xref: /linux/mm/internal.h (revision 28df0642abbf6d66908a2858922a7e4b21cdd8c2)
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/tracepoint-defs.h>
14 
15 /*
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.
20  */
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|\
24 			__GFP_ATOMIC)
25 
26 /* The GFP flags allowed during early boot */
27 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
28 
29 /* Control allocation cpuset and node placement constraints */
30 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
31 
32 /* Do not use these with a slab allocator */
33 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
34 
35 void page_writeback_init(void);
36 
37 vm_fault_t do_swap_page(struct vm_fault *vmf);
38 
39 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
40 		unsigned long floor, unsigned long ceiling);
41 
42 static inline bool can_madv_lru_vma(struct vm_area_struct *vma)
43 {
44 	return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
45 }
46 
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);
51 
52 extern unsigned int __do_page_cache_readahead(struct address_space *mapping,
53 		struct file *filp, pgoff_t offset, unsigned long nr_to_read,
54 		unsigned long lookahead_size);
55 
56 /*
57  * Submit IO for the read-ahead request in file_ra_state.
58  */
59 static inline unsigned long ra_submit(struct file_ra_state *ra,
60 		struct address_space *mapping, struct file *filp)
61 {
62 	return __do_page_cache_readahead(mapping, filp,
63 					ra->start, ra->size, ra->async_size);
64 }
65 
66 /*
67  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
68  * a count of one.
69  */
70 static inline void set_page_refcounted(struct page *page)
71 {
72 	VM_BUG_ON_PAGE(PageTail(page), page);
73 	VM_BUG_ON_PAGE(page_ref_count(page), page);
74 	set_page_count(page, 1);
75 }
76 
77 extern unsigned long highest_memmap_pfn;
78 
79 /*
80  * Maximum number of reclaim retries without progress before the OOM
81  * killer is consider the only way forward.
82  */
83 #define MAX_RECLAIM_RETRIES 16
84 
85 /*
86  * in mm/vmscan.c:
87  */
88 extern int isolate_lru_page(struct page *page);
89 extern void putback_lru_page(struct page *page);
90 
91 /*
92  * in mm/rmap.c:
93  */
94 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
95 
96 /*
97  * in mm/page_alloc.c
98  */
99 
100 /*
101  * Structure for holding the mostly immutable allocation parameters passed
102  * between functions involved in allocations, including the alloc_pages*
103  * family of functions.
104  *
105  * nodemask, migratetype and high_zoneidx are initialized only once in
106  * __alloc_pages_nodemask() and then never change.
107  *
108  * zonelist, preferred_zone and classzone_idx are set first in
109  * __alloc_pages_nodemask() for the fast path, and might be later changed
110  * in __alloc_pages_slowpath(). All other functions pass the whole strucure
111  * by a const pointer.
112  */
113 struct alloc_context {
114 	struct zonelist *zonelist;
115 	nodemask_t *nodemask;
116 	struct zoneref *preferred_zoneref;
117 	int migratetype;
118 	enum zone_type high_zoneidx;
119 	bool spread_dirty_pages;
120 };
121 
122 #define ac_classzone_idx(ac) zonelist_zone_idx(ac->preferred_zoneref)
123 
124 /*
125  * Locate the struct page for both the matching buddy in our
126  * pair (buddy1) and the combined O(n+1) page they form (page).
127  *
128  * 1) Any buddy B1 will have an order O twin B2 which satisfies
129  * the following equation:
130  *     B2 = B1 ^ (1 << O)
131  * For example, if the starting buddy (buddy2) is #8 its order
132  * 1 buddy is #10:
133  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
134  *
135  * 2) Any buddy B will have an order O+1 parent P which
136  * satisfies the following equation:
137  *     P = B & ~(1 << O)
138  *
139  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
140  */
141 static inline unsigned long
142 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
143 {
144 	return page_pfn ^ (1 << order);
145 }
146 
147 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
148 				unsigned long end_pfn, struct zone *zone);
149 
150 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
151 				unsigned long end_pfn, struct zone *zone)
152 {
153 	if (zone->contiguous)
154 		return pfn_to_page(start_pfn);
155 
156 	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
157 }
158 
159 extern int __isolate_free_page(struct page *page, unsigned int order);
160 extern void memblock_free_pages(struct page *page, unsigned long pfn,
161 					unsigned int order);
162 extern void __free_pages_core(struct page *page, unsigned int order);
163 extern void prep_compound_page(struct page *page, unsigned int order);
164 extern void post_alloc_hook(struct page *page, unsigned int order,
165 					gfp_t gfp_flags);
166 extern int user_min_free_kbytes;
167 
168 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
169 
170 /*
171  * in mm/compaction.c
172  */
173 /*
174  * compact_control is used to track pages being migrated and the free pages
175  * they are being migrated to during memory compaction. The free_pfn starts
176  * at the end of a zone and migrate_pfn begins at the start. Movable pages
177  * are moved to the end of a zone during a compaction run and the run
178  * completes when free_pfn <= migrate_pfn
179  */
180 struct compact_control {
181 	struct list_head freepages;	/* List of free pages to migrate to */
182 	struct list_head migratepages;	/* List of pages being migrated */
183 	unsigned int nr_freepages;	/* Number of isolated free pages */
184 	unsigned int nr_migratepages;	/* Number of pages to migrate */
185 	unsigned long free_pfn;		/* isolate_freepages search base */
186 	unsigned long migrate_pfn;	/* isolate_migratepages search base */
187 	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
188 	struct zone *zone;
189 	unsigned long total_migrate_scanned;
190 	unsigned long total_free_scanned;
191 	unsigned short fast_search_fail;/* failures to use free list searches */
192 	short search_order;		/* order to start a fast search at */
193 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
194 	int order;			/* order a direct compactor needs */
195 	int migratetype;		/* migratetype of direct compactor */
196 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
197 	const int classzone_idx;	/* zone index of a direct compactor */
198 	enum migrate_mode mode;		/* Async or sync migration mode */
199 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
200 	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
201 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
202 	bool direct_compaction;		/* False from kcompactd or /proc/... */
203 	bool whole_zone;		/* Whole zone should/has been scanned */
204 	bool contended;			/* Signal lock or sched contention */
205 	bool rescan;			/* Rescanning the same pageblock */
206 };
207 
208 /*
209  * Used in direct compaction when a page should be taken from the freelists
210  * immediately when one is created during the free path.
211  */
212 struct capture_control {
213 	struct compact_control *cc;
214 	struct page *page;
215 };
216 
217 unsigned long
218 isolate_freepages_range(struct compact_control *cc,
219 			unsigned long start_pfn, unsigned long end_pfn);
220 unsigned long
221 isolate_migratepages_range(struct compact_control *cc,
222 			   unsigned long low_pfn, unsigned long end_pfn);
223 int find_suitable_fallback(struct free_area *area, unsigned int order,
224 			int migratetype, bool only_stealable, bool *can_steal);
225 
226 #endif
227 
228 /*
229  * This function returns the order of a free page in the buddy system. In
230  * general, page_zone(page)->lock must be held by the caller to prevent the
231  * page from being allocated in parallel and returning garbage as the order.
232  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
233  * page cannot be allocated or merged in parallel. Alternatively, it must
234  * handle invalid values gracefully, and use page_order_unsafe() below.
235  */
236 static inline unsigned int page_order(struct page *page)
237 {
238 	/* PageBuddy() must be checked by the caller */
239 	return page_private(page);
240 }
241 
242 /*
243  * Like page_order(), but for callers who cannot afford to hold the zone lock.
244  * PageBuddy() should be checked first by the caller to minimize race window,
245  * and invalid values must be handled gracefully.
246  *
247  * READ_ONCE is used so that if the caller assigns the result into a local
248  * variable and e.g. tests it for valid range before using, the compiler cannot
249  * decide to remove the variable and inline the page_private(page) multiple
250  * times, potentially observing different values in the tests and the actual
251  * use of the result.
252  */
253 #define page_order_unsafe(page)		READ_ONCE(page_private(page))
254 
255 static inline bool is_cow_mapping(vm_flags_t flags)
256 {
257 	return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
258 }
259 
260 /*
261  * These three helpers classifies VMAs for virtual memory accounting.
262  */
263 
264 /*
265  * Executable code area - executable, not writable, not stack
266  */
267 static inline bool is_exec_mapping(vm_flags_t flags)
268 {
269 	return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
270 }
271 
272 /*
273  * Stack area - atomatically grows in one direction
274  *
275  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
276  * do_mmap() forbids all other combinations.
277  */
278 static inline bool is_stack_mapping(vm_flags_t flags)
279 {
280 	return (flags & VM_STACK) == VM_STACK;
281 }
282 
283 /*
284  * Data area - private, writable, not stack
285  */
286 static inline bool is_data_mapping(vm_flags_t flags)
287 {
288 	return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
289 }
290 
291 /* mm/util.c */
292 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
293 		struct vm_area_struct *prev, struct rb_node *rb_parent);
294 
295 #ifdef CONFIG_MMU
296 extern long populate_vma_page_range(struct vm_area_struct *vma,
297 		unsigned long start, unsigned long end, int *nonblocking);
298 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
299 			unsigned long start, unsigned long end);
300 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
301 {
302 	munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
303 }
304 
305 /*
306  * must be called with vma's mmap_sem held for read or write, and page locked.
307  */
308 extern void mlock_vma_page(struct page *page);
309 extern unsigned int munlock_vma_page(struct page *page);
310 
311 /*
312  * Clear the page's PageMlocked().  This can be useful in a situation where
313  * we want to unconditionally remove a page from the pagecache -- e.g.,
314  * on truncation or freeing.
315  *
316  * It is legal to call this function for any page, mlocked or not.
317  * If called for a page that is still mapped by mlocked vmas, all we do
318  * is revert to lazy LRU behaviour -- semantics are not broken.
319  */
320 extern void clear_page_mlock(struct page *page);
321 
322 /*
323  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
324  * (because that does not go through the full procedure of migration ptes):
325  * to migrate the Mlocked page flag; update statistics.
326  */
327 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
328 {
329 	if (TestClearPageMlocked(page)) {
330 		int nr_pages = hpage_nr_pages(page);
331 
332 		/* Holding pmd lock, no change in irq context: __mod is safe */
333 		__mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
334 		SetPageMlocked(newpage);
335 		__mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
336 	}
337 }
338 
339 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
340 
341 /*
342  * At what user virtual address is page expected in @vma?
343  */
344 static inline unsigned long
345 __vma_address(struct page *page, struct vm_area_struct *vma)
346 {
347 	pgoff_t pgoff = page_to_pgoff(page);
348 	return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
349 }
350 
351 static inline unsigned long
352 vma_address(struct page *page, struct vm_area_struct *vma)
353 {
354 	unsigned long start, end;
355 
356 	start = __vma_address(page, vma);
357 	end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
358 
359 	/* page should be within @vma mapping range */
360 	VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma);
361 
362 	return max(start, vma->vm_start);
363 }
364 
365 #else /* !CONFIG_MMU */
366 static inline void clear_page_mlock(struct page *page) { }
367 static inline void mlock_vma_page(struct page *page) { }
368 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
369 
370 #endif /* !CONFIG_MMU */
371 
372 /*
373  * Return the mem_map entry representing the 'offset' subpage within
374  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
375  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
376  */
377 static inline struct page *mem_map_offset(struct page *base, int offset)
378 {
379 	if (unlikely(offset >= MAX_ORDER_NR_PAGES))
380 		return nth_page(base, offset);
381 	return base + offset;
382 }
383 
384 /*
385  * Iterator over all subpages within the maximally aligned gigantic
386  * page 'base'.  Handle any discontiguity in the mem_map.
387  */
388 static inline struct page *mem_map_next(struct page *iter,
389 						struct page *base, int offset)
390 {
391 	if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
392 		unsigned long pfn = page_to_pfn(base) + offset;
393 		if (!pfn_valid(pfn))
394 			return NULL;
395 		return pfn_to_page(pfn);
396 	}
397 	return iter + 1;
398 }
399 
400 /* Memory initialisation debug and verification */
401 enum mminit_level {
402 	MMINIT_WARNING,
403 	MMINIT_VERIFY,
404 	MMINIT_TRACE
405 };
406 
407 #ifdef CONFIG_DEBUG_MEMORY_INIT
408 
409 extern int mminit_loglevel;
410 
411 #define mminit_dprintk(level, prefix, fmt, arg...) \
412 do { \
413 	if (level < mminit_loglevel) { \
414 		if (level <= MMINIT_WARNING) \
415 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
416 		else \
417 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
418 	} \
419 } while (0)
420 
421 extern void mminit_verify_pageflags_layout(void);
422 extern void mminit_verify_zonelist(void);
423 #else
424 
425 static inline void mminit_dprintk(enum mminit_level level,
426 				const char *prefix, const char *fmt, ...)
427 {
428 }
429 
430 static inline void mminit_verify_pageflags_layout(void)
431 {
432 }
433 
434 static inline void mminit_verify_zonelist(void)
435 {
436 }
437 #endif /* CONFIG_DEBUG_MEMORY_INIT */
438 
439 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
440 #if defined(CONFIG_SPARSEMEM)
441 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
442 				unsigned long *end_pfn);
443 #else
444 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
445 				unsigned long *end_pfn)
446 {
447 }
448 #endif /* CONFIG_SPARSEMEM */
449 
450 #define NODE_RECLAIM_NOSCAN	-2
451 #define NODE_RECLAIM_FULL	-1
452 #define NODE_RECLAIM_SOME	0
453 #define NODE_RECLAIM_SUCCESS	1
454 
455 #ifdef CONFIG_NUMA
456 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
457 #else
458 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
459 				unsigned int order)
460 {
461 	return NODE_RECLAIM_NOSCAN;
462 }
463 #endif
464 
465 extern int hwpoison_filter(struct page *p);
466 
467 extern u32 hwpoison_filter_dev_major;
468 extern u32 hwpoison_filter_dev_minor;
469 extern u64 hwpoison_filter_flags_mask;
470 extern u64 hwpoison_filter_flags_value;
471 extern u64 hwpoison_filter_memcg;
472 extern u32 hwpoison_filter_enable;
473 
474 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
475         unsigned long, unsigned long,
476         unsigned long, unsigned long);
477 
478 extern void set_pageblock_order(void);
479 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
480 					    struct list_head *page_list);
481 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
482 #define ALLOC_WMARK_MIN		WMARK_MIN
483 #define ALLOC_WMARK_LOW		WMARK_LOW
484 #define ALLOC_WMARK_HIGH	WMARK_HIGH
485 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
486 
487 /* Mask to get the watermark bits */
488 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
489 
490 /*
491  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
492  * cannot assume a reduced access to memory reserves is sufficient for
493  * !MMU
494  */
495 #ifdef CONFIG_MMU
496 #define ALLOC_OOM		0x08
497 #else
498 #define ALLOC_OOM		ALLOC_NO_WATERMARKS
499 #endif
500 
501 #define ALLOC_HARDER		 0x10 /* try to alloc harder */
502 #define ALLOC_HIGH		 0x20 /* __GFP_HIGH set */
503 #define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
504 #define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
505 #ifdef CONFIG_ZONE_DMA32
506 #define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
507 #else
508 #define ALLOC_NOFRAGMENT	  0x0
509 #endif
510 #define ALLOC_KSWAPD		0x200 /* allow waking of kswapd */
511 
512 enum ttu_flags;
513 struct tlbflush_unmap_batch;
514 
515 
516 /*
517  * only for MM internal work items which do not depend on
518  * any allocations or locks which might depend on allocations
519  */
520 extern struct workqueue_struct *mm_percpu_wq;
521 
522 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
523 void try_to_unmap_flush(void);
524 void try_to_unmap_flush_dirty(void);
525 void flush_tlb_batched_pending(struct mm_struct *mm);
526 #else
527 static inline void try_to_unmap_flush(void)
528 {
529 }
530 static inline void try_to_unmap_flush_dirty(void)
531 {
532 }
533 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
534 {
535 }
536 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
537 
538 extern const struct trace_print_flags pageflag_names[];
539 extern const struct trace_print_flags vmaflag_names[];
540 extern const struct trace_print_flags gfpflag_names[];
541 
542 static inline bool is_migrate_highatomic(enum migratetype migratetype)
543 {
544 	return migratetype == MIGRATE_HIGHATOMIC;
545 }
546 
547 static inline bool is_migrate_highatomic_page(struct page *page)
548 {
549 	return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
550 }
551 
552 void setup_zone_pageset(struct zone *zone);
553 extern struct page *alloc_new_node_page(struct page *page, unsigned long node);
554 #endif	/* __MM_INTERNAL_H */
555