Line data Source code
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 :
16 : struct 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_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 :
53 : void page_writeback_init(void);
54 :
55 : /*
56 : * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
57 : * its nr_pages_mapped would be 0x400000: choose the COMPOUND_MAPPED bit
58 : * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE). Hugetlb currently
59 : * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
60 : */
61 : #define COMPOUND_MAPPED 0x800000
62 : #define FOLIO_PAGES_MAPPED (COMPOUND_MAPPED - 1)
63 :
64 : /*
65 : * How many individual pages have an elevated _mapcount. Excludes
66 : * the folio's entire_mapcount.
67 : */
68 : static inline int folio_nr_pages_mapped(struct folio *folio)
69 : {
70 0 : return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
71 : }
72 :
73 : static inline void *folio_raw_mapping(struct folio *folio)
74 : {
75 0 : unsigned long mapping = (unsigned long)folio->mapping;
76 :
77 0 : return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
78 : }
79 :
80 : void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
81 : int nr_throttled);
82 : static inline void acct_reclaim_writeback(struct folio *folio)
83 : {
84 0 : pg_data_t *pgdat = folio_pgdat(folio);
85 0 : int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
86 :
87 0 : if (nr_throttled)
88 0 : __acct_reclaim_writeback(pgdat, folio, nr_throttled);
89 : }
90 :
91 : static inline void wake_throttle_isolated(pg_data_t *pgdat)
92 : {
93 : wait_queue_head_t *wqh;
94 :
95 0 : wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
96 0 : if (waitqueue_active(wqh))
97 0 : wake_up(wqh);
98 : }
99 :
100 : vm_fault_t do_swap_page(struct vm_fault *vmf);
101 : void folio_rotate_reclaimable(struct folio *folio);
102 : bool __folio_end_writeback(struct folio *folio);
103 : void deactivate_file_folio(struct folio *folio);
104 : void folio_activate(struct folio *folio);
105 :
106 : void free_pgtables(struct mmu_gather *tlb, struct maple_tree *mt,
107 : struct vm_area_struct *start_vma, unsigned long floor,
108 : unsigned long ceiling, bool mm_wr_locked);
109 : void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
110 :
111 : struct zap_details;
112 : void unmap_page_range(struct mmu_gather *tlb,
113 : struct vm_area_struct *vma,
114 : unsigned long addr, unsigned long end,
115 : struct zap_details *details);
116 :
117 : void page_cache_ra_order(struct readahead_control *, struct file_ra_state *,
118 : unsigned int order);
119 : void force_page_cache_ra(struct readahead_control *, unsigned long nr);
120 : static inline void force_page_cache_readahead(struct address_space *mapping,
121 : struct file *file, pgoff_t index, unsigned long nr_to_read)
122 : {
123 0 : DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
124 0 : force_page_cache_ra(&ractl, nr_to_read);
125 : }
126 :
127 : unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
128 : pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
129 : unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
130 : pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
131 : void filemap_free_folio(struct address_space *mapping, struct folio *folio);
132 : int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
133 : bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
134 : loff_t end);
135 : long invalidate_inode_page(struct page *page);
136 : unsigned long invalidate_mapping_pagevec(struct address_space *mapping,
137 : pgoff_t start, pgoff_t end, unsigned long *nr_pagevec);
138 :
139 : /**
140 : * folio_evictable - Test whether a folio is evictable.
141 : * @folio: The folio to test.
142 : *
143 : * Test whether @folio is evictable -- i.e., should be placed on
144 : * active/inactive lists vs unevictable list.
145 : *
146 : * Reasons folio might not be evictable:
147 : * 1. folio's mapping marked unevictable
148 : * 2. One of the pages in the folio is part of an mlocked VMA
149 : */
150 0 : static inline bool folio_evictable(struct folio *folio)
151 : {
152 : bool ret;
153 :
154 : /* Prevent address_space of inode and swap cache from being freed */
155 : rcu_read_lock();
156 0 : ret = !mapping_unevictable(folio_mapping(folio)) &&
157 0 : !folio_test_mlocked(folio);
158 : rcu_read_unlock();
159 0 : return ret;
160 : }
161 :
162 : /*
163 : * Turn a non-refcounted page (->_refcount == 0) into refcounted with
164 : * a count of one.
165 : */
166 : static inline void set_page_refcounted(struct page *page)
167 : {
168 : VM_BUG_ON_PAGE(PageTail(page), page);
169 : VM_BUG_ON_PAGE(page_ref_count(page), page);
170 44778 : set_page_count(page, 1);
171 : }
172 :
173 : extern unsigned long highest_memmap_pfn;
174 :
175 : /*
176 : * Maximum number of reclaim retries without progress before the OOM
177 : * killer is consider the only way forward.
178 : */
179 : #define MAX_RECLAIM_RETRIES 16
180 :
181 : /*
182 : * in mm/early_ioremap.c
183 : */
184 : pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr,
185 : unsigned long size, pgprot_t prot);
186 :
187 : /*
188 : * in mm/vmscan.c:
189 : */
190 : bool isolate_lru_page(struct page *page);
191 : bool folio_isolate_lru(struct folio *folio);
192 : void putback_lru_page(struct page *page);
193 : void folio_putback_lru(struct folio *folio);
194 : extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
195 :
196 : /*
197 : * in mm/rmap.c:
198 : */
199 : pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
200 :
201 : /*
202 : * in mm/page_alloc.c
203 : */
204 : #define K(x) ((x) << (PAGE_SHIFT-10))
205 :
206 : extern char * const zone_names[MAX_NR_ZONES];
207 :
208 : /* perform sanity checks on struct pages being allocated or freed */
209 : DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
210 :
211 44278 : static inline bool is_check_pages_enabled(void)
212 : {
213 343510 : return static_branch_unlikely(&check_pages_enabled);
214 : }
215 :
216 : /*
217 : * Structure for holding the mostly immutable allocation parameters passed
218 : * between functions involved in allocations, including the alloc_pages*
219 : * family of functions.
220 : *
221 : * nodemask, migratetype and highest_zoneidx are initialized only once in
222 : * __alloc_pages() and then never change.
223 : *
224 : * zonelist, preferred_zone and highest_zoneidx are set first in
225 : * __alloc_pages() for the fast path, and might be later changed
226 : * in __alloc_pages_slowpath(). All other functions pass the whole structure
227 : * by a const pointer.
228 : */
229 : struct alloc_context {
230 : struct zonelist *zonelist;
231 : nodemask_t *nodemask;
232 : struct zoneref *preferred_zoneref;
233 : int migratetype;
234 :
235 : /*
236 : * highest_zoneidx represents highest usable zone index of
237 : * the allocation request. Due to the nature of the zone,
238 : * memory on lower zone than the highest_zoneidx will be
239 : * protected by lowmem_reserve[highest_zoneidx].
240 : *
241 : * highest_zoneidx is also used by reclaim/compaction to limit
242 : * the target zone since higher zone than this index cannot be
243 : * usable for this allocation request.
244 : */
245 : enum zone_type highest_zoneidx;
246 : bool spread_dirty_pages;
247 : };
248 :
249 : /*
250 : * This function returns the order of a free page in the buddy system. In
251 : * general, page_zone(page)->lock must be held by the caller to prevent the
252 : * page from being allocated in parallel and returning garbage as the order.
253 : * If a caller does not hold page_zone(page)->lock, it must guarantee that the
254 : * page cannot be allocated or merged in parallel. Alternatively, it must
255 : * handle invalid values gracefully, and use buddy_order_unsafe() below.
256 : */
257 : static inline unsigned int buddy_order(struct page *page)
258 : {
259 : /* PageBuddy() must be checked by the caller */
260 1388 : return page_private(page);
261 : }
262 :
263 : /*
264 : * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
265 : * PageBuddy() should be checked first by the caller to minimize race window,
266 : * and invalid values must be handled gracefully.
267 : *
268 : * READ_ONCE is used so that if the caller assigns the result into a local
269 : * variable and e.g. tests it for valid range before using, the compiler cannot
270 : * decide to remove the variable and inline the page_private(page) multiple
271 : * times, potentially observing different values in the tests and the actual
272 : * use of the result.
273 : */
274 : #define buddy_order_unsafe(page) READ_ONCE(page_private(page))
275 :
276 : /*
277 : * This function checks whether a page is free && is the buddy
278 : * we can coalesce a page and its buddy if
279 : * (a) the buddy is not in a hole (check before calling!) &&
280 : * (b) the buddy is in the buddy system &&
281 : * (c) a page and its buddy have the same order &&
282 : * (d) a page and its buddy are in the same zone.
283 : *
284 : * For recording whether a page is in the buddy system, we set PageBuddy.
285 : * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
286 : *
287 : * For recording page's order, we use page_private(page).
288 : */
289 : static inline bool page_is_buddy(struct page *page, struct page *buddy,
290 : unsigned int order)
291 : {
292 5638 : if (!page_is_guard(buddy) && !PageBuddy(buddy))
293 : return false;
294 :
295 2768 : if (buddy_order(buddy) != order)
296 : return false;
297 :
298 : /*
299 : * zone check is done late to avoid uselessly calculating
300 : * zone/node ids for pages that could never merge.
301 : */
302 4113 : if (page_zone_id(page) != page_zone_id(buddy))
303 : return false;
304 :
305 : VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
306 :
307 : return true;
308 : }
309 :
310 : /*
311 : * Locate the struct page for both the matching buddy in our
312 : * pair (buddy1) and the combined O(n+1) page they form (page).
313 : *
314 : * 1) Any buddy B1 will have an order O twin B2 which satisfies
315 : * the following equation:
316 : * B2 = B1 ^ (1 << O)
317 : * For example, if the starting buddy (buddy2) is #8 its order
318 : * 1 buddy is #10:
319 : * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
320 : *
321 : * 2) Any buddy B will have an order O+1 parent P which
322 : * satisfies the following equation:
323 : * P = B & ~(1 << O)
324 : *
325 : * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
326 : */
327 : static inline unsigned long
328 : __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
329 : {
330 2819 : return page_pfn ^ (1 << order);
331 : }
332 :
333 : /*
334 : * Find the buddy of @page and validate it.
335 : * @page: The input page
336 : * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
337 : * function is used in the performance-critical __free_one_page().
338 : * @order: The order of the page
339 : * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
340 : * page_to_pfn().
341 : *
342 : * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
343 : * not the same as @page. The validation is necessary before use it.
344 : *
345 : * Return: the found buddy page or NULL if not found.
346 : */
347 : static inline struct page *find_buddy_page_pfn(struct page *page,
348 : unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
349 : {
350 2819 : unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
351 : struct page *buddy;
352 :
353 2819 : buddy = page + (__buddy_pfn - pfn);
354 : if (buddy_pfn)
355 1874 : *buddy_pfn = __buddy_pfn;
356 :
357 2819 : if (page_is_buddy(page, buddy, order))
358 : return buddy;
359 : return NULL;
360 : }
361 :
362 : extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
363 : unsigned long end_pfn, struct zone *zone);
364 :
365 : static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
366 : unsigned long end_pfn, struct zone *zone)
367 : {
368 0 : if (zone->contiguous)
369 0 : return pfn_to_page(start_pfn);
370 :
371 0 : return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
372 : }
373 :
374 : extern int __isolate_free_page(struct page *page, unsigned int order);
375 : extern void __putback_isolated_page(struct page *page, unsigned int order,
376 : int mt);
377 : extern void memblock_free_pages(struct page *page, unsigned long pfn,
378 : unsigned int order);
379 : extern void __free_pages_core(struct page *page, unsigned int order);
380 :
381 : static inline void prep_compound_head(struct page *page, unsigned int order)
382 : {
383 102 : struct folio *folio = (struct folio *)page;
384 :
385 102 : set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
386 102 : set_compound_order(page, order);
387 204 : atomic_set(&folio->_entire_mapcount, -1);
388 204 : atomic_set(&folio->_nr_pages_mapped, 0);
389 204 : atomic_set(&folio->_pincount, 0);
390 : }
391 :
392 : static inline void prep_compound_tail(struct page *head, int tail_idx)
393 : {
394 184 : struct page *p = head + tail_idx;
395 :
396 184 : p->mapping = TAIL_MAPPING;
397 184 : set_compound_head(p, head);
398 368 : set_page_private(p, 0);
399 : }
400 :
401 : extern void prep_compound_page(struct page *page, unsigned int order);
402 :
403 : extern void post_alloc_hook(struct page *page, unsigned int order,
404 : gfp_t gfp_flags);
405 : extern int user_min_free_kbytes;
406 :
407 : extern void free_unref_page(struct page *page, unsigned int order);
408 : extern void free_unref_page_list(struct list_head *list);
409 :
410 : extern void zone_pcp_reset(struct zone *zone);
411 : extern void zone_pcp_disable(struct zone *zone);
412 : extern void zone_pcp_enable(struct zone *zone);
413 : extern void zone_pcp_init(struct zone *zone);
414 :
415 : extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
416 : phys_addr_t min_addr,
417 : int nid, bool exact_nid);
418 :
419 : int split_free_page(struct page *free_page,
420 : unsigned int order, unsigned long split_pfn_offset);
421 :
422 : /*
423 : * This will have no effect, other than possibly generating a warning, if the
424 : * caller passes in a non-large folio.
425 : */
426 : static inline void folio_set_order(struct folio *folio, unsigned int order)
427 : {
428 : if (WARN_ON_ONCE(!folio_test_large(folio)))
429 : return;
430 :
431 : folio->_folio_order = order;
432 : #ifdef CONFIG_64BIT
433 : /*
434 : * When hugetlb dissolves a folio, we need to clear the tail
435 : * page, rather than setting nr_pages to 1.
436 : */
437 : folio->_folio_nr_pages = order ? 1U << order : 0;
438 : #endif
439 : }
440 :
441 : #if defined CONFIG_COMPACTION || defined CONFIG_CMA
442 :
443 : /*
444 : * in mm/compaction.c
445 : */
446 : /*
447 : * compact_control is used to track pages being migrated and the free pages
448 : * they are being migrated to during memory compaction. The free_pfn starts
449 : * at the end of a zone and migrate_pfn begins at the start. Movable pages
450 : * are moved to the end of a zone during a compaction run and the run
451 : * completes when free_pfn <= migrate_pfn
452 : */
453 : struct compact_control {
454 : struct list_head freepages; /* List of free pages to migrate to */
455 : struct list_head migratepages; /* List of pages being migrated */
456 : unsigned int nr_freepages; /* Number of isolated free pages */
457 : unsigned int nr_migratepages; /* Number of pages to migrate */
458 : unsigned long free_pfn; /* isolate_freepages search base */
459 : /*
460 : * Acts as an in/out parameter to page isolation for migration.
461 : * isolate_migratepages uses it as a search base.
462 : * isolate_migratepages_block will update the value to the next pfn
463 : * after the last isolated one.
464 : */
465 : unsigned long migrate_pfn;
466 : unsigned long fast_start_pfn; /* a pfn to start linear scan from */
467 : struct zone *zone;
468 : unsigned long total_migrate_scanned;
469 : unsigned long total_free_scanned;
470 : unsigned short fast_search_fail;/* failures to use free list searches */
471 : short search_order; /* order to start a fast search at */
472 : const gfp_t gfp_mask; /* gfp mask of a direct compactor */
473 : int order; /* order a direct compactor needs */
474 : int migratetype; /* migratetype of direct compactor */
475 : const unsigned int alloc_flags; /* alloc flags of a direct compactor */
476 : const int highest_zoneidx; /* zone index of a direct compactor */
477 : enum migrate_mode mode; /* Async or sync migration mode */
478 : bool ignore_skip_hint; /* Scan blocks even if marked skip */
479 : bool no_set_skip_hint; /* Don't mark blocks for skipping */
480 : bool ignore_block_suitable; /* Scan blocks considered unsuitable */
481 : bool direct_compaction; /* False from kcompactd or /proc/... */
482 : bool proactive_compaction; /* kcompactd proactive compaction */
483 : bool whole_zone; /* Whole zone should/has been scanned */
484 : bool contended; /* Signal lock contention */
485 : bool finish_pageblock; /* Scan the remainder of a pageblock. Used
486 : * when there are potentially transient
487 : * isolation or migration failures to
488 : * ensure forward progress.
489 : */
490 : bool alloc_contig; /* alloc_contig_range allocation */
491 : };
492 :
493 : /*
494 : * Used in direct compaction when a page should be taken from the freelists
495 : * immediately when one is created during the free path.
496 : */
497 : struct capture_control {
498 : struct compact_control *cc;
499 : struct page *page;
500 : };
501 :
502 : unsigned long
503 : isolate_freepages_range(struct compact_control *cc,
504 : unsigned long start_pfn, unsigned long end_pfn);
505 : int
506 : isolate_migratepages_range(struct compact_control *cc,
507 : unsigned long low_pfn, unsigned long end_pfn);
508 :
509 : int __alloc_contig_migrate_range(struct compact_control *cc,
510 : unsigned long start, unsigned long end);
511 :
512 : /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
513 : void init_cma_reserved_pageblock(struct page *page);
514 :
515 : #endif /* CONFIG_COMPACTION || CONFIG_CMA */
516 :
517 : int find_suitable_fallback(struct free_area *area, unsigned int order,
518 : int migratetype, bool only_stealable, bool *can_steal);
519 :
520 : static inline bool free_area_empty(struct free_area *area, int migratetype)
521 : {
522 348 : return list_empty(&area->free_list[migratetype]);
523 : }
524 :
525 : /*
526 : * These three helpers classifies VMAs for virtual memory accounting.
527 : */
528 :
529 : /*
530 : * Executable code area - executable, not writable, not stack
531 : */
532 : static inline bool is_exec_mapping(vm_flags_t flags)
533 : {
534 0 : return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
535 : }
536 :
537 : /*
538 : * Stack area - automatically grows in one direction
539 : *
540 : * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
541 : * do_mmap() forbids all other combinations.
542 : */
543 : static inline bool is_stack_mapping(vm_flags_t flags)
544 : {
545 0 : return (flags & VM_STACK) == VM_STACK;
546 : }
547 :
548 : /*
549 : * Data area - private, writable, not stack
550 : */
551 : static inline bool is_data_mapping(vm_flags_t flags)
552 : {
553 0 : return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
554 : }
555 :
556 : /* mm/util.c */
557 : struct anon_vma *folio_anon_vma(struct folio *folio);
558 :
559 : #ifdef CONFIG_MMU
560 : void unmap_mapping_folio(struct folio *folio);
561 : extern long populate_vma_page_range(struct vm_area_struct *vma,
562 : unsigned long start, unsigned long end, int *locked);
563 : extern long faultin_vma_page_range(struct vm_area_struct *vma,
564 : unsigned long start, unsigned long end,
565 : bool write, int *locked);
566 : extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
567 : unsigned long len);
568 : /*
569 : * mlock_vma_folio() and munlock_vma_folio():
570 : * should be called with vma's mmap_lock held for read or write,
571 : * under page table lock for the pte/pmd being added or removed.
572 : *
573 : * mlock is usually called at the end of page_add_*_rmap(), munlock at
574 : * the end of page_remove_rmap(); but new anon folios are managed by
575 : * folio_add_lru_vma() calling mlock_new_folio().
576 : *
577 : * @compound is used to include pmd mappings of THPs, but filter out
578 : * pte mappings of THPs, which cannot be consistently counted: a pte
579 : * mapping of the THP head cannot be distinguished by the page alone.
580 : */
581 : void mlock_folio(struct folio *folio);
582 0 : static inline void mlock_vma_folio(struct folio *folio,
583 : struct vm_area_struct *vma, bool compound)
584 : {
585 : /*
586 : * The VM_SPECIAL check here serves two purposes.
587 : * 1) VM_IO check prevents migration from double-counting during mlock.
588 : * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
589 : * is never left set on a VM_SPECIAL vma, there is an interval while
590 : * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
591 : * still be set while VM_SPECIAL bits are added: so ignore it then.
592 : */
593 0 : if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) &&
594 0 : (compound || !folio_test_large(folio)))
595 0 : mlock_folio(folio);
596 0 : }
597 :
598 : void munlock_folio(struct folio *folio);
599 0 : static inline void munlock_vma_folio(struct folio *folio,
600 : struct vm_area_struct *vma, bool compound)
601 : {
602 0 : if (unlikely(vma->vm_flags & VM_LOCKED) &&
603 0 : (compound || !folio_test_large(folio)))
604 0 : munlock_folio(folio);
605 0 : }
606 :
607 : void mlock_new_folio(struct folio *folio);
608 : bool need_mlock_drain(int cpu);
609 : void mlock_drain_local(void);
610 : void mlock_drain_remote(int cpu);
611 :
612 : extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
613 :
614 : /*
615 : * Return the start of user virtual address at the specific offset within
616 : * a vma.
617 : */
618 : static inline unsigned long
619 : vma_pgoff_address(pgoff_t pgoff, unsigned long nr_pages,
620 : struct vm_area_struct *vma)
621 : {
622 : unsigned long address;
623 :
624 0 : if (pgoff >= vma->vm_pgoff) {
625 0 : address = vma->vm_start +
626 0 : ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
627 : /* Check for address beyond vma (or wrapped through 0?) */
628 0 : if (address < vma->vm_start || address >= vma->vm_end)
629 0 : address = -EFAULT;
630 0 : } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
631 : /* Test above avoids possibility of wrap to 0 on 32-bit */
632 0 : address = vma->vm_start;
633 : } else {
634 : address = -EFAULT;
635 : }
636 : return address;
637 : }
638 :
639 : /*
640 : * Return the start of user virtual address of a page within a vma.
641 : * Returns -EFAULT if all of the page is outside the range of vma.
642 : * If page is a compound head, the entire compound page is considered.
643 : */
644 : static inline unsigned long
645 0 : vma_address(struct page *page, struct vm_area_struct *vma)
646 : {
647 : VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
648 0 : return vma_pgoff_address(page_to_pgoff(page), compound_nr(page), vma);
649 : }
650 :
651 : /*
652 : * Then at what user virtual address will none of the range be found in vma?
653 : * Assumes that vma_address() already returned a good starting address.
654 : */
655 : static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
656 : {
657 0 : struct vm_area_struct *vma = pvmw->vma;
658 : pgoff_t pgoff;
659 : unsigned long address;
660 :
661 : /* Common case, plus ->pgoff is invalid for KSM */
662 0 : if (pvmw->nr_pages == 1)
663 0 : return pvmw->address + PAGE_SIZE;
664 :
665 0 : pgoff = pvmw->pgoff + pvmw->nr_pages;
666 0 : address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
667 : /* Check for address beyond vma (or wrapped through 0?) */
668 0 : if (address < vma->vm_start || address > vma->vm_end)
669 0 : address = vma->vm_end;
670 : return address;
671 : }
672 :
673 0 : static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
674 : struct file *fpin)
675 : {
676 0 : int flags = vmf->flags;
677 :
678 0 : if (fpin)
679 : return fpin;
680 :
681 : /*
682 : * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
683 : * anything, so we only pin the file and drop the mmap_lock if only
684 : * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
685 : */
686 0 : if (fault_flag_allow_retry_first(flags) &&
687 0 : !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
688 0 : fpin = get_file(vmf->vma->vm_file);
689 0 : mmap_read_unlock(vmf->vma->vm_mm);
690 : }
691 : return fpin;
692 : }
693 : #else /* !CONFIG_MMU */
694 : static inline void unmap_mapping_folio(struct folio *folio) { }
695 : static inline void mlock_new_folio(struct folio *folio) { }
696 : static inline bool need_mlock_drain(int cpu) { return false; }
697 : static inline void mlock_drain_local(void) { }
698 : static inline void mlock_drain_remote(int cpu) { }
699 : static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
700 : {
701 : }
702 : #endif /* !CONFIG_MMU */
703 :
704 : /* Memory initialisation debug and verification */
705 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
706 : DECLARE_STATIC_KEY_TRUE(deferred_pages);
707 :
708 : bool __init deferred_grow_zone(struct zone *zone, unsigned int order);
709 : #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
710 :
711 : enum mminit_level {
712 : MMINIT_WARNING,
713 : MMINIT_VERIFY,
714 : MMINIT_TRACE
715 : };
716 :
717 : #ifdef CONFIG_DEBUG_MEMORY_INIT
718 :
719 : extern int mminit_loglevel;
720 :
721 : #define mminit_dprintk(level, prefix, fmt, arg...) \
722 : do { \
723 : if (level < mminit_loglevel) { \
724 : if (level <= MMINIT_WARNING) \
725 : pr_warn("mminit::" prefix " " fmt, ##arg); \
726 : else \
727 : printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
728 : } \
729 : } while (0)
730 :
731 : extern void mminit_verify_pageflags_layout(void);
732 : extern void mminit_verify_zonelist(void);
733 : #else
734 :
735 : static inline void mminit_dprintk(enum mminit_level level,
736 : const char *prefix, const char *fmt, ...)
737 : {
738 : }
739 :
740 : static inline void mminit_verify_pageflags_layout(void)
741 : {
742 : }
743 :
744 : static inline void mminit_verify_zonelist(void)
745 : {
746 : }
747 : #endif /* CONFIG_DEBUG_MEMORY_INIT */
748 :
749 : #define NODE_RECLAIM_NOSCAN -2
750 : #define NODE_RECLAIM_FULL -1
751 : #define NODE_RECLAIM_SOME 0
752 : #define NODE_RECLAIM_SUCCESS 1
753 :
754 : #ifdef CONFIG_NUMA
755 : extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
756 : extern int find_next_best_node(int node, nodemask_t *used_node_mask);
757 : #else
758 : static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
759 : unsigned int order)
760 : {
761 : return NODE_RECLAIM_NOSCAN;
762 : }
763 : static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
764 : {
765 : return NUMA_NO_NODE;
766 : }
767 : #endif
768 :
769 : /*
770 : * mm/memory-failure.c
771 : */
772 : extern int hwpoison_filter(struct page *p);
773 :
774 : extern u32 hwpoison_filter_dev_major;
775 : extern u32 hwpoison_filter_dev_minor;
776 : extern u64 hwpoison_filter_flags_mask;
777 : extern u64 hwpoison_filter_flags_value;
778 : extern u64 hwpoison_filter_memcg;
779 : extern u32 hwpoison_filter_enable;
780 :
781 : extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
782 : unsigned long, unsigned long,
783 : unsigned long, unsigned long);
784 :
785 : extern void set_pageblock_order(void);
786 : unsigned long reclaim_pages(struct list_head *folio_list);
787 : unsigned int reclaim_clean_pages_from_list(struct zone *zone,
788 : struct list_head *folio_list);
789 : /* The ALLOC_WMARK bits are used as an index to zone->watermark */
790 : #define ALLOC_WMARK_MIN WMARK_MIN
791 : #define ALLOC_WMARK_LOW WMARK_LOW
792 : #define ALLOC_WMARK_HIGH WMARK_HIGH
793 : #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
794 :
795 : /* Mask to get the watermark bits */
796 : #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
797 :
798 : /*
799 : * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
800 : * cannot assume a reduced access to memory reserves is sufficient for
801 : * !MMU
802 : */
803 : #ifdef CONFIG_MMU
804 : #define ALLOC_OOM 0x08
805 : #else
806 : #define ALLOC_OOM ALLOC_NO_WATERMARKS
807 : #endif
808 :
809 : #define ALLOC_NON_BLOCK 0x10 /* Caller cannot block. Allow access
810 : * to 25% of the min watermark or
811 : * 62.5% if __GFP_HIGH is set.
812 : */
813 : #define ALLOC_MIN_RESERVE 0x20 /* __GFP_HIGH set. Allow access to 50%
814 : * of the min watermark.
815 : */
816 : #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
817 : #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
818 : #ifdef CONFIG_ZONE_DMA32
819 : #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
820 : #else
821 : #define ALLOC_NOFRAGMENT 0x0
822 : #endif
823 : #define ALLOC_HIGHATOMIC 0x200 /* Allows access to MIGRATE_HIGHATOMIC */
824 : #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
825 :
826 : /* Flags that allow allocations below the min watermark. */
827 : #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
828 :
829 : enum ttu_flags;
830 : struct tlbflush_unmap_batch;
831 :
832 :
833 : /*
834 : * only for MM internal work items which do not depend on
835 : * any allocations or locks which might depend on allocations
836 : */
837 : extern struct workqueue_struct *mm_percpu_wq;
838 :
839 : #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
840 : void try_to_unmap_flush(void);
841 : void try_to_unmap_flush_dirty(void);
842 : void flush_tlb_batched_pending(struct mm_struct *mm);
843 : #else
844 : static inline void try_to_unmap_flush(void)
845 : {
846 : }
847 : static inline void try_to_unmap_flush_dirty(void)
848 : {
849 : }
850 : static inline void flush_tlb_batched_pending(struct mm_struct *mm)
851 : {
852 : }
853 : #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
854 :
855 : extern const struct trace_print_flags pageflag_names[];
856 : extern const struct trace_print_flags pagetype_names[];
857 : extern const struct trace_print_flags vmaflag_names[];
858 : extern const struct trace_print_flags gfpflag_names[];
859 :
860 : static inline bool is_migrate_highatomic(enum migratetype migratetype)
861 : {
862 : return migratetype == MIGRATE_HIGHATOMIC;
863 : }
864 :
865 0 : static inline bool is_migrate_highatomic_page(struct page *page)
866 : {
867 0 : return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
868 : }
869 :
870 : void setup_zone_pageset(struct zone *zone);
871 :
872 : struct migration_target_control {
873 : int nid; /* preferred node id */
874 : nodemask_t *nmask;
875 : gfp_t gfp_mask;
876 : };
877 :
878 : /*
879 : * mm/filemap.c
880 : */
881 : size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
882 : struct folio *folio, loff_t fpos, size_t size);
883 :
884 : /*
885 : * mm/vmalloc.c
886 : */
887 : #ifdef CONFIG_MMU
888 : void __init vmalloc_init(void);
889 : int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
890 : pgprot_t prot, struct page **pages, unsigned int page_shift);
891 : #else
892 : static inline void vmalloc_init(void)
893 : {
894 : }
895 :
896 : static inline
897 : int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
898 : pgprot_t prot, struct page **pages, unsigned int page_shift)
899 : {
900 : return -EINVAL;
901 : }
902 : #endif
903 :
904 : int __must_check __vmap_pages_range_noflush(unsigned long addr,
905 : unsigned long end, pgprot_t prot,
906 : struct page **pages, unsigned int page_shift);
907 :
908 : void vunmap_range_noflush(unsigned long start, unsigned long end);
909 :
910 : void __vunmap_range_noflush(unsigned long start, unsigned long end);
911 :
912 : int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
913 : unsigned long addr, int page_nid, int *flags);
914 :
915 : void free_zone_device_page(struct page *page);
916 : int migrate_device_coherent_page(struct page *page);
917 :
918 : /*
919 : * mm/gup.c
920 : */
921 : struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags);
922 : int __must_check try_grab_page(struct page *page, unsigned int flags);
923 :
924 : enum {
925 : /* mark page accessed */
926 : FOLL_TOUCH = 1 << 16,
927 : /* a retry, previous pass started an IO */
928 : FOLL_TRIED = 1 << 17,
929 : /* we are working on non-current tsk/mm */
930 : FOLL_REMOTE = 1 << 18,
931 : /* pages must be released via unpin_user_page */
932 : FOLL_PIN = 1 << 19,
933 : /* gup_fast: prevent fall-back to slow gup */
934 : FOLL_FAST_ONLY = 1 << 20,
935 : /* allow unlocking the mmap lock */
936 : FOLL_UNLOCKABLE = 1 << 21,
937 : };
938 :
939 : /*
940 : * Indicates for which pages that are write-protected in the page table,
941 : * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
942 : * GUP pin will remain consistent with the pages mapped into the page tables
943 : * of the MM.
944 : *
945 : * Temporary unmapping of PageAnonExclusive() pages or clearing of
946 : * PageAnonExclusive() has to protect against concurrent GUP:
947 : * * Ordinary GUP: Using the PT lock
948 : * * GUP-fast and fork(): mm->write_protect_seq
949 : * * GUP-fast and KSM or temporary unmapping (swap, migration): see
950 : * page_try_share_anon_rmap()
951 : *
952 : * Must be called with the (sub)page that's actually referenced via the
953 : * page table entry, which might not necessarily be the head page for a
954 : * PTE-mapped THP.
955 : *
956 : * If the vma is NULL, we're coming from the GUP-fast path and might have
957 : * to fallback to the slow path just to lookup the vma.
958 : */
959 0 : static inline bool gup_must_unshare(struct vm_area_struct *vma,
960 : unsigned int flags, struct page *page)
961 : {
962 : /*
963 : * FOLL_WRITE is implicitly handled correctly as the page table entry
964 : * has to be writable -- and if it references (part of) an anonymous
965 : * folio, that part is required to be marked exclusive.
966 : */
967 0 : if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
968 : return false;
969 : /*
970 : * Note: PageAnon(page) is stable until the page is actually getting
971 : * freed.
972 : */
973 0 : if (!PageAnon(page)) {
974 : /*
975 : * We only care about R/O long-term pining: R/O short-term
976 : * pinning does not have the semantics to observe successive
977 : * changes through the process page tables.
978 : */
979 0 : if (!(flags & FOLL_LONGTERM))
980 : return false;
981 :
982 : /* We really need the vma ... */
983 0 : if (!vma)
984 : return true;
985 :
986 : /*
987 : * ... because we only care about writable private ("COW")
988 : * mappings where we have to break COW early.
989 : */
990 0 : return is_cow_mapping(vma->vm_flags);
991 : }
992 :
993 : /* Paired with a memory barrier in page_try_share_anon_rmap(). */
994 : if (IS_ENABLED(CONFIG_HAVE_FAST_GUP))
995 : smp_rmb();
996 :
997 : /*
998 : * Note that PageKsm() pages cannot be exclusive, and consequently,
999 : * cannot get pinned.
1000 : */
1001 0 : return !PageAnonExclusive(page);
1002 : }
1003 :
1004 : extern bool mirrored_kernelcore;
1005 :
1006 : static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1007 : {
1008 : /*
1009 : * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1010 : * enablements, because when without soft-dirty being compiled in,
1011 : * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1012 : * will be constantly true.
1013 : */
1014 : if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY))
1015 : return false;
1016 :
1017 : /*
1018 : * Soft-dirty is kind of special: its tracking is enabled when the
1019 : * vma flags not set.
1020 : */
1021 : return !(vma->vm_flags & VM_SOFTDIRTY);
1022 : }
1023 :
1024 : /*
1025 : * VMA Iterator functions shared between nommu and mmap
1026 : */
1027 : static inline int vma_iter_prealloc(struct vma_iterator *vmi)
1028 : {
1029 0 : return mas_preallocate(&vmi->mas, GFP_KERNEL);
1030 : }
1031 :
1032 : static inline void vma_iter_clear(struct vma_iterator *vmi,
1033 : unsigned long start, unsigned long end)
1034 : {
1035 0 : mas_set_range(&vmi->mas, start, end - 1);
1036 0 : mas_store_prealloc(&vmi->mas, NULL);
1037 : }
1038 :
1039 : static inline struct vm_area_struct *vma_iter_load(struct vma_iterator *vmi)
1040 : {
1041 0 : return mas_walk(&vmi->mas);
1042 : }
1043 :
1044 : /* Store a VMA with preallocated memory */
1045 0 : static inline void vma_iter_store(struct vma_iterator *vmi,
1046 : struct vm_area_struct *vma)
1047 : {
1048 :
1049 : #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
1050 : if (WARN_ON(vmi->mas.node != MAS_START && vmi->mas.index > vma->vm_start)) {
1051 : printk("%lu > %lu\n", vmi->mas.index, vma->vm_start);
1052 : printk("store of vma %lu-%lu", vma->vm_start, vma->vm_end);
1053 : printk("into slot %lu-%lu", vmi->mas.index, vmi->mas.last);
1054 : mt_dump(vmi->mas.tree);
1055 : }
1056 : if (WARN_ON(vmi->mas.node != MAS_START && vmi->mas.last < vma->vm_start)) {
1057 : printk("%lu < %lu\n", vmi->mas.last, vma->vm_start);
1058 : printk("store of vma %lu-%lu", vma->vm_start, vma->vm_end);
1059 : printk("into slot %lu-%lu", vmi->mas.index, vmi->mas.last);
1060 : mt_dump(vmi->mas.tree);
1061 : }
1062 : #endif
1063 :
1064 0 : if (vmi->mas.node != MAS_START &&
1065 0 : ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
1066 : vma_iter_invalidate(vmi);
1067 :
1068 0 : vmi->mas.index = vma->vm_start;
1069 0 : vmi->mas.last = vma->vm_end - 1;
1070 0 : mas_store_prealloc(&vmi->mas, vma);
1071 0 : }
1072 :
1073 0 : static inline int vma_iter_store_gfp(struct vma_iterator *vmi,
1074 : struct vm_area_struct *vma, gfp_t gfp)
1075 : {
1076 0 : if (vmi->mas.node != MAS_START &&
1077 0 : ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
1078 : vma_iter_invalidate(vmi);
1079 :
1080 0 : vmi->mas.index = vma->vm_start;
1081 0 : vmi->mas.last = vma->vm_end - 1;
1082 0 : mas_store_gfp(&vmi->mas, vma, gfp);
1083 0 : if (unlikely(mas_is_err(&vmi->mas)))
1084 : return -ENOMEM;
1085 :
1086 0 : return 0;
1087 : }
1088 :
1089 : /*
1090 : * VMA lock generalization
1091 : */
1092 : struct vma_prepare {
1093 : struct vm_area_struct *vma;
1094 : struct vm_area_struct *adj_next;
1095 : struct file *file;
1096 : struct address_space *mapping;
1097 : struct anon_vma *anon_vma;
1098 : struct vm_area_struct *insert;
1099 : struct vm_area_struct *remove;
1100 : struct vm_area_struct *remove2;
1101 : };
1102 : #endif /* __MM_INTERNAL_H */
|
