xref: /linux/mm/internal.h (revision a89988a6e00c5a099ee23619cd91dc8dc7ec9328)
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/fs.h>
15 #include <linux/mm.h>
16 #include <linux/pagemap.h>
17 #include <linux/tracepoint-defs.h>
18 
19 /*
20  * The set of flags that only affect watermark checking and reclaim
21  * behaviour. This is used by the MM to obey the caller constraints
22  * about IO, FS and watermark checking while ignoring placement
23  * hints such as HIGHMEM usage.
24  */
25 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
26 			__GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
27 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
28 			__GFP_ATOMIC)
29 
30 /* The GFP flags allowed during early boot */
31 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
32 
33 /* Control allocation cpuset and node placement constraints */
34 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
35 
36 /* Do not use these with a slab allocator */
37 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
38 
39 void page_writeback_init(void);
40 
41 int do_swap_page(struct vm_fault *vmf);
42 
43 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
44 		unsigned long floor, unsigned long ceiling);
45 
46 static inline bool can_madv_dontneed_vma(struct vm_area_struct *vma)
47 {
48 	return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
49 }
50 
51 void unmap_page_range(struct mmu_gather *tlb,
52 			     struct vm_area_struct *vma,
53 			     unsigned long addr, unsigned long end,
54 			     struct zap_details *details);
55 
56 extern int __do_page_cache_readahead(struct address_space *mapping,
57 		struct file *filp, pgoff_t offset, unsigned long nr_to_read,
58 		unsigned long lookahead_size);
59 
60 /*
61  * Submit IO for the read-ahead request in file_ra_state.
62  */
63 static inline unsigned long ra_submit(struct file_ra_state *ra,
64 		struct address_space *mapping, struct file *filp)
65 {
66 	return __do_page_cache_readahead(mapping, filp,
67 					ra->start, ra->size, ra->async_size);
68 }
69 
70 /*
71  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
72  * a count of one.
73  */
74 static inline void set_page_refcounted(struct page *page)
75 {
76 	VM_BUG_ON_PAGE(PageTail(page), page);
77 	VM_BUG_ON_PAGE(page_ref_count(page), page);
78 	set_page_count(page, 1);
79 }
80 
81 extern unsigned long highest_memmap_pfn;
82 
83 /*
84  * Maximum number of reclaim retries without progress before the OOM
85  * killer is consider the only way forward.
86  */
87 #define MAX_RECLAIM_RETRIES 16
88 
89 /*
90  * in mm/vmscan.c:
91  */
92 extern int isolate_lru_page(struct page *page);
93 extern void putback_lru_page(struct page *page);
94 
95 /*
96  * in mm/rmap.c:
97  */
98 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
99 
100 /*
101  * in mm/page_alloc.c
102  */
103 
104 /*
105  * Structure for holding the mostly immutable allocation parameters passed
106  * between functions involved in allocations, including the alloc_pages*
107  * family of functions.
108  *
109  * nodemask, migratetype and high_zoneidx are initialized only once in
110  * __alloc_pages_nodemask() and then never change.
111  *
112  * zonelist, preferred_zone and classzone_idx are set first in
113  * __alloc_pages_nodemask() for the fast path, and might be later changed
114  * in __alloc_pages_slowpath(). All other functions pass the whole strucure
115  * by a const pointer.
116  */
117 struct alloc_context {
118 	struct zonelist *zonelist;
119 	nodemask_t *nodemask;
120 	struct zoneref *preferred_zoneref;
121 	int migratetype;
122 	enum zone_type high_zoneidx;
123 	bool spread_dirty_pages;
124 };
125 
126 #define ac_classzone_idx(ac) zonelist_zone_idx(ac->preferred_zoneref)
127 
128 /*
129  * Locate the struct page for both the matching buddy in our
130  * pair (buddy1) and the combined O(n+1) page they form (page).
131  *
132  * 1) Any buddy B1 will have an order O twin B2 which satisfies
133  * the following equation:
134  *     B2 = B1 ^ (1 << O)
135  * For example, if the starting buddy (buddy2) is #8 its order
136  * 1 buddy is #10:
137  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
138  *
139  * 2) Any buddy B will have an order O+1 parent P which
140  * satisfies the following equation:
141  *     P = B & ~(1 << O)
142  *
143  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
144  */
145 static inline unsigned long
146 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
147 {
148 	return page_pfn ^ (1 << order);
149 }
150 
151 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
152 				unsigned long end_pfn, struct zone *zone);
153 
154 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
155 				unsigned long end_pfn, struct zone *zone)
156 {
157 	if (zone->contiguous)
158 		return pfn_to_page(start_pfn);
159 
160 	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
161 }
162 
163 extern int __isolate_free_page(struct page *page, unsigned int order);
164 extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
165 					unsigned int order);
166 extern void prep_compound_page(struct page *page, unsigned int order);
167 extern void post_alloc_hook(struct page *page, unsigned int order,
168 					gfp_t gfp_flags);
169 extern int user_min_free_kbytes;
170 
171 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
172 
173 /*
174  * in mm/compaction.c
175  */
176 /*
177  * compact_control is used to track pages being migrated and the free pages
178  * they are being migrated to during memory compaction. The free_pfn starts
179  * at the end of a zone and migrate_pfn begins at the start. Movable pages
180  * are moved to the end of a zone during a compaction run and the run
181  * completes when free_pfn <= migrate_pfn
182  */
183 struct compact_control {
184 	struct list_head freepages;	/* List of free pages to migrate to */
185 	struct list_head migratepages;	/* List of pages being migrated */
186 	unsigned long nr_freepages;	/* Number of isolated free pages */
187 	unsigned long nr_migratepages;	/* Number of pages to migrate */
188 	unsigned long total_migrate_scanned;
189 	unsigned long total_free_scanned;
190 	unsigned long free_pfn;		/* isolate_freepages search base */
191 	unsigned long migrate_pfn;	/* isolate_migratepages search base */
192 	unsigned long last_migrated_pfn;/* Not yet flushed page being freed */
193 	enum migrate_mode mode;		/* Async or sync migration mode */
194 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
195 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
196 	bool direct_compaction;		/* False from kcompactd or /proc/... */
197 	bool whole_zone;		/* Whole zone should/has been scanned */
198 	int order;			/* order a direct compactor needs */
199 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
200 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
201 	const int classzone_idx;	/* zone index of a direct compactor */
202 	struct zone *zone;
203 	bool contended;			/* Signal lock or sched contention */
204 };
205 
206 unsigned long
207 isolate_freepages_range(struct compact_control *cc,
208 			unsigned long start_pfn, unsigned long end_pfn);
209 unsigned long
210 isolate_migratepages_range(struct compact_control *cc,
211 			   unsigned long low_pfn, unsigned long end_pfn);
212 int find_suitable_fallback(struct free_area *area, unsigned int order,
213 			int migratetype, bool only_stealable, bool *can_steal);
214 
215 #endif
216 
217 /*
218  * This function returns the order of a free page in the buddy system. In
219  * general, page_zone(page)->lock must be held by the caller to prevent the
220  * page from being allocated in parallel and returning garbage as the order.
221  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
222  * page cannot be allocated or merged in parallel. Alternatively, it must
223  * handle invalid values gracefully, and use page_order_unsafe() below.
224  */
225 static inline unsigned int page_order(struct page *page)
226 {
227 	/* PageBuddy() must be checked by the caller */
228 	return page_private(page);
229 }
230 
231 /*
232  * Like page_order(), but for callers who cannot afford to hold the zone lock.
233  * PageBuddy() should be checked first by the caller to minimize race window,
234  * and invalid values must be handled gracefully.
235  *
236  * READ_ONCE is used so that if the caller assigns the result into a local
237  * variable and e.g. tests it for valid range before using, the compiler cannot
238  * decide to remove the variable and inline the page_private(page) multiple
239  * times, potentially observing different values in the tests and the actual
240  * use of the result.
241  */
242 #define page_order_unsafe(page)		READ_ONCE(page_private(page))
243 
244 static inline bool is_cow_mapping(vm_flags_t flags)
245 {
246 	return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
247 }
248 
249 /*
250  * These three helpers classifies VMAs for virtual memory accounting.
251  */
252 
253 /*
254  * Executable code area - executable, not writable, not stack
255  */
256 static inline bool is_exec_mapping(vm_flags_t flags)
257 {
258 	return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
259 }
260 
261 /*
262  * Stack area - atomatically grows in one direction
263  *
264  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
265  * do_mmap() forbids all other combinations.
266  */
267 static inline bool is_stack_mapping(vm_flags_t flags)
268 {
269 	return (flags & VM_STACK) == VM_STACK;
270 }
271 
272 /*
273  * Data area - private, writable, not stack
274  */
275 static inline bool is_data_mapping(vm_flags_t flags)
276 {
277 	return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
278 }
279 
280 /* mm/util.c */
281 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
282 		struct vm_area_struct *prev, struct rb_node *rb_parent);
283 
284 #ifdef CONFIG_MMU
285 extern long populate_vma_page_range(struct vm_area_struct *vma,
286 		unsigned long start, unsigned long end, int *nonblocking);
287 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
288 			unsigned long start, unsigned long end);
289 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
290 {
291 	munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
292 }
293 
294 /*
295  * must be called with vma's mmap_sem held for read or write, and page locked.
296  */
297 extern void mlock_vma_page(struct page *page);
298 extern unsigned int munlock_vma_page(struct page *page);
299 
300 /*
301  * Clear the page's PageMlocked().  This can be useful in a situation where
302  * we want to unconditionally remove a page from the pagecache -- e.g.,
303  * on truncation or freeing.
304  *
305  * It is legal to call this function for any page, mlocked or not.
306  * If called for a page that is still mapped by mlocked vmas, all we do
307  * is revert to lazy LRU behaviour -- semantics are not broken.
308  */
309 extern void clear_page_mlock(struct page *page);
310 
311 /*
312  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
313  * (because that does not go through the full procedure of migration ptes):
314  * to migrate the Mlocked page flag; update statistics.
315  */
316 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
317 {
318 	if (TestClearPageMlocked(page)) {
319 		int nr_pages = hpage_nr_pages(page);
320 
321 		/* Holding pmd lock, no change in irq context: __mod is safe */
322 		__mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
323 		SetPageMlocked(newpage);
324 		__mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
325 	}
326 }
327 
328 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
329 
330 /*
331  * At what user virtual address is page expected in @vma?
332  */
333 static inline unsigned long
334 __vma_address(struct page *page, struct vm_area_struct *vma)
335 {
336 	pgoff_t pgoff = page_to_pgoff(page);
337 	return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
338 }
339 
340 static inline unsigned long
341 vma_address(struct page *page, struct vm_area_struct *vma)
342 {
343 	unsigned long start, end;
344 
345 	start = __vma_address(page, vma);
346 	end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
347 
348 	/* page should be within @vma mapping range */
349 	VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma);
350 
351 	return max(start, vma->vm_start);
352 }
353 
354 #else /* !CONFIG_MMU */
355 static inline void clear_page_mlock(struct page *page) { }
356 static inline void mlock_vma_page(struct page *page) { }
357 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
358 
359 #endif /* !CONFIG_MMU */
360 
361 /*
362  * Return the mem_map entry representing the 'offset' subpage within
363  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
364  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
365  */
366 static inline struct page *mem_map_offset(struct page *base, int offset)
367 {
368 	if (unlikely(offset >= MAX_ORDER_NR_PAGES))
369 		return nth_page(base, offset);
370 	return base + offset;
371 }
372 
373 /*
374  * Iterator over all subpages within the maximally aligned gigantic
375  * page 'base'.  Handle any discontiguity in the mem_map.
376  */
377 static inline struct page *mem_map_next(struct page *iter,
378 						struct page *base, int offset)
379 {
380 	if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
381 		unsigned long pfn = page_to_pfn(base) + offset;
382 		if (!pfn_valid(pfn))
383 			return NULL;
384 		return pfn_to_page(pfn);
385 	}
386 	return iter + 1;
387 }
388 
389 /*
390  * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
391  * so all functions starting at paging_init should be marked __init
392  * in those cases. SPARSEMEM, however, allows for memory hotplug,
393  * and alloc_bootmem_node is not used.
394  */
395 #ifdef CONFIG_SPARSEMEM
396 #define __paginginit __meminit
397 #else
398 #define __paginginit __init
399 #endif
400 
401 /* Memory initialisation debug and verification */
402 enum mminit_level {
403 	MMINIT_WARNING,
404 	MMINIT_VERIFY,
405 	MMINIT_TRACE
406 };
407 
408 #ifdef CONFIG_DEBUG_MEMORY_INIT
409 
410 extern int mminit_loglevel;
411 
412 #define mminit_dprintk(level, prefix, fmt, arg...) \
413 do { \
414 	if (level < mminit_loglevel) { \
415 		if (level <= MMINIT_WARNING) \
416 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
417 		else \
418 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
419 	} \
420 } while (0)
421 
422 extern void mminit_verify_pageflags_layout(void);
423 extern void mminit_verify_zonelist(void);
424 #else
425 
426 static inline void mminit_dprintk(enum mminit_level level,
427 				const char *prefix, const char *fmt, ...)
428 {
429 }
430 
431 static inline void mminit_verify_pageflags_layout(void)
432 {
433 }
434 
435 static inline void mminit_verify_zonelist(void)
436 {
437 }
438 #endif /* CONFIG_DEBUG_MEMORY_INIT */
439 
440 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
441 #if defined(CONFIG_SPARSEMEM)
442 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
443 				unsigned long *end_pfn);
444 #else
445 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
446 				unsigned long *end_pfn)
447 {
448 }
449 #endif /* CONFIG_SPARSEMEM */
450 
451 #define NODE_RECLAIM_NOSCAN	-2
452 #define NODE_RECLAIM_FULL	-1
453 #define NODE_RECLAIM_SOME	0
454 #define NODE_RECLAIM_SUCCESS	1
455 
456 extern int hwpoison_filter(struct page *p);
457 
458 extern u32 hwpoison_filter_dev_major;
459 extern u32 hwpoison_filter_dev_minor;
460 extern u64 hwpoison_filter_flags_mask;
461 extern u64 hwpoison_filter_flags_value;
462 extern u64 hwpoison_filter_memcg;
463 extern u32 hwpoison_filter_enable;
464 
465 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
466         unsigned long, unsigned long,
467         unsigned long, unsigned long);
468 
469 extern void set_pageblock_order(void);
470 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
471 					    struct list_head *page_list);
472 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
473 #define ALLOC_WMARK_MIN		WMARK_MIN
474 #define ALLOC_WMARK_LOW		WMARK_LOW
475 #define ALLOC_WMARK_HIGH	WMARK_HIGH
476 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
477 
478 /* Mask to get the watermark bits */
479 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
480 
481 #define ALLOC_HARDER		0x10 /* try to alloc harder */
482 #define ALLOC_HIGH		0x20 /* __GFP_HIGH set */
483 #define ALLOC_CPUSET		0x40 /* check for correct cpuset */
484 #define ALLOC_CMA		0x80 /* allow allocations from CMA areas */
485 
486 enum ttu_flags;
487 struct tlbflush_unmap_batch;
488 
489 
490 /*
491  * only for MM internal work items which do not depend on
492  * any allocations or locks which might depend on allocations
493  */
494 extern struct workqueue_struct *mm_percpu_wq;
495 
496 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
497 void try_to_unmap_flush(void);
498 void try_to_unmap_flush_dirty(void);
499 #else
500 static inline void try_to_unmap_flush(void)
501 {
502 }
503 static inline void try_to_unmap_flush_dirty(void)
504 {
505 }
506 
507 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
508 
509 extern const struct trace_print_flags pageflag_names[];
510 extern const struct trace_print_flags vmaflag_names[];
511 extern const struct trace_print_flags gfpflag_names[];
512 
513 static inline bool is_migrate_highatomic(enum migratetype migratetype)
514 {
515 	return migratetype == MIGRATE_HIGHATOMIC;
516 }
517 
518 static inline bool is_migrate_highatomic_page(struct page *page)
519 {
520 	return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
521 }
522 
523 #endif	/* __MM_INTERNAL_H */
524