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 vm_fault_t do_swap_page(struct vm_fault *vmf);
39 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
40 unsigned long floor, unsigned long ceiling);
42 static inline bool can_madv_lru_vma(struct vm_area_struct *vma)
44 return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
47 void unmap_page_range(struct mmu_gather *tlb,
48 struct vm_area_struct *vma,
49 unsigned long addr, unsigned long end,
50 struct zap_details *details);
52 void do_page_cache_ra(struct readahead_control *, unsigned long nr_to_read,
53 unsigned long lookahead_size);
54 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
55 static inline void force_page_cache_readahead(struct address_space *mapping,
56 struct file *file, pgoff_t index, unsigned long nr_to_read)
58 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
59 force_page_cache_ra(&ractl, nr_to_read);
62 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
63 pgoff_t end, struct pagevec *pvec, pgoff_t *indices);
66 * page_evictable - test whether a page is evictable
67 * @page: the page to test
69 * Test whether page is evictable--i.e., should be placed on active/inactive
70 * lists vs unevictable list.
72 * Reasons page might not be evictable:
73 * (1) page's mapping marked unevictable
74 * (2) page is part of an mlocked VMA
77 static inline bool page_evictable(struct page *page)
81 /* Prevent address_space of inode and swap cache from being freed */
83 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
89 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
92 static inline void set_page_refcounted(struct page *page)
94 VM_BUG_ON_PAGE(PageTail(page), page);
95 VM_BUG_ON_PAGE(page_ref_count(page), page);
96 set_page_count(page, 1);
100 * When kernel touch the user page, the user page may be have been marked
101 * poison but still mapped in user space, if without this page, the kernel
102 * can guarantee the data integrity and operation success, the kernel is
103 * better to check the posion status and avoid touching it, be good not to
104 * panic, coredump for process fatal signal is a sample case matching this
105 * scenario. Or if kernel can't guarantee the data integrity, it's better
106 * not to call this function, let kernel touch the poison page and get to
109 static inline bool is_page_poisoned(struct page *page)
111 if (PageHWPoison(page))
113 else if (PageHuge(page) && PageHWPoison(compound_head(page)))
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);
143 * Structure for holding the mostly immutable allocation parameters passed
144 * between functions involved in allocations, including the alloc_pages*
145 * family of functions.
147 * nodemask, migratetype and highest_zoneidx are initialized only once in
148 * __alloc_pages() and then never change.
150 * zonelist, preferred_zone and highest_zoneidx are set first in
151 * __alloc_pages() for the fast path, and might be later changed
152 * in __alloc_pages_slowpath(). All other functions pass the whole structure
153 * by a const pointer.
155 struct alloc_context {
156 struct zonelist *zonelist;
157 nodemask_t *nodemask;
158 struct zoneref *preferred_zoneref;
162 * highest_zoneidx represents highest usable zone index of
163 * the allocation request. Due to the nature of the zone,
164 * memory on lower zone than the highest_zoneidx will be
165 * protected by lowmem_reserve[highest_zoneidx].
167 * highest_zoneidx is also used by reclaim/compaction to limit
168 * the target zone since higher zone than this index cannot be
169 * usable for this allocation request.
171 enum zone_type highest_zoneidx;
172 bool spread_dirty_pages;
176 * Locate the struct page for both the matching buddy in our
177 * pair (buddy1) and the combined O(n+1) page they form (page).
179 * 1) Any buddy B1 will have an order O twin B2 which satisfies
180 * the following equation:
182 * For example, if the starting buddy (buddy2) is #8 its order
184 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
186 * 2) Any buddy B will have an order O+1 parent P which
187 * satisfies the following equation:
190 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
192 static inline unsigned long
193 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
195 return page_pfn ^ (1 << order);
198 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
199 unsigned long end_pfn, struct zone *zone);
201 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
202 unsigned long end_pfn, struct zone *zone)
204 if (zone->contiguous)
205 return pfn_to_page(start_pfn);
207 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
210 extern int __isolate_free_page(struct page *page, unsigned int order);
211 extern void __putback_isolated_page(struct page *page, unsigned int order,
213 extern void memblock_free_pages(struct page *page, unsigned long pfn,
215 extern void __free_pages_core(struct page *page, unsigned int order);
216 extern void prep_compound_page(struct page *page, unsigned int order);
217 extern void post_alloc_hook(struct page *page, unsigned int order,
219 extern int user_min_free_kbytes;
221 extern void free_unref_page(struct page *page);
222 extern void free_unref_page_list(struct list_head *list);
224 extern void zone_pcp_update(struct zone *zone);
225 extern void zone_pcp_reset(struct zone *zone);
226 extern void zone_pcp_disable(struct zone *zone);
227 extern void zone_pcp_enable(struct zone *zone);
229 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
235 * compact_control is used to track pages being migrated and the free pages
236 * they are being migrated to during memory compaction. The free_pfn starts
237 * at the end of a zone and migrate_pfn begins at the start. Movable pages
238 * are moved to the end of a zone during a compaction run and the run
239 * completes when free_pfn <= migrate_pfn
241 struct compact_control {
242 struct list_head freepages; /* List of free pages to migrate to */
243 struct list_head migratepages; /* List of pages being migrated */
244 unsigned int nr_freepages; /* Number of isolated free pages */
245 unsigned int nr_migratepages; /* Number of pages to migrate */
246 unsigned long free_pfn; /* isolate_freepages search base */
247 unsigned long migrate_pfn; /* isolate_migratepages search base */
248 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
250 unsigned long total_migrate_scanned;
251 unsigned long total_free_scanned;
252 unsigned short fast_search_fail;/* failures to use free list searches */
253 short search_order; /* order to start a fast search at */
254 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
255 int order; /* order a direct compactor needs */
256 int migratetype; /* migratetype of direct compactor */
257 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
258 const int highest_zoneidx; /* zone index of a direct compactor */
259 enum migrate_mode mode; /* Async or sync migration mode */
260 bool ignore_skip_hint; /* Scan blocks even if marked skip */
261 bool no_set_skip_hint; /* Don't mark blocks for skipping */
262 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
263 bool direct_compaction; /* False from kcompactd or /proc/... */
264 bool proactive_compaction; /* kcompactd proactive compaction */
265 bool whole_zone; /* Whole zone should/has been scanned */
266 bool contended; /* Signal lock or sched contention */
267 bool rescan; /* Rescanning the same pageblock */
268 bool alloc_contig; /* alloc_contig_range allocation */
272 * Used in direct compaction when a page should be taken from the freelists
273 * immediately when one is created during the free path.
275 struct capture_control {
276 struct compact_control *cc;
281 isolate_freepages_range(struct compact_control *cc,
282 unsigned long start_pfn, unsigned long end_pfn);
284 isolate_migratepages_range(struct compact_control *cc,
285 unsigned long low_pfn, unsigned long end_pfn);
286 int find_suitable_fallback(struct free_area *area, unsigned int order,
287 int migratetype, bool only_stealable, bool *can_steal);
292 * This function returns the order of a free page in the buddy system. In
293 * general, page_zone(page)->lock must be held by the caller to prevent the
294 * page from being allocated in parallel and returning garbage as the order.
295 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
296 * page cannot be allocated or merged in parallel. Alternatively, it must
297 * handle invalid values gracefully, and use buddy_order_unsafe() below.
299 static inline unsigned int buddy_order(struct page *page)
301 /* PageBuddy() must be checked by the caller */
302 return page_private(page);
306 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
307 * PageBuddy() should be checked first by the caller to minimize race window,
308 * and invalid values must be handled gracefully.
310 * READ_ONCE is used so that if the caller assigns the result into a local
311 * variable and e.g. tests it for valid range before using, the compiler cannot
312 * decide to remove the variable and inline the page_private(page) multiple
313 * times, potentially observing different values in the tests and the actual
316 #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
319 * These three helpers classifies VMAs for virtual memory accounting.
323 * Executable code area - executable, not writable, not stack
325 static inline bool is_exec_mapping(vm_flags_t flags)
327 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
331 * Stack area - atomatically grows in one direction
333 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
334 * do_mmap() forbids all other combinations.
336 static inline bool is_stack_mapping(vm_flags_t flags)
338 return (flags & VM_STACK) == VM_STACK;
342 * Data area - private, writable, not stack
344 static inline bool is_data_mapping(vm_flags_t flags)
346 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
350 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
351 struct vm_area_struct *prev);
352 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
355 extern long populate_vma_page_range(struct vm_area_struct *vma,
356 unsigned long start, unsigned long end, int *nonblocking);
357 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
358 unsigned long start, unsigned long end);
359 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
361 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
365 * must be called with vma's mmap_lock held for read or write, and page locked.
367 extern void mlock_vma_page(struct page *page);
368 extern unsigned int munlock_vma_page(struct page *page);
371 * Clear the page's PageMlocked(). This can be useful in a situation where
372 * we want to unconditionally remove a page from the pagecache -- e.g.,
373 * on truncation or freeing.
375 * It is legal to call this function for any page, mlocked or not.
376 * If called for a page that is still mapped by mlocked vmas, all we do
377 * is revert to lazy LRU behaviour -- semantics are not broken.
379 extern void clear_page_mlock(struct page *page);
382 * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
383 * (because that does not go through the full procedure of migration ptes):
384 * to migrate the Mlocked page flag; update statistics.
386 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
388 if (TestClearPageMlocked(page)) {
389 int nr_pages = thp_nr_pages(page);
391 /* Holding pmd lock, no change in irq context: __mod is safe */
392 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
393 SetPageMlocked(newpage);
394 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
398 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
401 * At what user virtual address is page expected in @vma?
403 static inline unsigned long
404 __vma_address(struct page *page, struct vm_area_struct *vma)
406 pgoff_t pgoff = page_to_pgoff(page);
407 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
410 static inline unsigned long
411 vma_address(struct page *page, struct vm_area_struct *vma)
413 unsigned long start, end;
415 start = __vma_address(page, vma);
416 end = start + thp_size(page) - PAGE_SIZE;
418 /* page should be within @vma mapping range */
419 VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma);
421 return max(start, vma->vm_start);
424 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
427 int flags = vmf->flags;
433 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
434 * anything, so we only pin the file and drop the mmap_lock if only
435 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
437 if (fault_flag_allow_retry_first(flags) &&
438 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
439 fpin = get_file(vmf->vma->vm_file);
440 mmap_read_unlock(vmf->vma->vm_mm);
445 #else /* !CONFIG_MMU */
446 static inline void clear_page_mlock(struct page *page) { }
447 static inline void mlock_vma_page(struct page *page) { }
448 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
449 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
452 #endif /* !CONFIG_MMU */
455 * Return the mem_map entry representing the 'offset' subpage within
456 * the maximally aligned gigantic page 'base'. Handle any discontiguity
457 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
459 static inline struct page *mem_map_offset(struct page *base, int offset)
461 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
462 return nth_page(base, offset);
463 return base + offset;
467 * Iterator over all subpages within the maximally aligned gigantic
468 * page 'base'. Handle any discontiguity in the mem_map.
470 static inline struct page *mem_map_next(struct page *iter,
471 struct page *base, int offset)
473 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
474 unsigned long pfn = page_to_pfn(base) + offset;
477 return pfn_to_page(pfn);
482 /* Memory initialisation debug and verification */
489 #ifdef CONFIG_DEBUG_MEMORY_INIT
491 extern int mminit_loglevel;
493 #define mminit_dprintk(level, prefix, fmt, arg...) \
495 if (level < mminit_loglevel) { \
496 if (level <= MMINIT_WARNING) \
497 pr_warn("mminit::" prefix " " fmt, ##arg); \
499 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
503 extern void mminit_verify_pageflags_layout(void);
504 extern void mminit_verify_zonelist(void);
507 static inline void mminit_dprintk(enum mminit_level level,
508 const char *prefix, const char *fmt, ...)
512 static inline void mminit_verify_pageflags_layout(void)
516 static inline void mminit_verify_zonelist(void)
519 #endif /* CONFIG_DEBUG_MEMORY_INIT */
521 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
522 #if defined(CONFIG_SPARSEMEM)
523 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
524 unsigned long *end_pfn);
526 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
527 unsigned long *end_pfn)
530 #endif /* CONFIG_SPARSEMEM */
532 #define NODE_RECLAIM_NOSCAN -2
533 #define NODE_RECLAIM_FULL -1
534 #define NODE_RECLAIM_SOME 0
535 #define NODE_RECLAIM_SUCCESS 1
538 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
540 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
543 return NODE_RECLAIM_NOSCAN;
547 extern int hwpoison_filter(struct page *p);
549 extern u32 hwpoison_filter_dev_major;
550 extern u32 hwpoison_filter_dev_minor;
551 extern u64 hwpoison_filter_flags_mask;
552 extern u64 hwpoison_filter_flags_value;
553 extern u64 hwpoison_filter_memcg;
554 extern u32 hwpoison_filter_enable;
556 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
557 unsigned long, unsigned long,
558 unsigned long, unsigned long);
560 extern void set_pageblock_order(void);
561 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
562 struct list_head *page_list);
563 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
564 #define ALLOC_WMARK_MIN WMARK_MIN
565 #define ALLOC_WMARK_LOW WMARK_LOW
566 #define ALLOC_WMARK_HIGH WMARK_HIGH
567 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
569 /* Mask to get the watermark bits */
570 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
573 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
574 * cannot assume a reduced access to memory reserves is sufficient for
578 #define ALLOC_OOM 0x08
580 #define ALLOC_OOM ALLOC_NO_WATERMARKS
583 #define ALLOC_HARDER 0x10 /* try to alloc harder */
584 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
585 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
586 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
587 #ifdef CONFIG_ZONE_DMA32
588 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
590 #define ALLOC_NOFRAGMENT 0x0
592 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
595 struct tlbflush_unmap_batch;
599 * only for MM internal work items which do not depend on
600 * any allocations or locks which might depend on allocations
602 extern struct workqueue_struct *mm_percpu_wq;
604 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
605 void try_to_unmap_flush(void);
606 void try_to_unmap_flush_dirty(void);
607 void flush_tlb_batched_pending(struct mm_struct *mm);
609 static inline void try_to_unmap_flush(void)
612 static inline void try_to_unmap_flush_dirty(void)
615 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
618 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
620 extern const struct trace_print_flags pageflag_names[];
621 extern const struct trace_print_flags vmaflag_names[];
622 extern const struct trace_print_flags gfpflag_names[];
624 static inline bool is_migrate_highatomic(enum migratetype migratetype)
626 return migratetype == MIGRATE_HIGHATOMIC;
629 static inline bool is_migrate_highatomic_page(struct page *page)
631 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
634 void setup_zone_pageset(struct zone *zone);
636 struct migration_target_control {
637 int nid; /* preferred node id */
646 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
647 pgprot_t prot, struct page **pages, unsigned int page_shift);
650 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
651 pgprot_t prot, struct page **pages, unsigned int page_shift)
657 void vunmap_range_noflush(unsigned long start, unsigned long end);
659 #endif /* __MM_INTERNAL_H */