);
571 void calculate_min_free_kbytes(void);
572 int __meminit init_per_zone_wmark_min(void);
573 void page_alloc_sysctl_init(void);
574 
575 /*
576  * Structure for holding the mostly immutable allocation parameters passed
577  * between functions involved in allocations, including the alloc_pages*
578  * family of functions.
579  *
580  * nodemask, migratetype and highest_zoneidx are initialized only once in
581  * __alloc_pages() and then never change.
582  *
583  * zonelist, preferred_zone and highest_zoneidx are set first in
584  * __alloc_pages() for the fast path, and might be later changed
585  * in __alloc_pages_slowpath(). All other functions pass the whole structure
586  * by a const pointer.
587  */
588 struct alloc_context {
589 	struct zonelist *zonelist;
590 	nodemask_t *nodemask;
591 	struct zoneref *preferred_zoneref;
592 	int migratetype;
593 
594 	/*
595 	 * highest_zoneidx represents highest usable zone index of
596 	 * the allocation request. Due to the nature of the zone,
597 	 * memory on lower zone than the highest_zoneidx will be
598 	 * protected by lowmem_reserve[highest_zoneidx].
599 	 *
600 	 * highest_zoneidx is also used by reclaim/compaction to limit
601 	 * the target zone since higher zone than this index cannot be
602 	 * usable for this allocation request.
603 	 */
604 	enum zone_type highest_zoneidx;
605 	bool spread_dirty_pages;
606 };
607 
608 /*
609  * This function returns the order of a free page in the buddy system. In
610  * general, page_zone(page)->lock must be held by the caller to prevent the
611  * page from being allocated in parallel and returning garbage as the order.
612  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
613  * page cannot be allocated or merged in parallel. Alternatively, it must
614  * handle invalid values gracefully, and use buddy_order_unsafe() below.
615  */
616 static inline unsigned int buddy_order(struct page *page)
617 {
618 	/* PageBuddy() must be checked by the caller */
619 	return page_private(page);
620 }
621 
622 /*
623  * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
624  * PageBuddy() should be checked first by the caller to minimize race window,
625  * and invalid values must be handled gracefully.
626  *
627  * READ_ONCE is used so that if the caller assigns the result into a local
628  * variable and e.g. tests it for valid range before using, the compiler cannot
629  * decide to remove the variable and inline the page_private(page) multiple
630  * times, potentially observing different values in the tests and the actual
631  * use of the result.
632  */
633 #define buddy_order_unsafe(page)	READ_ONCE(page_private(page))
634 
635 /*
636  * This function checks whether a page is free && is the buddy
637  * we can coalesce a page and its buddy if
638  * (a) the buddy is not in a hole (check before calling!) &&
639  * (b) the buddy is in the buddy system &&
640  * (c) a page and its buddy have the same order &&
641  * (d) a page and its buddy are in the same zone.
642  *
643  * For recording whether a page is in the buddy system, we set PageBuddy.
644  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
645  *
646  * For recording page's order, we use page_private(page).
647  */
648 static inline bool page_is_buddy(struct page *page, struct page *buddy,
649 				 unsigned int order)
650 {
651 	if (!page_is_guard(buddy) && !PageBuddy(buddy))
652 		return false;
653 
654 	if (buddy_order(buddy) != order)
655 		return false;
656 
657 	/*
658 	 * zone check is done late to avoid uselessly calculating
659 	 * zone/node ids for pages that could never merge.
660 	 */
661 	if (page_zone_id(page) != page_zone_id(buddy))
662 		return false;
663 
664 	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
665 
666 	return true;
667 }
668 
669 /*
670  * Locate the struct page for both the matching buddy in our
671  * pair (buddy1) and the combined O(n+1) page they form (page).
672  *
673  * 1) Any buddy B1 will have an order O twin B2 which satisfies
674  * the following equation:
675  *     B2 = B1 ^ (1 << O)
676  * For example, if the starting buddy (buddy2) is #8 its order
677  * 1 buddy is #10:
678  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
679  *
680  * 2) Any buddy B will have an order O+1 parent P which
681  * satisfies the following equation:
682  *     P = B & ~(1 << O)
683  *
684  * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER
685  */
686 static inline unsigned long
687 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
688 {
689 	return page_pfn ^ (1 << order);
690 }
691 
692 /*
693  * Find the buddy of @page and validate it.
694  * @page: The input page
695  * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
696  *       function is used in the performance-critical __free_one_page().
697  * @order: The order of the page
698  * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
699  *             page_to_pfn().
700  *
701  * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
702  * not the same as @page. The validation is necessary before use it.
703  *
704  * Return: the found buddy page or NULL if not found.
705  */
706 static inline struct page *find_buddy_page_pfn(struct page *page,
707 			unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
708 {
709 	unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
710 	struct page *buddy;
711 
712 	buddy = page + (__buddy_pfn - pfn);
713 	if (buddy_pfn)
714 		*buddy_pfn = __buddy_pfn;
715 
716 	if (page_is_buddy(page, buddy, order))
717 		return buddy;
718 	return NULL;
719 }
720 
721 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
722 				unsigned long end_pfn, struct zone *zone);
723 
724 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
725 				unsigned long end_pfn, struct zone *zone)
726 {
727 	if (zone->contiguous)
728 		return pfn_to_page(start_pfn);
729 
730 	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
731 }
732 
733 void set_zone_contiguous(struct zone *zone);
734 bool pfn_range_intersects_zones(int nid, unsigned long start_pfn,
735 			   unsigned long nr_pages);
736 
737 static inline void clear_zone_contiguous(struct zone *zone)
738 {
739 	zone->contiguous = false;
740 }
741 
742 extern int __isolate_free_page(struct page *page, unsigned int order);
743 extern void __putback_isolated_page(struct page *page, unsigned int order,
744 				    int mt);
745 extern void memblock_free_pages(struct page *page, unsigned long pfn,
746 					unsigned int order);
747 extern void __free_pages_core(struct page *page, unsigned int order,
748 		enum meminit_context context);
749 
750 /*
751  * This will have no effect, other than possibly generating a warning, if the
752  * caller passes in a non-large folio.
753  */
754 static inline void folio_set_order(struct folio *folio, unsigned int order)
755 {
756 	if (WARN_ON_ONCE(!order || !folio_test_large(folio)))
757 		return;
758 
759 	folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order;
760 #ifdef NR_PAGES_IN_LARGE_FOLIO
761 	folio->_nr_pages = 1U << order;
762 #endif
763 }
764 
765 bool __folio_unqueue_deferred_split(struct folio *folio);
766 static inline bool folio_unqueue_deferred_split(struct folio *folio)
767 {
768 	if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio))
769 		return false;
770 
771 	/*
772 	 * At this point, there is no one trying to add the folio to
773 	 * deferred_list. If folio is not in deferred_list, it's safe
774 	 * to check without acquiring the split_queue_lock.
775 	 */
776 	if (data_race(list_empty(&folio->_deferred_list)))
777 		return false;
778 
779 	return __folio_unqueue_deferred_split(folio);
780 }
781 
782 static inline struct folio *page_rmappable_folio(struct page *page)
783 {
784 	struct folio *folio = (struct folio *)page;
785 
786 	if (folio && folio_test_large(folio))
787 		folio_set_large_rmappable(folio);
788 	return folio;
789 }
790 
791 static inline void prep_compound_head(struct page *page, unsigned int order)
792 {
793 	struct folio *folio = (struct folio *)page;
794 
795 	folio_set_order(folio, order);
796 	atomic_set(&folio->_large_mapcount, -1);
797 	if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
798 		atomic_set(&folio->_nr_pages_mapped, 0);
799 	if (IS_ENABLED(CONFIG_MM_ID)) {
800 		folio->_mm_ids = 0;
801 		folio->_mm_id_mapcount[0] = -1;
802 		folio->_mm_id_mapcount[1] = -1;
803 	}
804 	if (IS_ENABLED(CONFIG_64BIT) || order > 1) {
805 		atomic_set(&folio->_pincount, 0);
806 		atomic_set(&folio->_entire_mapcount, -1);
807 	}
808 	if (order > 1)
809 		INIT_LIST_HEAD(&folio->_deferred_list);
810 }
811 
812 static inline void prep_compound_tail(struct page *head, int tail_idx)
813 {
814 	struct page *p = head + tail_idx;
815 
816 	p->mapping = TAIL_MAPPING;
817 	set_compound_head(p, head);
818 	set_page_private(p, 0);
819 }
820 
821 void post_alloc_hook(struct page *page, unsigned int order, gfp_t gfp_flags);
822 extern bool free_pages_prepare(struct page *page, unsigned int order);
823 
824 extern int user_min_free_kbytes;
825 
826 struct page *__alloc_frozen_pages_noprof(gfp_t, unsigned int order, int nid,
827 		nodemask_t *);
828 #define __alloc_frozen_pages(...) \
829 	alloc_hooks(__alloc_frozen_pages_noprof(__VA_ARGS__))
830 void free_frozen_pages(struct page *page, unsigned int order);
831 void free_unref_folios(struct folio_batch *fbatch);
832 
833 #ifdef CONFIG_NUMA
834 struct page *alloc_frozen_pages_noprof(gfp_t, unsigned int order);
835 #else
836 static inline struct page *alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order)
837 {
838 	return __alloc_frozen_pages_noprof(gfp, order, numa_node_id(), NULL);
839 }
840 #endif
841 
842 #define alloc_frozen_pages(...) \
843 	alloc_hooks(alloc_frozen_pages_noprof(__VA_ARGS__))
844 
845 extern void zone_pcp_reset(struct zone *zone);
846 extern void zone_pcp_disable(struct zone *zone);
847 extern void zone_pcp_enable(struct zone *zone);
848 extern void zone_pcp_init(struct zone *zone);
849 
850 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
851 			  phys_addr_t min_addr,
852 			  int nid, bool exact_nid);
853 
854 void memmap_init_range(unsigned long, int, unsigned long, unsigned long,
855 		unsigned long, enum meminit_context, struct vmem_altmap *, int,
856 		bool);
857 
858 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
859 
860 /*
861  * in mm/compaction.c
862  */
863 /*
864  * compact_control is used to track pages being migrated and the free pages
865  * they are being migrated to during memory compaction. The free_pfn starts
866  * at the end of a zone and migrate_pfn begins at the start. Movable pages
867  * are moved to the end of a zone during a compaction run and the run
868  * completes when free_pfn <= migrate_pfn
869  */
870 struct compact_control {
871 	struct list_head freepages[NR_PAGE_ORDERS];	/* List of free pages to migrate to */
872 	struct list_head migratepages;	/* List of pages being migrated */
873 	unsigned int nr_freepages;	/* Number of isolated free pages */
874 	unsigned int nr_migratepages;	/* Number of pages to migrate */
875 	unsigned long free_pfn;		/* isolate_freepages search base */
876 	/*
877 	 * Acts as an in/out parameter to page isolation for migration.
878 	 * isolate_migratepages uses it as a search base.
879 	 * isolate_migratepages_block will update the value to the next pfn
880 	 * after the last isolated one.
881 	 */
882 	unsigned long migrate_pfn;
883 	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
884 	struct zone *zone;
885 	unsigned long total_migrate_scanned;
886 	unsigned long total_free_scanned;
887 	unsigned short fast_search_fail;/* failures to use free list searches */
888 	short search_order;		/* order to start a fast search at */
889 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
890 	int order;			/* order a direct compactor needs */
891 	int migratetype;		/* migratetype of direct compactor */
892 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
893 	const int highest_zoneidx;	/* zone index of a direct compactor */
894 	enum migrate_mode mode;		/* Async or sync migration mode */
895 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
896 	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
897 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
898 	bool direct_compaction;		/* False from kcompactd or /proc/... */
899 	bool proactive_compaction;	/* kcompactd proactive compaction */
900 	bool whole_zone;		/* Whole zone should/has been scanned */
901 	bool contended;			/* Signal lock contention */
902 	bool finish_pageblock;		/* Scan the remainder of a pageblock. Used
903 					 * when there are potentially transient
904 					 * isolation or migration failures to
905 					 * ensure forward progress.
906 					 */
907 	bool alloc_contig;		/* alloc_contig_range allocation */
908 };
909 
910 /*
911  * Used in direct compaction when a page should be taken from the freelists
912  * immediately when one is created during the free path.
913  */
914 struct capture_control {
915 	struct compact_control *cc;
916 	struct page *page;
917 };
918 
919 unsigned long
920 isolate_freepages_range(struct compact_control *cc,
921 			unsigned long start_pfn, unsigned long end_pfn);
922 int
923 isolate_migratepages_range(struct compact_control *cc,
924 			   unsigned long low_pfn, unsigned long end_pfn);
925 
926 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
927 void init_cma_reserved_pageblock(struct page *page);
928 
929 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
930 
931 struct cma;
932 
933 #ifdef CONFIG_CMA
934 void *cma_reserve_early(struct cma *cma, unsigned long size);
935 void init_cma_pageblock(struct page *page);
936 #else
937 static inline void *cma_reserve_early(struct cma *cma, unsigned long size)
938 {
939 	return NULL;
940 }
941 static inline void init_cma_pageblock(struct page *page)
942 {
943 }
944 #endif
945 
946 
947 int find_suitable_fallback(struct free_area *area, unsigned int order,
948 			   int migratetype, bool claimable);
949 
950 static inline bool free_area_empty(struct free_area *area, int migratetype)
951 {
952 	return list_empty(&area->free_list[migratetype]);
953 }
954 
955 /* mm/util.c */
956 struct anon_vma *folio_anon_vma(const struct folio *folio);
957 
958 #ifdef CONFIG_MMU
959 void unmap_mapping_folio(struct folio *folio);
960 extern long populate_vma_page_range(struct vm_area_struct *vma,
961 		unsigned long start, unsigned long end, int *locked);
962 extern long faultin_page_range(struct mm_struct *mm, unsigned long start,
963 		unsigned long end, bool write, int *locked);
964 extern bool mlock_future_ok(struct mm_struct *mm, vm_flags_t vm_flags,
965 			       unsigned long bytes);
966 
967 /*
968  * NOTE: This function can't tell whether the folio is "fully mapped" in the
969  * range.
970  * "fully mapped" means all the pages of folio is associated with the page
971  * table of range while this function just check whether the folio range is
972  * within the range [start, end). Function caller needs to do page table
973  * check if it cares about the page table association.
974  *
975  * Typical usage (like mlock or madvise) is:
976  * Caller knows at least 1 page of folio is associated with page table of VMA
977  * and the range [start, end) is intersect with the VMA range. Caller wants
978  * to know whether the folio is fully associated with the range. It calls
979  * this function to check whether the folio is in the range first. Then checks
980  * the page table to know whether the folio is fully mapped to the range.
981  */
982 static inline bool
983 folio_within_range(struct folio *folio, struct vm_area_struct *vma,
984 		unsigned long start, unsigned long end)
985 {
986 	pgoff_t pgoff, addr;
987 	unsigned long vma_pglen = vma_pages(vma);
988 
989 	VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio);
990 	if (start > end)
991 		return false;
992 
993 	if (start < vma->vm_start)
994 		start = vma->vm_start;
995 
996 	if (end > vma->vm_end)
997 		end = vma->vm_end;
998 
999 	pgoff = folio_pgoff(folio);
1000 
1001 	/* if folio start address is not in vma range */
1002 	if (!in_range(pgoff, vma->vm_pgoff, vma_pglen))
1003 		return false;
1004 
1005 	addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1006 
1007 	return !(addr < start || end - addr < folio_size(folio));
1008 }
1009 
1010 static inline bool
1011 folio_within_vma(struct folio *folio, struct vm_area_struct *vma)
1012 {
1013 	return folio_within_range(folio, vma, vma->vm_start, vma->vm_end);
1014 }
1015 
1016 /*
1017  * mlock_vma_folio() and munlock_vma_folio():
1018  * should be called with vma's mmap_lock held for read or write,
1019  * under page table lock for the pte/pmd being added or removed.
1020  *
1021  * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at
1022  * the end of folio_remove_rmap_*(); but new anon folios are managed by
1023  * folio_add_lru_vma() calling mlock_new_folio().
1024  */
1025 void mlock_folio(struct folio *folio);
1026 static inline void mlock_vma_folio(struct folio *folio,
1027 				struct vm_area_struct *vma)
1028 {
1029 	/*
1030 	 * The VM_SPECIAL check here serves two purposes.
1031 	 * 1) VM_IO check prevents migration from double-counting during mlock.
1032 	 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
1033 	 *    is never left set on a VM_SPECIAL vma, there is an interval while
1034 	 *    file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
1035 	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
1036 	 */
1037 	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED))
1038 		mlock_folio(folio);
1039 }
1040 
1041 void munlock_folio(struct folio *folio);
1042 static inline void munlock_vma_folio(struct folio *folio,
1043 					struct vm_area_struct *vma)
1044 {
1045 	/*
1046 	 * munlock if the function is called. Ideally, we should only
1047 	 * do munlock if any page of folio is unmapped from VMA and
1048 	 * cause folio not fully mapped to VMA.
1049 	 *
1050 	 * But it's not easy to confirm that's the situation. So we
1051 	 * always munlock the folio and page reclaim will correct it
1052 	 * if it's wrong.
1053 	 */
1054 	if (unlikely(vma->vm_flags & VM_LOCKED))
1055 		munlock_folio(folio);
1056 }
1057 
1058 void mlock_new_folio(struct folio *folio);
1059 bool need_mlock_drain(int cpu);
1060 void mlock_drain_local(void);
1061 void mlock_drain_remote(int cpu);
1062 
1063 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
1064 
1065 /**
1066  * vma_address - Find the virtual address a page range is mapped at
1067  * @vma: The vma which maps this object.
1068  * @pgoff: The page offset within its object.
1069  * @nr_pages: The number of pages to consider.
1070  *
1071  * If any page in this range is mapped by this VMA, return the first address
1072  * where any of these pages appear.  Otherwise, return -EFAULT.
1073  */
1074 static inline unsigned long vma_address(const struct vm_area_struct *vma,
1075 		pgoff_t pgoff, unsigned long nr_pages)
1076 {
1077 	unsigned long address;
1078 
1079 	if (pgoff >= vma->vm_pgoff) {
1080 		address = vma->vm_start +
1081 			((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1082 		/* Check for address beyond vma (or wrapped through 0?) */
1083 		if (address < vma->vm_start || address >= vma->vm_end)
1084 			address = -EFAULT;
1085 	} else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
1086 		/* Test above avoids possibility of wrap to 0 on 32-bit */
1087 		address = vma->vm_start;
1088 	} else {
1089 		address = -EFAULT;
1090 	}
1091 	return address;
1092 }
1093 
1094 /*
1095  * Then at what user virtual address will none of the range be found in vma?
1096  * Assumes that vma_address() already returned a good starting address.
1097  */
1098 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
1099 {
1100 	struct vm_area_struct *vma = pvmw->vma;
1101 	pgoff_t pgoff;
1102 	unsigned long address;
1103 
1104 	/* Common case, plus ->pgoff is invalid for KSM */
1105 	if (pvmw->nr_pages == 1)
1106 		return pvmw->address + PAGE_SIZE;
1107 
1108 	pgoff = pvmw->pgoff + pvmw->nr_pages;
1109 	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1110 	/* Check for address beyond vma (or wrapped through 0?) */
1111 	if (address < vma->vm_start || address > vma->vm_end)
1112 		address = vma->vm_end;
1113 	return address;
1114 }
1115 
1116 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
1117 						    struct file *fpin)
1118 {
1119 	int flags = vmf->flags;
1120 
1121 	if (fpin)
1122 		return fpin;
1123 
1124 	/*
1125 	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
1126 	 * anything, so we only pin the file and drop the mmap_lock if only
1127 	 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
1128 	 */
1129 	if (fault_flag_allow_retry_first(flags) &&
1130 	    !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
1131 		fpin = get_file(vmf->vma->vm_file);
1132 		release_fault_lock(vmf);
1133 	}
1134 	return fpin;
1135 }
1136 #else /* !CONFIG_MMU */
1137 static inline void unmap_mapping_folio(struct folio *folio) { }
1138 static inline void mlock_new_folio(struct folio *folio) { }
1139 static inline bool need_mlock_drain(int cpu) { return false; }
1140 static inline void mlock_drain_local(void) { }
1141 static inline void mlock_drain_remote(int cpu) { }
1142 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
1143 {
1144 }
1145 #endif /* !CONFIG_MMU */
1146 
1147 /* Memory initialisation debug and verification */
1148 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1149 DECLARE_STATIC_KEY_TRUE(deferred_pages);
1150 
1151 bool __init deferred_grow_zone(struct zone *zone, unsigned int order);
1152 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1153 
1154 void init_deferred_page(unsigned long pfn, int nid);
1155 
1156 enum mminit_level {
1157 	MMINIT_WARNING,
1158 	MMINIT_VERIFY,
1159 	MMINIT_TRACE
1160 };
1161 
1162 #ifdef CONFIG_DEBUG_MEMORY_INIT
1163 
1164 extern int mminit_loglevel;
1165 
1166 #define mminit_dprintk(level, prefix, fmt, arg...) \
1167 do { \
1168 	if (level < mminit_loglevel) { \
1169 		if (level <= MMINIT_WARNING) \
1170 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
1171 		else \
1172 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
1173 	} \
1174 } while (0)
1175 
1176 extern void mminit_verify_pageflags_layout(void);
1177 extern void mminit_verify_zonelist(void);
1178 #else
1179 
1180 static inline void mminit_dprintk(enum mminit_level level,
1181 				const char *prefix, const char *fmt, ...)
1182 {
1183 }
1184 
1185 static inline void mminit_verify_pageflags_layout(void)
1186 {
1187 }
1188 
1189 static inline void mminit_verify_zonelist(void)
1190 {
1191 }
1192 #endif /* CONFIG_DEBUG_MEMORY_INIT */
1193 
1194 #define NODE_RECLAIM_NOSCAN	-2
1195 #define NODE_RECLAIM_FULL	-1
1196 #define NODE_RECLAIM_SOME	0
1197 #define NODE_RECLAIM_SUCCESS	1
1198 
1199 #ifdef CONFIG_NUMA
1200 extern int node_reclaim_mode;
1201 
1202 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
1203 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
1204 #else
1205 #define node_reclaim_mode 0
1206 
1207 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
1208 				unsigned int order)
1209 {
1210 	return NODE_RECLAIM_NOSCAN;
1211 }
1212 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
1213 {
1214 	return NUMA_NO_NODE;
1215 }
1216 #endif
1217 
1218 static inline bool node_reclaim_enabled(void)
1219 {
1220 	/* Is any node_reclaim_mode bit set? */
1221 	return node_reclaim_mode & (RECLAIM_ZONE|RECLAIM_WRITE|RECLAIM_UNMAP);
1222 }
1223 
1224 /*
1225  * mm/memory-failure.c
1226  */
1227 #ifdef CONFIG_MEMORY_FAILURE
1228 int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill);
1229 void shake_folio(struct folio *folio);
1230 extern int hwpoison_filter(struct page *p);
1231 
1232 extern u32 hwpoison_filter_dev_major;
1233 extern u32 hwpoison_filter_dev_minor;
1234 extern u64 hwpoison_filter_flags_mask;
1235 extern u64 hwpoison_filter_flags_value;
1236 extern u64 hwpoison_filter_memcg;
1237 extern u32 hwpoison_filter_enable;
1238 #define MAGIC_HWPOISON	0x48575053U	/* HWPS */
1239 void SetPageHWPoisonTakenOff(struct page *page);
1240 void ClearPageHWPoisonTakenOff(struct page *page);
1241 bool take_page_off_buddy(struct page *page);
1242 bool put_page_back_buddy(struct page *page);
1243 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
1244 void add_to_kill_ksm(struct task_struct *tsk, const struct page *p,
1245 		     struct vm_area_struct *vma, struct list_head *to_kill,
1246 		     unsigned long ksm_addr);
1247 unsigned long page_mapped_in_vma(const struct page *page,
1248 		struct vm_area_struct *vma);
1249 
1250 #else
1251 static inline int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill)
1252 {
1253 	return -EBUSY;
1254 }
1255 #endif
1256 
1257 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
1258         unsigned long, unsigned long,
1259         unsigned long, unsigned long);
1260 
1261 extern void set_pageblock_order(void);
1262 unsigned long reclaim_pages(struct list_head *folio_list);
1263 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
1264 					    struct list_head *folio_list);
1265 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1266 #define ALLOC_WMARK_MIN		WMARK_MIN
1267 #define ALLOC_WMARK_LOW		WMARK_LOW
1268 #define ALLOC_WMARK_HIGH	WMARK_HIGH
1269 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
1270 
1271 /* Mask to get the watermark bits */
1272 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
1273 
1274 /*
1275  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
1276  * cannot assume a reduced access to memory reserves is sufficient for
1277  * !MMU
1278  */
1279 #ifdef CONFIG_MMU
1280 #define ALLOC_OOM		0x08
1281 #else
1282 #define ALLOC_OOM		ALLOC_NO_WATERMARKS
1283 #endif
1284 
1285 #define ALLOC_NON_BLOCK		 0x10 /* Caller cannot block. Allow access
1286 				       * to 25% of the min watermark or
1287 				       * 62.5% if __GFP_HIGH is set.
1288 				       */
1289 #define ALLOC_MIN_RESERVE	 0x20 /* __GFP_HIGH set. Allow access to 50%
1290 				       * of the min watermark.
1291 				       */
1292 #define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
1293 #define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
1294 #ifdef CONFIG_ZONE_DMA32
1295 #define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
1296 #else
1297 #define ALLOC_NOFRAGMENT	  0x0
1298 #endif
1299 #define ALLOC_HIGHATOMIC	0x200 /* Allows access to MIGRATE_HIGHATOMIC */
1300 #define ALLOC_TRYLOCK		0x400 /* Only use spin_trylock in allocation path */
1301 #define ALLOC_KSWAPD		0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
1302 
1303 /* Flags that allow allocations below the min watermark. */
1304 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
1305 
1306 enum ttu_flags;
1307 struct tlbflush_unmap_batch;
1308 
1309 
1310 /*
1311  * only for MM internal work items which do not depend on
1312  * any allocations or locks which might depend on allocations
1313  */
1314 extern struct workqueue_struct *mm_percpu_wq;
1315 
1316 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1317 void try_to_unmap_flush(void);
1318 void try_to_unmap_flush_dirty(void);
1319 void flush_tlb_batched_pending(struct mm_struct *mm);
1320 #else
1321 static inline void try_to_unmap_flush(void)
1322 {
1323 }
1324 static inline void try_to_unmap_flush_dirty(void)
1325 {
1326 }
1327 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
1328 {
1329 }
1330 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
1331 
1332 extern const struct trace_print_flags pageflag_names[];
1333 extern const struct trace_print_flags vmaflag_names[];
1334 extern const struct trace_print_flags gfpflag_names[];
1335 
1336 static inline bool is_migrate_highatomic(enum migratetype migratetype)
1337 {
1338 	return migratetype == MIGRATE_HIGHATOMIC;
1339 }
1340 
1341 void setup_zone_pageset(struct zone *zone);
1342 
1343 struct migration_target_control {
1344 	int nid;		/* preferred node id */
1345 	nodemask_t *nmask;
1346 	gfp_t gfp_mask;
1347 	enum migrate_reason reason;
1348 };
1349 
1350 /*
1351  * mm/filemap.c
1352  */
1353 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
1354 			      struct folio *folio, loff_t fpos, size_t size);
1355 
1356 /*
1357  * mm/vmalloc.c
1358  */
1359 #ifdef CONFIG_MMU
1360 void __init vmalloc_init(void);
1361 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1362                 pgprot_t prot, struct page **pages, unsigned int page_shift);
1363 unsigned int get_vm_area_page_order(struct vm_struct *vm);
1364 #else
1365 static inline void vmalloc_init(void)
1366 {
1367 }
1368 
1369 static inline
1370 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1371                 pgprot_t prot, struct page **pages, unsigned int page_shift)
1372 {
1373 	return -EINVAL;
1374 }
1375 #endif
1376 
1377 int __must_check __vmap_pages_range_noflush(unsigned long addr,
1378 			       unsigned long end, pgprot_t prot,
1379 			       struct page **pages, unsigned int page_shift);
1380 
1381 void vunmap_range_noflush(unsigned long start, unsigned long end);
1382 
1383 void __vunmap_range_noflush(unsigned long start, unsigned long end);
1384 
1385 int numa_migrate_check(struct folio *folio, struct vm_fault *vmf,
1386 		      unsigned long addr, int *flags, bool writable,
1387 		      int *last_cpupid);
1388 
1389 void free_zone_device_folio(struct folio *folio);
1390 int migrate_device_coherent_folio(struct folio *folio);
1391 
1392 struct vm_struct *__get_vm_area_node(unsigned long size,
1393 				     unsigned long align, unsigned long shift,
1394 				     vm_flags_t vm_flags, unsigned long start,
1395 				     unsigned long end, int node, gfp_t gfp_mask,
1396 				     const void *caller);
1397 
1398 /*
1399  * mm/gup.c
1400  */
1401 int __must_check try_grab_folio(struct folio *folio, int refs,
1402 				unsigned int flags);
1403 
1404 /*
1405  * mm/huge_memory.c
1406  */
1407 void touch_pud(struct vm_area_struct *vma, unsigned long addr,
1408 	       pud_t *pud, bool write);
1409 void touch_pmd(struct vm_area_struct *vma, unsigned long addr,
1410 	       pmd_t *pmd, bool write);
1411 
1412 /*
1413  * Parses a string with mem suffixes into its order. Useful to parse kernel
1414  * parameters.
1415  */
1416 static inline int get_order_from_str(const char *size_str,
1417 				     unsigned long valid_orders)
1418 {
1419 	unsigned long size;
1420 	char *endptr;
1421 	int order;
1422 
1423 	size = memparse(size_str, &endptr);
1424 
1425 	if (!is_power_of_2(size))
1426 		return -EINVAL;
1427 	order = get_order(size);
1428 	if (BIT(order) & ~valid_orders)
1429 		return -EINVAL;
1430 
1431 	return order;
1432 }
1433 
1434 enum {
1435 	/* mark page accessed */
1436 	FOLL_TOUCH = 1 << 16,
1437 	/* a retry, previous pass started an IO */
1438 	FOLL_TRIED = 1 << 17,
1439 	/* we are working on non-current tsk/mm */
1440 	FOLL_REMOTE = 1 << 18,
1441 	/* pages must be released via unpin_user_page */
1442 	FOLL_PIN = 1 << 19,
1443 	/* gup_fast: prevent fall-back to slow gup */
1444 	FOLL_FAST_ONLY = 1 << 20,
1445 	/* allow unlocking the mmap lock */
1446 	FOLL_UNLOCKABLE = 1 << 21,
1447 	/* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */
1448 	FOLL_MADV_POPULATE = 1 << 22,
1449 };
1450 
1451 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \
1452 			    FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \
1453 			    FOLL_MADV_POPULATE)
1454 
1455 /*
1456  * Indicates for which pages that are write-protected in the page table,
1457  * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
1458  * GUP pin will remain consistent with the pages mapped into the page tables
1459  * of the MM.
1460  *
1461  * Temporary unmapping of PageAnonExclusive() pages or clearing of
1462  * PageAnonExclusive() has to protect against concurrent GUP:
1463  * * Ordinary GUP: Using the PT lock
1464  * * GUP-fast and fork(): mm->write_protect_seq
1465  * * GUP-fast and KSM or temporary unmapping (swap, migration): see
1466  *    folio_try_share_anon_rmap_*()
1467  *
1468  * Must be called with the (sub)page that's actually referenced via the
1469  * page table entry, which might not necessarily be the head page for a
1470  * PTE-mapped THP.
1471  *
1472  * If the vma is NULL, we're coming from the GUP-fast path and might have
1473  * to fallback to the slow path just to lookup the vma.
1474  */
1475 static inline bool gup_must_unshare(struct vm_area_struct *vma,
1476 				    unsigned int flags, struct page *page)
1477 {
1478 	/*
1479 	 * FOLL_WRITE is implicitly handled correctly as the page table entry
1480 	 * has to be writable -- and if it references (part of) an anonymous
1481 	 * folio, that part is required to be marked exclusive.
1482 	 */
1483 	if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
1484 		return false;
1485 	/*
1486 	 * Note: PageAnon(page) is stable until the page is actually getting
1487 	 * freed.
1488 	 */
1489 	if (!PageAnon(page)) {
1490 		/*
1491 		 * We only care about R/O long-term pining: R/O short-term
1492 		 * pinning does not have the semantics to observe successive
1493 		 * changes through the process page tables.
1494 		 */
1495 		if (!(flags & FOLL_LONGTERM))
1496 			return false;
1497 
1498 		/* We really need the vma ... */
1499 		if (!vma)
1500 			return true;
1501 
1502 		/*
1503 		 * ... because we only care about writable private ("COW")
1504 		 * mappings where we have to break COW early.
1505 		 */
1506 		return is_cow_mapping(vma->vm_flags);
1507 	}
1508 
1509 	/* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */
1510 	if (IS_ENABLED(CONFIG_HAVE_GUP_FAST))
1511 		smp_rmb();
1512 
1513 	/*
1514 	 * Note that KSM pages cannot be exclusive, and consequently,
1515 	 * cannot get pinned.
1516 	 */
1517 	return !PageAnonExclusive(page);
1518 }
1519 
1520 extern bool mirrored_kernelcore;
1521 bool memblock_has_mirror(void);
1522 void memblock_free_all(void);
1523 
1524 static __always_inline void vma_set_range(struct vm_area_struct *vma,
1525 					  unsigned long start, unsigned long end,
1526 					  pgoff_t pgoff)
1527 {
1528 	vma->vm_start = start;
1529 	vma->vm_end = end;
1530 	vma->vm_pgoff = pgoff;
1531 }
1532 
1533 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1534 {
1535 	/*
1536 	 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1537 	 * enablements, because when without soft-dirty being compiled in,
1538 	 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1539 	 * will be constantly true.
1540 	 */
1541 	if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
1542 		return false;
1543 
1544 	/*
1545 	 * Soft-dirty is kind of special: its tracking is enabled when the
1546 	 * vma flags not set.
1547 	 */
1548 	return !(vma->vm_flags & VM_SOFTDIRTY);
1549 }
1550 
1551 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd)
1552 {
1553 	return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd);
1554 }
1555 
1556 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte)
1557 {
1558 	return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte);
1559 }
1560 
1561 void __meminit __init_single_page(struct page *page, unsigned long pfn,
1562 				unsigned long zone, int nid);
1563 void __meminit __init_page_from_nid(unsigned long pfn, int nid);
1564 
1565 /* shrinker related functions */
1566 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg,
1567 			  int priority);
1568 
1569 #ifdef CONFIG_SHRINKER_DEBUG
1570 static inline __printf(2, 0) int shrinker_debugfs_name_alloc(
1571 			struct shrinker *shrinker, const char *fmt, va_list ap)
1572 {
1573 	shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap);
1574 
1575 	return shrinker->name ? 0 : -ENOMEM;
1576 }
1577 
1578 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1579 {
1580 	kfree_const(shrinker->name);
1581 	shrinker->name = NULL;
1582 }
1583 
1584 extern int shrinker_debugfs_add(struct shrinker *shrinker);
1585 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1586 					      int *debugfs_id);
1587 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1588 				    int debugfs_id);
1589 #else /* CONFIG_SHRINKER_DEBUG */
1590 static inline int shrinker_debugfs_add(struct shrinker *shrinker)
1591 {
1592 	return 0;
1593 }
1594 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker,
1595 					      const char *fmt, va_list ap)
1596 {
1597 	return 0;
1598 }
1599 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1600 {
1601 }
1602 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1603 						     int *debugfs_id)
1604 {
1605 	*debugfs_id = -1;
1606 	return NULL;
1607 }
1608 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1609 					   int debugfs_id)
1610 {
1611 }
1612 #endif /* CONFIG_SHRINKER_DEBUG */
1613 
1614 /* Only track the nodes of mappings with shadow entries */
1615 void workingset_update_node(struct xa_node *node);
1616 extern struct list_lru shadow_nodes;
1617 #define mapping_set_update(xas, mapping) do {			\
1618 	if (!dax_mapping(mapping) && !shmem_mapping(mapping)) {	\
1619 		xas_set_update(xas, workingset_update_node);	\
1620 		xas_set_lru(xas, &shadow_nodes);		\
1621 	}							\
1622 } while (0)
1623 
1624 /* mremap.c */
1625 unsigned long move_page_tables(struct pagetable_move_control *pmc);
1626 
1627 #ifdef CONFIG_UNACCEPTED_MEMORY
1628 void accept_page(struct page *page);
1629 #else /* CONFIG_UNACCEPTED_MEMORY */
1630 static inline void accept_page(struct page *page)
1631 {
1632 }
1633 #endif /* CONFIG_UNACCEPTED_MEMORY */
1634 
1635 /* pagewalk.c */
1636 int walk_page_range_mm(struct mm_struct *mm, unsigned long start,
1637 		unsigned long end, const struct mm_walk_ops *ops,
1638 		void *private);
1639 int walk_page_range_debug(struct mm_struct *mm, unsigned long start,
1640 			  unsigned long end, const struct mm_walk_ops *ops,
1641 			  pgd_t *pgd, void *private);
1642 
1643 /* pt_reclaim.c */
1644 bool try_get_and_clear_pmd(struct mm_struct *mm, pmd_t *pmd, pmd_t *pmdval);
1645 void free_pte(struct mm_struct *mm, unsigned long addr, struct mmu_gather *tlb,
1646 	      pmd_t pmdval);
1647 void try_to_free_pte(struct mm_struct *mm, pmd_t *pmd, unsigned long addr,
1648 		     struct mmu_gather *tlb);
1649 
1650 #ifdef CONFIG_PT_RECLAIM
1651 bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
1652 			   struct zap_details *details);
1653 #else
1654 static inline bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
1655 					 struct zap_details *details)
1656 {
1657 	return false;
1658 }
1659 #endif /* CONFIG_PT_RECLAIM */
1660 
1661 void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm);
1662 int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm);
1663 
1664 #endif	/* __MM_INTERNAL_H */
1665