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 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);
41 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
43 static inline bool can_madv_lru_vma(struct vm_area_struct *vma)
45 return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
48 void unmap_page_range(struct mmu_gather *tlb,
49 struct vm_area_struct *vma,
50 unsigned long addr, unsigned long end,
51 struct zap_details *details);
53 void do_page_cache_ra(struct readahead_control *, unsigned long nr_to_read,
54 unsigned long lookahead_size);
55 void force_page_cache_ra(struct readahead_control *, unsigned long nr);
56 static inline void force_page_cache_readahead(struct address_space *mapping,
57 struct file *file, pgoff_t index, unsigned long nr_to_read)
59 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
60 force_page_cache_ra(&ractl, nr_to_read);
63 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
64 pgoff_t end, struct pagevec *pvec, pgoff_t *indices);
67 * page_evictable - test whether a page is evictable
68 * @page: the page to test
70 * Test whether page is evictable--i.e., should be placed on active/inactive
71 * lists vs unevictable list.
73 * Reasons page might not be evictable:
74 * (1) page's mapping marked unevictable
75 * (2) page is part of an mlocked VMA
78 static inline bool page_evictable(struct page *page)
82 /* Prevent address_space of inode and swap cache from being freed */
84 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
90 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
93 static inline void set_page_refcounted(struct page *page)
95 VM_BUG_ON_PAGE(PageTail(page), page);
96 VM_BUG_ON_PAGE(page_ref_count(page), page);
97 set_page_count(page, 1);
100 extern unsigned long highest_memmap_pfn;
103 * Maximum number of reclaim retries without progress before the OOM
104 * killer is consider the only way forward.
106 #define MAX_RECLAIM_RETRIES 16
111 extern int isolate_lru_page(struct page *page);
112 extern void putback_lru_page(struct page *page);
117 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
120 * in mm/memcontrol.c:
122 extern bool cgroup_memory_nokmem;
129 * Structure for holding the mostly immutable allocation parameters passed
130 * between functions involved in allocations, including the alloc_pages*
131 * family of functions.
133 * nodemask, migratetype and highest_zoneidx are initialized only once in
134 * __alloc_pages() and then never change.
136 * zonelist, preferred_zone and highest_zoneidx are set first in
137 * __alloc_pages() for the fast path, and might be later changed
138 * in __alloc_pages_slowpath(). All other functions pass the whole structure
139 * by a const pointer.
141 struct alloc_context {
142 struct zonelist *zonelist;
143 nodemask_t *nodemask;
144 struct zoneref *preferred_zoneref;
148 * highest_zoneidx represents highest usable zone index of
149 * the allocation request. Due to the nature of the zone,
150 * memory on lower zone than the highest_zoneidx will be
151 * protected by lowmem_reserve[highest_zoneidx].
153 * highest_zoneidx is also used by reclaim/compaction to limit
154 * the target zone since higher zone than this index cannot be
155 * usable for this allocation request.
157 enum zone_type highest_zoneidx;
158 bool spread_dirty_pages;
162 * Locate the struct page for both the matching buddy in our
163 * pair (buddy1) and the combined O(n+1) page they form (page).
165 * 1) Any buddy B1 will have an order O twin B2 which satisfies
166 * the following equation:
168 * For example, if the starting buddy (buddy2) is #8 its order
170 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
172 * 2) Any buddy B will have an order O+1 parent P which
173 * satisfies the following equation:
176 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
178 static inline unsigned long
179 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
181 return page_pfn ^ (1 << order);
184 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
185 unsigned long end_pfn, struct zone *zone);
187 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
188 unsigned long end_pfn, struct zone *zone)
190 if (zone->contiguous)
191 return pfn_to_page(start_pfn);
193 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
196 extern int __isolate_free_page(struct page *page, unsigned int order);
197 extern void __putback_isolated_page(struct page *page, unsigned int order,
199 extern void memblock_free_pages(struct page *page, unsigned long pfn,
201 extern void __free_pages_core(struct page *page, unsigned int order);
202 extern void prep_compound_page(struct page *page, unsigned int order);
203 extern void post_alloc_hook(struct page *page, unsigned int order,
205 extern int user_min_free_kbytes;
207 extern void free_unref_page(struct page *page, unsigned int order);
208 extern void free_unref_page_list(struct list_head *list);
210 extern void zone_pcp_update(struct zone *zone, int cpu_online);
211 extern void zone_pcp_reset(struct zone *zone);
212 extern void zone_pcp_disable(struct zone *zone);
213 extern void zone_pcp_enable(struct zone *zone);
215 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
216 phys_addr_t min_addr,
217 int nid, bool exact_nid);
219 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
225 * compact_control is used to track pages being migrated and the free pages
226 * they are being migrated to during memory compaction. The free_pfn starts
227 * at the end of a zone and migrate_pfn begins at the start. Movable pages
228 * are moved to the end of a zone during a compaction run and the run
229 * completes when free_pfn <= migrate_pfn
231 struct compact_control {
232 struct list_head freepages; /* List of free pages to migrate to */
233 struct list_head migratepages; /* List of pages being migrated */
234 unsigned int nr_freepages; /* Number of isolated free pages */
235 unsigned int nr_migratepages; /* Number of pages to migrate */
236 unsigned long free_pfn; /* isolate_freepages search base */
238 * Acts as an in/out parameter to page isolation for migration.
239 * isolate_migratepages uses it as a search base.
240 * isolate_migratepages_block will update the value to the next pfn
241 * after the last isolated one.
243 unsigned long migrate_pfn;
244 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
246 unsigned long total_migrate_scanned;
247 unsigned long total_free_scanned;
248 unsigned short fast_search_fail;/* failures to use free list searches */
249 short search_order; /* order to start a fast search at */
250 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
251 int order; /* order a direct compactor needs */
252 int migratetype; /* migratetype of direct compactor */
253 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
254 const int highest_zoneidx; /* zone index of a direct compactor */
255 enum migrate_mode mode; /* Async or sync migration mode */
256 bool ignore_skip_hint; /* Scan blocks even if marked skip */
257 bool no_set_skip_hint; /* Don't mark blocks for skipping */
258 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
259 bool direct_compaction; /* False from kcompactd or /proc/... */
260 bool proactive_compaction; /* kcompactd proactive compaction */
261 bool whole_zone; /* Whole zone should/has been scanned */
262 bool contended; /* Signal lock or sched contention */
263 bool rescan; /* Rescanning the same pageblock */
264 bool alloc_contig; /* alloc_contig_range allocation */
268 * Used in direct compaction when a page should be taken from the freelists
269 * immediately when one is created during the free path.
271 struct capture_control {
272 struct compact_control *cc;
277 isolate_freepages_range(struct compact_control *cc,
278 unsigned long start_pfn, unsigned long end_pfn);
280 isolate_migratepages_range(struct compact_control *cc,
281 unsigned long low_pfn, unsigned long end_pfn);
283 int find_suitable_fallback(struct free_area *area, unsigned int order,
284 int migratetype, bool only_stealable, bool *can_steal);
287 * This function returns the order of a free page in the buddy system. In
288 * general, page_zone(page)->lock must be held by the caller to prevent the
289 * page from being allocated in parallel and returning garbage as the order.
290 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
291 * page cannot be allocated or merged in parallel. Alternatively, it must
292 * handle invalid values gracefully, and use buddy_order_unsafe() below.
294 static inline unsigned int buddy_order(struct page *page)
296 /* PageBuddy() must be checked by the caller */
297 return page_private(page);
301 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
302 * PageBuddy() should be checked first by the caller to minimize race window,
303 * and invalid values must be handled gracefully.
305 * READ_ONCE is used so that if the caller assigns the result into a local
306 * variable and e.g. tests it for valid range before using, the compiler cannot
307 * decide to remove the variable and inline the page_private(page) multiple
308 * times, potentially observing different values in the tests and the actual
311 #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
314 * These three helpers classifies VMAs for virtual memory accounting.
318 * Executable code area - executable, not writable, not stack
320 static inline bool is_exec_mapping(vm_flags_t flags)
322 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
326 * Stack area - automatically grows in one direction
328 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
329 * do_mmap() forbids all other combinations.
331 static inline bool is_stack_mapping(vm_flags_t flags)
333 return (flags & VM_STACK) == VM_STACK;
337 * Data area - private, writable, not stack
339 static inline bool is_data_mapping(vm_flags_t flags)
341 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
345 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
346 struct vm_area_struct *prev);
347 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
350 extern long populate_vma_page_range(struct vm_area_struct *vma,
351 unsigned long start, unsigned long end, int *locked);
352 extern long faultin_vma_page_range(struct vm_area_struct *vma,
353 unsigned long start, unsigned long end,
354 bool write, int *locked);
355 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
356 unsigned long start, unsigned long end);
357 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
359 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
363 * must be called with vma's mmap_lock held for read or write, and page locked.
365 extern void mlock_vma_page(struct page *page);
366 extern unsigned int munlock_vma_page(struct page *page);
368 extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
372 * Clear the page's PageMlocked(). This can be useful in a situation where
373 * we want to unconditionally remove a page from the pagecache -- e.g.,
374 * on truncation or freeing.
376 * It is legal to call this function for any page, mlocked or not.
377 * If called for a page that is still mapped by mlocked vmas, all we do
378 * is revert to lazy LRU behaviour -- semantics are not broken.
380 extern void clear_page_mlock(struct page *page);
382 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
385 * At what user virtual address is page expected in vma?
386 * Returns -EFAULT if all of the page is outside the range of vma.
387 * If page is a compound head, the entire compound page is considered.
389 static inline unsigned long
390 vma_address(struct page *page, struct vm_area_struct *vma)
393 unsigned long address;
395 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
396 pgoff = page_to_pgoff(page);
397 if (pgoff >= vma->vm_pgoff) {
398 address = vma->vm_start +
399 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
400 /* Check for address beyond vma (or wrapped through 0?) */
401 if (address < vma->vm_start || address >= vma->vm_end)
403 } else if (PageHead(page) &&
404 pgoff + compound_nr(page) - 1 >= vma->vm_pgoff) {
405 /* Test above avoids possibility of wrap to 0 on 32-bit */
406 address = vma->vm_start;
414 * Then at what user virtual address will none of the page be found in vma?
415 * Assumes that vma_address() already returned a good starting address.
416 * If page is a compound head, the entire compound page is considered.
418 static inline unsigned long
419 vma_address_end(struct page *page, struct vm_area_struct *vma)
422 unsigned long address;
424 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
425 pgoff = page_to_pgoff(page) + compound_nr(page);
426 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
427 /* Check for address beyond vma (or wrapped through 0?) */
428 if (address < vma->vm_start || address > vma->vm_end)
429 address = vma->vm_end;
433 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
436 int flags = vmf->flags;
442 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
443 * anything, so we only pin the file and drop the mmap_lock if only
444 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
446 if (fault_flag_allow_retry_first(flags) &&
447 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
448 fpin = get_file(vmf->vma->vm_file);
449 mmap_read_unlock(vmf->vma->vm_mm);
454 #else /* !CONFIG_MMU */
455 static inline void clear_page_mlock(struct page *page) { }
456 static inline void mlock_vma_page(struct page *page) { }
457 static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
460 #endif /* !CONFIG_MMU */
463 * Return the mem_map entry representing the 'offset' subpage within
464 * the maximally aligned gigantic page 'base'. Handle any discontiguity
465 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
467 static inline struct page *mem_map_offset(struct page *base, int offset)
469 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
470 return nth_page(base, offset);
471 return base + offset;
475 * Iterator over all subpages within the maximally aligned gigantic
476 * page 'base'. Handle any discontiguity in the mem_map.
478 static inline struct page *mem_map_next(struct page *iter,
479 struct page *base, int offset)
481 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
482 unsigned long pfn = page_to_pfn(base) + offset;
485 return pfn_to_page(pfn);
490 /* Memory initialisation debug and verification */
497 #ifdef CONFIG_DEBUG_MEMORY_INIT
499 extern int mminit_loglevel;
501 #define mminit_dprintk(level, prefix, fmt, arg...) \
503 if (level < mminit_loglevel) { \
504 if (level <= MMINIT_WARNING) \
505 pr_warn("mminit::" prefix " " fmt, ##arg); \
507 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
511 extern void mminit_verify_pageflags_layout(void);
512 extern void mminit_verify_zonelist(void);
515 static inline void mminit_dprintk(enum mminit_level level,
516 const char *prefix, const char *fmt, ...)
520 static inline void mminit_verify_pageflags_layout(void)
524 static inline void mminit_verify_zonelist(void)
527 #endif /* CONFIG_DEBUG_MEMORY_INIT */
529 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
530 #if defined(CONFIG_SPARSEMEM)
531 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
532 unsigned long *end_pfn);
534 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
535 unsigned long *end_pfn)
538 #endif /* CONFIG_SPARSEMEM */
540 #define NODE_RECLAIM_NOSCAN -2
541 #define NODE_RECLAIM_FULL -1
542 #define NODE_RECLAIM_SOME 0
543 #define NODE_RECLAIM_SUCCESS 1
546 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
547 extern int find_next_best_node(int node, nodemask_t *used_node_mask);
549 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
552 return NODE_RECLAIM_NOSCAN;
554 static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
560 extern int hwpoison_filter(struct page *p);
562 extern u32 hwpoison_filter_dev_major;
563 extern u32 hwpoison_filter_dev_minor;
564 extern u64 hwpoison_filter_flags_mask;
565 extern u64 hwpoison_filter_flags_value;
566 extern u64 hwpoison_filter_memcg;
567 extern u32 hwpoison_filter_enable;
569 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
570 unsigned long, unsigned long,
571 unsigned long, unsigned long);
573 extern void set_pageblock_order(void);
574 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
575 struct list_head *page_list);
576 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
577 #define ALLOC_WMARK_MIN WMARK_MIN
578 #define ALLOC_WMARK_LOW WMARK_LOW
579 #define ALLOC_WMARK_HIGH WMARK_HIGH
580 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
582 /* Mask to get the watermark bits */
583 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
586 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
587 * cannot assume a reduced access to memory reserves is sufficient for
591 #define ALLOC_OOM 0x08
593 #define ALLOC_OOM ALLOC_NO_WATERMARKS
596 #define ALLOC_HARDER 0x10 /* try to alloc harder */
597 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
598 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
599 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
600 #ifdef CONFIG_ZONE_DMA32
601 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
603 #define ALLOC_NOFRAGMENT 0x0
605 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
608 struct tlbflush_unmap_batch;
612 * only for MM internal work items which do not depend on
613 * any allocations or locks which might depend on allocations
615 extern struct workqueue_struct *mm_percpu_wq;
617 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
618 void try_to_unmap_flush(void);
619 void try_to_unmap_flush_dirty(void);
620 void flush_tlb_batched_pending(struct mm_struct *mm);
622 static inline void try_to_unmap_flush(void)
625 static inline void try_to_unmap_flush_dirty(void)
628 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
631 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
633 extern const struct trace_print_flags pageflag_names[];
634 extern const struct trace_print_flags vmaflag_names[];
635 extern const struct trace_print_flags gfpflag_names[];
637 static inline bool is_migrate_highatomic(enum migratetype migratetype)
639 return migratetype == MIGRATE_HIGHATOMIC;
642 static inline bool is_migrate_highatomic_page(struct page *page)
644 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
647 void setup_zone_pageset(struct zone *zone);
649 struct migration_target_control {
650 int nid; /* preferred node id */
659 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
660 pgprot_t prot, struct page **pages, unsigned int page_shift);
663 int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
664 pgprot_t prot, struct page **pages, unsigned int page_shift)
670 void vunmap_range_noflush(unsigned long start, unsigned long end);
672 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
673 unsigned long addr, int page_nid, int *flags);
675 #endif /* __MM_INTERNAL_H */