xref: /linux/mm/internal.h (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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 
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_REPEAT|__GFP_NOFAIL|\
26 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC)
27 
28 /* The GFP flags allowed during early boot */
29 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
30 
31 /* Control allocation cpuset and node placement constraints */
32 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
33 
34 /* Do not use these with a slab allocator */
35 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
36 
37 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
38 		unsigned long floor, unsigned long ceiling);
39 
40 static inline void set_page_count(struct page *page, int v)
41 {
42 	atomic_set(&page->_count, v);
43 }
44 
45 extern int __do_page_cache_readahead(struct address_space *mapping,
46 		struct file *filp, pgoff_t offset, unsigned long nr_to_read,
47 		unsigned long lookahead_size);
48 
49 /*
50  * Submit IO for the read-ahead request in file_ra_state.
51  */
52 static inline unsigned long ra_submit(struct file_ra_state *ra,
53 		struct address_space *mapping, struct file *filp)
54 {
55 	return __do_page_cache_readahead(mapping, filp,
56 					ra->start, ra->size, ra->async_size);
57 }
58 
59 /*
60  * Turn a non-refcounted page (->_count == 0) into refcounted with
61  * a count of one.
62  */
63 static inline void set_page_refcounted(struct page *page)
64 {
65 	VM_BUG_ON_PAGE(PageTail(page), page);
66 	VM_BUG_ON_PAGE(atomic_read(&page->_count), page);
67 	set_page_count(page, 1);
68 }
69 
70 extern unsigned long highest_memmap_pfn;
71 
72 /*
73  * in mm/vmscan.c:
74  */
75 extern int isolate_lru_page(struct page *page);
76 extern void putback_lru_page(struct page *page);
77 extern bool zone_reclaimable(struct zone *zone);
78 
79 /*
80  * in mm/rmap.c:
81  */
82 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
83 
84 /*
85  * in mm/page_alloc.c
86  */
87 
88 /*
89  * Structure for holding the mostly immutable allocation parameters passed
90  * between functions involved in allocations, including the alloc_pages*
91  * family of functions.
92  *
93  * nodemask, migratetype and high_zoneidx are initialized only once in
94  * __alloc_pages_nodemask() and then never change.
95  *
96  * zonelist, preferred_zone and classzone_idx are set first in
97  * __alloc_pages_nodemask() for the fast path, and might be later changed
98  * in __alloc_pages_slowpath(). All other functions pass the whole strucure
99  * by a const pointer.
100  */
101 struct alloc_context {
102 	struct zonelist *zonelist;
103 	nodemask_t *nodemask;
104 	struct zone *preferred_zone;
105 	int classzone_idx;
106 	int migratetype;
107 	enum zone_type high_zoneidx;
108 	bool spread_dirty_pages;
109 };
110 
111 /*
112  * Locate the struct page for both the matching buddy in our
113  * pair (buddy1) and the combined O(n+1) page they form (page).
114  *
115  * 1) Any buddy B1 will have an order O twin B2 which satisfies
116  * the following equation:
117  *     B2 = B1 ^ (1 << O)
118  * For example, if the starting buddy (buddy2) is #8 its order
119  * 1 buddy is #10:
120  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
121  *
122  * 2) Any buddy B will have an order O+1 parent P which
123  * satisfies the following equation:
124  *     P = B & ~(1 << O)
125  *
126  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
127  */
128 static inline unsigned long
129 __find_buddy_index(unsigned long page_idx, unsigned int order)
130 {
131 	return page_idx ^ (1 << order);
132 }
133 
134 extern int __isolate_free_page(struct page *page, unsigned int order);
135 extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
136 					unsigned int order);
137 extern void prep_compound_page(struct page *page, unsigned int order);
138 #ifdef CONFIG_MEMORY_FAILURE
139 extern bool is_free_buddy_page(struct page *page);
140 #endif
141 extern int user_min_free_kbytes;
142 
143 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
144 
145 /*
146  * in mm/compaction.c
147  */
148 /*
149  * compact_control is used to track pages being migrated and the free pages
150  * they are being migrated to during memory compaction. The free_pfn starts
151  * at the end of a zone and migrate_pfn begins at the start. Movable pages
152  * are moved to the end of a zone during a compaction run and the run
153  * completes when free_pfn <= migrate_pfn
154  */
155 struct compact_control {
156 	struct list_head freepages;	/* List of free pages to migrate to */
157 	struct list_head migratepages;	/* List of pages being migrated */
158 	unsigned long nr_freepages;	/* Number of isolated free pages */
159 	unsigned long nr_migratepages;	/* Number of pages to migrate */
160 	unsigned long free_pfn;		/* isolate_freepages search base */
161 	unsigned long migrate_pfn;	/* isolate_migratepages search base */
162 	unsigned long last_migrated_pfn;/* Not yet flushed page being freed */
163 	enum migrate_mode mode;		/* Async or sync migration mode */
164 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
165 	int order;			/* order a direct compactor needs */
166 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
167 	const int alloc_flags;		/* alloc flags of a direct compactor */
168 	const int classzone_idx;	/* zone index of a direct compactor */
169 	struct zone *zone;
170 	int contended;			/* Signal need_sched() or lock
171 					 * contention detected during
172 					 * compaction
173 					 */
174 };
175 
176 unsigned long
177 isolate_freepages_range(struct compact_control *cc,
178 			unsigned long start_pfn, unsigned long end_pfn);
179 unsigned long
180 isolate_migratepages_range(struct compact_control *cc,
181 			   unsigned long low_pfn, unsigned long end_pfn);
182 int find_suitable_fallback(struct free_area *area, unsigned int order,
183 			int migratetype, bool only_stealable, bool *can_steal);
184 
185 #endif
186 
187 /*
188  * This function returns the order of a free page in the buddy system. In
189  * general, page_zone(page)->lock must be held by the caller to prevent the
190  * page from being allocated in parallel and returning garbage as the order.
191  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
192  * page cannot be allocated or merged in parallel. Alternatively, it must
193  * handle invalid values gracefully, and use page_order_unsafe() below.
194  */
195 static inline unsigned int page_order(struct page *page)
196 {
197 	/* PageBuddy() must be checked by the caller */
198 	return page_private(page);
199 }
200 
201 /*
202  * Like page_order(), but for callers who cannot afford to hold the zone lock.
203  * PageBuddy() should be checked first by the caller to minimize race window,
204  * and invalid values must be handled gracefully.
205  *
206  * READ_ONCE is used so that if the caller assigns the result into a local
207  * variable and e.g. tests it for valid range before using, the compiler cannot
208  * decide to remove the variable and inline the page_private(page) multiple
209  * times, potentially observing different values in the tests and the actual
210  * use of the result.
211  */
212 #define page_order_unsafe(page)		READ_ONCE(page_private(page))
213 
214 static inline bool is_cow_mapping(vm_flags_t flags)
215 {
216 	return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
217 }
218 
219 /*
220  * These three helpers classifies VMAs for virtual memory accounting.
221  */
222 
223 /*
224  * Executable code area - executable, not writable, not stack
225  */
226 static inline bool is_exec_mapping(vm_flags_t flags)
227 {
228 	return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
229 }
230 
231 /*
232  * Stack area - atomatically grows in one direction
233  *
234  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
235  * do_mmap() forbids all other combinations.
236  */
237 static inline bool is_stack_mapping(vm_flags_t flags)
238 {
239 	return (flags & VM_STACK) == VM_STACK;
240 }
241 
242 /*
243  * Data area - private, writable, not stack
244  */
245 static inline bool is_data_mapping(vm_flags_t flags)
246 {
247 	return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
248 }
249 
250 /* mm/util.c */
251 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
252 		struct vm_area_struct *prev, struct rb_node *rb_parent);
253 
254 #ifdef CONFIG_MMU
255 extern long populate_vma_page_range(struct vm_area_struct *vma,
256 		unsigned long start, unsigned long end, int *nonblocking);
257 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
258 			unsigned long start, unsigned long end);
259 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
260 {
261 	munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
262 }
263 
264 /*
265  * must be called with vma's mmap_sem held for read or write, and page locked.
266  */
267 extern void mlock_vma_page(struct page *page);
268 extern unsigned int munlock_vma_page(struct page *page);
269 
270 /*
271  * Clear the page's PageMlocked().  This can be useful in a situation where
272  * we want to unconditionally remove a page from the pagecache -- e.g.,
273  * on truncation or freeing.
274  *
275  * It is legal to call this function for any page, mlocked or not.
276  * If called for a page that is still mapped by mlocked vmas, all we do
277  * is revert to lazy LRU behaviour -- semantics are not broken.
278  */
279 extern void clear_page_mlock(struct page *page);
280 
281 /*
282  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
283  * (because that does not go through the full procedure of migration ptes):
284  * to migrate the Mlocked page flag; update statistics.
285  */
286 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
287 {
288 	if (TestClearPageMlocked(page)) {
289 		int nr_pages = hpage_nr_pages(page);
290 
291 		/* Holding pmd lock, no change in irq context: __mod is safe */
292 		__mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
293 		SetPageMlocked(newpage);
294 		__mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
295 	}
296 }
297 
298 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
299 
300 /*
301  * At what user virtual address is page expected in @vma?
302  */
303 static inline unsigned long
304 __vma_address(struct page *page, struct vm_area_struct *vma)
305 {
306 	pgoff_t pgoff = page_to_pgoff(page);
307 	return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
308 }
309 
310 static inline unsigned long
311 vma_address(struct page *page, struct vm_area_struct *vma)
312 {
313 	unsigned long address = __vma_address(page, vma);
314 
315 	/* page should be within @vma mapping range */
316 	VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
317 
318 	return address;
319 }
320 
321 #else /* !CONFIG_MMU */
322 static inline void clear_page_mlock(struct page *page) { }
323 static inline void mlock_vma_page(struct page *page) { }
324 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
325 
326 #endif /* !CONFIG_MMU */
327 
328 /*
329  * Return the mem_map entry representing the 'offset' subpage within
330  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
331  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
332  */
333 static inline struct page *mem_map_offset(struct page *base, int offset)
334 {
335 	if (unlikely(offset >= MAX_ORDER_NR_PAGES))
336 		return nth_page(base, offset);
337 	return base + offset;
338 }
339 
340 /*
341  * Iterator over all subpages within the maximally aligned gigantic
342  * page 'base'.  Handle any discontiguity in the mem_map.
343  */
344 static inline struct page *mem_map_next(struct page *iter,
345 						struct page *base, int offset)
346 {
347 	if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
348 		unsigned long pfn = page_to_pfn(base) + offset;
349 		if (!pfn_valid(pfn))
350 			return NULL;
351 		return pfn_to_page(pfn);
352 	}
353 	return iter + 1;
354 }
355 
356 /*
357  * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
358  * so all functions starting at paging_init should be marked __init
359  * in those cases. SPARSEMEM, however, allows for memory hotplug,
360  * and alloc_bootmem_node is not used.
361  */
362 #ifdef CONFIG_SPARSEMEM
363 #define __paginginit __meminit
364 #else
365 #define __paginginit __init
366 #endif
367 
368 /* Memory initialisation debug and verification */
369 enum mminit_level {
370 	MMINIT_WARNING,
371 	MMINIT_VERIFY,
372 	MMINIT_TRACE
373 };
374 
375 #ifdef CONFIG_DEBUG_MEMORY_INIT
376 
377 extern int mminit_loglevel;
378 
379 #define mminit_dprintk(level, prefix, fmt, arg...) \
380 do { \
381 	if (level < mminit_loglevel) { \
382 		if (level <= MMINIT_WARNING) \
383 			printk(KERN_WARNING "mminit::" prefix " " fmt, ##arg); \
384 		else \
385 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
386 	} \
387 } while (0)
388 
389 extern void mminit_verify_pageflags_layout(void);
390 extern void mminit_verify_zonelist(void);
391 #else
392 
393 static inline void mminit_dprintk(enum mminit_level level,
394 				const char *prefix, const char *fmt, ...)
395 {
396 }
397 
398 static inline void mminit_verify_pageflags_layout(void)
399 {
400 }
401 
402 static inline void mminit_verify_zonelist(void)
403 {
404 }
405 #endif /* CONFIG_DEBUG_MEMORY_INIT */
406 
407 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
408 #if defined(CONFIG_SPARSEMEM)
409 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
410 				unsigned long *end_pfn);
411 #else
412 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
413 				unsigned long *end_pfn)
414 {
415 }
416 #endif /* CONFIG_SPARSEMEM */
417 
418 #define ZONE_RECLAIM_NOSCAN	-2
419 #define ZONE_RECLAIM_FULL	-1
420 #define ZONE_RECLAIM_SOME	0
421 #define ZONE_RECLAIM_SUCCESS	1
422 
423 extern int hwpoison_filter(struct page *p);
424 
425 extern u32 hwpoison_filter_dev_major;
426 extern u32 hwpoison_filter_dev_minor;
427 extern u64 hwpoison_filter_flags_mask;
428 extern u64 hwpoison_filter_flags_value;
429 extern u64 hwpoison_filter_memcg;
430 extern u32 hwpoison_filter_enable;
431 
432 extern unsigned long vm_mmap_pgoff(struct file *, unsigned long,
433         unsigned long, unsigned long,
434         unsigned long, unsigned long);
435 
436 extern void set_pageblock_order(void);
437 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
438 					    struct list_head *page_list);
439 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
440 #define ALLOC_WMARK_MIN		WMARK_MIN
441 #define ALLOC_WMARK_LOW		WMARK_LOW
442 #define ALLOC_WMARK_HIGH	WMARK_HIGH
443 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
444 
445 /* Mask to get the watermark bits */
446 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
447 
448 #define ALLOC_HARDER		0x10 /* try to alloc harder */
449 #define ALLOC_HIGH		0x20 /* __GFP_HIGH set */
450 #define ALLOC_CPUSET		0x40 /* check for correct cpuset */
451 #define ALLOC_CMA		0x80 /* allow allocations from CMA areas */
452 #define ALLOC_FAIR		0x100 /* fair zone allocation */
453 
454 enum ttu_flags;
455 struct tlbflush_unmap_batch;
456 
457 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
458 void try_to_unmap_flush(void);
459 void try_to_unmap_flush_dirty(void);
460 #else
461 static inline void try_to_unmap_flush(void)
462 {
463 }
464 static inline void try_to_unmap_flush_dirty(void)
465 {
466 }
467 
468 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
469 #endif	/* __MM_INTERNAL_H */
470