Loading...
1/* internal.h: mm/ internal definitions
2 *
3 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11#ifndef __MM_INTERNAL_H
12#define __MM_INTERNAL_H
13
14#include <linux/mm.h>
15
16void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
17 unsigned long floor, unsigned long ceiling);
18
19static inline void set_page_count(struct page *page, int v)
20{
21 atomic_set(&page->_count, v);
22}
23
24/*
25 * Turn a non-refcounted page (->_count == 0) into refcounted with
26 * a count of one.
27 */
28static inline void set_page_refcounted(struct page *page)
29{
30 VM_BUG_ON(PageTail(page));
31 VM_BUG_ON(atomic_read(&page->_count));
32 set_page_count(page, 1);
33}
34
35static inline void __put_page(struct page *page)
36{
37 atomic_dec(&page->_count);
38}
39
40extern unsigned long highest_memmap_pfn;
41
42/*
43 * in mm/vmscan.c:
44 */
45extern int isolate_lru_page(struct page *page);
46extern void putback_lru_page(struct page *page);
47
48/*
49 * in mm/page_alloc.c
50 */
51extern void __free_pages_bootmem(struct page *page, unsigned int order);
52extern void prep_compound_page(struct page *page, unsigned long order);
53#ifdef CONFIG_MEMORY_FAILURE
54extern bool is_free_buddy_page(struct page *page);
55#endif
56
57
58/*
59 * function for dealing with page's order in buddy system.
60 * zone->lock is already acquired when we use these.
61 * So, we don't need atomic page->flags operations here.
62 */
63static inline unsigned long page_order(struct page *page)
64{
65 /* PageBuddy() must be checked by the caller */
66 return page_private(page);
67}
68
69/* mm/util.c */
70void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
71 struct vm_area_struct *prev, struct rb_node *rb_parent);
72
73#ifdef CONFIG_MMU
74extern long mlock_vma_pages_range(struct vm_area_struct *vma,
75 unsigned long start, unsigned long end);
76extern void munlock_vma_pages_range(struct vm_area_struct *vma,
77 unsigned long start, unsigned long end);
78static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
79{
80 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
81}
82
83/*
84 * Called only in fault path via page_evictable() for a new page
85 * to determine if it's being mapped into a LOCKED vma.
86 * If so, mark page as mlocked.
87 */
88static inline int is_mlocked_vma(struct vm_area_struct *vma, struct page *page)
89{
90 VM_BUG_ON(PageLRU(page));
91
92 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
93 return 0;
94
95 if (!TestSetPageMlocked(page)) {
96 inc_zone_page_state(page, NR_MLOCK);
97 count_vm_event(UNEVICTABLE_PGMLOCKED);
98 }
99 return 1;
100}
101
102/*
103 * must be called with vma's mmap_sem held for read or write, and page locked.
104 */
105extern void mlock_vma_page(struct page *page);
106extern void munlock_vma_page(struct page *page);
107
108/*
109 * Clear the page's PageMlocked(). This can be useful in a situation where
110 * we want to unconditionally remove a page from the pagecache -- e.g.,
111 * on truncation or freeing.
112 *
113 * It is legal to call this function for any page, mlocked or not.
114 * If called for a page that is still mapped by mlocked vmas, all we do
115 * is revert to lazy LRU behaviour -- semantics are not broken.
116 */
117extern void __clear_page_mlock(struct page *page);
118static inline void clear_page_mlock(struct page *page)
119{
120 if (unlikely(TestClearPageMlocked(page)))
121 __clear_page_mlock(page);
122}
123
124/*
125 * mlock_migrate_page - called only from migrate_page_copy() to
126 * migrate the Mlocked page flag; update statistics.
127 */
128static inline void mlock_migrate_page(struct page *newpage, struct page *page)
129{
130 if (TestClearPageMlocked(page)) {
131 unsigned long flags;
132
133 local_irq_save(flags);
134 __dec_zone_page_state(page, NR_MLOCK);
135 SetPageMlocked(newpage);
136 __inc_zone_page_state(newpage, NR_MLOCK);
137 local_irq_restore(flags);
138 }
139}
140
141#ifdef CONFIG_TRANSPARENT_HUGEPAGE
142extern unsigned long vma_address(struct page *page,
143 struct vm_area_struct *vma);
144#endif
145#else /* !CONFIG_MMU */
146static inline int is_mlocked_vma(struct vm_area_struct *v, struct page *p)
147{
148 return 0;
149}
150static inline void clear_page_mlock(struct page *page) { }
151static inline void mlock_vma_page(struct page *page) { }
152static inline void mlock_migrate_page(struct page *new, struct page *old) { }
153
154#endif /* !CONFIG_MMU */
155
156/*
157 * Return the mem_map entry representing the 'offset' subpage within
158 * the maximally aligned gigantic page 'base'. Handle any discontiguity
159 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
160 */
161static inline struct page *mem_map_offset(struct page *base, int offset)
162{
163 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
164 return pfn_to_page(page_to_pfn(base) + offset);
165 return base + offset;
166}
167
168/*
169 * Iterator over all subpages within the maximally aligned gigantic
170 * page 'base'. Handle any discontiguity in the mem_map.
171 */
172static inline struct page *mem_map_next(struct page *iter,
173 struct page *base, int offset)
174{
175 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
176 unsigned long pfn = page_to_pfn(base) + offset;
177 if (!pfn_valid(pfn))
178 return NULL;
179 return pfn_to_page(pfn);
180 }
181 return iter + 1;
182}
183
184/*
185 * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
186 * so all functions starting at paging_init should be marked __init
187 * in those cases. SPARSEMEM, however, allows for memory hotplug,
188 * and alloc_bootmem_node is not used.
189 */
190#ifdef CONFIG_SPARSEMEM
191#define __paginginit __meminit
192#else
193#define __paginginit __init
194#endif
195
196/* Memory initialisation debug and verification */
197enum mminit_level {
198 MMINIT_WARNING,
199 MMINIT_VERIFY,
200 MMINIT_TRACE
201};
202
203#ifdef CONFIG_DEBUG_MEMORY_INIT
204
205extern int mminit_loglevel;
206
207#define mminit_dprintk(level, prefix, fmt, arg...) \
208do { \
209 if (level < mminit_loglevel) { \
210 printk(level <= MMINIT_WARNING ? KERN_WARNING : KERN_DEBUG); \
211 printk(KERN_CONT "mminit::" prefix " " fmt, ##arg); \
212 } \
213} while (0)
214
215extern void mminit_verify_pageflags_layout(void);
216extern void mminit_verify_page_links(struct page *page,
217 enum zone_type zone, unsigned long nid, unsigned long pfn);
218extern void mminit_verify_zonelist(void);
219
220#else
221
222static inline void mminit_dprintk(enum mminit_level level,
223 const char *prefix, const char *fmt, ...)
224{
225}
226
227static inline void mminit_verify_pageflags_layout(void)
228{
229}
230
231static inline void mminit_verify_page_links(struct page *page,
232 enum zone_type zone, unsigned long nid, unsigned long pfn)
233{
234}
235
236static inline void mminit_verify_zonelist(void)
237{
238}
239#endif /* CONFIG_DEBUG_MEMORY_INIT */
240
241/* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
242#if defined(CONFIG_SPARSEMEM)
243extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
244 unsigned long *end_pfn);
245#else
246static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
247 unsigned long *end_pfn)
248{
249}
250#endif /* CONFIG_SPARSEMEM */
251
252#define ZONE_RECLAIM_NOSCAN -2
253#define ZONE_RECLAIM_FULL -1
254#define ZONE_RECLAIM_SOME 0
255#define ZONE_RECLAIM_SUCCESS 1
256#endif
257
258extern int hwpoison_filter(struct page *p);
259
260extern u32 hwpoison_filter_dev_major;
261extern u32 hwpoison_filter_dev_minor;
262extern u64 hwpoison_filter_flags_mask;
263extern u64 hwpoison_filter_flags_value;
264extern u64 hwpoison_filter_memcg;
265extern u32 hwpoison_filter_enable;
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/* internal.h: mm/ internal definitions
3 *
4 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7#ifndef __MM_INTERNAL_H
8#define __MM_INTERNAL_H
9
10#include <linux/fs.h>
11#include <linux/mm.h>
12#include <linux/pagemap.h>
13#include <linux/rmap.h>
14#include <linux/tracepoint-defs.h>
15
16struct folio_batch;
17
18/*
19 * The set of flags that only affect watermark checking and reclaim
20 * behaviour. This is used by the MM to obey the caller constraints
21 * about IO, FS and watermark checking while ignoring placement
22 * hints such as HIGHMEM usage.
23 */
24#define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
25 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
26 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
27 __GFP_ATOMIC|__GFP_NOLOCKDEP)
28
29/* The GFP flags allowed during early boot */
30#define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
31
32/* Control allocation cpuset and node placement constraints */
33#define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
34
35/* Do not use these with a slab allocator */
36#define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
37
38/*
39 * Different from WARN_ON_ONCE(), no warning will be issued
40 * when we specify __GFP_NOWARN.
41 */
42#define WARN_ON_ONCE_GFP(cond, gfp) ({ \
43 static bool __section(".data.once") __warned; \
44 int __ret_warn_once = !!(cond); \
45 \
46 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
47 __warned = true; \
48 WARN_ON(1); \
49 } \
50 unlikely(__ret_warn_once); \
51})
52
53void page_writeback_init(void);
54
55static inline void *folio_raw_mapping(struct folio *folio)
56{
57 unsigned long mapping = (unsigned long)folio->mapping;
58
59 return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
60}
61
62void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
63 int nr_throttled);
64static inline void acct_reclaim_writeback(struct folio *folio)
65{
66 pg_data_t *pgdat = folio_pgdat(folio);
67 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
68
69 if (nr_throttled)
70 __acct_reclaim_writeback(pgdat, folio, nr_throttled);
71}
72
73static inline void wake_throttle_isolated(pg_data_t *pgdat)
74{
75 wait_queue_head_t *wqh;
76
77 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
78 if (waitqueue_active(wqh))
79 wake_up(wqh);
80}
81
82vm_fault_t do_swap_page(struct vm_fault *vmf);
83void folio_rotate_reclaimable(struct folio *folio);
84bool __folio_end_writeback(struct folio *folio);
85void deactivate_file_folio(struct folio *folio);
86void folio_activate(struct folio *folio);
87
88void free_pgtables(struct mmu_gather *tlb, struct maple_tree *mt,
89 struct vm_area_struct *start_vma, unsigned long floor,
90 unsigned long ceiling);
91void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
92
93struct zap_details;
94void unmap_page_range(struct mmu_gather *tlb,
95 struct vm_area_struct *vma,
96 unsigned long addr, unsigned long end,
97 struct zap_details *details);
98
99void page_cache_ra_order(struct readahead_control *, struct file_ra_state *,
100 unsigned int order);
101void force_page_cache_ra(struct readahead_control *, unsigned long nr);
102static inline void force_page_cache_readahead(struct address_space *mapping,
103 struct file *file, pgoff_t index, unsigned long nr_to_read)
104{
105 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
106 force_page_cache_ra(&ractl, nr_to_read);
107}
108
109unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
110 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
111unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
112 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
113void filemap_free_folio(struct address_space *mapping, struct folio *folio);
114int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
115bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
116 loff_t end);
117long invalidate_inode_page(struct page *page);
118unsigned long invalidate_mapping_pagevec(struct address_space *mapping,
119 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec);
120
121/**
122 * folio_evictable - Test whether a folio is evictable.
123 * @folio: The folio to test.
124 *
125 * Test whether @folio is evictable -- i.e., should be placed on
126 * active/inactive lists vs unevictable list.
127 *
128 * Reasons folio might not be evictable:
129 * 1. folio's mapping marked unevictable
130 * 2. One of the pages in the folio is part of an mlocked VMA
131 */
132static inline bool folio_evictable(struct folio *folio)
133{
134 bool ret;
135
136 /* Prevent address_space of inode and swap cache from being freed */
137 rcu_read_lock();
138 ret = !mapping_unevictable(folio_mapping(folio)) &&
139 !folio_test_mlocked(folio);
140 rcu_read_unlock();
141 return ret;
142}
143
144static inline bool page_evictable(struct page *page)
145{
146 bool ret;
147
148 /* Prevent address_space of inode and swap cache from being freed */
149 rcu_read_lock();
150 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
151 rcu_read_unlock();
152 return ret;
153}
154
155/*
156 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
157 * a count of one.
158 */
159static inline void set_page_refcounted(struct page *page)
160{
161 VM_BUG_ON_PAGE(PageTail(page), page);
162 VM_BUG_ON_PAGE(page_ref_count(page), page);
163 set_page_count(page, 1);
164}
165
166extern unsigned long highest_memmap_pfn;
167
168/*
169 * Maximum number of reclaim retries without progress before the OOM
170 * killer is consider the only way forward.
171 */
172#define MAX_RECLAIM_RETRIES 16
173
174/*
175 * in mm/early_ioremap.c
176 */
177pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr,
178 unsigned long size, pgprot_t prot);
179
180/*
181 * in mm/vmscan.c:
182 */
183int isolate_lru_page(struct page *page);
184int folio_isolate_lru(struct folio *folio);
185void putback_lru_page(struct page *page);
186void folio_putback_lru(struct folio *folio);
187extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
188
189/*
190 * in mm/rmap.c:
191 */
192pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
193
194/*
195 * in mm/page_alloc.c
196 */
197
198/*
199 * Structure for holding the mostly immutable allocation parameters passed
200 * between functions involved in allocations, including the alloc_pages*
201 * family of functions.
202 *
203 * nodemask, migratetype and highest_zoneidx are initialized only once in
204 * __alloc_pages() and then never change.
205 *
206 * zonelist, preferred_zone and highest_zoneidx are set first in
207 * __alloc_pages() for the fast path, and might be later changed
208 * in __alloc_pages_slowpath(). All other functions pass the whole structure
209 * by a const pointer.
210 */
211struct alloc_context {
212 struct zonelist *zonelist;
213 nodemask_t *nodemask;
214 struct zoneref *preferred_zoneref;
215 int migratetype;
216
217 /*
218 * highest_zoneidx represents highest usable zone index of
219 * the allocation request. Due to the nature of the zone,
220 * memory on lower zone than the highest_zoneidx will be
221 * protected by lowmem_reserve[highest_zoneidx].
222 *
223 * highest_zoneidx is also used by reclaim/compaction to limit
224 * the target zone since higher zone than this index cannot be
225 * usable for this allocation request.
226 */
227 enum zone_type highest_zoneidx;
228 bool spread_dirty_pages;
229};
230
231/*
232 * This function returns the order of a free page in the buddy system. In
233 * general, page_zone(page)->lock must be held by the caller to prevent the
234 * page from being allocated in parallel and returning garbage as the order.
235 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
236 * page cannot be allocated or merged in parallel. Alternatively, it must
237 * handle invalid values gracefully, and use buddy_order_unsafe() below.
238 */
239static inline unsigned int buddy_order(struct page *page)
240{
241 /* PageBuddy() must be checked by the caller */
242 return page_private(page);
243}
244
245/*
246 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
247 * PageBuddy() should be checked first by the caller to minimize race window,
248 * and invalid values must be handled gracefully.
249 *
250 * READ_ONCE is used so that if the caller assigns the result into a local
251 * variable and e.g. tests it for valid range before using, the compiler cannot
252 * decide to remove the variable and inline the page_private(page) multiple
253 * times, potentially observing different values in the tests and the actual
254 * use of the result.
255 */
256#define buddy_order_unsafe(page) READ_ONCE(page_private(page))
257
258/*
259 * This function checks whether a page is free && is the buddy
260 * we can coalesce a page and its buddy if
261 * (a) the buddy is not in a hole (check before calling!) &&
262 * (b) the buddy is in the buddy system &&
263 * (c) a page and its buddy have the same order &&
264 * (d) a page and its buddy are in the same zone.
265 *
266 * For recording whether a page is in the buddy system, we set PageBuddy.
267 * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
268 *
269 * For recording page's order, we use page_private(page).
270 */
271static inline bool page_is_buddy(struct page *page, struct page *buddy,
272 unsigned int order)
273{
274 if (!page_is_guard(buddy) && !PageBuddy(buddy))
275 return false;
276
277 if (buddy_order(buddy) != order)
278 return false;
279
280 /*
281 * zone check is done late to avoid uselessly calculating
282 * zone/node ids for pages that could never merge.
283 */
284 if (page_zone_id(page) != page_zone_id(buddy))
285 return false;
286
287 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
288
289 return true;
290}
291
292/*
293 * Locate the struct page for both the matching buddy in our
294 * pair (buddy1) and the combined O(n+1) page they form (page).
295 *
296 * 1) Any buddy B1 will have an order O twin B2 which satisfies
297 * the following equation:
298 * B2 = B1 ^ (1 << O)
299 * For example, if the starting buddy (buddy2) is #8 its order
300 * 1 buddy is #10:
301 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
302 *
303 * 2) Any buddy B will have an order O+1 parent P which
304 * satisfies the following equation:
305 * P = B & ~(1 << O)
306 *
307 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
308 */
309static inline unsigned long
310__find_buddy_pfn(unsigned long page_pfn, unsigned int order)
311{
312 return page_pfn ^ (1 << order);
313}
314
315/*
316 * Find the buddy of @page and validate it.
317 * @page: The input page
318 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
319 * function is used in the performance-critical __free_one_page().
320 * @order: The order of the page
321 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
322 * page_to_pfn().
323 *
324 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
325 * not the same as @page. The validation is necessary before use it.
326 *
327 * Return: the found buddy page or NULL if not found.
328 */
329static inline struct page *find_buddy_page_pfn(struct page *page,
330 unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
331{
332 unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
333 struct page *buddy;
334
335 buddy = page + (__buddy_pfn - pfn);
336 if (buddy_pfn)
337 *buddy_pfn = __buddy_pfn;
338
339 if (page_is_buddy(page, buddy, order))
340 return buddy;
341 return NULL;
342}
343
344extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
345 unsigned long end_pfn, struct zone *zone);
346
347static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
348 unsigned long end_pfn, struct zone *zone)
349{
350 if (zone->contiguous)
351 return pfn_to_page(start_pfn);
352
353 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
354}
355
356extern int __isolate_free_page(struct page *page, unsigned int order);
357extern void __putback_isolated_page(struct page *page, unsigned int order,
358 int mt);
359extern void memblock_free_pages(struct page *page, unsigned long pfn,
360 unsigned int order);
361extern void __free_pages_core(struct page *page, unsigned int order);
362extern void prep_compound_page(struct page *page, unsigned int order);
363extern void post_alloc_hook(struct page *page, unsigned int order,
364 gfp_t gfp_flags);
365extern int user_min_free_kbytes;
366
367extern void free_unref_page(struct page *page, unsigned int order);
368extern void free_unref_page_list(struct list_head *list);
369
370extern void zone_pcp_reset(struct zone *zone);
371extern void zone_pcp_disable(struct zone *zone);
372extern void zone_pcp_enable(struct zone *zone);
373
374extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
375 phys_addr_t min_addr,
376 int nid, bool exact_nid);
377
378int split_free_page(struct page *free_page,
379 unsigned int order, unsigned long split_pfn_offset);
380
381#if defined CONFIG_COMPACTION || defined CONFIG_CMA
382
383/*
384 * in mm/compaction.c
385 */
386/*
387 * compact_control is used to track pages being migrated and the free pages
388 * they are being migrated to during memory compaction. The free_pfn starts
389 * at the end of a zone and migrate_pfn begins at the start. Movable pages
390 * are moved to the end of a zone during a compaction run and the run
391 * completes when free_pfn <= migrate_pfn
392 */
393struct compact_control {
394 struct list_head freepages; /* List of free pages to migrate to */
395 struct list_head migratepages; /* List of pages being migrated */
396 unsigned int nr_freepages; /* Number of isolated free pages */
397 unsigned int nr_migratepages; /* Number of pages to migrate */
398 unsigned long free_pfn; /* isolate_freepages search base */
399 /*
400 * Acts as an in/out parameter to page isolation for migration.
401 * isolate_migratepages uses it as a search base.
402 * isolate_migratepages_block will update the value to the next pfn
403 * after the last isolated one.
404 */
405 unsigned long migrate_pfn;
406 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
407 struct zone *zone;
408 unsigned long total_migrate_scanned;
409 unsigned long total_free_scanned;
410 unsigned short fast_search_fail;/* failures to use free list searches */
411 short search_order; /* order to start a fast search at */
412 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
413 int order; /* order a direct compactor needs */
414 int migratetype; /* migratetype of direct compactor */
415 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
416 const int highest_zoneidx; /* zone index of a direct compactor */
417 enum migrate_mode mode; /* Async or sync migration mode */
418 bool ignore_skip_hint; /* Scan blocks even if marked skip */
419 bool no_set_skip_hint; /* Don't mark blocks for skipping */
420 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
421 bool direct_compaction; /* False from kcompactd or /proc/... */
422 bool proactive_compaction; /* kcompactd proactive compaction */
423 bool whole_zone; /* Whole zone should/has been scanned */
424 bool contended; /* Signal lock contention */
425 bool rescan; /* Rescanning the same pageblock */
426 bool alloc_contig; /* alloc_contig_range allocation */
427};
428
429/*
430 * Used in direct compaction when a page should be taken from the freelists
431 * immediately when one is created during the free path.
432 */
433struct capture_control {
434 struct compact_control *cc;
435 struct page *page;
436};
437
438unsigned long
439isolate_freepages_range(struct compact_control *cc,
440 unsigned long start_pfn, unsigned long end_pfn);
441int
442isolate_migratepages_range(struct compact_control *cc,
443 unsigned long low_pfn, unsigned long end_pfn);
444
445int __alloc_contig_migrate_range(struct compact_control *cc,
446 unsigned long start, unsigned long end);
447#endif
448int find_suitable_fallback(struct free_area *area, unsigned int order,
449 int migratetype, bool only_stealable, bool *can_steal);
450
451/*
452 * These three helpers classifies VMAs for virtual memory accounting.
453 */
454
455/*
456 * Executable code area - executable, not writable, not stack
457 */
458static inline bool is_exec_mapping(vm_flags_t flags)
459{
460 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
461}
462
463/*
464 * Stack area - automatically grows in one direction
465 *
466 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
467 * do_mmap() forbids all other combinations.
468 */
469static inline bool is_stack_mapping(vm_flags_t flags)
470{
471 return (flags & VM_STACK) == VM_STACK;
472}
473
474/*
475 * Data area - private, writable, not stack
476 */
477static inline bool is_data_mapping(vm_flags_t flags)
478{
479 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
480}
481
482/* mm/util.c */
483struct anon_vma *folio_anon_vma(struct folio *folio);
484
485#ifdef CONFIG_MMU
486void unmap_mapping_folio(struct folio *folio);
487extern long populate_vma_page_range(struct vm_area_struct *vma,
488 unsigned long start, unsigned long end, int *locked);
489extern long faultin_vma_page_range(struct vm_area_struct *vma,
490 unsigned long start, unsigned long end,
491 bool write, int *locked);
492extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
493 unsigned long len);
494/*
495 * mlock_vma_page() and munlock_vma_page():
496 * should be called with vma's mmap_lock held for read or write,
497 * under page table lock for the pte/pmd being added or removed.
498 *
499 * mlock is usually called at the end of page_add_*_rmap(),
500 * munlock at the end of page_remove_rmap(); but new anon
501 * pages are managed by lru_cache_add_inactive_or_unevictable()
502 * calling mlock_new_page().
503 *
504 * @compound is used to include pmd mappings of THPs, but filter out
505 * pte mappings of THPs, which cannot be consistently counted: a pte
506 * mapping of the THP head cannot be distinguished by the page alone.
507 */
508void mlock_folio(struct folio *folio);
509static inline void mlock_vma_folio(struct folio *folio,
510 struct vm_area_struct *vma, bool compound)
511{
512 /*
513 * The VM_SPECIAL check here serves two purposes.
514 * 1) VM_IO check prevents migration from double-counting during mlock.
515 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
516 * is never left set on a VM_SPECIAL vma, there is an interval while
517 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
518 * still be set while VM_SPECIAL bits are added: so ignore it then.
519 */
520 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) &&
521 (compound || !folio_test_large(folio)))
522 mlock_folio(folio);
523}
524
525static inline void mlock_vma_page(struct page *page,
526 struct vm_area_struct *vma, bool compound)
527{
528 mlock_vma_folio(page_folio(page), vma, compound);
529}
530
531void munlock_page(struct page *page);
532static inline void munlock_vma_page(struct page *page,
533 struct vm_area_struct *vma, bool compound)
534{
535 if (unlikely(vma->vm_flags & VM_LOCKED) &&
536 (compound || !PageTransCompound(page)))
537 munlock_page(page);
538}
539void mlock_new_page(struct page *page);
540bool need_mlock_page_drain(int cpu);
541void mlock_page_drain_local(void);
542void mlock_page_drain_remote(int cpu);
543
544extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
545
546/*
547 * Return the start of user virtual address at the specific offset within
548 * a vma.
549 */
550static inline unsigned long
551vma_pgoff_address(pgoff_t pgoff, unsigned long nr_pages,
552 struct vm_area_struct *vma)
553{
554 unsigned long address;
555
556 if (pgoff >= vma->vm_pgoff) {
557 address = vma->vm_start +
558 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
559 /* Check for address beyond vma (or wrapped through 0?) */
560 if (address < vma->vm_start || address >= vma->vm_end)
561 address = -EFAULT;
562 } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
563 /* Test above avoids possibility of wrap to 0 on 32-bit */
564 address = vma->vm_start;
565 } else {
566 address = -EFAULT;
567 }
568 return address;
569}
570
571/*
572 * Return the start of user virtual address of a page within a vma.
573 * Returns -EFAULT if all of the page is outside the range of vma.
574 * If page is a compound head, the entire compound page is considered.
575 */
576static inline unsigned long
577vma_address(struct page *page, struct vm_area_struct *vma)
578{
579 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
580 return vma_pgoff_address(page_to_pgoff(page), compound_nr(page), vma);
581}
582
583/*
584 * Then at what user virtual address will none of the range be found in vma?
585 * Assumes that vma_address() already returned a good starting address.
586 */
587static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
588{
589 struct vm_area_struct *vma = pvmw->vma;
590 pgoff_t pgoff;
591 unsigned long address;
592
593 /* Common case, plus ->pgoff is invalid for KSM */
594 if (pvmw->nr_pages == 1)
595 return pvmw->address + PAGE_SIZE;
596
597 pgoff = pvmw->pgoff + pvmw->nr_pages;
598 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
599 /* Check for address beyond vma (or wrapped through 0?) */
600 if (address < vma->vm_start || address > vma->vm_end)
601 address = vma->vm_end;
602 return address;
603}
604
605static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
606 struct file *fpin)
607{
608 int flags = vmf->flags;
609
610 if (fpin)
611 return fpin;
612
613 /*
614 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
615 * anything, so we only pin the file and drop the mmap_lock if only
616 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
617 */
618 if (fault_flag_allow_retry_first(flags) &&
619 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
620 fpin = get_file(vmf->vma->vm_file);
621 mmap_read_unlock(vmf->vma->vm_mm);
622 }
623 return fpin;
624}
625#else /* !CONFIG_MMU */
626static inline void unmap_mapping_folio(struct folio *folio) { }
627static inline void mlock_vma_page(struct page *page,
628 struct vm_area_struct *vma, bool compound) { }
629static inline void munlock_vma_page(struct page *page,
630 struct vm_area_struct *vma, bool compound) { }
631static inline void mlock_new_page(struct page *page) { }
632static inline bool need_mlock_page_drain(int cpu) { return false; }
633static inline void mlock_page_drain_local(void) { }
634static inline void mlock_page_drain_remote(int cpu) { }
635static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
636{
637}
638#endif /* !CONFIG_MMU */
639
640/* Memory initialisation debug and verification */
641enum mminit_level {
642 MMINIT_WARNING,
643 MMINIT_VERIFY,
644 MMINIT_TRACE
645};
646
647#ifdef CONFIG_DEBUG_MEMORY_INIT
648
649extern int mminit_loglevel;
650
651#define mminit_dprintk(level, prefix, fmt, arg...) \
652do { \
653 if (level < mminit_loglevel) { \
654 if (level <= MMINIT_WARNING) \
655 pr_warn("mminit::" prefix " " fmt, ##arg); \
656 else \
657 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
658 } \
659} while (0)
660
661extern void mminit_verify_pageflags_layout(void);
662extern void mminit_verify_zonelist(void);
663#else
664
665static inline void mminit_dprintk(enum mminit_level level,
666 const char *prefix, const char *fmt, ...)
667{
668}
669
670static inline void mminit_verify_pageflags_layout(void)
671{
672}
673
674static inline void mminit_verify_zonelist(void)
675{
676}
677#endif /* CONFIG_DEBUG_MEMORY_INIT */
678
679#define NODE_RECLAIM_NOSCAN -2
680#define NODE_RECLAIM_FULL -1
681#define NODE_RECLAIM_SOME 0
682#define NODE_RECLAIM_SUCCESS 1
683
684#ifdef CONFIG_NUMA
685extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
686extern int find_next_best_node(int node, nodemask_t *used_node_mask);
687#else
688static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
689 unsigned int order)
690{
691 return NODE_RECLAIM_NOSCAN;
692}
693static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
694{
695 return NUMA_NO_NODE;
696}
697#endif
698
699/*
700 * mm/memory-failure.c
701 */
702extern int hwpoison_filter(struct page *p);
703
704extern u32 hwpoison_filter_dev_major;
705extern u32 hwpoison_filter_dev_minor;
706extern u64 hwpoison_filter_flags_mask;
707extern u64 hwpoison_filter_flags_value;
708extern u64 hwpoison_filter_memcg;
709extern u32 hwpoison_filter_enable;
710
711extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
712 unsigned long, unsigned long,
713 unsigned long, unsigned long);
714
715extern void set_pageblock_order(void);
716unsigned int reclaim_clean_pages_from_list(struct zone *zone,
717 struct list_head *page_list);
718/* The ALLOC_WMARK bits are used as an index to zone->watermark */
719#define ALLOC_WMARK_MIN WMARK_MIN
720#define ALLOC_WMARK_LOW WMARK_LOW
721#define ALLOC_WMARK_HIGH WMARK_HIGH
722#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
723
724/* Mask to get the watermark bits */
725#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
726
727/*
728 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
729 * cannot assume a reduced access to memory reserves is sufficient for
730 * !MMU
731 */
732#ifdef CONFIG_MMU
733#define ALLOC_OOM 0x08
734#else
735#define ALLOC_OOM ALLOC_NO_WATERMARKS
736#endif
737
738#define ALLOC_HARDER 0x10 /* try to alloc harder */
739#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
740#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
741#define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
742#ifdef CONFIG_ZONE_DMA32
743#define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
744#else
745#define ALLOC_NOFRAGMENT 0x0
746#endif
747#define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
748
749enum ttu_flags;
750struct tlbflush_unmap_batch;
751
752
753/*
754 * only for MM internal work items which do not depend on
755 * any allocations or locks which might depend on allocations
756 */
757extern struct workqueue_struct *mm_percpu_wq;
758
759#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
760void try_to_unmap_flush(void);
761void try_to_unmap_flush_dirty(void);
762void flush_tlb_batched_pending(struct mm_struct *mm);
763#else
764static inline void try_to_unmap_flush(void)
765{
766}
767static inline void try_to_unmap_flush_dirty(void)
768{
769}
770static inline void flush_tlb_batched_pending(struct mm_struct *mm)
771{
772}
773#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
774
775extern const struct trace_print_flags pageflag_names[];
776extern const struct trace_print_flags vmaflag_names[];
777extern const struct trace_print_flags gfpflag_names[];
778
779static inline bool is_migrate_highatomic(enum migratetype migratetype)
780{
781 return migratetype == MIGRATE_HIGHATOMIC;
782}
783
784static inline bool is_migrate_highatomic_page(struct page *page)
785{
786 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
787}
788
789void setup_zone_pageset(struct zone *zone);
790
791struct migration_target_control {
792 int nid; /* preferred node id */
793 nodemask_t *nmask;
794 gfp_t gfp_mask;
795};
796
797/*
798 * mm/vmalloc.c
799 */
800#ifdef CONFIG_MMU
801int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
802 pgprot_t prot, struct page **pages, unsigned int page_shift);
803#else
804static inline
805int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
806 pgprot_t prot, struct page **pages, unsigned int page_shift)
807{
808 return -EINVAL;
809}
810#endif
811
812int __vmap_pages_range_noflush(unsigned long addr, unsigned long end,
813 pgprot_t prot, struct page **pages,
814 unsigned int page_shift);
815
816void vunmap_range_noflush(unsigned long start, unsigned long end);
817
818void __vunmap_range_noflush(unsigned long start, unsigned long end);
819
820int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
821 unsigned long addr, int page_nid, int *flags);
822
823void free_zone_device_page(struct page *page);
824int migrate_device_coherent_page(struct page *page);
825
826/*
827 * mm/gup.c
828 */
829struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags);
830
831extern bool mirrored_kernelcore;
832
833static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
834{
835 /*
836 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
837 * enablements, because when without soft-dirty being compiled in,
838 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
839 * will be constantly true.
840 */
841 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
842 return false;
843
844 /*
845 * Soft-dirty is kind of special: its tracking is enabled when the
846 * vma flags not set.
847 */
848 return !(vma->vm_flags & VM_SOFTDIRTY);
849}
850
851#endif /* __MM_INTERNAL_H */