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 void __acct_reclaim_writeback(pg_data_t *pgdat, struct page *page,
39 static inline void acct_reclaim_writeback(struct page *page)
41 pg_data_t *pgdat = page_pgdat(page);
42 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
45 __acct_reclaim_writeback(pgdat, page, nr_throttled);
48 static inline void wake_throttle_isolated(pg_data_t *pgdat)
50 wait_queue_head_t *wqh;
52 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
53 if (waitqueue_active(wqh))
57 vm_fault_t do_swap_page(struct vm_fault *vmf);
59 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
60 unsigned long floor, unsigned long ceiling);
61 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
63 static inline bool can_madv_lru_vma(struct vm_area_struct *vma)
65 return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
68 void unmap_page_range(struct mmu_gather *tlb,
69 struct vm_area_struct *vma,
70 unsigned long addr, unsigned long end,
71 struct zap_details *details);
73 void do_page_cache_ra(struct readahead_control *, unsigned long nr_to_read,
74 unsigned long lookahead_size);
75 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
76 static inline void force_page_cache_readahead(struct address_space *mapping,
77 struct file *file, pgoff_t index, unsigned long nr_to_read)
79 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
80 force_page_cache_ra(&ractl, nr_to_read);
83 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
84 pgoff_t end, struct pagevec *pvec, pgoff_t *indices);
87 * page_evictable - test whether a page is evictable
88 * @page: the page to test
90 * Test whether page is evictable--i.e., should be placed on active/inactive
91 * lists vs unevictable list.
93 * Reasons page might not be evictable:
94 * (1) page's mapping marked unevictable
95 * (2) page is part of an mlocked VMA
98 static inline bool page_evictable(struct page *page)
102 /* Prevent address_space of inode and swap cache from being freed */
104 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
110 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
113 static inline void set_page_refcounted(struct page *page)
115 VM_BUG_ON_PAGE(PageTail(page), page);
116 VM_BUG_ON_PAGE(page_ref_count(page), page);
117 set_page_count(page, 1);
120 extern unsigned long highest_memmap_pfn;
123 * Maximum number of reclaim retries without progress before the OOM
124 * killer is consider the only way forward.
126 #define MAX_RECLAIM_RETRIES 16
131 extern int isolate_lru_page(struct page *page);
132 extern void putback_lru_page(struct page *page);
133 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
138 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
141 * in mm/memcontrol.c:
143 extern bool cgroup_memory_nokmem;
150 * Structure for holding the mostly immutable allocation parameters passed
151 * between functions involved in allocations, including the alloc_pages*
152 * family of functions.
154 * nodemask, migratetype and highest_zoneidx are initialized only once in
155 * __alloc_pages() and then never change.
157 * zonelist, preferred_zone and highest_zoneidx are set first in
158 * __alloc_pages() for the fast path, and might be later changed
159 * in __alloc_pages_slowpath(). All other functions pass the whole structure
160 * by a const pointer.
162 struct alloc_context {
163 struct zonelist *zonelist;
164 nodemask_t *nodemask;
165 struct zoneref *preferred_zoneref;
169 * highest_zoneidx represents highest usable zone index of
170 * the allocation request. Due to the nature of the zone,
171 * memory on lower zone than the highest_zoneidx will be
172 * protected by lowmem_reserve[highest_zoneidx].
174 * highest_zoneidx is also used by reclaim/compaction to limit
175 * the target zone since higher zone than this index cannot be
176 * usable for this allocation request.
178 enum zone_type highest_zoneidx;
179 bool spread_dirty_pages;
183 * Locate the struct page for both the matching buddy in our
184 * pair (buddy1) and the combined O(n+1) page they form (page).
186 * 1) Any buddy B1 will have an order O twin B2 which satisfies
187 * the following equation:
189 * For example, if the starting buddy (buddy2) is #8 its order
191 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
193 * 2) Any buddy B will have an order O+1 parent P which
194 * satisfies the following equation:
197 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
199 static inline unsigned long
200 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
202 return page_pfn ^ (1 << order);
205 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
206 unsigned long end_pfn, struct zone *zone);
208 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
209 unsigned long end_pfn, struct zone *zone)
211 if (zone->contiguous)
212 return pfn_to_page(start_pfn);
214 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
217 extern int __isolate_free_page(struct page *page, unsigned int order);
218 extern void __putback_isolated_page(struct page *page, unsigned int order,
220 extern void memblock_free_pages(struct page *page, unsigned long pfn,
222 extern void __free_pages_core(struct page *page, unsigned int order);
223 extern void prep_compound_page(struct page *page, unsigned int order);
224 extern void post_alloc_hook(struct page *page, unsigned int order,
226 extern int user_min_free_kbytes;
228 extern void free_unref_page(struct page *page, unsigned int order);
229 extern void free_unref_page_list(struct list_head *list);
231 extern void zone_pcp_update(struct zone *zone, int cpu_online);
232 extern void zone_pcp_reset(struct zone *zone);
233 extern void zone_pcp_disable(struct zone *zone);
234 extern void zone_pcp_enable(struct zone *zone);
236 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
237 phys_addr_t min_addr,
238 int nid, bool exact_nid);
240 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
246 * compact_control is used to track pages being migrated and the free pages
247 * they are being migrated to during memory compaction. The free_pfn starts
248 * at the end of a zone and migrate_pfn begins at the start. Movable pages
249 * are moved to the end of a zone during a compaction run and the run
250 * completes when free_pfn <= migrate_pfn
252 struct compact_control {
253 struct list_head freepages; /* List of free pages to migrate to */
254 struct list_head migratepages; /* List of pages being migrated */
255 unsigned int nr_freepages; /* Number of isolated free pages */
256 unsigned int nr_migratepages; /* Number of pages to migrate */
257 unsigned long free_pfn; /* isolate_freepages search base */
259 * Acts as an in/out parameter to page isolation for migration.
260 * isolate_migratepages uses it as a search base.
261 * isolate_migratepages_block will update the value to the next pfn
262 * after the last isolated one.
264 unsigned long migrate_pfn;
265 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
267 unsigned long total_migrate_scanned;
268 unsigned long total_free_scanned;
269 unsigned short fast_search_fail;/* failures to use free list searches */
270 short search_order; /* order to start a fast search at */
271 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
272 int order; /* order a direct compactor needs */
273 int migratetype; /* migratetype of direct compactor */
274 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
275 const int highest_zoneidx; /* zone index of a direct compactor */
276 enum migrate_mode mode; /* Async or sync migration mode */
277 bool ignore_skip_hint; /* Scan blocks even if marked skip */
278 bool no_set_skip_hint; /* Don't mark blocks for skipping */
279 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
280 bool direct_compaction; /* False from kcompactd or /proc/... */
281 bool proactive_compaction; /* kcompactd proactive compaction */
282 bool whole_zone; /* Whole zone should/has been scanned */
283 bool contended; /* Signal lock or sched contention */
284 bool rescan; /* Rescanning the same pageblock */
285 bool alloc_contig; /* alloc_contig_range allocation */
289 * Used in direct compaction when a page should be taken from the freelists
290 * immediately when one is created during the free path.
292 struct capture_control {
293 struct compact_control *cc;
298 isolate_freepages_range(struct compact_control *cc,
299 unsigned long start_pfn, unsigned long end_pfn);
301 isolate_migratepages_range(struct compact_control *cc,
302 unsigned long low_pfn, unsigned long end_pfn);
304 int find_suitable_fallback(struct free_area *area, unsigned int order,
305 int migratetype, bool only_stealable, bool *can_steal);
308 * This function returns the order of a free page in the buddy system. In
309 * general, page_zone(page)->lock must be held by the caller to prevent the
310 * page from being allocated in parallel and returning garbage as the order.
311 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
312 * page cannot be allocated or merged in parallel. Alternatively, it must
313 * handle invalid values gracefully, and use buddy_order_unsafe() below.
315 static inline unsigned int buddy_order(struct page *page)
317 /* PageBuddy() must be checked by the caller */
318 return page_private(page);
322 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
323 * PageBuddy() should be checked first by the caller to minimize race window,
324 * and invalid values must be handled gracefully.
326 * READ_ONCE is used so that if the caller assigns the result into a local
327 * variable and e.g. tests it for valid range before using, the compiler cannot
328 * decide to remove the variable and inline the page_private(page) multiple
329 * times, potentially observing different values in the tests and the actual
332 #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
335 * These three helpers classifies VMAs for virtual memory accounting.
339 * Executable code area - executable, not writable, not stack
341 static inline bool is_exec_mapping(vm_flags_t flags)
343 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
347 * Stack area - automatically grows in one direction
349 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
350 * do_mmap() forbids all other combinations.
352 static inline bool is_stack_mapping(vm_flags_t flags)
354 return (flags & VM_STACK) == VM_STACK;
358 * Data area - private, writable, not stack
360 static inline bool is_data_mapping(vm_flags_t flags)
362 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
366 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
367 struct vm_area_struct *prev);
368 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
371 extern long populate_vma_page_range(struct vm_area_struct *vma,
372 unsigned long start, unsigned long end, int *locked);
373 extern long faultin_vma_page_range(struct vm_area_struct *vma,
374 unsigned long start, unsigned long end,
375 bool write, int *locked);
376 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
377 unsigned long start, unsigned long end);
378 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
380 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
384 * must be called with vma's mmap_lock held for read or write, and page locked.
386 extern void mlock_vma_page(struct page *page);
387 extern unsigned int munlock_vma_page(struct page *page);
389 extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
393 * Clear the page's PageMlocked(). This can be useful in a situation where
394 * we want to unconditionally remove a page from the pagecache -- e.g.,
395 * on truncation or freeing.
397 * It is legal to call this function for any page, mlocked or not.
398 * If called for a page that is still mapped by mlocked vmas, all we do
399 * is revert to lazy LRU behaviour -- semantics are not broken.
401 extern void clear_page_mlock(struct page *page);
403 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
406 * At what user virtual address is page expected in vma?
407 * Returns -EFAULT if all of the page is outside the range of vma.
408 * If page is a compound head, the entire compound page is considered.
410 static inline unsigned long
411 vma_address(struct page *page, struct vm_area_struct *vma)
414 unsigned long address;
416 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
417 pgoff = page_to_pgoff(page);
418 if (pgoff >= vma->vm_pgoff) {
419 address = vma->vm_start +
420 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
421 /* Check for address beyond vma (or wrapped through 0?) */
422 if (address < vma->vm_start || address >= vma->vm_end)
424 } else if (PageHead(page) &&
425 pgoff + compound_nr(page) - 1 >= vma->vm_pgoff) {
426 /* Test above avoids possibility of wrap to 0 on 32-bit */
427 address = vma->vm_start;
435 * Then at what user virtual address will none of the page be found in vma?
436 * Assumes that vma_address() already returned a good starting address.
437 * If page is a compound head, the entire compound page is considered.
439 static inline unsigned long
440 vma_address_end(struct page *page, struct vm_area_struct *vma)
443 unsigned long address;
445 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
446 pgoff = page_to_pgoff(page) + compound_nr(page);
447 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
448 /* Check for address beyond vma (or wrapped through 0?) */
449 if (address < vma->vm_start || address > vma->vm_end)
450 address = vma->vm_end;
454 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
457 int flags = vmf->flags;
463 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
464 * anything, so we only pin the file and drop the mmap_lock if only
465 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
467 if (fault_flag_allow_retry_first(flags) &&
468 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
469 fpin = get_file(vmf->vma->vm_file);
470 mmap_read_unlock(vmf->vma->vm_mm);
475 #else /* !CONFIG_MMU */
476 static inline void clear_page_mlock(struct page *page) { }
477 static inline void mlock_vma_page(struct page *page) { }
478 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
481 #endif /* !CONFIG_MMU */
484 * Return the mem_map entry representing the 'offset' subpage within
485 * the maximally aligned gigantic page 'base'. Handle any discontiguity
486 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
488 static inline struct page *mem_map_offset(struct page *base, int offset)
490 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
491 return nth_page(base, offset);
492 return base + offset;
496 * Iterator over all subpages within the maximally aligned gigantic
497 * page 'base'. Handle any discontiguity in the mem_map.
499 static inline struct page *mem_map_next(struct page *iter,
500 struct page *base, int offset)
502 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
503 unsigned long pfn = page_to_pfn(base) + offset;
506 return pfn_to_page(pfn);
511 /* Memory initialisation debug and verification */
518 #ifdef CONFIG_DEBUG_MEMORY_INIT
520 extern int mminit_loglevel;
522 #define mminit_dprintk(level, prefix, fmt, arg...) \
524 if (level < mminit_loglevel) { \
525 if (level <= MMINIT_WARNING) \
526 pr_warn("mminit::" prefix " " fmt, ##arg); \
528 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
532 extern void mminit_verify_pageflags_layout(void);
533 extern void mminit_verify_zonelist(void);
536 static inline void mminit_dprintk(enum mminit_level level,
537 const char *prefix, const char *fmt, ...)
541 static inline void mminit_verify_pageflags_layout(void)
545 static inline void mminit_verify_zonelist(void)
548 #endif /* CONFIG_DEBUG_MEMORY_INIT */
550 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
551 #if defined(CONFIG_SPARSEMEM)
552 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
553 unsigned long *end_pfn);
555 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
556 unsigned long *end_pfn)
559 #endif /* CONFIG_SPARSEMEM */
561 #define NODE_RECLAIM_NOSCAN -2
562 #define NODE_RECLAIM_FULL -1
563 #define NODE_RECLAIM_SOME 0
564 #define NODE_RECLAIM_SUCCESS 1
567 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
568 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
570 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
573 return NODE_RECLAIM_NOSCAN;
575 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
581 extern int hwpoison_filter(struct page *p);
583 extern u32 hwpoison_filter_dev_major;
584 extern u32 hwpoison_filter_dev_minor;
585 extern u64 hwpoison_filter_flags_mask;
586 extern u64 hwpoison_filter_flags_value;
587 extern u64 hwpoison_filter_memcg;
588 extern u32 hwpoison_filter_enable;
590 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
591 unsigned long, unsigned long,
592 unsigned long, unsigned long);
594 extern void set_pageblock_order(void);
595 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
596 struct list_head *page_list);
597 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
598 #define ALLOC_WMARK_MIN WMARK_MIN
599 #define ALLOC_WMARK_LOW WMARK_LOW
600 #define ALLOC_WMARK_HIGH WMARK_HIGH
601 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
603 /* Mask to get the watermark bits */
604 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
607 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
608 * cannot assume a reduced access to memory reserves is sufficient for
612 #define ALLOC_OOM 0x08
614 #define ALLOC_OOM ALLOC_NO_WATERMARKS
617 #define ALLOC_HARDER 0x10 /* try to alloc harder */
618 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
619 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
620 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
621 #ifdef CONFIG_ZONE_DMA32
622 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
624 #define ALLOC_NOFRAGMENT 0x0
626 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
629 struct tlbflush_unmap_batch;
633 * only for MM internal work items which do not depend on
634 * any allocations or locks which might depend on allocations
636 extern struct workqueue_struct *mm_percpu_wq;
638 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
639 void try_to_unmap_flush(void);
640 void try_to_unmap_flush_dirty(void);
641 void flush_tlb_batched_pending(struct mm_struct *mm);
643 static inline void try_to_unmap_flush(void)
646 static inline void try_to_unmap_flush_dirty(void)
649 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
652 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
654 extern const struct trace_print_flags pageflag_names[];
655 extern const struct trace_print_flags vmaflag_names[];
656 extern const struct trace_print_flags gfpflag_names[];
658 static inline bool is_migrate_highatomic(enum migratetype migratetype)
660 return migratetype == MIGRATE_HIGHATOMIC;
663 static inline bool is_migrate_highatomic_page(struct page *page)
665 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
668 void setup_zone_pageset(struct zone *zone);
670 struct migration_target_control {
671 int nid; /* preferred node id */
680 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
681 pgprot_t prot, struct page **pages, unsigned int page_shift);
684 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
685 pgprot_t prot, struct page **pages, unsigned int page_shift)
691 void vunmap_range_noflush(unsigned long start, unsigned long end);
693 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
694 unsigned long addr, int page_nid, int *flags);
696 #endif /* __MM_INTERNAL_H */