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.
7 #ifndef __MM_INTERNAL_H
8 #define __MM_INTERNAL_H
12 #include <linux/pagemap.h>
13 #include <linux/tracepoint-defs.h>
18 * The set of flags that only affect watermark checking and reclaim
19 * behaviour. This is used by the MM to obey the caller constraints
20 * about IO, FS and watermark checking while ignoring placement
21 * hints such as HIGHMEM usage.
23 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
24 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
25 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
26 __GFP_ATOMIC|__GFP_NOLOCKDEP)
28 /* The GFP flags allowed during early boot */
29 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
31 /* Control allocation cpuset and node placement constraints */
32 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
34 /* Do not use these with a slab allocator */
35 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
37 void page_writeback_init(void);
39 static inline void *folio_raw_mapping(struct folio *folio)
41 unsigned long mapping = (unsigned long)folio->mapping;
43 return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
46 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
48 static inline void acct_reclaim_writeback(struct folio *folio)
50 pg_data_t *pgdat = folio_pgdat(folio);
51 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
54 __acct_reclaim_writeback(pgdat, folio, nr_throttled);
57 static inline void wake_throttle_isolated(pg_data_t *pgdat)
59 wait_queue_head_t *wqh;
61 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
62 if (waitqueue_active(wqh))
66 vm_fault_t do_swap_page(struct vm_fault *vmf);
67 void folio_rotate_reclaimable(struct folio *folio);
68 bool __folio_end_writeback(struct folio *folio);
70 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
71 unsigned long floor, unsigned long ceiling);
72 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
74 static inline bool can_madv_lru_vma(struct vm_area_struct *vma)
76 return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
80 void unmap_page_range(struct mmu_gather *tlb,
81 struct vm_area_struct *vma,
82 unsigned long addr, unsigned long end,
83 struct zap_details *details);
85 void do_page_cache_ra(struct readahead_control *, unsigned long nr_to_read,
86 unsigned long lookahead_size);
87 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
88 static inline void force_page_cache_readahead(struct address_space *mapping,
89 struct file *file, pgoff_t index, unsigned long nr_to_read)
91 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
92 force_page_cache_ra(&ractl, nr_to_read);
95 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
96 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
97 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
98 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
99 void filemap_free_folio(struct address_space *mapping, struct folio *folio);
100 int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
101 bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
105 * folio_evictable - Test whether a folio is evictable.
106 * @folio: The folio to test.
108 * Test whether @folio is evictable -- i.e., should be placed on
109 * active/inactive lists vs unevictable list.
111 * Reasons folio might not be evictable:
112 * 1. folio's mapping marked unevictable
113 * 2. One of the pages in the folio is part of an mlocked VMA
115 static inline bool folio_evictable(struct folio *folio)
119 /* Prevent address_space of inode and swap cache from being freed */
121 ret = !mapping_unevictable(folio_mapping(folio)) &&
122 !folio_test_mlocked(folio);
127 static inline bool page_evictable(struct page *page)
131 /* Prevent address_space of inode and swap cache from being freed */
133 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
139 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
142 static inline void set_page_refcounted(struct page *page)
144 VM_BUG_ON_PAGE(PageTail(page), page);
145 VM_BUG_ON_PAGE(page_ref_count(page), page);
146 set_page_count(page, 1);
149 extern unsigned long highest_memmap_pfn;
152 * Maximum number of reclaim retries without progress before the OOM
153 * killer is consider the only way forward.
155 #define MAX_RECLAIM_RETRIES 16
160 extern int isolate_lru_page(struct page *page);
161 extern void putback_lru_page(struct page *page);
162 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
167 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
174 * Structure for holding the mostly immutable allocation parameters passed
175 * between functions involved in allocations, including the alloc_pages*
176 * family of functions.
178 * nodemask, migratetype and highest_zoneidx are initialized only once in
179 * __alloc_pages() and then never change.
181 * zonelist, preferred_zone and highest_zoneidx are set first in
182 * __alloc_pages() for the fast path, and might be later changed
183 * in __alloc_pages_slowpath(). All other functions pass the whole structure
184 * by a const pointer.
186 struct alloc_context {
187 struct zonelist *zonelist;
188 nodemask_t *nodemask;
189 struct zoneref *preferred_zoneref;
193 * highest_zoneidx represents highest usable zone index of
194 * the allocation request. Due to the nature of the zone,
195 * memory on lower zone than the highest_zoneidx will be
196 * protected by lowmem_reserve[highest_zoneidx].
198 * highest_zoneidx is also used by reclaim/compaction to limit
199 * the target zone since higher zone than this index cannot be
200 * usable for this allocation request.
202 enum zone_type highest_zoneidx;
203 bool spread_dirty_pages;
207 * Locate the struct page for both the matching buddy in our
208 * pair (buddy1) and the combined O(n+1) page they form (page).
210 * 1) Any buddy B1 will have an order O twin B2 which satisfies
211 * the following equation:
213 * For example, if the starting buddy (buddy2) is #8 its order
215 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
217 * 2) Any buddy B will have an order O+1 parent P which
218 * satisfies the following equation:
221 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
223 static inline unsigned long
224 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
226 return page_pfn ^ (1 << order);
229 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
230 unsigned long end_pfn, struct zone *zone);
232 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
233 unsigned long end_pfn, struct zone *zone)
235 if (zone->contiguous)
236 return pfn_to_page(start_pfn);
238 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
241 extern int __isolate_free_page(struct page *page, unsigned int order);
242 extern void __putback_isolated_page(struct page *page, unsigned int order,
244 extern void memblock_free_pages(struct page *page, unsigned long pfn,
246 extern void __free_pages_core(struct page *page, unsigned int order);
247 extern void prep_compound_page(struct page *page, unsigned int order);
248 extern void post_alloc_hook(struct page *page, unsigned int order,
250 extern int user_min_free_kbytes;
252 extern void free_unref_page(struct page *page, unsigned int order);
253 extern void free_unref_page_list(struct list_head *list);
255 extern void zone_pcp_update(struct zone *zone, int cpu_online);
256 extern void zone_pcp_reset(struct zone *zone);
257 extern void zone_pcp_disable(struct zone *zone);
258 extern void zone_pcp_enable(struct zone *zone);
260 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
261 phys_addr_t min_addr,
262 int nid, bool exact_nid);
264 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
270 * compact_control is used to track pages being migrated and the free pages
271 * they are being migrated to during memory compaction. The free_pfn starts
272 * at the end of a zone and migrate_pfn begins at the start. Movable pages
273 * are moved to the end of a zone during a compaction run and the run
274 * completes when free_pfn <= migrate_pfn
276 struct compact_control {
277 struct list_head freepages; /* List of free pages to migrate to */
278 struct list_head migratepages; /* List of pages being migrated */
279 unsigned int nr_freepages; /* Number of isolated free pages */
280 unsigned int nr_migratepages; /* Number of pages to migrate */
281 unsigned long free_pfn; /* isolate_freepages search base */
283 * Acts as an in/out parameter to page isolation for migration.
284 * isolate_migratepages uses it as a search base.
285 * isolate_migratepages_block will update the value to the next pfn
286 * after the last isolated one.
288 unsigned long migrate_pfn;
289 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
291 unsigned long total_migrate_scanned;
292 unsigned long total_free_scanned;
293 unsigned short fast_search_fail;/* failures to use free list searches */
294 short search_order; /* order to start a fast search at */
295 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
296 int order; /* order a direct compactor needs */
297 int migratetype; /* migratetype of direct compactor */
298 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
299 const int highest_zoneidx; /* zone index of a direct compactor */
300 enum migrate_mode mode; /* Async or sync migration mode */
301 bool ignore_skip_hint; /* Scan blocks even if marked skip */
302 bool no_set_skip_hint; /* Don't mark blocks for skipping */
303 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
304 bool direct_compaction; /* False from kcompactd or /proc/... */
305 bool proactive_compaction; /* kcompactd proactive compaction */
306 bool whole_zone; /* Whole zone should/has been scanned */
307 bool contended; /* Signal lock or sched contention */
308 bool rescan; /* Rescanning the same pageblock */
309 bool alloc_contig; /* alloc_contig_range allocation */
313 * Used in direct compaction when a page should be taken from the freelists
314 * immediately when one is created during the free path.
316 struct capture_control {
317 struct compact_control *cc;
322 isolate_freepages_range(struct compact_control *cc,
323 unsigned long start_pfn, unsigned long end_pfn);
325 isolate_migratepages_range(struct compact_control *cc,
326 unsigned long low_pfn, unsigned long end_pfn);
328 int find_suitable_fallback(struct free_area *area, unsigned int order,
329 int migratetype, bool only_stealable, bool *can_steal);
332 * This function returns the order of a free page in the buddy system. In
333 * general, page_zone(page)->lock must be held by the caller to prevent the
334 * page from being allocated in parallel and returning garbage as the order.
335 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
336 * page cannot be allocated or merged in parallel. Alternatively, it must
337 * handle invalid values gracefully, and use buddy_order_unsafe() below.
339 static inline unsigned int buddy_order(struct page *page)
341 /* PageBuddy() must be checked by the caller */
342 return page_private(page);
346 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
347 * PageBuddy() should be checked first by the caller to minimize race window,
348 * and invalid values must be handled gracefully.
350 * READ_ONCE is used so that if the caller assigns the result into a local
351 * variable and e.g. tests it for valid range before using, the compiler cannot
352 * decide to remove the variable and inline the page_private(page) multiple
353 * times, potentially observing different values in the tests and the actual
356 #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
359 * These three helpers classifies VMAs for virtual memory accounting.
363 * Executable code area - executable, not writable, not stack
365 static inline bool is_exec_mapping(vm_flags_t flags)
367 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
371 * Stack area - automatically grows in one direction
373 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
374 * do_mmap() forbids all other combinations.
376 static inline bool is_stack_mapping(vm_flags_t flags)
378 return (flags & VM_STACK) == VM_STACK;
382 * Data area - private, writable, not stack
384 static inline bool is_data_mapping(vm_flags_t flags)
386 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
390 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
391 struct vm_area_struct *prev);
392 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
395 void unmap_mapping_folio(struct folio *folio);
396 extern long populate_vma_page_range(struct vm_area_struct *vma,
397 unsigned long start, unsigned long end, int *locked);
398 extern long faultin_vma_page_range(struct vm_area_struct *vma,
399 unsigned long start, unsigned long end,
400 bool write, int *locked);
401 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
402 unsigned long start, unsigned long end);
403 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
405 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
409 * must be called with vma's mmap_lock held for read or write, and page locked.
411 extern void mlock_vma_page(struct page *page);
412 extern unsigned int munlock_vma_page(struct page *page);
414 extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
418 * Clear the page's PageMlocked(). This can be useful in a situation where
419 * we want to unconditionally remove a page from the pagecache -- e.g.,
420 * on truncation or freeing.
422 * It is legal to call this function for any page, mlocked or not.
423 * If called for a page that is still mapped by mlocked vmas, all we do
424 * is revert to lazy LRU behaviour -- semantics are not broken.
426 extern void clear_page_mlock(struct page *page);
428 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
431 * At what user virtual address is page expected in vma?
432 * Returns -EFAULT if all of the page is outside the range of vma.
433 * If page is a compound head, the entire compound page is considered.
435 static inline unsigned long
436 vma_address(struct page *page, struct vm_area_struct *vma)
439 unsigned long address;
441 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
442 pgoff = page_to_pgoff(page);
443 if (pgoff >= vma->vm_pgoff) {
444 address = vma->vm_start +
445 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
446 /* Check for address beyond vma (or wrapped through 0?) */
447 if (address < vma->vm_start || address >= vma->vm_end)
449 } else if (PageHead(page) &&
450 pgoff + compound_nr(page) - 1 >= vma->vm_pgoff) {
451 /* Test above avoids possibility of wrap to 0 on 32-bit */
452 address = vma->vm_start;
460 * Then at what user virtual address will none of the page be found in vma?
461 * Assumes that vma_address() already returned a good starting address.
462 * If page is a compound head, the entire compound page is considered.
464 static inline unsigned long
465 vma_address_end(struct page *page, struct vm_area_struct *vma)
468 unsigned long address;
470 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
471 pgoff = page_to_pgoff(page) + compound_nr(page);
472 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
473 /* Check for address beyond vma (or wrapped through 0?) */
474 if (address < vma->vm_start || address > vma->vm_end)
475 address = vma->vm_end;
479 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
482 int flags = vmf->flags;
488 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
489 * anything, so we only pin the file and drop the mmap_lock if only
490 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
492 if (fault_flag_allow_retry_first(flags) &&
493 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
494 fpin = get_file(vmf->vma->vm_file);
495 mmap_read_unlock(vmf->vma->vm_mm);
499 #else /* !CONFIG_MMU */
500 static inline void unmap_mapping_folio(struct folio *folio) { }
501 static inline void clear_page_mlock(struct page *page) { }
502 static inline void mlock_vma_page(struct page *page) { }
503 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
506 #endif /* !CONFIG_MMU */
509 * Return the mem_map entry representing the 'offset' subpage within
510 * the maximally aligned gigantic page 'base'. Handle any discontiguity
511 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
513 static inline struct page *mem_map_offset(struct page *base, int offset)
515 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
516 return nth_page(base, offset);
517 return base + offset;
521 * Iterator over all subpages within the maximally aligned gigantic
522 * page 'base'. Handle any discontiguity in the mem_map.
524 static inline struct page *mem_map_next(struct page *iter,
525 struct page *base, int offset)
527 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
528 unsigned long pfn = page_to_pfn(base) + offset;
531 return pfn_to_page(pfn);
536 /* Memory initialisation debug and verification */
543 #ifdef CONFIG_DEBUG_MEMORY_INIT
545 extern int mminit_loglevel;
547 #define mminit_dprintk(level, prefix, fmt, arg...) \
549 if (level < mminit_loglevel) { \
550 if (level <= MMINIT_WARNING) \
551 pr_warn("mminit::" prefix " " fmt, ##arg); \
553 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
557 extern void mminit_verify_pageflags_layout(void);
558 extern void mminit_verify_zonelist(void);
561 static inline void mminit_dprintk(enum mminit_level level,
562 const char *prefix, const char *fmt, ...)
566 static inline void mminit_verify_pageflags_layout(void)
570 static inline void mminit_verify_zonelist(void)
573 #endif /* CONFIG_DEBUG_MEMORY_INIT */
575 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
576 #if defined(CONFIG_SPARSEMEM)
577 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
578 unsigned long *end_pfn);
580 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
581 unsigned long *end_pfn)
584 #endif /* CONFIG_SPARSEMEM */
586 #define NODE_RECLAIM_NOSCAN -2
587 #define NODE_RECLAIM_FULL -1
588 #define NODE_RECLAIM_SOME 0
589 #define NODE_RECLAIM_SUCCESS 1
592 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
593 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
595 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
598 return NODE_RECLAIM_NOSCAN;
600 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
606 extern int hwpoison_filter(struct page *p);
608 extern u32 hwpoison_filter_dev_major;
609 extern u32 hwpoison_filter_dev_minor;
610 extern u64 hwpoison_filter_flags_mask;
611 extern u64 hwpoison_filter_flags_value;
612 extern u64 hwpoison_filter_memcg;
613 extern u32 hwpoison_filter_enable;
615 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
616 unsigned long, unsigned long,
617 unsigned long, unsigned long);
619 extern void set_pageblock_order(void);
620 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
621 struct list_head *page_list);
622 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
623 #define ALLOC_WMARK_MIN WMARK_MIN
624 #define ALLOC_WMARK_LOW WMARK_LOW
625 #define ALLOC_WMARK_HIGH WMARK_HIGH
626 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
628 /* Mask to get the watermark bits */
629 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
632 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
633 * cannot assume a reduced access to memory reserves is sufficient for
637 #define ALLOC_OOM 0x08
639 #define ALLOC_OOM ALLOC_NO_WATERMARKS
642 #define ALLOC_HARDER 0x10 /* try to alloc harder */
643 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
644 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
645 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
646 #ifdef CONFIG_ZONE_DMA32
647 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
649 #define ALLOC_NOFRAGMENT 0x0
651 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
654 struct tlbflush_unmap_batch;
658 * only for MM internal work items which do not depend on
659 * any allocations or locks which might depend on allocations
661 extern struct workqueue_struct *mm_percpu_wq;
663 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
664 void try_to_unmap_flush(void);
665 void try_to_unmap_flush_dirty(void);
666 void flush_tlb_batched_pending(struct mm_struct *mm);
668 static inline void try_to_unmap_flush(void)
671 static inline void try_to_unmap_flush_dirty(void)
674 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
677 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
679 extern const struct trace_print_flags pageflag_names[];
680 extern const struct trace_print_flags vmaflag_names[];
681 extern const struct trace_print_flags gfpflag_names[];
683 static inline bool is_migrate_highatomic(enum migratetype migratetype)
685 return migratetype == MIGRATE_HIGHATOMIC;
688 static inline bool is_migrate_highatomic_page(struct page *page)
690 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
693 void setup_zone_pageset(struct zone *zone);
695 struct migration_target_control {
696 int nid; /* preferred node id */
705 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
706 pgprot_t prot, struct page **pages, unsigned int page_shift);
709 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
710 pgprot_t prot, struct page **pages, unsigned int page_shift)
716 void vunmap_range_noflush(unsigned long start, unsigned long end);
718 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
719 unsigned long addr, int page_nid, int *flags);
721 #endif /* __MM_INTERNAL_H */