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/tracepoint-defs.h>
16 * The set of flags that only affect watermark checking and reclaim
17 * behaviour. This is used by the MM to obey the caller constraints
18 * about IO, FS and watermark checking while ignoring placement
19 * hints such as HIGHMEM usage.
21 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
22 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
23 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
26 /* The GFP flags allowed during early boot */
27 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
29 /* Control allocation cpuset and node placement constraints */
30 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
32 /* Do not use these with a slab allocator */
33 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
35 void page_writeback_init(void);
37 static inline void *folio_raw_mapping(struct folio *folio)
39 unsigned long mapping = (unsigned long)folio->mapping;
41 return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
44 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
46 static inline void acct_reclaim_writeback(struct folio *folio)
48 pg_data_t *pgdat = folio_pgdat(folio);
49 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
52 __acct_reclaim_writeback(pgdat, folio, nr_throttled);
55 static inline void wake_throttle_isolated(pg_data_t *pgdat)
57 wait_queue_head_t *wqh;
59 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
60 if (waitqueue_active(wqh))
64 vm_fault_t do_swap_page(struct vm_fault *vmf);
65 void folio_rotate_reclaimable(struct folio *folio);
66 bool __folio_end_writeback(struct folio *folio);
68 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
69 unsigned long floor, unsigned long ceiling);
70 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
72 static inline bool can_madv_lru_vma(struct vm_area_struct *vma)
74 return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
77 void unmap_page_range(struct mmu_gather *tlb,
78 struct vm_area_struct *vma,
79 unsigned long addr, unsigned long end,
80 struct zap_details *details);
82 void do_page_cache_ra(struct readahead_control *, unsigned long nr_to_read,
83 unsigned long lookahead_size);
84 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
85 static inline void force_page_cache_readahead(struct address_space *mapping,
86 struct file *file, pgoff_t index, unsigned long nr_to_read)
88 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
89 force_page_cache_ra(&ractl, nr_to_read);
92 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
93 pgoff_t end, struct pagevec *pvec, pgoff_t *indices);
96 * folio_evictable - Test whether a folio is evictable.
97 * @folio: The folio to test.
99 * Test whether @folio is evictable -- i.e., should be placed on
100 * active/inactive lists vs unevictable list.
102 * Reasons folio might not be evictable:
103 * 1. folio's mapping marked unevictable
104 * 2. One of the pages in the folio is part of an mlocked VMA
106 static inline bool folio_evictable(struct folio *folio)
110 /* Prevent address_space of inode and swap cache from being freed */
112 ret = !mapping_unevictable(folio_mapping(folio)) &&
113 !folio_test_mlocked(folio);
118 static inline bool page_evictable(struct page *page)
122 /* Prevent address_space of inode and swap cache from being freed */
124 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
130 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
133 static inline void set_page_refcounted(struct page *page)
135 VM_BUG_ON_PAGE(PageTail(page), page);
136 VM_BUG_ON_PAGE(page_ref_count(page), page);
137 set_page_count(page, 1);
140 extern unsigned long highest_memmap_pfn;
143 * Maximum number of reclaim retries without progress before the OOM
144 * killer is consider the only way forward.
146 #define MAX_RECLAIM_RETRIES 16
151 extern int isolate_lru_page(struct page *page);
152 extern void putback_lru_page(struct page *page);
153 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
158 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
165 * Structure for holding the mostly immutable allocation parameters passed
166 * between functions involved in allocations, including the alloc_pages*
167 * family of functions.
169 * nodemask, migratetype and highest_zoneidx are initialized only once in
170 * __alloc_pages() and then never change.
172 * zonelist, preferred_zone and highest_zoneidx are set first in
173 * __alloc_pages() for the fast path, and might be later changed
174 * in __alloc_pages_slowpath(). All other functions pass the whole structure
175 * by a const pointer.
177 struct alloc_context {
178 struct zonelist *zonelist;
179 nodemask_t *nodemask;
180 struct zoneref *preferred_zoneref;
184 * highest_zoneidx represents highest usable zone index of
185 * the allocation request. Due to the nature of the zone,
186 * memory on lower zone than the highest_zoneidx will be
187 * protected by lowmem_reserve[highest_zoneidx].
189 * highest_zoneidx is also used by reclaim/compaction to limit
190 * the target zone since higher zone than this index cannot be
191 * usable for this allocation request.
193 enum zone_type highest_zoneidx;
194 bool spread_dirty_pages;
198 * Locate the struct page for both the matching buddy in our
199 * pair (buddy1) and the combined O(n+1) page they form (page).
201 * 1) Any buddy B1 will have an order O twin B2 which satisfies
202 * the following equation:
204 * For example, if the starting buddy (buddy2) is #8 its order
206 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
208 * 2) Any buddy B will have an order O+1 parent P which
209 * satisfies the following equation:
212 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
214 static inline unsigned long
215 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
217 return page_pfn ^ (1 << order);
220 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
221 unsigned long end_pfn, struct zone *zone);
223 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
224 unsigned long end_pfn, struct zone *zone)
226 if (zone->contiguous)
227 return pfn_to_page(start_pfn);
229 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
232 extern int __isolate_free_page(struct page *page, unsigned int order);
233 extern void __putback_isolated_page(struct page *page, unsigned int order,
235 extern void memblock_free_pages(struct page *page, unsigned long pfn,
237 extern void __free_pages_core(struct page *page, unsigned int order);
238 extern void prep_compound_page(struct page *page, unsigned int order);
239 extern void post_alloc_hook(struct page *page, unsigned int order,
241 extern int user_min_free_kbytes;
243 extern void free_unref_page(struct page *page, unsigned int order);
244 extern void free_unref_page_list(struct list_head *list);
246 extern void zone_pcp_update(struct zone *zone, int cpu_online);
247 extern void zone_pcp_reset(struct zone *zone);
248 extern void zone_pcp_disable(struct zone *zone);
249 extern void zone_pcp_enable(struct zone *zone);
251 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
252 phys_addr_t min_addr,
253 int nid, bool exact_nid);
255 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
261 * compact_control is used to track pages being migrated and the free pages
262 * they are being migrated to during memory compaction. The free_pfn starts
263 * at the end of a zone and migrate_pfn begins at the start. Movable pages
264 * are moved to the end of a zone during a compaction run and the run
265 * completes when free_pfn <= migrate_pfn
267 struct compact_control {
268 struct list_head freepages; /* List of free pages to migrate to */
269 struct list_head migratepages; /* List of pages being migrated */
270 unsigned int nr_freepages; /* Number of isolated free pages */
271 unsigned int nr_migratepages; /* Number of pages to migrate */
272 unsigned long free_pfn; /* isolate_freepages search base */
274 * Acts as an in/out parameter to page isolation for migration.
275 * isolate_migratepages uses it as a search base.
276 * isolate_migratepages_block will update the value to the next pfn
277 * after the last isolated one.
279 unsigned long migrate_pfn;
280 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
282 unsigned long total_migrate_scanned;
283 unsigned long total_free_scanned;
284 unsigned short fast_search_fail;/* failures to use free list searches */
285 short search_order; /* order to start a fast search at */
286 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
287 int order; /* order a direct compactor needs */
288 int migratetype; /* migratetype of direct compactor */
289 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
290 const int highest_zoneidx; /* zone index of a direct compactor */
291 enum migrate_mode mode; /* Async or sync migration mode */
292 bool ignore_skip_hint; /* Scan blocks even if marked skip */
293 bool no_set_skip_hint; /* Don't mark blocks for skipping */
294 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
295 bool direct_compaction; /* False from kcompactd or /proc/... */
296 bool proactive_compaction; /* kcompactd proactive compaction */
297 bool whole_zone; /* Whole zone should/has been scanned */
298 bool contended; /* Signal lock or sched contention */
299 bool rescan; /* Rescanning the same pageblock */
300 bool alloc_contig; /* alloc_contig_range allocation */
304 * Used in direct compaction when a page should be taken from the freelists
305 * immediately when one is created during the free path.
307 struct capture_control {
308 struct compact_control *cc;
313 isolate_freepages_range(struct compact_control *cc,
314 unsigned long start_pfn, unsigned long end_pfn);
316 isolate_migratepages_range(struct compact_control *cc,
317 unsigned long low_pfn, unsigned long end_pfn);
319 int find_suitable_fallback(struct free_area *area, unsigned int order,
320 int migratetype, bool only_stealable, bool *can_steal);
323 * This function returns the order of a free page in the buddy system. In
324 * general, page_zone(page)->lock must be held by the caller to prevent the
325 * page from being allocated in parallel and returning garbage as the order.
326 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
327 * page cannot be allocated or merged in parallel. Alternatively, it must
328 * handle invalid values gracefully, and use buddy_order_unsafe() below.
330 static inline unsigned int buddy_order(struct page *page)
332 /* PageBuddy() must be checked by the caller */
333 return page_private(page);
337 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
338 * PageBuddy() should be checked first by the caller to minimize race window,
339 * and invalid values must be handled gracefully.
341 * READ_ONCE is used so that if the caller assigns the result into a local
342 * variable and e.g. tests it for valid range before using, the compiler cannot
343 * decide to remove the variable and inline the page_private(page) multiple
344 * times, potentially observing different values in the tests and the actual
347 #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
350 * These three helpers classifies VMAs for virtual memory accounting.
354 * Executable code area - executable, not writable, not stack
356 static inline bool is_exec_mapping(vm_flags_t flags)
358 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
362 * Stack area - automatically grows in one direction
364 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
365 * do_mmap() forbids all other combinations.
367 static inline bool is_stack_mapping(vm_flags_t flags)
369 return (flags & VM_STACK) == VM_STACK;
373 * Data area - private, writable, not stack
375 static inline bool is_data_mapping(vm_flags_t flags)
377 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
381 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
382 struct vm_area_struct *prev);
383 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
386 extern long populate_vma_page_range(struct vm_area_struct *vma,
387 unsigned long start, unsigned long end, int *locked);
388 extern long faultin_vma_page_range(struct vm_area_struct *vma,
389 unsigned long start, unsigned long end,
390 bool write, int *locked);
391 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
392 unsigned long start, unsigned long end);
393 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
395 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
399 * must be called with vma's mmap_lock held for read or write, and page locked.
401 extern void mlock_vma_page(struct page *page);
402 extern unsigned int munlock_vma_page(struct page *page);
404 extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
408 * Clear the page's PageMlocked(). This can be useful in a situation where
409 * we want to unconditionally remove a page from the pagecache -- e.g.,
410 * on truncation or freeing.
412 * It is legal to call this function for any page, mlocked or not.
413 * If called for a page that is still mapped by mlocked vmas, all we do
414 * is revert to lazy LRU behaviour -- semantics are not broken.
416 extern void clear_page_mlock(struct page *page);
418 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
421 * At what user virtual address is page expected in vma?
422 * Returns -EFAULT if all of the page is outside the range of vma.
423 * If page is a compound head, the entire compound page is considered.
425 static inline unsigned long
426 vma_address(struct page *page, struct vm_area_struct *vma)
429 unsigned long address;
431 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
432 pgoff = page_to_pgoff(page);
433 if (pgoff >= vma->vm_pgoff) {
434 address = vma->vm_start +
435 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
436 /* Check for address beyond vma (or wrapped through 0?) */
437 if (address < vma->vm_start || address >= vma->vm_end)
439 } else if (PageHead(page) &&
440 pgoff + compound_nr(page) - 1 >= vma->vm_pgoff) {
441 /* Test above avoids possibility of wrap to 0 on 32-bit */
442 address = vma->vm_start;
450 * Then at what user virtual address will none of the page be found in vma?
451 * Assumes that vma_address() already returned a good starting address.
452 * If page is a compound head, the entire compound page is considered.
454 static inline unsigned long
455 vma_address_end(struct page *page, struct vm_area_struct *vma)
458 unsigned long address;
460 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
461 pgoff = page_to_pgoff(page) + compound_nr(page);
462 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
463 /* Check for address beyond vma (or wrapped through 0?) */
464 if (address < vma->vm_start || address > vma->vm_end)
465 address = vma->vm_end;
469 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
472 int flags = vmf->flags;
478 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
479 * anything, so we only pin the file and drop the mmap_lock if only
480 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
482 if (fault_flag_allow_retry_first(flags) &&
483 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
484 fpin = get_file(vmf->vma->vm_file);
485 mmap_read_unlock(vmf->vma->vm_mm);
490 #else /* !CONFIG_MMU */
491 static inline void clear_page_mlock(struct page *page) { }
492 static inline void mlock_vma_page(struct page *page) { }
493 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
496 #endif /* !CONFIG_MMU */
499 * Return the mem_map entry representing the 'offset' subpage within
500 * the maximally aligned gigantic page 'base'. Handle any discontiguity
501 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
503 static inline struct page *mem_map_offset(struct page *base, int offset)
505 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
506 return nth_page(base, offset);
507 return base + offset;
511 * Iterator over all subpages within the maximally aligned gigantic
512 * page 'base'. Handle any discontiguity in the mem_map.
514 static inline struct page *mem_map_next(struct page *iter,
515 struct page *base, int offset)
517 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
518 unsigned long pfn = page_to_pfn(base) + offset;
521 return pfn_to_page(pfn);
526 /* Memory initialisation debug and verification */
533 #ifdef CONFIG_DEBUG_MEMORY_INIT
535 extern int mminit_loglevel;
537 #define mminit_dprintk(level, prefix, fmt, arg...) \
539 if (level < mminit_loglevel) { \
540 if (level <= MMINIT_WARNING) \
541 pr_warn("mminit::" prefix " " fmt, ##arg); \
543 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
547 extern void mminit_verify_pageflags_layout(void);
548 extern void mminit_verify_zonelist(void);
551 static inline void mminit_dprintk(enum mminit_level level,
552 const char *prefix, const char *fmt, ...)
556 static inline void mminit_verify_pageflags_layout(void)
560 static inline void mminit_verify_zonelist(void)
563 #endif /* CONFIG_DEBUG_MEMORY_INIT */
565 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
566 #if defined(CONFIG_SPARSEMEM)
567 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
568 unsigned long *end_pfn);
570 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
571 unsigned long *end_pfn)
574 #endif /* CONFIG_SPARSEMEM */
576 #define NODE_RECLAIM_NOSCAN -2
577 #define NODE_RECLAIM_FULL -1
578 #define NODE_RECLAIM_SOME 0
579 #define NODE_RECLAIM_SUCCESS 1
582 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
583 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
585 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
588 return NODE_RECLAIM_NOSCAN;
590 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
596 extern int hwpoison_filter(struct page *p);
598 extern u32 hwpoison_filter_dev_major;
599 extern u32 hwpoison_filter_dev_minor;
600 extern u64 hwpoison_filter_flags_mask;
601 extern u64 hwpoison_filter_flags_value;
602 extern u64 hwpoison_filter_memcg;
603 extern u32 hwpoison_filter_enable;
605 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
606 unsigned long, unsigned long,
607 unsigned long, unsigned long);
609 extern void set_pageblock_order(void);
610 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
611 struct list_head *page_list);
612 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
613 #define ALLOC_WMARK_MIN WMARK_MIN
614 #define ALLOC_WMARK_LOW WMARK_LOW
615 #define ALLOC_WMARK_HIGH WMARK_HIGH
616 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
618 /* Mask to get the watermark bits */
619 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
622 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
623 * cannot assume a reduced access to memory reserves is sufficient for
627 #define ALLOC_OOM 0x08
629 #define ALLOC_OOM ALLOC_NO_WATERMARKS
632 #define ALLOC_HARDER 0x10 /* try to alloc harder */
633 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
634 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
635 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
636 #ifdef CONFIG_ZONE_DMA32
637 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
639 #define ALLOC_NOFRAGMENT 0x0
641 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
644 struct tlbflush_unmap_batch;
648 * only for MM internal work items which do not depend on
649 * any allocations or locks which might depend on allocations
651 extern struct workqueue_struct *mm_percpu_wq;
653 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
654 void try_to_unmap_flush(void);
655 void try_to_unmap_flush_dirty(void);
656 void flush_tlb_batched_pending(struct mm_struct *mm);
658 static inline void try_to_unmap_flush(void)
661 static inline void try_to_unmap_flush_dirty(void)
664 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
667 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
669 extern const struct trace_print_flags pageflag_names[];
670 extern const struct trace_print_flags vmaflag_names[];
671 extern const struct trace_print_flags gfpflag_names[];
673 static inline bool is_migrate_highatomic(enum migratetype migratetype)
675 return migratetype == MIGRATE_HIGHATOMIC;
678 static inline bool is_migrate_highatomic_page(struct page *page)
680 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
683 void setup_zone_pageset(struct zone *zone);
685 struct migration_target_control {
686 int nid; /* preferred node id */
695 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
696 pgprot_t prot, struct page **pages, unsigned int page_shift);
699 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
700 pgprot_t prot, struct page **pages, unsigned int page_shift)
706 void vunmap_range_noflush(unsigned long start, unsigned long end);
708 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
709 unsigned long addr, int page_nid, int *flags);
711 #endif /* __MM_INTERNAL_H */