projects/linux·blob master · refs

1657 lines · 50.2 KiB

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 ([email protected])
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/mmu_notifier.h>
15#include <linux/pagemap.h>
16#include <linux/pagewalk.h>
17#include <linux/rmap.h>
18#include <linux/swap.h>
19#include <linux/leafops.h>
20#include <linux/tracepoint-defs.h>
21
22/* Internal core VMA manipulation functions. */
23#include "vma.h"
24
25struct folio_batch;
26struct hstate;
27
28struct huge_bootmem_page {
29 struct list_head list;
30 struct hstate *hstate;
31 unsigned long flags;
32};
33
34/* mm/workingset.c */
35bool workingset_test_recent(void *shadow, bool file, bool *workingset,
36 bool flush);
37void workingset_age_nonresident(struct lruvec *lruvec, unsigned long nr_pages);
38void *workingset_eviction(struct folio *folio,
39 struct mem_cgroup *target_memcg);
40void workingset_refault(struct folio *folio, void *shadow);
41void workingset_activation(struct folio *folio);
42
43/* mm/folio.c */
44void folio_add_lru_vma(struct folio *folio, struct vm_area_struct *vma);
45
46static inline bool folio_may_be_lru_cached(const struct folio *folio)
47{
48 /*
49 * Holding PMD-sized folios in per-CPU LRU cache unbalances accounting.
50 * Holding small numbers of low-order mTHP folios in per-CPU LRU cache
51 * will be sensible, but nobody has implemented and tested that yet.
52 */
53 return !folio_test_large(folio);
54}
55
56static inline void lru_cache_enable(void)
57{
58 atomic_dec(&lru_disable_count);
59}
60
61void lru_cache_disable(void);
62void lru_add_drain(void);
63void lru_add_drain_cpu(int cpu);
64void lru_add_drain_cpu_zone(struct zone *zone);
65void folio_deactivate(struct folio *folio);
66void folio_mark_lazyfree(struct folio *folio);
67
68/* mm/vmscan.c */
69unsigned long zone_reclaimable_pages(struct zone *zone);
70unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
71 gfp_t gfp_mask, const nodemask_t *mask);
72unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru,
73 int zone_idx);
74
75#define MEMCG_RECLAIM_MAY_SWAP (1 << 1)
76#define MEMCG_RECLAIM_PROACTIVE (1 << 2)
77#define MIN_SWAPPINESS 0
78#define MAX_SWAPPINESS 200
79
80/* Just reclaim from anon folios in proactive memory reclaim */
81#define SWAPPINESS_ANON_ONLY (MAX_SWAPPINESS + 1)
82
83unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
84 unsigned long nr_pages,
85 gfp_t gfp_mask,
86 unsigned int reclaim_options,
87 int *swappiness);
88unsigned long mem_cgroup_shrink_node(struct mem_cgroup *memcg,
89 gfp_t gfp_mask, bool noswap,
90 pg_data_t *pgdat,
91 unsigned long *nr_scanned);
92
93#ifdef CONFIG_NUMA
94extern int sysctl_min_unmapped_ratio;
95extern int sysctl_min_slab_ratio;
96#endif
97
98/*
99 * Maintains state across a page table move. The operation assumes both source
100 * and destination VMAs already exist and are specified by the user.
101 *
102 * Partial moves are permitted, but the old and new ranges must both reside
103 * within a VMA.
104 *
105 * mmap lock must be held in write and VMA write locks must be held on any VMA
106 * that is visible.
107 *
108 * Use the PAGETABLE_MOVE() macro to initialise this struct.
109 *
110 * The old_addr and new_addr fields are updated as the page table move is
111 * executed.
112 *
113 * NOTE: The page table move is affected by reading from [old_addr, old_end),
114 * and old_addr may be updated for better page table alignment, so len_in
115 * represents the length of the range being copied as specified by the user.
116 */
117struct pagetable_move_control {
118 struct vm_area_struct *old; /* Source VMA. */
119 struct vm_area_struct *new; /* Destination VMA. */
120 unsigned long old_addr; /* Address from which the move begins. */
121 unsigned long old_end; /* Exclusive address at which old range ends. */
122 unsigned long new_addr; /* Address to move page tables to. */
123 unsigned long len_in; /* Bytes to remap specified by user. */
124
125 bool need_rmap_locks; /* Do rmap locks need to be taken? */
126 bool for_stack; /* Is this an early temp stack being moved? */
127};
128
129#define PAGETABLE_MOVE(name, old_, new_, old_addr_, new_addr_, len_) \
130 struct pagetable_move_control name = { \
131 .old = old_, \
132 .new = new_, \
133 .old_addr = old_addr_, \
134 .old_end = (old_addr_) + (len_), \
135 .new_addr = new_addr_, \
136 .len_in = len_, \
137 }
138
139/*
140 * The set of flags that only affect watermark checking and reclaim
141 * behaviour. This is used by the MM to obey the caller constraints
142 * about IO, FS and watermark checking while ignoring placement
143 * hints such as HIGHMEM usage.
144 */
145#define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
146 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
147 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
148 __GFP_NOLOCKDEP)
149
150/* The GFP flags allowed during early boot */
151#define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
152
153/* Control allocation cpuset and node placement constraints */
154#define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
155
156/* Do not use these with a slab allocator */
157#define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
158
159/*
160 * Different from WARN_ON_ONCE(), no warning will be issued
161 * when we specify __GFP_NOWARN.
162 */
163#define WARN_ON_ONCE_GFP(cond, gfp) ({ \
164 static bool __section(".data..once") __warned; \
165 int __ret_warn_once = !!(cond); \
166 \
167 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
168 __warned = true; \
169 WARN_ON(1); \
170 } \
171 unlikely(__ret_warn_once); \
172})
173
174void page_writeback_init(void);
175
176/*
177 * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
178 * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit
179 * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE). Hugetlb currently
180 * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
181 */
182#define ENTIRELY_MAPPED 0x800000
183#define FOLIO_PAGES_MAPPED (ENTIRELY_MAPPED - 1)
184
185/*
186 * Flags passed to __show_mem() and show_free_areas() to suppress output in
187 * various contexts.
188 */
189#define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */
190
191/*
192 * How many individual pages have an elevated _mapcount. Excludes
193 * the folio's entire_mapcount.
194 *
195 * Don't use this function outside of debugging code.
196 */
197static inline int folio_nr_pages_mapped(const struct folio *folio)
198{
199 if (IS_ENABLED(CONFIG_NO_PAGE_MAPCOUNT))
200 return -1;
201 return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
202}
203
204/*
205 * Retrieve the first entry of a folio based on a provided entry within the
206 * folio. We cannot rely on folio->swap as there is no guarantee that it has
207 * been initialized. Used for calling arch_swap_restore()
208 */
209static inline swp_entry_t folio_swap(swp_entry_t entry,
210 const struct folio *folio)
211{
212 swp_entry_t swap = {
213 .val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)),
214 };
215
216 return swap;
217}
218
219static inline void *folio_raw_mapping(const struct folio *folio)
220{
221 unsigned long mapping = (unsigned long)folio->mapping;
222
223 return (void *)(mapping & ~FOLIO_MAPPING_FLAGS);
224}
225
226/*
227 * This is a file-backed mapping, and is about to be memory mapped - invoke its
228 * mmap hook and safely handle error conditions. On error, VMA hooks will be
229 * mutated.
230 *
231 * @file: File which backs the mapping.
232 * @vma: VMA which we are mapping.
233 *
234 * Returns: 0 if success, error otherwise.
235 */
236static inline int mmap_file(struct file *file, struct vm_area_struct *vma)
237{
238 int err = vfs_mmap(file, vma);
239
240 if (likely(!err))
241 return 0;
242
243 /*
244 * OK, we tried to call the file hook for mmap(), but an error
245 * arose. The mapping is in an inconsistent state and we must not invoke
246 * any further hooks on it.
247 */
248 vma->vm_ops = &vma_dummy_vm_ops;
249
250 return err;
251}
252
253/*
254 * If the VMA has a close hook then close it, and since closing it might leave
255 * it in an inconsistent state which makes the use of any hooks suspect, clear
256 * them down by installing dummy empty hooks.
257 */
258static inline void vma_close(struct vm_area_struct *vma)
259{
260 if (vma->vm_ops && vma->vm_ops->close) {
261 vma->vm_ops->close(vma);
262
263 /*
264 * The mapping is in an inconsistent state, and no further hooks
265 * may be invoked upon it.
266 */
267 vma->vm_ops = &vma_dummy_vm_ops;
268 }
269}
270
271/* unmap_vmas is in mm/memory.c */
272void unmap_vmas(struct mmu_gather *tlb, struct unmap_desc *unmap);
273
274#ifdef CONFIG_MMU
275
276bool cond_install_uffd_wp_ptes(struct vm_area_struct *vma,
277 unsigned long addr, pte_t *ptep, pte_t pte,
278 unsigned long nr_ptes);
279
280static inline void get_anon_vma(struct anon_vma *anon_vma)
281{
282 atomic_inc(&anon_vma->refcount);
283}
284
285void __put_anon_vma(struct anon_vma *anon_vma);
286
287static inline void put_anon_vma(struct anon_vma *anon_vma)
288{
289 if (atomic_dec_and_test(&anon_vma->refcount))
290 __put_anon_vma(anon_vma);
291}
292
293static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
294{
295 down_write(&anon_vma->root->rwsem);
296}
297
298static inline int anon_vma_trylock_write(struct anon_vma *anon_vma)
299{
300 return down_write_trylock(&anon_vma->root->rwsem);
301}
302
303static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
304{
305 up_write(&anon_vma->root->rwsem);
306}
307
308static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
309{
310 down_read(&anon_vma->root->rwsem);
311}
312
313static inline int anon_vma_trylock_read(struct anon_vma *anon_vma)
314{
315 return down_read_trylock(&anon_vma->root->rwsem);
316}
317
318static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
319{
320 up_read(&anon_vma->root->rwsem);
321}
322
323struct anon_vma *folio_get_anon_vma(const struct folio *folio);
324
325/* Operations which modify VMAs. */
326enum vma_operation {
327 VMA_OP_SPLIT,
328 VMA_OP_MERGE_UNFAULTED,
329 VMA_OP_REMAP,
330 VMA_OP_FORK,
331};
332
333int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src,
334 enum vma_operation operation);
335int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma);
336int __anon_vma_prepare(struct vm_area_struct *vma);
337void unlink_anon_vmas(struct vm_area_struct *vma);
338
339static inline int anon_vma_prepare(struct vm_area_struct *vma)
340{
341 if (likely(vma->anon_vma))
342 return 0;
343
344 return __anon_vma_prepare(vma);
345}
346
347/* Flags for folio_pte_batch(). */
348typedef int __bitwise fpb_t;
349
350/* Compare PTEs respecting the dirty bit. */
351#define FPB_RESPECT_DIRTY ((__force fpb_t)BIT(0))
352
353/* Compare PTEs respecting the soft-dirty bit. */
354#define FPB_RESPECT_SOFT_DIRTY ((__force fpb_t)BIT(1))
355
356/* Compare PTEs respecting the writable bit. */
357#define FPB_RESPECT_WRITE ((__force fpb_t)BIT(2))
358
359/*
360 * Merge PTE write bits: if any PTE in the batch is writable, modify the
361 * PTE at @ptentp to be writable.
362 */
363#define FPB_MERGE_WRITE ((__force fpb_t)BIT(3))
364
365/*
366 * Merge PTE young and dirty bits: if any PTE in the batch is young or dirty,
367 * modify the PTE at @ptentp to be young or dirty, respectively.
368 */
369#define FPB_MERGE_YOUNG_DIRTY ((__force fpb_t)BIT(4))
370
371static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
372{
373 if (!(flags & FPB_RESPECT_DIRTY))
374 pte = pte_mkclean(pte);
375 if (likely(!(flags & FPB_RESPECT_SOFT_DIRTY)))
376 pte = pte_clear_soft_dirty(pte);
377 if (likely(!(flags & FPB_RESPECT_WRITE)))
378 pte = pte_wrprotect(pte);
379 return pte_mkold(pte);
380}
381
382/**
383 * folio_pte_batch_flags - detect a PTE batch for a large folio
384 * @folio: The large folio to detect a PTE batch for.
385 * @vma: The VMA. Only relevant with FPB_MERGE_WRITE, otherwise can be NULL.
386 * @ptep: Page table pointer for the first entry.
387 * @ptentp: Pointer to a COPY of the first page table entry whose flags this
388 * function updates based on @flags if appropriate.
389 * @max_nr: The maximum number of table entries to consider.
390 * @flags: Flags to modify the PTE batch semantics.
391 *
392 * Detect a PTE batch: consecutive (present) PTEs that map consecutive
393 * pages of the same large folio in a single VMA and a single page table.
394 *
395 * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN,
396 * the accessed bit, writable bit, dirty bit (unless FPB_RESPECT_DIRTY is set)
397 * and soft-dirty bit (unless FPB_RESPECT_SOFT_DIRTY is set).
398 *
399 * @ptep must map any page of the folio. max_nr must be at least one and
400 * must be limited by the caller so scanning cannot exceed a single VMA and
401 * a single page table.
402 *
403 * Depending on the FPB_MERGE_* flags, the pte stored at @ptentp will
404 * be updated: it's crucial that a pointer to a COPY of the first
405 * page table entry, obtained through ptep_get(), is provided as @ptentp.
406 *
407 * This function will be inlined to optimize based on the input parameters;
408 * consider using folio_pte_batch() instead if applicable.
409 *
410 * Return: the number of table entries in the batch.
411 */
412static inline unsigned int folio_pte_batch_flags(struct folio *folio,
413 struct vm_area_struct *vma, pte_t *ptep, pte_t *ptentp,
414 unsigned int max_nr, fpb_t flags)
415{
416 bool any_writable = false, any_young = false, any_dirty = false;
417 pte_t expected_pte, pte = *ptentp;
418 unsigned int nr, cur_nr;
419
420 VM_WARN_ON_FOLIO(!pte_present(pte), folio);
421 VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio);
422 VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio);
423 /*
424 * Ensure this is a pointer to a copy not a pointer into a page table.
425 * If this is a stack value, it won't be a valid virtual address, but
426 * that's fine because it also cannot be pointing into the page table.
427 */
428 VM_WARN_ON(virt_addr_valid(ptentp) && PageTable(virt_to_page(ptentp)));
429
430 /* Limit max_nr to the actual remaining PFNs in the folio we could batch. */
431 max_nr = min_t(unsigned long, max_nr,
432 folio_pfn(folio) + folio_nr_pages(folio) - pte_pfn(pte));
433
434 nr = pte_batch_hint(ptep, pte);
435 expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags);
436 ptep = ptep + nr;
437
438 while (nr < max_nr) {
439 pte = ptep_get(ptep);
440
441 if (!pte_same(__pte_batch_clear_ignored(pte, flags), expected_pte))
442 break;
443
444 if (flags & FPB_MERGE_WRITE)
445 any_writable |= pte_write(pte);
446 if (flags & FPB_MERGE_YOUNG_DIRTY) {
447 any_young |= pte_young(pte);
448 any_dirty |= pte_dirty(pte);
449 }
450
451 cur_nr = pte_batch_hint(ptep, pte);
452 expected_pte = pte_advance_pfn(expected_pte, cur_nr);
453 ptep += cur_nr;
454 nr += cur_nr;
455 }
456
457 if (any_writable)
458 *ptentp = pte_mkwrite(*ptentp, vma);
459 if (any_young)
460 *ptentp = pte_mkyoung(*ptentp);
461 if (any_dirty)
462 *ptentp = pte_mkdirty(*ptentp);
463
464 return min(nr, max_nr);
465}
466
467unsigned int folio_pte_batch(struct folio *folio, pte_t *ptep, pte_t pte,
468 unsigned int max_nr);
469
470/**
471 * pte_move_swp_offset - Move the swap entry offset field of a swap pte
472 * forward or backward by delta
473 * @pte: The initial pte state; must be a swap entry
474 * @delta: The direction and the offset we are moving; forward if delta
475 * is positive; backward if delta is negative
476 *
477 * Moves the swap offset, while maintaining all other fields, including
478 * swap type, and any swp pte bits. The resulting pte is returned.
479 */
480static inline pte_t pte_move_swp_offset(pte_t pte, long delta)
481{
482 const softleaf_t entry = softleaf_from_pte(pte);
483 pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry),
484 (swp_offset(entry) + delta)));
485
486 if (pte_swp_soft_dirty(pte))
487 new = pte_swp_mksoft_dirty(new);
488 if (pte_swp_exclusive(pte))
489 new = pte_swp_mkexclusive(new);
490 if (pte_swp_uffd(pte))
491 new = pte_swp_mkuffd(new);
492
493 return new;
494}
495
496
497/**
498 * pte_next_swp_offset - Increment the swap entry offset field of a swap pte.
499 * @pte: The initial pte state; must be a swap entry.
500 *
501 * Increments the swap offset, while maintaining all other fields, including
502 * swap type, and any swp pte bits. The resulting pte is returned.
503 */
504static inline pte_t pte_next_swp_offset(pte_t pte)
505{
506 return pte_move_swp_offset(pte, 1);
507}
508
509/**
510 * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries
511 * @start_ptep: Page table pointer for the first entry.
512 * @max_nr: The maximum number of table entries to consider.
513 * @pte: Page table entry for the first entry.
514 *
515 * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs
516 * containing swap entries all with consecutive offsets and targeting the same
517 * swap type, all with matching swp pte bits.
518 *
519 * max_nr must be at least one and must be limited by the caller so scanning
520 * cannot exceed a single page table.
521 *
522 * Return: the number of table entries in the batch.
523 */
524static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte)
525{
526 pte_t expected_pte = pte_next_swp_offset(pte);
527 const pte_t *end_ptep = start_ptep + max_nr;
528 pte_t *ptep = start_ptep + 1;
529
530 VM_WARN_ON(max_nr < 1);
531 VM_WARN_ON(!softleaf_is_swap(softleaf_from_pte(pte)));
532
533 while (ptep < end_ptep) {
534 pte = ptep_get(ptep);
535
536 if (!pte_same(pte, expected_pte))
537 break;
538 expected_pte = pte_next_swp_offset(expected_pte);
539 ptep++;
540 }
541
542 return ptep - start_ptep;
543}
544#endif /* CONFIG_MMU */
545
546void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
547 int nr_throttled);
548static inline void acct_reclaim_writeback(struct folio *folio)
549{
550 pg_data_t *pgdat = folio_pgdat(folio);
551 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
552
553 if (nr_throttled)
554 __acct_reclaim_writeback(pgdat, folio, nr_throttled);
555}
556
557static inline void wake_throttle_isolated(pg_data_t *pgdat)
558{
559 wait_queue_head_t *wqh;
560
561 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
562 if (waitqueue_active(wqh))
563 wake_up(wqh);
564}
565
566vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf);
567static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf)
568{
569 vm_fault_t ret = __vmf_anon_prepare(vmf);
570
571 if (unlikely(ret & VM_FAULT_RETRY))
572 vma_end_read(vmf->vma);
573 return ret;
574}
575
576vm_fault_t do_swap_page(struct vm_fault *vmf);
577void folio_rotate_reclaimable(struct folio *folio);
578bool __folio_end_writeback(struct folio *folio);
579void deactivate_file_folio(struct folio *folio);
580void folio_activate(struct folio *folio);
581
582void free_pgtables(struct mmu_gather *tlb, struct unmap_desc *desc);
583
584void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
585
586/**
587 * sync_with_folio_pmd_zap - sync with concurrent zapping of a folio PMD
588 * @mm: The mm_struct.
589 * @pmdp: Pointer to the pmd that was found to be pmd_none().
590 *
591 * When we find a pmd_none() while unmapping a folio without holding the PTL,
592 * zap_huge_pmd() may have cleared the PMD but not yet modified the folio to
593 * indicate that it's unmapped. Skipping the PMD without synchronization could
594 * make folio unmapping code assume that unmapping failed.
595 *
596 * Wait for concurrent zapping to complete by grabbing the PTL.
597 */
598static inline void sync_with_folio_pmd_zap(struct mm_struct *mm, pmd_t *pmdp)
599{
600 spinlock_t *ptl = pmd_lock(mm, pmdp);
601
602 spin_unlock(ptl);
603}
604
605struct zap_details;
606void zap_vma_range_batched(struct mmu_gather *tlb,
607 struct vm_area_struct *vma, unsigned long addr,
608 unsigned long size, struct zap_details *details);
609int zap_vma_for_reaping(struct vm_area_struct *vma);
610int folio_unmap_invalidate(struct address_space *mapping, struct folio *folio,
611 gfp_t gfp);
612
613void page_cache_ra_order(struct readahead_control *, struct file_ra_state *);
614void force_page_cache_ra(struct readahead_control *, unsigned long nr);
615static inline void force_page_cache_readahead(struct address_space *mapping,
616 struct file *file, pgoff_t index, unsigned long nr_to_read)
617{
618 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
619 force_page_cache_ra(&ractl, nr_to_read);
620}
621
622unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
623 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
624unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
625 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
626int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
627bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
628 loff_t end);
629long mapping_evict_folio(struct address_space *mapping, struct folio *folio);
630unsigned long mapping_try_invalidate(struct address_space *mapping,
631 pgoff_t start, pgoff_t end, unsigned long *nr_failed);
632
633/**
634 * folio_evictable - Test whether a folio is evictable.
635 * @folio: The folio to test.
636 *
637 * Test whether @folio is evictable -- i.e., should be placed on
638 * active/inactive lists vs unevictable list.
639 *
640 * Reasons folio might not be evictable:
641 * 1. folio's mapping marked unevictable
642 * 2. One of the pages in the folio is part of an mlocked VMA
643 */
644static inline bool folio_evictable(struct folio *folio)
645{
646 bool ret;
647
648 /* Prevent address_space of inode and swap cache from being freed */
649 rcu_read_lock();
650 ret = !mapping_unevictable(folio_mapping(folio)) &&
651 !folio_test_mlocked(folio);
652 rcu_read_unlock();
653 return ret;
654}
655
656/*
657 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
658 * a count of one.
659 */
660static inline void set_page_refcounted(struct page *page)
661{
662 VM_BUG_ON_PAGE(PageTail(page), page);
663 VM_BUG_ON_PAGE(page_ref_count(page), page);
664 set_page_count(page, 1);
665}
666
667static inline void set_pages_refcounted(struct page *page, unsigned long nr_pages)
668{
669 unsigned long pfn = page_to_pfn(page);
670
671 for (; nr_pages--; pfn++)
672 set_page_refcounted(pfn_to_page(pfn));
673}
674
675/*
676 * Return true if a folio needs ->release_folio() calling upon it.
677 */
678static inline bool folio_needs_release(struct folio *folio)
679{
680 struct address_space *mapping = folio_mapping(folio);
681
682 return folio_has_private(folio) ||
683 (mapping && mapping_release_always(mapping));
684}
685
686extern unsigned long highest_memmap_pfn;
687
688/*
689 * Maximum number of reclaim retries without progress before the OOM
690 * killer is consider the only way forward.
691 */
692#define MAX_RECLAIM_RETRIES 16
693
694/*
695 * in mm/vmscan.c:
696 */
697bool folio_isolate_lru(struct folio *folio);
698void folio_putback_lru(struct folio *folio);
699extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
700int user_proactive_reclaim(char *buf,
701 struct mem_cgroup *memcg, pg_data_t *pgdat);
702
703/*
704 * in mm/rmap.c:
705 */
706pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
707
708/*
709 * in mm/khugepaged.c
710 */
711void set_recommended_min_free_kbytes(void);
712
713/*
714 * in mm/page_alloc.c
715 */
716#define K(x) ((x) << (PAGE_SHIFT-10))
717
718extern char * const zone_names[MAX_NR_ZONES];
719
720extern int min_free_kbytes;
721extern int defrag_mode;
722
723void setup_per_zone_wmarks(void);
724void calculate_min_free_kbytes(void);
725int __meminit init_per_zone_wmark_min(void);
726
727extern int __isolate_free_page(struct page *page, unsigned int order);
728extern void __putback_isolated_page(struct page *page, unsigned int order,
729 int mt);
730
731/*
732 * This will have no effect, other than possibly generating a warning, if the
733 * caller passes in a non-large folio.
734 */
735static inline void folio_set_order(struct folio *folio, unsigned int order)
736{
737 if (WARN_ON_ONCE(!order || !folio_test_large(folio)))
738 return;
739 VM_WARN_ON_ONCE(order > MAX_FOLIO_ORDER);
740
741 folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order;
742#ifdef NR_PAGES_IN_LARGE_FOLIO
743 folio->_nr_pages = 1U << order;
744#endif
745}
746
747bool __folio_unqueue_deferred_split(struct folio *folio);
748static inline bool folio_unqueue_deferred_split(struct folio *folio)
749{
750 if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio))
751 return false;
752
753 /*
754 * At this point, there is no one trying to add the folio to
755 * deferred_list. If folio is not in deferred_list, it's safe
756 * to check without acquiring the list_lru lock.
757 */
758 if (data_race(list_empty(&folio->_deferred_list)))
759 return false;
760
761 return __folio_unqueue_deferred_split(folio);
762}
763
764static inline struct folio *page_rmappable_folio(struct page *page)
765{
766 struct folio *folio = (struct folio *)page;
767
768 if (folio && folio_test_large(folio))
769 folio_set_large_rmappable(folio);
770 return folio;
771}
772
773static inline void prep_compound_head(struct page *page, unsigned int order)
774{
775 struct folio *folio = (struct folio *)page;
776
777 folio_set_order(folio, order);
778 atomic_set(&folio->_large_mapcount, -1);
779 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
780 atomic_set(&folio->_nr_pages_mapped, 0);
781 if (IS_ENABLED(CONFIG_MM_ID)) {
782 folio->_mm_ids = 0;
783 folio->_mm_id_mapcount[0] = -1;
784 folio->_mm_id_mapcount[1] = -1;
785 }
786 if (IS_ENABLED(CONFIG_64BIT) || order > 1) {
787 atomic_set(&folio->_pincount, 0);
788 atomic_set(&folio->_entire_mapcount, -1);
789 }
790 if (order > 1)
791 INIT_LIST_HEAD(&folio->_deferred_list);
792}
793
794static inline void prep_compound_tail(struct page *tail,
795 const struct page *head, unsigned int order)
796{
797 tail->mapping = TAIL_MAPPING;
798 set_compound_head(tail, head, order);
799 VM_WARN_ON_ONCE(tail->private);
800}
801
802static inline void init_compound_tail(struct page *tail,
803 const struct page *head, unsigned int order, struct zone *zone)
804{
805 atomic_set(&tail->_mapcount, -1);
806 set_page_node(tail, zone_to_nid(zone));
807 set_page_zone(tail, zone_idx(zone));
808 prep_compound_tail(tail, head, order);
809}
810
811#if defined CONFIG_COMPACTION || defined CONFIG_CMA
812
813/*
814 * in mm/compaction.c
815 */
816/*
817 * compact_control is used to track pages being migrated and the free pages
818 * they are being migrated to during memory compaction. The free_pfn starts
819 * at the end of a zone and migrate_pfn begins at the start. Movable pages
820 * are moved to the end of a zone during a compaction run and the run
821 * completes when free_pfn <= migrate_pfn
822 */
823struct compact_control {
824 struct list_head freepages[NR_PAGE_ORDERS]; /* List of free pages to migrate to */
825 struct list_head migratepages; /* List of pages being migrated */
826 unsigned int nr_freepages; /* Number of isolated free pages */
827 unsigned int nr_migratepages; /* Number of pages to migrate */
828 unsigned long free_pfn; /* isolate_freepages search base */
829 /*
830 * Acts as an in/out parameter to page isolation for migration.
831 * isolate_migratepages uses it as a search base.
832 * isolate_migratepages_block will update the value to the next pfn
833 * after the last isolated one.
834 */
835 unsigned long migrate_pfn;
836 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
837 struct zone *zone;
838 unsigned long total_migrate_scanned;
839 unsigned long total_free_scanned;
840 unsigned short fast_search_fail;/* failures to use free list searches */
841 short search_order; /* order to start a fast search at */
842 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
843 int order; /* order a direct compactor needs */
844 int migratetype; /* migratetype of direct compactor */
845 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
846 const int highest_zoneidx; /* zone index of a direct compactor */
847 enum migrate_mode mode; /* Async or sync migration mode */
848 bool ignore_skip_hint; /* Scan blocks even if marked skip */
849 bool no_set_skip_hint; /* Don't mark blocks for skipping */
850 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
851 bool direct_compaction; /* False from kcompactd or /proc/... */
852 bool proactive_compaction; /* kcompactd proactive compaction */
853 bool whole_zone; /* Whole zone should/has been scanned */
854 bool contended; /* Signal lock contention */
855 bool finish_pageblock; /* Scan the remainder of a pageblock. Used
856 * when there are potentially transient
857 * isolation or migration failures to
858 * ensure forward progress.
859 */
860 bool alloc_contig; /* alloc_contig_range allocation */
861};
862
863/*
864 * Used in direct compaction when a page should be taken from the freelists
865 * immediately when one is created during the free path.
866 */
867struct capture_control {
868 struct zone *zone;
869 int migratetype;
870 /*
871 * Allocation request order. May differ from the compaction
872 * order: defrag_mode promotes sub-block allocations to
873 * pageblock-order compaction; capture still matches at the
874 * original allocation order so prep_new_page() is consistent.
875 */
876 int order;
877 struct page *page;
878};
879
880unsigned long
881isolate_freepages_range(struct compact_control *cc,
882 unsigned long start_pfn, unsigned long end_pfn);
883int
884isolate_migratepages_range(struct compact_control *cc,
885 unsigned long low_pfn, unsigned long end_pfn);
886
887#endif /* CONFIG_COMPACTION || CONFIG_CMA */
888
889struct cma;
890
891#ifdef CONFIG_CMA
892bool cma_validate_zones(struct cma *cma);
893void *cma_reserve_early(struct cma *cma, unsigned long size);
894#else
895static inline bool cma_validate_zones(struct cma *cma)
896{
897 return false;
898}
899static inline void *cma_reserve_early(struct cma *cma, unsigned long size)
900{
901 return NULL;
902}
903#endif
904
905/* mm/util.c */
906struct anon_vma *folio_anon_vma(const struct folio *folio);
907
908#ifdef CONFIG_MMU
909void unmap_mapping_folio(struct folio *folio);
910extern long populate_vma_page_range(struct vm_area_struct *vma,
911 unsigned long start, unsigned long end, int *locked);
912extern long faultin_page_range(struct mm_struct *mm, unsigned long start,
913 unsigned long end, bool write, int *locked);
914bool mlock_future_ok(const struct mm_struct *mm, bool is_vma_locked,
915 unsigned long bytes);
916
917/*
918 * NOTE: This function can't tell whether the folio is "fully mapped" in the
919 * range.
920 * "fully mapped" means all the pages of folio is associated with the page
921 * table of range while this function just check whether the folio range is
922 * within the range [start, end). Function caller needs to do page table
923 * check if it cares about the page table association.
924 *
925 * Typical usage (like mlock or madvise) is:
926 * Caller knows at least 1 page of folio is associated with page table of VMA
927 * and the range [start, end) is intersect with the VMA range. Caller wants
928 * to know whether the folio is fully associated with the range. It calls
929 * this function to check whether the folio is in the range first. Then checks
930 * the page table to know whether the folio is fully mapped to the range.
931 */
932static inline bool
933folio_within_range(struct folio *folio, struct vm_area_struct *vma,
934 unsigned long start, unsigned long end)
935{
936 const unsigned long vma_pglen = vma_pages(vma);
937 pgoff_t pgoff_folio, pgoff_vma_start;
938 unsigned long addr;
939
940 VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio);
941 if (start > end)
942 return false;
943
944 pgoff_folio = folio_pgoff(folio);
945 pgoff_vma_start = folio_test_anon(folio) ?
946 vma_start_anon_pgoff(vma) : vma_start_pgoff(vma);
947
948 if (start < vma->vm_start)
949 start = vma->vm_start;
950
951 if (end > vma->vm_end)
952 end = vma->vm_end;
953
954 /* if folio start address is not in vma range */
955 if (!in_range(pgoff_folio, pgoff_vma_start, vma_pglen))
956 return false;
957
958 addr = vma->vm_start + ((pgoff_folio - pgoff_vma_start) << PAGE_SHIFT);
959
960 return !(addr < start || end - addr < folio_size(folio));
961}
962
963static inline bool
964folio_within_vma(struct folio *folio, struct vm_area_struct *vma)
965{
966 return folio_within_range(folio, vma, vma->vm_start, vma->vm_end);
967}
968
969/*
970 * mlock_vma_folio() and munlock_vma_folio():
971 * should be called with vma's mmap_lock held for read or write,
972 * under page table lock for the pte/pmd being added or removed.
973 *
974 * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at
975 * the end of folio_remove_rmap_*(); but new anon folios are managed by
976 * folio_add_lru_vma() calling mlock_new_folio().
977 */
978void mlock_folio(struct folio *folio);
979static inline void mlock_vma_folio(struct folio *folio,
980 struct vm_area_struct *vma)
981{
982 /*
983 * The VM_SPECIAL check here serves two purposes.
984 * 1) VM_IO check prevents migration from double-counting during mlock.
985 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
986 * is never left set on a VM_SPECIAL vma, there is an interval while
987 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
988 * still be set while VM_SPECIAL bits are added: so ignore it then.
989 */
990 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED))
991 mlock_folio(folio);
992}
993
994void munlock_folio(struct folio *folio);
995static inline void munlock_vma_folio(struct folio *folio,
996 struct vm_area_struct *vma)
997{
998 /*
999 * munlock if the function is called. Ideally, we should only
1000 * do munlock if any page of folio is unmapped from VMA and
1001 * cause folio not fully mapped to VMA.
1002 *
1003 * But it's not easy to confirm that's the situation. So we
1004 * always munlock the folio and page reclaim will correct it
1005 * if it's wrong.
1006 */
1007 if (unlikely(vma->vm_flags & VM_LOCKED))
1008 munlock_folio(folio);
1009}
1010
1011void mlock_new_folio(struct folio *folio);
1012bool need_mlock_drain(int cpu);
1013void mlock_drain_local(void);
1014void mlock_drain_remote(int cpu);
1015
1016extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
1017
1018static inline unsigned long __vma_address(const struct vm_area_struct *vma,
1019 pgoff_t pgoff, pgoff_t pgoff_start, unsigned long nr_pages)
1020{
1021 unsigned long address;
1022
1023 if (pgoff >= pgoff_start) {
1024 address = vma->vm_start +
1025 ((pgoff - pgoff_start) << PAGE_SHIFT);
1026 /* Check for address beyond vma (or wrapped through 0?) */
1027 if (address < vma->vm_start || address >= vma->vm_end)
1028 address = -EFAULT;
1029 } else if (pgoff + nr_pages - 1 >= pgoff_start) {
1030 /* Test above avoids possibility of wrap to 0 on 32-bit */
1031 address = vma->vm_start;
1032 } else {
1033 address = -EFAULT;
1034 }
1035 return address;
1036}
1037
1038/**
1039 * vma_filebacked_address - Find the virtual address a file-backed page range is
1040 * mapped at.
1041 * @vma: The vma which maps this object.
1042 * @pgoff: The page offset within its object.
1043 * @nr_pages: The number of pages to consider.
1044 *
1045 * Returns: If any page in this range is mapped by this VMA, return the first
1046 * address where any of these pages appear. Otherwise, return -EFAULT.
1047 */
1048static inline unsigned long vma_filebacked_address(const struct vm_area_struct *vma,
1049 pgoff_t pgoff, unsigned long nr_pages)
1050{
1051 VM_WARN_ON_ONCE(vma_is_anonymous(vma));
1052
1053 return __vma_address(vma, pgoff, vma_start_pgoff(vma), nr_pages);
1054}
1055
1056/**
1057 * vma_anon_address - Find the virtual address an anonymous page range is mapped
1058 * at.
1059 * @vma: The vma which maps this object.
1060 * @pgoff_anon: The anonymous page index belonging to the folio.
1061 * @nr_pages: The number of pages to consider.
1062 *
1063 * This is only valid for anonymous or MAP_PRIVATE-mapped file-backed VMAs.
1064 *
1065 * Returns: If any page in this range is mapped by this VMA, return the first
1066 * address where any of these pages appear. Otherwise, return -EFAULT.
1067 */
1068static inline unsigned long vma_anon_address(const struct vm_area_struct *vma,
1069 pgoff_t pgoff_anon, unsigned long nr_pages)
1070{
1071 VM_WARN_ON_ONCE(!vma_is_cow_mapping(vma));
1072
1073 return __vma_address(vma, pgoff_anon, vma_start_anon_pgoff(vma), nr_pages);
1074}
1075
1076/*
1077 * At what user virtual address will none of the range be found in vma?
1078 * Assumes that vma_address() already returned a good starting address.
1079 */
1080static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
1081{
1082 const pgoff_t pgoff_end = pvmw->pgoff + pvmw->nr_pages;
1083 const struct vm_area_struct *vma = pvmw->vma;
1084 pgoff_t pgoff_vma_start;
1085 unsigned long address;
1086
1087 /* Common case, plus ->pgoff is invalid for KSM */
1088 if (pvmw->nr_pages == 1)
1089 return pvmw->address + PAGE_SIZE;
1090
1091 if (pvmw->pgoff_is_anon)
1092 pgoff_vma_start = vma_start_anon_pgoff(vma);
1093 else
1094 pgoff_vma_start = vma_start_pgoff(vma);
1095
1096 address = vma->vm_start +
1097 ((pgoff_end - pgoff_vma_start) << PAGE_SHIFT);
1098 /* Check for address beyond vma (or wrapped through 0?) */
1099 if (address < vma->vm_start || address > vma->vm_end)
1100 address = vma->vm_end;
1101 return address;
1102}
1103
1104static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
1105 struct file *fpin)
1106{
1107 int flags = vmf->flags;
1108
1109 if (fpin)
1110 return fpin;
1111
1112 /*
1113 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
1114 * anything, so we only pin the file and drop the mmap_lock if only
1115 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
1116 */
1117 if (fault_flag_allow_retry_first(flags) &&
1118 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
1119 fpin = get_file(vmf->vma->vm_file);
1120 release_fault_lock(vmf);
1121 }
1122 return fpin;
1123}
1124
1125static inline bool vma_supports_mlock(const struct vm_area_struct *vma)
1126{
1127 if (vma_test_any_mask(vma, VMA_SPECIAL_FLAGS))
1128 return false;
1129 if (vma_test_single_mask(vma, VMA_DROPPABLE))
1130 return false;
1131 if (vma_is_dax(vma) || is_vm_hugetlb_page(vma))
1132 return false;
1133 return vma != get_gate_vma(current->mm);
1134}
1135
1136#else /* !CONFIG_MMU */
1137static inline void unmap_mapping_folio(struct folio *folio) { }
1138static inline void mlock_new_folio(struct folio *folio) { }
1139static inline bool need_mlock_drain(int cpu) { return false; }
1140static inline void mlock_drain_local(void) { }
1141static inline void mlock_drain_remote(int cpu) { }
1142#endif /* !CONFIG_MMU */
1143
1144#ifdef CONFIG_NUMA
1145extern int node_reclaim_mode;
1146
1147extern unsigned long node_reclaim(struct pglist_data *pgdat,
1148 gfp_t gfp_mask, unsigned int order);
1149extern int find_next_best_node(int node, nodemask_t *used_node_mask);
1150#else
1151#define node_reclaim_mode 0
1152
1153static inline unsigned long node_reclaim(struct pglist_data *pgdat,
1154 gfp_t mask, unsigned int order)
1155{
1156 return 0;
1157}
1158static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
1159{
1160 return NUMA_NO_NODE;
1161}
1162#endif
1163
1164static inline bool node_reclaim_enabled(void)
1165{
1166 /* Is any node_reclaim_mode bit set? */
1167 return node_reclaim_mode & (RECLAIM_ZONE|RECLAIM_WRITE|RECLAIM_UNMAP);
1168}
1169
1170/*
1171 * mm/memory-failure.c
1172 */
1173#ifdef CONFIG_MEMORY_FAILURE
1174int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill);
1175void shake_folio(struct folio *folio);
1176typedef int hwpoison_filter_func_t(struct page *p);
1177void hwpoison_filter_register(hwpoison_filter_func_t *filter);
1178void hwpoison_filter_unregister(void);
1179
1180#define MAGIC_HWPOISON 0x48575053U /* HWPS */
1181void SetPageHWPoisonTakenOff(struct page *page);
1182void ClearPageHWPoisonTakenOff(struct page *page);
1183bool take_page_off_buddy(struct page *page);
1184bool put_page_back_buddy(struct page *page);
1185struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
1186void add_to_kill_ksm(struct task_struct *tsk, const struct page *p,
1187 struct vm_area_struct *vma, struct list_head *to_kill,
1188 unsigned long ksm_addr);
1189unsigned long page_mapped_in_vma(const struct page *page,
1190 struct vm_area_struct *vma);
1191
1192#else
1193static inline int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill)
1194{
1195 return -EBUSY;
1196}
1197#endif
1198
1199extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
1200 unsigned long, unsigned long,
1201 unsigned long, unsigned long);
1202
1203unsigned long reclaim_pages(struct list_head *folio_list);
1204unsigned int reclaim_clean_pages_from_list(struct zone *zone,
1205 struct list_head *folio_list);
1206
1207enum ttu_flags;
1208struct tlbflush_unmap_batch;
1209
1210
1211/*
1212 * only for MM internal work items which do not depend on
1213 * any allocations or locks which might depend on allocations
1214 */
1215extern struct workqueue_struct *mm_percpu_wq;
1216
1217#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1218void try_to_unmap_flush(void);
1219void try_to_unmap_flush_dirty(void);
1220void flush_tlb_batched_pending(struct mm_struct *mm);
1221#else
1222static inline void try_to_unmap_flush(void)
1223{
1224}
1225static inline void try_to_unmap_flush_dirty(void)
1226{
1227}
1228static inline void flush_tlb_batched_pending(struct mm_struct *mm)
1229{
1230}
1231#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
1232
1233extern const struct trace_print_flags pageflag_names[];
1234extern const struct trace_print_flags vmaflag_names[];
1235extern const struct trace_print_flags gfpflag_names[];
1236
1237void setup_zone_pageset(struct zone *zone);
1238
1239struct migration_target_control {
1240 int nid; /* preferred node id */
1241 nodemask_t *nmask;
1242 gfp_t gfp_mask;
1243 enum migrate_reason reason;
1244};
1245
1246/*
1247 * mm/filemap.c
1248 */
1249size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
1250 struct folio *folio, loff_t fpos, size_t size);
1251
1252static inline bool vma_is_single_threaded_private(struct vm_area_struct *vma)
1253{
1254 if (vma->vm_flags & VM_SHARED)
1255 return false;
1256
1257 return atomic_read(&vma->vm_mm->mm_users) == 1;
1258}
1259
1260#ifdef CONFIG_NUMA_BALANCING
1261bool folio_can_map_prot_numa(struct folio *folio, struct vm_area_struct *vma,
1262 bool is_private_single_threaded);
1263
1264#else
1265static inline bool folio_can_map_prot_numa(struct folio *folio,
1266 struct vm_area_struct *vma, bool is_private_single_threaded)
1267{
1268 return false;
1269}
1270#endif
1271
1272int numa_migrate_check(struct folio *folio, struct vm_fault *vmf,
1273 unsigned long addr, int *flags, bool writable,
1274 int *last_cpupid);
1275
1276void free_zone_device_folio(struct folio *folio);
1277int migrate_device_coherent_folio(struct folio *folio);
1278
1279/*
1280 * mm/gup.c
1281 */
1282int __must_check try_grab_folio(struct folio *folio, int refs,
1283 unsigned int flags);
1284
1285/*
1286 * mm/huge_memory.c
1287 */
1288void touch_pud(struct vm_area_struct *vma, unsigned long addr,
1289 pud_t *pud, bool write);
1290bool touch_pmd(struct vm_area_struct *vma, unsigned long addr,
1291 pmd_t *pmd, bool write);
1292
1293/*
1294 * Parses a string with mem suffixes into its order. Useful to parse kernel
1295 * parameters.
1296 */
1297static inline int get_order_from_str(const char *size_str,
1298 unsigned long valid_orders)
1299{
1300 unsigned long size;
1301 char *endptr;
1302 int order;
1303
1304 size = memparse(size_str, &endptr);
1305
1306 if (!is_power_of_2(size))
1307 return -EINVAL;
1308 order = get_order(size);
1309 if (BIT(order) & ~valid_orders)
1310 return -EINVAL;
1311
1312 return order;
1313}
1314
1315enum {
1316 /* mark page accessed */
1317 FOLL_TOUCH = 1 << 16,
1318 /* a retry, previous pass started an IO */
1319 FOLL_TRIED = 1 << 17,
1320 /* we are working on non-current tsk/mm */
1321 FOLL_REMOTE = 1 << 18,
1322 /* pages must be released via unpin_user_page */
1323 FOLL_PIN = 1 << 19,
1324 /* gup_fast: prevent fall-back to slow gup */
1325 FOLL_FAST_ONLY = 1 << 20,
1326 /* allow unlocking the mmap lock */
1327 FOLL_UNLOCKABLE = 1 << 21,
1328 /* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */
1329 FOLL_MADV_POPULATE = 1 << 22,
1330};
1331
1332#define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \
1333 FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \
1334 FOLL_MADV_POPULATE)
1335
1336/*
1337 * Indicates for which pages that are write-protected in the page table,
1338 * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
1339 * GUP pin will remain consistent with the pages mapped into the page tables
1340 * of the MM.
1341 *
1342 * Temporary unmapping of PageAnonExclusive() pages or clearing of
1343 * PageAnonExclusive() has to protect against concurrent GUP:
1344 * * Ordinary GUP: Using the PT lock
1345 * * GUP-fast and fork(): mm->write_protect_seq
1346 * * GUP-fast and KSM or temporary unmapping (swap, migration): see
1347 * folio_try_share_anon_rmap_*()
1348 *
1349 * Must be called with the (sub)page that's actually referenced via the
1350 * page table entry, which might not necessarily be the head page for a
1351 * PTE-mapped THP.
1352 *
1353 * If the vma is NULL, we're coming from the GUP-fast path and might have
1354 * to fallback to the slow path just to lookup the vma.
1355 */
1356static inline bool gup_must_unshare(struct vm_area_struct *vma,
1357 unsigned int flags, struct page *page)
1358{
1359 /*
1360 * FOLL_WRITE is implicitly handled correctly as the page table entry
1361 * has to be writable -- and if it references (part of) an anonymous
1362 * folio, that part is required to be marked exclusive.
1363 */
1364 if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
1365 return false;
1366 /*
1367 * Note: PageAnon(page) is stable until the page is actually getting
1368 * freed.
1369 */
1370 if (!PageAnon(page)) {
1371 /*
1372 * We only care about R/O long-term pining: R/O short-term
1373 * pinning does not have the semantics to observe successive
1374 * changes through the process page tables.
1375 */
1376 if (!(flags & FOLL_LONGTERM))
1377 return false;
1378
1379 /* We really need the vma ... */
1380 if (!vma)
1381 return true;
1382
1383 /*
1384 * ... because we only care about writable private ("COW")
1385 * mappings where we have to break COW early.
1386 */
1387 return vma_is_cow_mapping(vma);
1388 }
1389
1390 /* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */
1391 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST))
1392 smp_rmb();
1393
1394 /*
1395 * Note that KSM pages cannot be exclusive, and consequently,
1396 * cannot get pinned.
1397 */
1398 return !PageAnonExclusive(page);
1399}
1400
1401
1402static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1403{
1404 /*
1405 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1406 * enablements, because when without soft-dirty being compiled in,
1407 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1408 * will be constantly true.
1409 */
1410 if (!pgtable_supports_soft_dirty())
1411 return false;
1412
1413 /*
1414 * Soft-dirty is kind of special: its tracking is enabled when the
1415 * vma flags not set.
1416 */
1417 return !(vma->vm_flags & VM_SOFTDIRTY);
1418}
1419
1420static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd)
1421{
1422 return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd);
1423}
1424
1425static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte)
1426{
1427 return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte);
1428}
1429
1430/* shrinker related functions */
1431unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg,
1432 int priority);
1433
1434int shmem_add_to_page_cache(struct folio *folio,
1435 struct address_space *mapping,
1436 pgoff_t index, void *expected, gfp_t gfp);
1437int shmem_inode_acct_blocks(struct inode *inode, long pages);
1438bool shmem_recalc_inode(struct inode *inode, long alloced, long swapped);
1439
1440#ifdef CONFIG_SHRINKER_DEBUG
1441static inline __printf(2, 0) int shrinker_debugfs_name_alloc(
1442 struct shrinker *shrinker, const char *fmt, va_list ap)
1443{
1444 shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap);
1445
1446 return shrinker->name ? 0 : -ENOMEM;
1447}
1448
1449static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1450{
1451 kfree_const(shrinker->name);
1452 shrinker->name = NULL;
1453}
1454
1455extern int shrinker_debugfs_add(struct shrinker *shrinker);
1456extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1457 int *debugfs_id);
1458extern void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1459 int debugfs_id);
1460#else /* CONFIG_SHRINKER_DEBUG */
1461static inline int shrinker_debugfs_add(struct shrinker *shrinker)
1462{
1463 return 0;
1464}
1465static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker,
1466 const char *fmt, va_list ap)
1467{
1468 return 0;
1469}
1470static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1471{
1472}
1473static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1474 int *debugfs_id)
1475{
1476 *debugfs_id = -1;
1477 return NULL;
1478}
1479static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1480 int debugfs_id)
1481{
1482}
1483#endif /* CONFIG_SHRINKER_DEBUG */
1484
1485/* Only track the nodes of mappings with shadow entries */
1486void workingset_update_node(struct xa_node *node);
1487extern struct list_lru shadow_nodes;
1488#define mapping_set_update(xas, mapping) do { \
1489 if (!dax_mapping(mapping) && !shmem_mapping(mapping)) { \
1490 xas_set_update(xas, workingset_update_node); \
1491 xas_set_lru(xas, &shadow_nodes); \
1492 } \
1493} while (0)
1494
1495/* mremap.c */
1496unsigned long move_page_tables(struct pagetable_move_control *pmc);
1497
1498#ifdef CONFIG_UNACCEPTED_MEMORY
1499void accept_page(struct page *page);
1500#else /* CONFIG_UNACCEPTED_MEMORY */
1501static inline void accept_page(struct page *page)
1502{
1503}
1504#endif /* CONFIG_UNACCEPTED_MEMORY */
1505
1506/* pagewalk.c */
1507int walk_page_range_mm_unsafe(struct mm_struct *mm, unsigned long start,
1508 unsigned long end, const struct mm_walk_ops *ops,
1509 void *private);
1510int walk_page_range_vma_unsafe(struct vm_area_struct *vma, unsigned long start,
1511 unsigned long end, const struct mm_walk_ops *ops,
1512 void *private);
1513int walk_page_range_debug(struct mm_struct *mm, unsigned long start,
1514 unsigned long end, const struct mm_walk_ops *ops,
1515 pgd_t *pgd, void *private);
1516
1517void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm);
1518int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm);
1519
1520int remap_pfn_range_prepare(struct vm_area_desc *desc);
1521int remap_pfn_range_complete(struct vm_area_struct *vma,
1522 struct mmap_action *action);
1523int simple_ioremap_prepare(struct vm_area_desc *desc);
1524
1525static inline int io_remap_pfn_range_prepare(struct vm_area_desc *desc)
1526{
1527 struct mmap_action *action = &desc->action;
1528 const unsigned long orig_pfn = action->remap.start_pfn;
1529 const pgprot_t orig_pgprot = action->remap.pgprot;
1530 const unsigned long size = action->remap.size;
1531 const unsigned long pfn = io_remap_pfn_range_pfn(orig_pfn, size);
1532 int err;
1533
1534 action->remap.start_pfn = pfn;
1535 action->remap.pgprot = pgprot_decrypted(orig_pgprot);
1536 err = remap_pfn_range_prepare(desc);
1537 if (err)
1538 return err;
1539
1540 /* Remap does the actual work. */
1541 action->type = MMAP_REMAP_PFN;
1542 return 0;
1543}
1544
1545/*
1546 * When we succeed an mmap action or just before we unmap a VMA on error, we
1547 * need to ensure any rmap lock held is released. On unmap it's required to
1548 * avoid a deadlock.
1549 */
1550static inline void maybe_rmap_unlock_action(struct vm_area_struct *vma,
1551 struct mmap_action *action)
1552{
1553 struct file *file;
1554
1555 if (!action->hide_from_rmap_until_complete)
1556 return;
1557
1558 VM_WARN_ON_ONCE(vma_is_anonymous(vma));
1559 file = vma->vm_file;
1560 i_mmap_unlock_write(file->f_mapping);
1561 action->hide_from_rmap_until_complete = false;
1562}
1563
1564#ifdef CONFIG_MMU_NOTIFIER
1565static inline bool clear_flush_young_ptes_notify(struct vm_area_struct *vma,
1566 unsigned long addr, pte_t *ptep, unsigned int nr)
1567{
1568 bool young;
1569
1570 young = clear_flush_young_ptes(vma, addr, ptep, nr);
1571 young |= mmu_notifier_clear_flush_young(vma->vm_mm, addr,
1572 addr + nr * PAGE_SIZE);
1573 return young;
1574}
1575
1576static inline bool pmdp_clear_flush_young_notify(struct vm_area_struct *vma,
1577 unsigned long addr, pmd_t *pmdp)
1578{
1579 bool young;
1580
1581 young = pmdp_clear_flush_young(vma, addr, pmdp);
1582 young |= mmu_notifier_clear_flush_young(vma->vm_mm, addr, addr + PMD_SIZE);
1583 return young;
1584}
1585
1586static inline bool test_and_clear_young_ptes_notify(struct vm_area_struct *vma,
1587 unsigned long addr, pte_t *ptep, unsigned int nr)
1588{
1589 bool young;
1590
1591 young = test_and_clear_young_ptes(vma, addr, ptep, nr);
1592 young |= mmu_notifier_clear_young(vma->vm_mm, addr, addr + nr * PAGE_SIZE);
1593 return young;
1594}
1595
1596static inline bool pmdp_test_and_clear_young_notify(struct vm_area_struct *vma,
1597 unsigned long addr, pmd_t *pmdp)
1598{
1599 bool young;
1600
1601 young = pmdp_test_and_clear_young(vma, addr, pmdp);
1602 young |= mmu_notifier_clear_young(vma->vm_mm, addr, addr + PMD_SIZE);
1603 return young;
1604}
1605
1606#else /* CONFIG_MMU_NOTIFIER */
1607
1608#define clear_flush_young_ptes_notify clear_flush_young_ptes
1609#define pmdp_clear_flush_young_notify pmdp_clear_flush_young
1610#define test_and_clear_young_ptes_notify test_and_clear_young_ptes
1611#define pmdp_test_and_clear_young_notify pmdp_test_and_clear_young
1612
1613#endif /* CONFIG_MMU_NOTIFIER */
1614
1615extern int sysctl_max_map_count;
1616static inline int get_sysctl_max_map_count(void)
1617{
1618 return READ_ONCE(sysctl_max_map_count);
1619}
1620
1621bool may_expand_vm(struct mm_struct *mm, const vma_flags_t *vma_flags,
1622 unsigned long npages);
1623
1624static inline void mm_prepare_for_swap_entries(struct mm_struct *mm)
1625{
1626 if (list_empty(&mm->mmlist)) {
1627 spin_lock(&mmlist_lock);
1628 if (list_empty(&mm->mmlist))
1629 list_add(&mm->mmlist, &init_mm.mmlist);
1630 spin_unlock(&mmlist_lock);
1631 }
1632}
1633
1634static inline bool can_spin_trylock(void)
1635{
1636 /*
1637 * In PREEMPT_RT spin_trylock() will call raw_spin_lock() which is
1638 * unsafe in NMI. If spin_trylock() is called from hard IRQ the current
1639 * task may be waiting for one rt_spin_lock, but rt_spin_trylock() will
1640 * mark the task as the owner of another rt_spin_lock which will
1641 * confuse PI logic, so return immediately if called from hard IRQ or
1642 * NMI.
1643 *
1644 * Note, irqs_disabled() case is ok. spin_trylock() can be called
1645 * from raw_spin_lock_irqsave region.
1646 */
1647 if (IS_ENABLED(CONFIG_PREEMPT_RT) && (in_nmi() || in_hardirq()))
1648 return false;
1649
1650 /* On UP, spin_trylock() always succeeds even when it is locked */
1651 if (!IS_ENABLED(CONFIG_SMP) && in_nmi())
1652 return false;
1653
1654 return true;
1655}
1656
1657#endif /* __MM_INTERNAL_H */