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>
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 vm_fault_t do_swap_page(struct vm_fault *vmf);
45 void folio_rotate_reclaimable(struct folio *folio);
46 bool __folio_end_writeback(struct folio *folio);
48 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
49 unsigned long floor, unsigned long ceiling);
51 static inline bool can_madv_lru_vma(struct vm_area_struct *vma)
53 return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
56 void unmap_page_range(struct mmu_gather *tlb,
57 struct vm_area_struct *vma,
58 unsigned long addr, unsigned long end,
59 struct zap_details *details);
61 void do_page_cache_ra(struct readahead_control *, unsigned long nr_to_read,
62 unsigned long lookahead_size);
63 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
64 static inline void force_page_cache_readahead(struct address_space *mapping,
65 struct file *file, pgoff_t index, unsigned long nr_to_read)
67 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
68 force_page_cache_ra(&ractl, nr_to_read);
71 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
72 pgoff_t end, struct pagevec *pvec, pgoff_t *indices);
75 * folio_evictable - Test whether a folio is evictable.
76 * @folio: The folio to test.
78 * Test whether @folio is evictable -- i.e., should be placed on
79 * active/inactive lists vs unevictable list.
81 * Reasons folio might not be evictable:
82 * 1. folio's mapping marked unevictable
83 * 2. One of the pages in the folio is part of an mlocked VMA
85 static inline bool folio_evictable(struct folio *folio)
89 /* Prevent address_space of inode and swap cache from being freed */
91 ret = !mapping_unevictable(folio_mapping(folio)) &&
92 !folio_test_mlocked(folio);
97 static inline bool page_evictable(struct page *page)
101 /* Prevent address_space of inode and swap cache from being freed */
103 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
109 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
112 static inline void set_page_refcounted(struct page *page)
114 VM_BUG_ON_PAGE(PageTail(page), page);
115 VM_BUG_ON_PAGE(page_ref_count(page), page);
116 set_page_count(page, 1);
119 extern unsigned long highest_memmap_pfn;
122 * Maximum number of reclaim retries without progress before the OOM
123 * killer is consider the only way forward.
125 #define MAX_RECLAIM_RETRIES 16
130 extern int isolate_lru_page(struct page *page);
131 extern void putback_lru_page(struct page *page);
136 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
139 * in mm/memcontrol.c:
141 extern bool cgroup_memory_nokmem;
148 * Structure for holding the mostly immutable allocation parameters passed
149 * between functions involved in allocations, including the alloc_pages*
150 * family of functions.
152 * nodemask, migratetype and highest_zoneidx are initialized only once in
153 * __alloc_pages() and then never change.
155 * zonelist, preferred_zone and highest_zoneidx are set first in
156 * __alloc_pages() for the fast path, and might be later changed
157 * in __alloc_pages_slowpath(). All other functions pass the whole structure
158 * by a const pointer.
160 struct alloc_context {
161 struct zonelist *zonelist;
162 nodemask_t *nodemask;
163 struct zoneref *preferred_zoneref;
167 * highest_zoneidx represents highest usable zone index of
168 * the allocation request. Due to the nature of the zone,
169 * memory on lower zone than the highest_zoneidx will be
170 * protected by lowmem_reserve[highest_zoneidx].
172 * highest_zoneidx is also used by reclaim/compaction to limit
173 * the target zone since higher zone than this index cannot be
174 * usable for this allocation request.
176 enum zone_type highest_zoneidx;
177 bool spread_dirty_pages;
181 * Locate the struct page for both the matching buddy in our
182 * pair (buddy1) and the combined O(n+1) page they form (page).
184 * 1) Any buddy B1 will have an order O twin B2 which satisfies
185 * the following equation:
187 * For example, if the starting buddy (buddy2) is #8 its order
189 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
191 * 2) Any buddy B will have an order O+1 parent P which
192 * satisfies the following equation:
195 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
197 static inline unsigned long
198 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
200 return page_pfn ^ (1 << order);
203 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
204 unsigned long end_pfn, struct zone *zone);
206 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
207 unsigned long end_pfn, struct zone *zone)
209 if (zone->contiguous)
210 return pfn_to_page(start_pfn);
212 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
215 extern int __isolate_free_page(struct page *page, unsigned int order);
216 extern void __putback_isolated_page(struct page *page, unsigned int order,
218 extern void memblock_free_pages(struct page *page, unsigned long pfn,
220 extern void __free_pages_core(struct page *page, unsigned int order);
221 extern void prep_compound_page(struct page *page, unsigned int order);
222 extern void post_alloc_hook(struct page *page, unsigned int order,
224 extern int user_min_free_kbytes;
226 extern void free_unref_page(struct page *page, unsigned int order);
227 extern void free_unref_page_list(struct list_head *list);
229 extern void zone_pcp_update(struct zone *zone, int cpu_online);
230 extern void zone_pcp_reset(struct zone *zone);
231 extern void zone_pcp_disable(struct zone *zone);
232 extern void zone_pcp_enable(struct zone *zone);
234 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
235 phys_addr_t min_addr,
236 int nid, bool exact_nid);
238 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
244 * compact_control is used to track pages being migrated and the free pages
245 * they are being migrated to during memory compaction. The free_pfn starts
246 * at the end of a zone and migrate_pfn begins at the start. Movable pages
247 * are moved to the end of a zone during a compaction run and the run
248 * completes when free_pfn <= migrate_pfn
250 struct compact_control {
251 struct list_head freepages; /* List of free pages to migrate to */
252 struct list_head migratepages; /* List of pages being migrated */
253 unsigned int nr_freepages; /* Number of isolated free pages */
254 unsigned int nr_migratepages; /* Number of pages to migrate */
255 unsigned long free_pfn; /* isolate_freepages search base */
257 * Acts as an in/out parameter to page isolation for migration.
258 * isolate_migratepages uses it as a search base.
259 * isolate_migratepages_block will update the value to the next pfn
260 * after the last isolated one.
262 unsigned long migrate_pfn;
263 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
265 unsigned long total_migrate_scanned;
266 unsigned long total_free_scanned;
267 unsigned short fast_search_fail;/* failures to use free list searches */
268 short search_order; /* order to start a fast search at */
269 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
270 int order; /* order a direct compactor needs */
271 int migratetype; /* migratetype of direct compactor */
272 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
273 const int highest_zoneidx; /* zone index of a direct compactor */
274 enum migrate_mode mode; /* Async or sync migration mode */
275 bool ignore_skip_hint; /* Scan blocks even if marked skip */
276 bool no_set_skip_hint; /* Don't mark blocks for skipping */
277 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
278 bool direct_compaction; /* False from kcompactd or /proc/... */
279 bool proactive_compaction; /* kcompactd proactive compaction */
280 bool whole_zone; /* Whole zone should/has been scanned */
281 bool contended; /* Signal lock or sched contention */
282 bool rescan; /* Rescanning the same pageblock */
283 bool alloc_contig; /* alloc_contig_range allocation */
287 * Used in direct compaction when a page should be taken from the freelists
288 * immediately when one is created during the free path.
290 struct capture_control {
291 struct compact_control *cc;
296 isolate_freepages_range(struct compact_control *cc,
297 unsigned long start_pfn, unsigned long end_pfn);
299 isolate_migratepages_range(struct compact_control *cc,
300 unsigned long low_pfn, unsigned long end_pfn);
302 int find_suitable_fallback(struct free_area *area, unsigned int order,
303 int migratetype, bool only_stealable, bool *can_steal);
306 * This function returns the order of a free page in the buddy system. In
307 * general, page_zone(page)->lock must be held by the caller to prevent the
308 * page from being allocated in parallel and returning garbage as the order.
309 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
310 * page cannot be allocated or merged in parallel. Alternatively, it must
311 * handle invalid values gracefully, and use buddy_order_unsafe() below.
313 static inline unsigned int buddy_order(struct page *page)
315 /* PageBuddy() must be checked by the caller */
316 return page_private(page);
320 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
321 * PageBuddy() should be checked first by the caller to minimize race window,
322 * and invalid values must be handled gracefully.
324 * READ_ONCE is used so that if the caller assigns the result into a local
325 * variable and e.g. tests it for valid range before using, the compiler cannot
326 * decide to remove the variable and inline the page_private(page) multiple
327 * times, potentially observing different values in the tests and the actual
330 #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
333 * These three helpers classifies VMAs for virtual memory accounting.
337 * Executable code area - executable, not writable, not stack
339 static inline bool is_exec_mapping(vm_flags_t flags)
341 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
345 * Stack area - automatically grows in one direction
347 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
348 * do_mmap() forbids all other combinations.
350 static inline bool is_stack_mapping(vm_flags_t flags)
352 return (flags & VM_STACK) == VM_STACK;
356 * Data area - private, writable, not stack
358 static inline bool is_data_mapping(vm_flags_t flags)
360 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
364 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
365 struct vm_area_struct *prev);
366 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
369 extern long populate_vma_page_range(struct vm_area_struct *vma,
370 unsigned long start, unsigned long end, int *locked);
371 extern long faultin_vma_page_range(struct vm_area_struct *vma,
372 unsigned long start, unsigned long end,
373 bool write, int *locked);
374 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
375 unsigned long start, unsigned long end);
376 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
378 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
382 * must be called with vma's mmap_lock held for read or write, and page locked.
384 extern void mlock_vma_page(struct page *page);
385 extern unsigned int munlock_vma_page(struct page *page);
387 extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
391 * Clear the page's PageMlocked(). This can be useful in a situation where
392 * we want to unconditionally remove a page from the pagecache -- e.g.,
393 * on truncation or freeing.
395 * It is legal to call this function for any page, mlocked or not.
396 * If called for a page that is still mapped by mlocked vmas, all we do
397 * is revert to lazy LRU behaviour -- semantics are not broken.
399 extern void clear_page_mlock(struct page *page);
401 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
404 * At what user virtual address is page expected in vma?
405 * Returns -EFAULT if all of the page is outside the range of vma.
406 * If page is a compound head, the entire compound page is considered.
408 static inline unsigned long
409 vma_address(struct page *page, struct vm_area_struct *vma)
412 unsigned long address;
414 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
415 pgoff = page_to_pgoff(page);
416 if (pgoff >= vma->vm_pgoff) {
417 address = vma->vm_start +
418 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
419 /* Check for address beyond vma (or wrapped through 0?) */
420 if (address < vma->vm_start || address >= vma->vm_end)
422 } else if (PageHead(page) &&
423 pgoff + compound_nr(page) - 1 >= vma->vm_pgoff) {
424 /* Test above avoids possibility of wrap to 0 on 32-bit */
425 address = vma->vm_start;
433 * Then at what user virtual address will none of the page be found in vma?
434 * Assumes that vma_address() already returned a good starting address.
435 * If page is a compound head, the entire compound page is considered.
437 static inline unsigned long
438 vma_address_end(struct page *page, struct vm_area_struct *vma)
441 unsigned long address;
443 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
444 pgoff = page_to_pgoff(page) + compound_nr(page);
445 address = vma->vm_start + ((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)
448 address = vma->vm_end;
452 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
455 int flags = vmf->flags;
461 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
462 * anything, so we only pin the file and drop the mmap_lock if only
463 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
465 if (fault_flag_allow_retry_first(flags) &&
466 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
467 fpin = get_file(vmf->vma->vm_file);
468 mmap_read_unlock(vmf->vma->vm_mm);
473 #else /* !CONFIG_MMU */
474 static inline void clear_page_mlock(struct page *page) { }
475 static inline void mlock_vma_page(struct page *page) { }
476 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
479 #endif /* !CONFIG_MMU */
482 * Return the mem_map entry representing the 'offset' subpage within
483 * the maximally aligned gigantic page 'base'. Handle any discontiguity
484 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
486 static inline struct page *mem_map_offset(struct page *base, int offset)
488 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
489 return nth_page(base, offset);
490 return base + offset;
494 * Iterator over all subpages within the maximally aligned gigantic
495 * page 'base'. Handle any discontiguity in the mem_map.
497 static inline struct page *mem_map_next(struct page *iter,
498 struct page *base, int offset)
500 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
501 unsigned long pfn = page_to_pfn(base) + offset;
504 return pfn_to_page(pfn);
509 /* Memory initialisation debug and verification */
516 #ifdef CONFIG_DEBUG_MEMORY_INIT
518 extern int mminit_loglevel;
520 #define mminit_dprintk(level, prefix, fmt, arg...) \
522 if (level < mminit_loglevel) { \
523 if (level <= MMINIT_WARNING) \
524 pr_warn("mminit::" prefix " " fmt, ##arg); \
526 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
530 extern void mminit_verify_pageflags_layout(void);
531 extern void mminit_verify_zonelist(void);
534 static inline void mminit_dprintk(enum mminit_level level,
535 const char *prefix, const char *fmt, ...)
539 static inline void mminit_verify_pageflags_layout(void)
543 static inline void mminit_verify_zonelist(void)
546 #endif /* CONFIG_DEBUG_MEMORY_INIT */
548 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
549 #if defined(CONFIG_SPARSEMEM)
550 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
551 unsigned long *end_pfn);
553 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
554 unsigned long *end_pfn)
557 #endif /* CONFIG_SPARSEMEM */
559 #define NODE_RECLAIM_NOSCAN -2
560 #define NODE_RECLAIM_FULL -1
561 #define NODE_RECLAIM_SOME 0
562 #define NODE_RECLAIM_SUCCESS 1
565 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
566 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
568 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
571 return NODE_RECLAIM_NOSCAN;
573 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
579 extern int hwpoison_filter(struct page *p);
581 extern u32 hwpoison_filter_dev_major;
582 extern u32 hwpoison_filter_dev_minor;
583 extern u64 hwpoison_filter_flags_mask;
584 extern u64 hwpoison_filter_flags_value;
585 extern u64 hwpoison_filter_memcg;
586 extern u32 hwpoison_filter_enable;
588 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
589 unsigned long, unsigned long,
590 unsigned long, unsigned long);
592 extern void set_pageblock_order(void);
593 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
594 struct list_head *page_list);
595 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
596 #define ALLOC_WMARK_MIN WMARK_MIN
597 #define ALLOC_WMARK_LOW WMARK_LOW
598 #define ALLOC_WMARK_HIGH WMARK_HIGH
599 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
601 /* Mask to get the watermark bits */
602 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
605 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
606 * cannot assume a reduced access to memory reserves is sufficient for
610 #define ALLOC_OOM 0x08
612 #define ALLOC_OOM ALLOC_NO_WATERMARKS
615 #define ALLOC_HARDER 0x10 /* try to alloc harder */
616 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
617 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
618 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
619 #ifdef CONFIG_ZONE_DMA32
620 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
622 #define ALLOC_NOFRAGMENT 0x0
624 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
627 struct tlbflush_unmap_batch;
631 * only for MM internal work items which do not depend on
632 * any allocations or locks which might depend on allocations
634 extern struct workqueue_struct *mm_percpu_wq;
636 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
637 void try_to_unmap_flush(void);
638 void try_to_unmap_flush_dirty(void);
639 void flush_tlb_batched_pending(struct mm_struct *mm);
641 static inline void try_to_unmap_flush(void)
644 static inline void try_to_unmap_flush_dirty(void)
647 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
650 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
652 extern const struct trace_print_flags pageflag_names[];
653 extern const struct trace_print_flags vmaflag_names[];
654 extern const struct trace_print_flags gfpflag_names[];
656 static inline bool is_migrate_highatomic(enum migratetype migratetype)
658 return migratetype == MIGRATE_HIGHATOMIC;
661 static inline bool is_migrate_highatomic_page(struct page *page)
663 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
666 void setup_zone_pageset(struct zone *zone);
668 struct migration_target_control {
669 int nid; /* preferred node id */
678 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
679 pgprot_t prot, struct page **pages, unsigned int page_shift);
682 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
683 pgprot_t prot, struct page **pages, unsigned int page_shift)
689 void vunmap_range_noflush(unsigned long start, unsigned long end);
691 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
692 unsigned long addr, int page_nid, int *flags);
694 #endif /* __MM_INTERNAL_H */