1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* internal.h: mm/ internal definitions
4 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
7 #ifndef __MM_INTERNAL_H
8 #define __MM_INTERNAL_H
12 #include <linux/pagemap.h>
13 #include <linux/rmap.h>
14 #include <linux/tracepoint-defs.h>
19 * The set of flags that only affect watermark checking and reclaim
20 * behaviour. This is used by the MM to obey the caller constraints
21 * about IO, FS and watermark checking while ignoring placement
22 * hints such as HIGHMEM usage.
24 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
25 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
26 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
29 /* The GFP flags allowed during early boot */
30 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
32 /* Control allocation cpuset and node placement constraints */
33 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
35 /* Do not use these with a slab allocator */
36 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
39 * Different from WARN_ON_ONCE(), no warning will be issued
40 * when we specify __GFP_NOWARN.
42 #define WARN_ON_ONCE_GFP(cond, gfp) ({ \
43 static bool __section(".data.once") __warned; \
44 int __ret_warn_once = !!(cond); \
46 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
50 unlikely(__ret_warn_once); \
53 void page_writeback_init(void);
56 * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
57 * its nr_pages_mapped would be 0x400000: choose the COMPOUND_MAPPED bit
58 * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE). Hugetlb currently
59 * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
61 #define COMPOUND_MAPPED 0x800000
62 #define FOLIO_PAGES_MAPPED (COMPOUND_MAPPED - 1)
65 * Flags passed to __show_mem() and show_free_areas() to suppress output in
68 #define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */
71 * How many individual pages have an elevated _mapcount. Excludes
72 * the folio's entire_mapcount.
74 static inline int folio_nr_pages_mapped(struct folio *folio)
76 return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
79 static inline void *folio_raw_mapping(struct folio *folio)
81 unsigned long mapping = (unsigned long)folio->mapping;
83 return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
86 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
88 static inline void acct_reclaim_writeback(struct folio *folio)
90 pg_data_t *pgdat = folio_pgdat(folio);
91 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
94 __acct_reclaim_writeback(pgdat, folio, nr_throttled);
97 static inline void wake_throttle_isolated(pg_data_t *pgdat)
99 wait_queue_head_t *wqh;
101 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
102 if (waitqueue_active(wqh))
106 vm_fault_t do_swap_page(struct vm_fault *vmf);
107 void folio_rotate_reclaimable(struct folio *folio);
108 bool __folio_end_writeback(struct folio *folio);
109 void deactivate_file_folio(struct folio *folio);
110 void folio_activate(struct folio *folio);
112 void free_pgtables(struct mmu_gather *tlb, struct maple_tree *mt,
113 struct vm_area_struct *start_vma, unsigned long floor,
114 unsigned long ceiling, bool mm_wr_locked);
115 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
118 void unmap_page_range(struct mmu_gather *tlb,
119 struct vm_area_struct *vma,
120 unsigned long addr, unsigned long end,
121 struct zap_details *details);
123 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *,
125 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
126 static inline void force_page_cache_readahead(struct address_space *mapping,
127 struct file *file, pgoff_t index, unsigned long nr_to_read)
129 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
130 force_page_cache_ra(&ractl, nr_to_read);
133 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
134 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
135 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
136 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
137 void filemap_free_folio(struct address_space *mapping, struct folio *folio);
138 int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
139 bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
141 long invalidate_inode_page(struct page *page);
142 unsigned long mapping_try_invalidate(struct address_space *mapping,
143 pgoff_t start, pgoff_t end, unsigned long *nr_failed);
146 * folio_evictable - Test whether a folio is evictable.
147 * @folio: The folio to test.
149 * Test whether @folio is evictable -- i.e., should be placed on
150 * active/inactive lists vs unevictable list.
152 * Reasons folio might not be evictable:
153 * 1. folio's mapping marked unevictable
154 * 2. One of the pages in the folio is part of an mlocked VMA
156 static inline bool folio_evictable(struct folio *folio)
160 /* Prevent address_space of inode and swap cache from being freed */
162 ret = !mapping_unevictable(folio_mapping(folio)) &&
163 !folio_test_mlocked(folio);
169 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
172 static inline void set_page_refcounted(struct page *page)
174 VM_BUG_ON_PAGE(PageTail(page), page);
175 VM_BUG_ON_PAGE(page_ref_count(page), page);
176 set_page_count(page, 1);
180 * Return true if a folio needs ->release_folio() calling upon it.
182 static inline bool folio_needs_release(struct folio *folio)
184 struct address_space *mapping = folio_mapping(folio);
186 return folio_has_private(folio) ||
187 (mapping && mapping_release_always(mapping));
190 extern unsigned long highest_memmap_pfn;
193 * Maximum number of reclaim retries without progress before the OOM
194 * killer is consider the only way forward.
196 #define MAX_RECLAIM_RETRIES 16
201 bool isolate_lru_page(struct page *page);
202 bool folio_isolate_lru(struct folio *folio);
203 void putback_lru_page(struct page *page);
204 void folio_putback_lru(struct folio *folio);
205 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
210 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
215 #define K(x) ((x) << (PAGE_SHIFT-10))
217 extern char * const zone_names[MAX_NR_ZONES];
219 /* perform sanity checks on struct pages being allocated or freed */
220 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
222 extern int min_free_kbytes;
224 void setup_per_zone_wmarks(void);
225 void calculate_min_free_kbytes(void);
226 int __meminit init_per_zone_wmark_min(void);
227 void page_alloc_sysctl_init(void);
230 * Structure for holding the mostly immutable allocation parameters passed
231 * between functions involved in allocations, including the alloc_pages*
232 * family of functions.
234 * nodemask, migratetype and highest_zoneidx are initialized only once in
235 * __alloc_pages() and then never change.
237 * zonelist, preferred_zone and highest_zoneidx are set first in
238 * __alloc_pages() for the fast path, and might be later changed
239 * in __alloc_pages_slowpath(). All other functions pass the whole structure
240 * by a const pointer.
242 struct alloc_context {
243 struct zonelist *zonelist;
244 nodemask_t *nodemask;
245 struct zoneref *preferred_zoneref;
249 * highest_zoneidx represents highest usable zone index of
250 * the allocation request. Due to the nature of the zone,
251 * memory on lower zone than the highest_zoneidx will be
252 * protected by lowmem_reserve[highest_zoneidx].
254 * highest_zoneidx is also used by reclaim/compaction to limit
255 * the target zone since higher zone than this index cannot be
256 * usable for this allocation request.
258 enum zone_type highest_zoneidx;
259 bool spread_dirty_pages;
263 * This function returns the order of a free page in the buddy system. In
264 * general, page_zone(page)->lock must be held by the caller to prevent the
265 * page from being allocated in parallel and returning garbage as the order.
266 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
267 * page cannot be allocated or merged in parallel. Alternatively, it must
268 * handle invalid values gracefully, and use buddy_order_unsafe() below.
270 static inline unsigned int buddy_order(struct page *page)
272 /* PageBuddy() must be checked by the caller */
273 return page_private(page);
277 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
278 * PageBuddy() should be checked first by the caller to minimize race window,
279 * and invalid values must be handled gracefully.
281 * READ_ONCE is used so that if the caller assigns the result into a local
282 * variable and e.g. tests it for valid range before using, the compiler cannot
283 * decide to remove the variable and inline the page_private(page) multiple
284 * times, potentially observing different values in the tests and the actual
287 #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
290 * This function checks whether a page is free && is the buddy
291 * we can coalesce a page and its buddy if
292 * (a) the buddy is not in a hole (check before calling!) &&
293 * (b) the buddy is in the buddy system &&
294 * (c) a page and its buddy have the same order &&
295 * (d) a page and its buddy are in the same zone.
297 * For recording whether a page is in the buddy system, we set PageBuddy.
298 * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
300 * For recording page's order, we use page_private(page).
302 static inline bool page_is_buddy(struct page *page, struct page *buddy,
305 if (!page_is_guard(buddy) && !PageBuddy(buddy))
308 if (buddy_order(buddy) != order)
312 * zone check is done late to avoid uselessly calculating
313 * zone/node ids for pages that could never merge.
315 if (page_zone_id(page) != page_zone_id(buddy))
318 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
324 * Locate the struct page for both the matching buddy in our
325 * pair (buddy1) and the combined O(n+1) page they form (page).
327 * 1) Any buddy B1 will have an order O twin B2 which satisfies
328 * the following equation:
330 * For example, if the starting buddy (buddy2) is #8 its order
332 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
334 * 2) Any buddy B will have an order O+1 parent P which
335 * satisfies the following equation:
338 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
340 static inline unsigned long
341 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
343 return page_pfn ^ (1 << order);
347 * Find the buddy of @page and validate it.
348 * @page: The input page
349 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
350 * function is used in the performance-critical __free_one_page().
351 * @order: The order of the page
352 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
355 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
356 * not the same as @page. The validation is necessary before use it.
358 * Return: the found buddy page or NULL if not found.
360 static inline struct page *find_buddy_page_pfn(struct page *page,
361 unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
363 unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
366 buddy = page + (__buddy_pfn - pfn);
368 *buddy_pfn = __buddy_pfn;
370 if (page_is_buddy(page, buddy, order))
375 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
376 unsigned long end_pfn, struct zone *zone);
378 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
379 unsigned long end_pfn, struct zone *zone)
381 if (zone->contiguous)
382 return pfn_to_page(start_pfn);
384 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
387 void set_zone_contiguous(struct zone *zone);
389 static inline void clear_zone_contiguous(struct zone *zone)
391 zone->contiguous = false;
394 extern int __isolate_free_page(struct page *page, unsigned int order);
395 extern void __putback_isolated_page(struct page *page, unsigned int order,
397 extern void memblock_free_pages(struct page *page, unsigned long pfn,
399 extern void __free_pages_core(struct page *page, unsigned int order);
402 * This will have no effect, other than possibly generating a warning, if the
403 * caller passes in a non-large folio.
405 static inline void folio_set_order(struct folio *folio, unsigned int order)
407 if (WARN_ON_ONCE(!order || !folio_test_large(folio)))
410 folio->_folio_order = order;
412 folio->_folio_nr_pages = 1U << order;
416 static inline void prep_compound_head(struct page *page, unsigned int order)
418 struct folio *folio = (struct folio *)page;
420 folio_set_compound_dtor(folio, COMPOUND_PAGE_DTOR);
421 folio_set_order(folio, order);
422 atomic_set(&folio->_entire_mapcount, -1);
423 atomic_set(&folio->_nr_pages_mapped, 0);
424 atomic_set(&folio->_pincount, 0);
427 static inline void prep_compound_tail(struct page *head, int tail_idx)
429 struct page *p = head + tail_idx;
431 p->mapping = TAIL_MAPPING;
432 set_compound_head(p, head);
433 set_page_private(p, 0);
436 extern void prep_compound_page(struct page *page, unsigned int order);
438 extern void post_alloc_hook(struct page *page, unsigned int order,
440 extern int user_min_free_kbytes;
442 extern void free_unref_page(struct page *page, unsigned int order);
443 extern void free_unref_page_list(struct list_head *list);
445 extern void zone_pcp_reset(struct zone *zone);
446 extern void zone_pcp_disable(struct zone *zone);
447 extern void zone_pcp_enable(struct zone *zone);
448 extern void zone_pcp_init(struct zone *zone);
450 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
451 phys_addr_t min_addr,
452 int nid, bool exact_nid);
454 void memmap_init_range(unsigned long, int, unsigned long, unsigned long,
455 unsigned long, enum meminit_context, struct vmem_altmap *, int);
458 int split_free_page(struct page *free_page,
459 unsigned int order, unsigned long split_pfn_offset);
461 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
467 * compact_control is used to track pages being migrated and the free pages
468 * they are being migrated to during memory compaction. The free_pfn starts
469 * at the end of a zone and migrate_pfn begins at the start. Movable pages
470 * are moved to the end of a zone during a compaction run and the run
471 * completes when free_pfn <= migrate_pfn
473 struct compact_control {
474 struct list_head freepages; /* List of free pages to migrate to */
475 struct list_head migratepages; /* List of pages being migrated */
476 unsigned int nr_freepages; /* Number of isolated free pages */
477 unsigned int nr_migratepages; /* Number of pages to migrate */
478 unsigned long free_pfn; /* isolate_freepages search base */
480 * Acts as an in/out parameter to page isolation for migration.
481 * isolate_migratepages uses it as a search base.
482 * isolate_migratepages_block will update the value to the next pfn
483 * after the last isolated one.
485 unsigned long migrate_pfn;
486 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
488 unsigned long total_migrate_scanned;
489 unsigned long total_free_scanned;
490 unsigned short fast_search_fail;/* failures to use free list searches */
491 short search_order; /* order to start a fast search at */
492 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
493 int order; /* order a direct compactor needs */
494 int migratetype; /* migratetype of direct compactor */
495 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
496 const int highest_zoneidx; /* zone index of a direct compactor */
497 enum migrate_mode mode; /* Async or sync migration mode */
498 bool ignore_skip_hint; /* Scan blocks even if marked skip */
499 bool no_set_skip_hint; /* Don't mark blocks for skipping */
500 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
501 bool direct_compaction; /* False from kcompactd or /proc/... */
502 bool proactive_compaction; /* kcompactd proactive compaction */
503 bool whole_zone; /* Whole zone should/has been scanned */
504 bool contended; /* Signal lock contention */
505 bool finish_pageblock; /* Scan the remainder of a pageblock. Used
506 * when there are potentially transient
507 * isolation or migration failures to
508 * ensure forward progress.
510 bool alloc_contig; /* alloc_contig_range allocation */
514 * Used in direct compaction when a page should be taken from the freelists
515 * immediately when one is created during the free path.
517 struct capture_control {
518 struct compact_control *cc;
523 isolate_freepages_range(struct compact_control *cc,
524 unsigned long start_pfn, unsigned long end_pfn);
526 isolate_migratepages_range(struct compact_control *cc,
527 unsigned long low_pfn, unsigned long end_pfn);
529 int __alloc_contig_migrate_range(struct compact_control *cc,
530 unsigned long start, unsigned long end);
532 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
533 void init_cma_reserved_pageblock(struct page *page);
535 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
537 int find_suitable_fallback(struct free_area *area, unsigned int order,
538 int migratetype, bool only_stealable, bool *can_steal);
540 static inline bool free_area_empty(struct free_area *area, int migratetype)
542 return list_empty(&area->free_list[migratetype]);
546 * These three helpers classifies VMAs for virtual memory accounting.
550 * Executable code area - executable, not writable, not stack
552 static inline bool is_exec_mapping(vm_flags_t flags)
554 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
558 * Stack area - automatically grows in one direction
560 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
561 * do_mmap() forbids all other combinations.
563 static inline bool is_stack_mapping(vm_flags_t flags)
565 return (flags & VM_STACK) == VM_STACK;
569 * Data area - private, writable, not stack
571 static inline bool is_data_mapping(vm_flags_t flags)
573 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
577 struct anon_vma *folio_anon_vma(struct folio *folio);
580 void unmap_mapping_folio(struct folio *folio);
581 extern long populate_vma_page_range(struct vm_area_struct *vma,
582 unsigned long start, unsigned long end, int *locked);
583 extern long faultin_vma_page_range(struct vm_area_struct *vma,
584 unsigned long start, unsigned long end,
585 bool write, int *locked);
586 extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags,
587 unsigned long bytes);
589 * mlock_vma_folio() and munlock_vma_folio():
590 * should be called with vma's mmap_lock held for read or write,
591 * under page table lock for the pte/pmd being added or removed.
593 * mlock is usually called at the end of page_add_*_rmap(), munlock at
594 * the end of page_remove_rmap(); but new anon folios are managed by
595 * folio_add_lru_vma() calling mlock_new_folio().
597 * @compound is used to include pmd mappings of THPs, but filter out
598 * pte mappings of THPs, which cannot be consistently counted: a pte
599 * mapping of the THP head cannot be distinguished by the page alone.
601 void mlock_folio(struct folio *folio);
602 static inline void mlock_vma_folio(struct folio *folio,
603 struct vm_area_struct *vma, bool compound)
606 * The VM_SPECIAL check here serves two purposes.
607 * 1) VM_IO check prevents migration from double-counting during mlock.
608 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
609 * is never left set on a VM_SPECIAL vma, there is an interval while
610 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
611 * still be set while VM_SPECIAL bits are added: so ignore it then.
613 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) &&
614 (compound || !folio_test_large(folio)))
618 void munlock_folio(struct folio *folio);
619 static inline void munlock_vma_folio(struct folio *folio,
620 struct vm_area_struct *vma, bool compound)
622 if (unlikely(vma->vm_flags & VM_LOCKED) &&
623 (compound || !folio_test_large(folio)))
624 munlock_folio(folio);
627 void mlock_new_folio(struct folio *folio);
628 bool need_mlock_drain(int cpu);
629 void mlock_drain_local(void);
630 void mlock_drain_remote(int cpu);
632 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
635 * Return the start of user virtual address at the specific offset within
638 static inline unsigned long
639 vma_pgoff_address(pgoff_t pgoff, unsigned long nr_pages,
640 struct vm_area_struct *vma)
642 unsigned long address;
644 if (pgoff >= vma->vm_pgoff) {
645 address = vma->vm_start +
646 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
647 /* Check for address beyond vma (or wrapped through 0?) */
648 if (address < vma->vm_start || address >= vma->vm_end)
650 } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
651 /* Test above avoids possibility of wrap to 0 on 32-bit */
652 address = vma->vm_start;
660 * Return the start of user virtual address of a page within a vma.
661 * Returns -EFAULT if all of the page is outside the range of vma.
662 * If page is a compound head, the entire compound page is considered.
664 static inline unsigned long
665 vma_address(struct page *page, struct vm_area_struct *vma)
667 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
668 return vma_pgoff_address(page_to_pgoff(page), compound_nr(page), vma);
672 * Then at what user virtual address will none of the range be found in vma?
673 * Assumes that vma_address() already returned a good starting address.
675 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
677 struct vm_area_struct *vma = pvmw->vma;
679 unsigned long address;
681 /* Common case, plus ->pgoff is invalid for KSM */
682 if (pvmw->nr_pages == 1)
683 return pvmw->address + PAGE_SIZE;
685 pgoff = pvmw->pgoff + pvmw->nr_pages;
686 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
687 /* Check for address beyond vma (or wrapped through 0?) */
688 if (address < vma->vm_start || address > vma->vm_end)
689 address = vma->vm_end;
693 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
696 int flags = vmf->flags;
702 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
703 * anything, so we only pin the file and drop the mmap_lock if only
704 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
706 if (fault_flag_allow_retry_first(flags) &&
707 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
708 fpin = get_file(vmf->vma->vm_file);
709 mmap_read_unlock(vmf->vma->vm_mm);
713 #else /* !CONFIG_MMU */
714 static inline void unmap_mapping_folio(struct folio *folio) { }
715 static inline void mlock_new_folio(struct folio *folio) { }
716 static inline bool need_mlock_drain(int cpu) { return false; }
717 static inline void mlock_drain_local(void) { }
718 static inline void mlock_drain_remote(int cpu) { }
719 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
722 #endif /* !CONFIG_MMU */
724 /* Memory initialisation debug and verification */
725 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
726 DECLARE_STATIC_KEY_TRUE(deferred_pages);
728 bool __init deferred_grow_zone(struct zone *zone, unsigned int order);
729 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
737 #ifdef CONFIG_DEBUG_MEMORY_INIT
739 extern int mminit_loglevel;
741 #define mminit_dprintk(level, prefix, fmt, arg...) \
743 if (level < mminit_loglevel) { \
744 if (level <= MMINIT_WARNING) \
745 pr_warn("mminit::" prefix " " fmt, ##arg); \
747 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
751 extern void mminit_verify_pageflags_layout(void);
752 extern void mminit_verify_zonelist(void);
755 static inline void mminit_dprintk(enum mminit_level level,
756 const char *prefix, const char *fmt, ...)
760 static inline void mminit_verify_pageflags_layout(void)
764 static inline void mminit_verify_zonelist(void)
767 #endif /* CONFIG_DEBUG_MEMORY_INIT */
769 #define NODE_RECLAIM_NOSCAN -2
770 #define NODE_RECLAIM_FULL -1
771 #define NODE_RECLAIM_SOME 0
772 #define NODE_RECLAIM_SUCCESS 1
775 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
776 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
778 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
781 return NODE_RECLAIM_NOSCAN;
783 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
790 * mm/memory-failure.c
792 extern int hwpoison_filter(struct page *p);
794 extern u32 hwpoison_filter_dev_major;
795 extern u32 hwpoison_filter_dev_minor;
796 extern u64 hwpoison_filter_flags_mask;
797 extern u64 hwpoison_filter_flags_value;
798 extern u64 hwpoison_filter_memcg;
799 extern u32 hwpoison_filter_enable;
801 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
802 unsigned long, unsigned long,
803 unsigned long, unsigned long);
805 extern void set_pageblock_order(void);
806 unsigned long reclaim_pages(struct list_head *folio_list);
807 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
808 struct list_head *folio_list);
809 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
810 #define ALLOC_WMARK_MIN WMARK_MIN
811 #define ALLOC_WMARK_LOW WMARK_LOW
812 #define ALLOC_WMARK_HIGH WMARK_HIGH
813 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
815 /* Mask to get the watermark bits */
816 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
819 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
820 * cannot assume a reduced access to memory reserves is sufficient for
824 #define ALLOC_OOM 0x08
826 #define ALLOC_OOM ALLOC_NO_WATERMARKS
829 #define ALLOC_NON_BLOCK 0x10 /* Caller cannot block. Allow access
830 * to 25% of the min watermark or
831 * 62.5% if __GFP_HIGH is set.
833 #define ALLOC_MIN_RESERVE 0x20 /* __GFP_HIGH set. Allow access to 50%
834 * of the min watermark.
836 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
837 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
838 #ifdef CONFIG_ZONE_DMA32
839 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
841 #define ALLOC_NOFRAGMENT 0x0
843 #define ALLOC_HIGHATOMIC 0x200 /* Allows access to MIGRATE_HIGHATOMIC */
844 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
846 /* Flags that allow allocations below the min watermark. */
847 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
850 struct tlbflush_unmap_batch;
854 * only for MM internal work items which do not depend on
855 * any allocations or locks which might depend on allocations
857 extern struct workqueue_struct *mm_percpu_wq;
859 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
860 void try_to_unmap_flush(void);
861 void try_to_unmap_flush_dirty(void);
862 void flush_tlb_batched_pending(struct mm_struct *mm);
864 static inline void try_to_unmap_flush(void)
867 static inline void try_to_unmap_flush_dirty(void)
870 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
873 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
875 extern const struct trace_print_flags pageflag_names[];
876 extern const struct trace_print_flags pagetype_names[];
877 extern const struct trace_print_flags vmaflag_names[];
878 extern const struct trace_print_flags gfpflag_names[];
880 static inline bool is_migrate_highatomic(enum migratetype migratetype)
882 return migratetype == MIGRATE_HIGHATOMIC;
885 static inline bool is_migrate_highatomic_page(struct page *page)
887 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
890 void setup_zone_pageset(struct zone *zone);
892 struct migration_target_control {
893 int nid; /* preferred node id */
901 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
902 struct folio *folio, loff_t fpos, size_t size);
908 void __init vmalloc_init(void);
909 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
910 pgprot_t prot, struct page **pages, unsigned int page_shift);
912 static inline void vmalloc_init(void)
917 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
918 pgprot_t prot, struct page **pages, unsigned int page_shift)
924 int __must_check __vmap_pages_range_noflush(unsigned long addr,
925 unsigned long end, pgprot_t prot,
926 struct page **pages, unsigned int page_shift);
928 void vunmap_range_noflush(unsigned long start, unsigned long end);
930 void __vunmap_range_noflush(unsigned long start, unsigned long end);
932 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
933 unsigned long addr, int page_nid, int *flags);
935 void free_zone_device_page(struct page *page);
936 int migrate_device_coherent_page(struct page *page);
941 struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags);
942 int __must_check try_grab_page(struct page *page, unsigned int flags);
945 /* mark page accessed */
946 FOLL_TOUCH = 1 << 16,
947 /* a retry, previous pass started an IO */
948 FOLL_TRIED = 1 << 17,
949 /* we are working on non-current tsk/mm */
950 FOLL_REMOTE = 1 << 18,
951 /* pages must be released via unpin_user_page */
953 /* gup_fast: prevent fall-back to slow gup */
954 FOLL_FAST_ONLY = 1 << 20,
955 /* allow unlocking the mmap lock */
956 FOLL_UNLOCKABLE = 1 << 21,
960 * Indicates for which pages that are write-protected in the page table,
961 * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
962 * GUP pin will remain consistent with the pages mapped into the page tables
965 * Temporary unmapping of PageAnonExclusive() pages or clearing of
966 * PageAnonExclusive() has to protect against concurrent GUP:
967 * * Ordinary GUP: Using the PT lock
968 * * GUP-fast and fork(): mm->write_protect_seq
969 * * GUP-fast and KSM or temporary unmapping (swap, migration): see
970 * page_try_share_anon_rmap()
972 * Must be called with the (sub)page that's actually referenced via the
973 * page table entry, which might not necessarily be the head page for a
976 * If the vma is NULL, we're coming from the GUP-fast path and might have
977 * to fallback to the slow path just to lookup the vma.
979 static inline bool gup_must_unshare(struct vm_area_struct *vma,
980 unsigned int flags, struct page *page)
983 * FOLL_WRITE is implicitly handled correctly as the page table entry
984 * has to be writable -- and if it references (part of) an anonymous
985 * folio, that part is required to be marked exclusive.
987 if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
990 * Note: PageAnon(page) is stable until the page is actually getting
993 if (!PageAnon(page)) {
995 * We only care about R/O long-term pining: R/O short-term
996 * pinning does not have the semantics to observe successive
997 * changes through the process page tables.
999 if (!(flags & FOLL_LONGTERM))
1002 /* We really need the vma ... */
1007 * ... because we only care about writable private ("COW")
1008 * mappings where we have to break COW early.
1010 return is_cow_mapping(vma->vm_flags);
1013 /* Paired with a memory barrier in page_try_share_anon_rmap(). */
1014 if (IS_ENABLED(CONFIG_HAVE_FAST_GUP))
1018 * Note that PageKsm() pages cannot be exclusive, and consequently,
1019 * cannot get pinned.
1021 return !PageAnonExclusive(page);
1024 extern bool mirrored_kernelcore;
1026 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1029 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1030 * enablements, because when without soft-dirty being compiled in,
1031 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1032 * will be constantly true.
1034 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
1038 * Soft-dirty is kind of special: its tracking is enabled when the
1039 * vma flags not set.
1041 return !(vma->vm_flags & VM_SOFTDIRTY);
1045 * VMA Iterator functions shared between nommu and mmap
1047 static inline int vma_iter_prealloc(struct vma_iterator *vmi)
1049 return mas_preallocate(&vmi->mas, GFP_KERNEL);
1052 static inline void vma_iter_clear(struct vma_iterator *vmi,
1053 unsigned long start, unsigned long end)
1055 mas_set_range(&vmi->mas, start, end - 1);
1056 mas_store_prealloc(&vmi->mas, NULL);
1059 static inline struct vm_area_struct *vma_iter_load(struct vma_iterator *vmi)
1061 return mas_walk(&vmi->mas);
1064 /* Store a VMA with preallocated memory */
1065 static inline void vma_iter_store(struct vma_iterator *vmi,
1066 struct vm_area_struct *vma)
1069 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
1070 if (MAS_WARN_ON(&vmi->mas, vmi->mas.node != MAS_START &&
1071 vmi->mas.index > vma->vm_start)) {
1072 pr_warn("%lx > %lx\n store vma %lx-%lx\n into slot %lx-%lx\n",
1073 vmi->mas.index, vma->vm_start, vma->vm_start,
1074 vma->vm_end, vmi->mas.index, vmi->mas.last);
1076 if (MAS_WARN_ON(&vmi->mas, vmi->mas.node != MAS_START &&
1077 vmi->mas.last < vma->vm_start)) {
1078 pr_warn("%lx < %lx\nstore vma %lx-%lx\ninto slot %lx-%lx\n",
1079 vmi->mas.last, vma->vm_start, vma->vm_start, vma->vm_end,
1080 vmi->mas.index, vmi->mas.last);
1084 if (vmi->mas.node != MAS_START &&
1085 ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
1086 vma_iter_invalidate(vmi);
1088 vmi->mas.index = vma->vm_start;
1089 vmi->mas.last = vma->vm_end - 1;
1090 mas_store_prealloc(&vmi->mas, vma);
1093 static inline int vma_iter_store_gfp(struct vma_iterator *vmi,
1094 struct vm_area_struct *vma, gfp_t gfp)
1096 if (vmi->mas.node != MAS_START &&
1097 ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
1098 vma_iter_invalidate(vmi);
1100 vmi->mas.index = vma->vm_start;
1101 vmi->mas.last = vma->vm_end - 1;
1102 mas_store_gfp(&vmi->mas, vma, gfp);
1103 if (unlikely(mas_is_err(&vmi->mas)))
1110 * VMA lock generalization
1112 struct vma_prepare {
1113 struct vm_area_struct *vma;
1114 struct vm_area_struct *adj_next;
1116 struct address_space *mapping;
1117 struct anon_vma *anon_vma;
1118 struct vm_area_struct *insert;
1119 struct vm_area_struct *remove;
1120 struct vm_area_struct *remove2;
1122 #endif /* __MM_INTERNAL_H */