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 *, struct file_ra_state *,
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, mapping, index);
60 force_page_cache_ra(&ractl, &file->f_ra, nr_to_read);
63 struct page *find_get_entry(struct address_space *mapping, pgoff_t index);
64 struct page *find_lock_entry(struct address_space *mapping, pgoff_t index);
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);
124 * Structure for holding the mostly immutable allocation parameters passed
125 * between functions involved in allocations, including the alloc_pages*
126 * family of functions.
128 * nodemask, migratetype and highest_zoneidx are initialized only once in
129 * __alloc_pages_nodemask() and then never change.
131 * zonelist, preferred_zone and highest_zoneidx are set first in
132 * __alloc_pages_nodemask() for the fast path, and might be later changed
133 * in __alloc_pages_slowpath(). All other functions pass the whole structure
134 * by a const pointer.
136 struct alloc_context {
137 struct zonelist *zonelist;
138 nodemask_t *nodemask;
139 struct zoneref *preferred_zoneref;
143 * highest_zoneidx represents highest usable zone index of
144 * the allocation request. Due to the nature of the zone,
145 * memory on lower zone than the highest_zoneidx will be
146 * protected by lowmem_reserve[highest_zoneidx].
148 * highest_zoneidx is also used by reclaim/compaction to limit
149 * the target zone since higher zone than this index cannot be
150 * usable for this allocation request.
152 enum zone_type highest_zoneidx;
153 bool spread_dirty_pages;
157 * Locate the struct page for both the matching buddy in our
158 * pair (buddy1) and the combined O(n+1) page they form (page).
160 * 1) Any buddy B1 will have an order O twin B2 which satisfies
161 * the following equation:
163 * For example, if the starting buddy (buddy2) is #8 its order
165 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
167 * 2) Any buddy B will have an order O+1 parent P which
168 * satisfies the following equation:
171 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
173 static inline unsigned long
174 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
176 return page_pfn ^ (1 << order);
179 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
180 unsigned long end_pfn, struct zone *zone);
182 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
183 unsigned long end_pfn, struct zone *zone)
185 if (zone->contiguous)
186 return pfn_to_page(start_pfn);
188 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
191 extern int __isolate_free_page(struct page *page, unsigned int order);
192 extern void __putback_isolated_page(struct page *page, unsigned int order,
194 extern void memblock_free_pages(struct page *page, unsigned long pfn,
196 extern void __free_pages_core(struct page *page, unsigned int order);
197 extern void prep_compound_page(struct page *page, unsigned int order);
198 extern void post_alloc_hook(struct page *page, unsigned int order,
200 extern int user_min_free_kbytes;
202 extern void zone_pcp_update(struct zone *zone);
203 extern void zone_pcp_reset(struct zone *zone);
205 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
211 * compact_control is used to track pages being migrated and the free pages
212 * they are being migrated to during memory compaction. The free_pfn starts
213 * at the end of a zone and migrate_pfn begins at the start. Movable pages
214 * are moved to the end of a zone during a compaction run and the run
215 * completes when free_pfn <= migrate_pfn
217 struct compact_control {
218 struct list_head freepages; /* List of free pages to migrate to */
219 struct list_head migratepages; /* List of pages being migrated */
220 unsigned int nr_freepages; /* Number of isolated free pages */
221 unsigned int nr_migratepages; /* Number of pages to migrate */
222 unsigned long free_pfn; /* isolate_freepages search base */
223 unsigned long migrate_pfn; /* isolate_migratepages search base */
224 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
226 unsigned long total_migrate_scanned;
227 unsigned long total_free_scanned;
228 unsigned short fast_search_fail;/* failures to use free list searches */
229 short search_order; /* order to start a fast search at */
230 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
231 int order; /* order a direct compactor needs */
232 int migratetype; /* migratetype of direct compactor */
233 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
234 const int highest_zoneidx; /* zone index of a direct compactor */
235 enum migrate_mode mode; /* Async or sync migration mode */
236 bool ignore_skip_hint; /* Scan blocks even if marked skip */
237 bool no_set_skip_hint; /* Don't mark blocks for skipping */
238 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
239 bool direct_compaction; /* False from kcompactd or /proc/... */
240 bool proactive_compaction; /* kcompactd proactive compaction */
241 bool whole_zone; /* Whole zone should/has been scanned */
242 bool contended; /* Signal lock or sched contention */
243 bool rescan; /* Rescanning the same pageblock */
244 bool alloc_contig; /* alloc_contig_range allocation */
248 * Used in direct compaction when a page should be taken from the freelists
249 * immediately when one is created during the free path.
251 struct capture_control {
252 struct compact_control *cc;
257 isolate_freepages_range(struct compact_control *cc,
258 unsigned long start_pfn, unsigned long end_pfn);
260 isolate_migratepages_range(struct compact_control *cc,
261 unsigned long low_pfn, unsigned long end_pfn);
262 int find_suitable_fallback(struct free_area *area, unsigned int order,
263 int migratetype, bool only_stealable, bool *can_steal);
268 * This function returns the order of a free page in the buddy system. In
269 * general, page_zone(page)->lock must be held by the caller to prevent the
270 * page from being allocated in parallel and returning garbage as the order.
271 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
272 * page cannot be allocated or merged in parallel. Alternatively, it must
273 * handle invalid values gracefully, and use buddy_order_unsafe() below.
275 static inline unsigned int buddy_order(struct page *page)
277 /* PageBuddy() must be checked by the caller */
278 return page_private(page);
282 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
283 * PageBuddy() should be checked first by the caller to minimize race window,
284 * and invalid values must be handled gracefully.
286 * READ_ONCE is used so that if the caller assigns the result into a local
287 * variable and e.g. tests it for valid range before using, the compiler cannot
288 * decide to remove the variable and inline the page_private(page) multiple
289 * times, potentially observing different values in the tests and the actual
292 #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
294 static inline bool is_cow_mapping(vm_flags_t flags)
296 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
300 * These three helpers classifies VMAs for virtual memory accounting.
304 * Executable code area - executable, not writable, not stack
306 static inline bool is_exec_mapping(vm_flags_t flags)
308 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
312 * Stack area - atomatically grows in one direction
314 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
315 * do_mmap() forbids all other combinations.
317 static inline bool is_stack_mapping(vm_flags_t flags)
319 return (flags & VM_STACK) == VM_STACK;
323 * Data area - private, writable, not stack
325 static inline bool is_data_mapping(vm_flags_t flags)
327 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
331 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
332 struct vm_area_struct *prev);
333 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
336 extern long populate_vma_page_range(struct vm_area_struct *vma,
337 unsigned long start, unsigned long end, int *nonblocking);
338 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
339 unsigned long start, unsigned long end);
340 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
342 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
346 * must be called with vma's mmap_lock held for read or write, and page locked.
348 extern void mlock_vma_page(struct page *page);
349 extern unsigned int munlock_vma_page(struct page *page);
352 * Clear the page's PageMlocked(). This can be useful in a situation where
353 * we want to unconditionally remove a page from the pagecache -- e.g.,
354 * on truncation or freeing.
356 * It is legal to call this function for any page, mlocked or not.
357 * If called for a page that is still mapped by mlocked vmas, all we do
358 * is revert to lazy LRU behaviour -- semantics are not broken.
360 extern void clear_page_mlock(struct page *page);
363 * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
364 * (because that does not go through the full procedure of migration ptes):
365 * to migrate the Mlocked page flag; update statistics.
367 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
369 if (TestClearPageMlocked(page)) {
370 int nr_pages = thp_nr_pages(page);
372 /* Holding pmd lock, no change in irq context: __mod is safe */
373 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
374 SetPageMlocked(newpage);
375 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
379 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
382 * At what user virtual address is page expected in vma?
383 * Returns -EFAULT if all of the page is outside the range of vma.
384 * If page is a compound head, the entire compound page is considered.
386 static inline unsigned long
387 vma_address(struct page *page, struct vm_area_struct *vma)
390 unsigned long address;
392 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
393 pgoff = page_to_pgoff(page);
394 if (pgoff >= vma->vm_pgoff) {
395 address = vma->vm_start +
396 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
397 /* Check for address beyond vma (or wrapped through 0?) */
398 if (address < vma->vm_start || address >= vma->vm_end)
400 } else if (PageHead(page) &&
401 pgoff + compound_nr(page) - 1 >= vma->vm_pgoff) {
402 /* Test above avoids possibility of wrap to 0 on 32-bit */
403 address = vma->vm_start;
411 * Then at what user virtual address will none of the page be found in vma?
412 * Assumes that vma_address() already returned a good starting address.
413 * If page is a compound head, the entire compound page is considered.
415 static inline unsigned long
416 vma_address_end(struct page *page, struct vm_area_struct *vma)
419 unsigned long address;
421 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
422 pgoff = page_to_pgoff(page) + compound_nr(page);
423 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
424 /* Check for address beyond vma (or wrapped through 0?) */
425 if (address < vma->vm_start || address > vma->vm_end)
426 address = vma->vm_end;
430 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
433 int flags = vmf->flags;
439 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
440 * anything, so we only pin the file and drop the mmap_lock if only
441 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
443 if (fault_flag_allow_retry_first(flags) &&
444 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
445 fpin = get_file(vmf->vma->vm_file);
446 mmap_read_unlock(vmf->vma->vm_mm);
451 #else /* !CONFIG_MMU */
452 static inline void clear_page_mlock(struct page *page) { }
453 static inline void mlock_vma_page(struct page *page) { }
454 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
456 #endif /* !CONFIG_MMU */
459 * Return the mem_map entry representing the 'offset' subpage within
460 * the maximally aligned gigantic page 'base'. Handle any discontiguity
461 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
463 static inline struct page *mem_map_offset(struct page *base, int offset)
465 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
466 return nth_page(base, offset);
467 return base + offset;
471 * Iterator over all subpages within the maximally aligned gigantic
472 * page 'base'. Handle any discontiguity in the mem_map.
474 static inline struct page *mem_map_next(struct page *iter,
475 struct page *base, int offset)
477 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
478 unsigned long pfn = page_to_pfn(base) + offset;
481 return pfn_to_page(pfn);
486 /* Memory initialisation debug and verification */
493 #ifdef CONFIG_DEBUG_MEMORY_INIT
495 extern int mminit_loglevel;
497 #define mminit_dprintk(level, prefix, fmt, arg...) \
499 if (level < mminit_loglevel) { \
500 if (level <= MMINIT_WARNING) \
501 pr_warn("mminit::" prefix " " fmt, ##arg); \
503 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
507 extern void mminit_verify_pageflags_layout(void);
508 extern void mminit_verify_zonelist(void);
511 static inline void mminit_dprintk(enum mminit_level level,
512 const char *prefix, const char *fmt, ...)
516 static inline void mminit_verify_pageflags_layout(void)
520 static inline void mminit_verify_zonelist(void)
523 #endif /* CONFIG_DEBUG_MEMORY_INIT */
525 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
526 #if defined(CONFIG_SPARSEMEM)
527 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
528 unsigned long *end_pfn);
530 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
531 unsigned long *end_pfn)
534 #endif /* CONFIG_SPARSEMEM */
536 #define NODE_RECLAIM_NOSCAN -2
537 #define NODE_RECLAIM_FULL -1
538 #define NODE_RECLAIM_SOME 0
539 #define NODE_RECLAIM_SUCCESS 1
542 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
544 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
547 return NODE_RECLAIM_NOSCAN;
551 extern int hwpoison_filter(struct page *p);
553 extern u32 hwpoison_filter_dev_major;
554 extern u32 hwpoison_filter_dev_minor;
555 extern u64 hwpoison_filter_flags_mask;
556 extern u64 hwpoison_filter_flags_value;
557 extern u64 hwpoison_filter_memcg;
558 extern u32 hwpoison_filter_enable;
560 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
561 unsigned long, unsigned long,
562 unsigned long, unsigned long);
564 extern void set_pageblock_order(void);
565 unsigned int reclaim_clean_pages_from_list(struct zone *zone,
566 struct list_head *page_list);
567 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
568 #define ALLOC_WMARK_MIN WMARK_MIN
569 #define ALLOC_WMARK_LOW WMARK_LOW
570 #define ALLOC_WMARK_HIGH WMARK_HIGH
571 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
573 /* Mask to get the watermark bits */
574 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
577 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
578 * cannot assume a reduced access to memory reserves is sufficient for
582 #define ALLOC_OOM 0x08
584 #define ALLOC_OOM ALLOC_NO_WATERMARKS
587 #define ALLOC_HARDER 0x10 /* try to alloc harder */
588 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
589 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
590 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
591 #ifdef CONFIG_ZONE_DMA32
592 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
594 #define ALLOC_NOFRAGMENT 0x0
596 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
599 struct tlbflush_unmap_batch;
603 * only for MM internal work items which do not depend on
604 * any allocations or locks which might depend on allocations
606 extern struct workqueue_struct *mm_percpu_wq;
608 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
609 void try_to_unmap_flush(void);
610 void try_to_unmap_flush_dirty(void);
611 void flush_tlb_batched_pending(struct mm_struct *mm);
613 static inline void try_to_unmap_flush(void)
616 static inline void try_to_unmap_flush_dirty(void)
619 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
622 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
624 extern const struct trace_print_flags pageflag_names[];
625 extern const struct trace_print_flags vmaflag_names[];
626 extern const struct trace_print_flags gfpflag_names[];
628 static inline bool is_migrate_highatomic(enum migratetype migratetype)
630 return migratetype == MIGRATE_HIGHATOMIC;
633 static inline bool is_migrate_highatomic_page(struct page *page)
635 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
638 void setup_zone_pageset(struct zone *zone);
640 extern atomic_long_t nr_phys_cont_pte_pages;
641 extern atomic_long_t nr_phys_huge_pmd_pages;
643 struct migration_target_control {
644 int nid; /* preferred node id */
649 #endif /* __MM_INTERNAL_H */