xref: /openbmc/linux/mm/internal.h (revision 021ed9f5)
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 extern void zone_pcp_update(struct zone *zone);
169 extern void zone_pcp_reset(struct zone *zone);
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 int nr_freepages;	/* Number of isolated free pages */
187 	unsigned int nr_migratepages;	/* Number of pages to migrate */
188 	unsigned long free_pfn;		/* isolate_freepages search base */
189 	unsigned long migrate_pfn;	/* isolate_migratepages search base */
190 	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
191 	struct zone *zone;
192 	unsigned long total_migrate_scanned;
193 	unsigned long total_free_scanned;
194 	unsigned short fast_search_fail;/* failures to use free list searches */
195 	short search_order;		/* order to start a fast search at */
196 	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
197 	int order;			/* order a direct compactor needs */
198 	int migratetype;		/* migratetype of direct compactor */
199 	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
200 	const int classzone_idx;	/* zone index of a direct compactor */
201 	enum migrate_mode mode;		/* Async or sync migration mode */
202 	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
203 	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
204 	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
205 	bool direct_compaction;		/* False from kcompactd or /proc/... */
206 	bool whole_zone;		/* Whole zone should/has been scanned */
207 	bool contended;			/* Signal lock or sched contention */
208 	bool rescan;			/* Rescanning the same pageblock */
209 };
210 
211 /*
212  * Used in direct compaction when a page should be taken from the freelists
213  * immediately when one is created during the free path.
214  */
215 struct capture_control {
216 	struct compact_control *cc;
217 	struct page *page;
218 };
219 
220 unsigned long
221 isolate_freepages_range(struct compact_control *cc,
222 			unsigned long start_pfn, unsigned long end_pfn);
223 unsigned long
224 isolate_migratepages_range(struct compact_control *cc,
225 			   unsigned long low_pfn, unsigned long end_pfn);
226 int find_suitable_fallback(struct free_area *area, unsigned int order,
227 			int migratetype, bool only_stealable, bool *can_steal);
228 
229 #endif
230 
231 /*
232  * This function returns the order of a free page in the buddy system. In
233  * general, page_zone(page)->lock must be held by the caller to prevent the
234  * page from being allocated in parallel and returning garbage as the order.
235  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
236  * page cannot be allocated or merged in parallel. Alternatively, it must
237  * handle invalid values gracefully, and use page_order_unsafe() below.
238  */
239 static inline unsigned int page_order(struct page *page)
240 {
241 	/* PageBuddy() must be checked by the caller */
242 	return page_private(page);
243 }
244 
245 /*
246  * Like page_order(), but for callers who cannot afford to hold the zone lock.
247  * PageBuddy() should be checked first by the caller to minimize race window,
248  * and invalid values must be handled gracefully.
249  *
250  * READ_ONCE is used so that if the caller assigns the result into a local
251  * variable and e.g. tests it for valid range before using, the compiler cannot
252  * decide to remove the variable and inline the page_private(page) multiple
253  * times, potentially observing different values in the tests and the actual
254  * use of the result.
255  */
256 #define page_order_unsafe(page)		READ_ONCE(page_private(page))
257 
258 static inline bool is_cow_mapping(vm_flags_t flags)
259 {
260 	return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
261 }
262 
263 /*
264  * These three helpers classifies VMAs for virtual memory accounting.
265  */
266 
267 /*
268  * Executable code area - executable, not writable, not stack
269  */
270 static inline bool is_exec_mapping(vm_flags_t flags)
271 {
272 	return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
273 }
274 
275 /*
276  * Stack area - atomatically grows in one direction
277  *
278  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
279  * do_mmap() forbids all other combinations.
280  */
281 static inline bool is_stack_mapping(vm_flags_t flags)
282 {
283 	return (flags & VM_STACK) == VM_STACK;
284 }
285 
286 /*
287  * Data area - private, writable, not stack
288  */
289 static inline bool is_data_mapping(vm_flags_t flags)
290 {
291 	return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
292 }
293 
294 /* mm/util.c */
295 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
296 		struct vm_area_struct *prev);
297 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
298 
299 #ifdef CONFIG_MMU
300 extern long populate_vma_page_range(struct vm_area_struct *vma,
301 		unsigned long start, unsigned long end, int *nonblocking);
302 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
303 			unsigned long start, unsigned long end);
304 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
305 {
306 	munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
307 }
308 
309 /*
310  * must be called with vma's mmap_sem held for read or write, and page locked.
311  */
312 extern void mlock_vma_page(struct page *page);
313 extern unsigned int munlock_vma_page(struct page *page);
314 
315 /*
316  * Clear the page's PageMlocked().  This can be useful in a situation where
317  * we want to unconditionally remove a page from the pagecache -- e.g.,
318  * on truncation or freeing.
319  *
320  * It is legal to call this function for any page, mlocked or not.
321  * If called for a page that is still mapped by mlocked vmas, all we do
322  * is revert to lazy LRU behaviour -- semantics are not broken.
323  */
324 extern void clear_page_mlock(struct page *page);
325 
326 /*
327  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
328  * (because that does not go through the full procedure of migration ptes):
329  * to migrate the Mlocked page flag; update statistics.
330  */
331 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
332 {
333 	if (TestClearPageMlocked(page)) {
334 		int nr_pages = hpage_nr_pages(page);
335 
336 		/* Holding pmd lock, no change in irq context: __mod is safe */
337 		__mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
338 		SetPageMlocked(newpage);
339 		__mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
340 	}
341 }
342 
343 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
344 
345 /*
346  * At what user virtual address is page expected in @vma?
347  */
348 static inline unsigned long
349 __vma_address(struct page *page, struct vm_area_struct *vma)
350 {
351 	pgoff_t pgoff = page_to_pgoff(page);
352 	return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
353 }
354 
355 static inline unsigned long
356 vma_address(struct page *page, struct vm_area_struct *vma)
357 {
358 	unsigned long start, end;
359 
360 	start = __vma_address(page, vma);
361 	end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
362 
363 	/* page should be within @vma mapping range */
364 	VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma);
365 
366 	return max(start, vma->vm_start);
367 }
368 
369 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
370 						    struct file *fpin)
371 {
372 	int flags = vmf->flags;
373 
374 	if (fpin)
375 		return fpin;
376 
377 	/*
378 	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
379 	 * anything, so we only pin the file and drop the mmap_sem if only
380 	 * FAULT_FLAG_ALLOW_RETRY is set.
381 	 */
382 	if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) ==
383 	    FAULT_FLAG_ALLOW_RETRY) {
384 		fpin = get_file(vmf->vma->vm_file);
385 		up_read(&vmf->vma->vm_mm->mmap_sem);
386 	}
387 	return fpin;
388 }
389 
390 #else /* !CONFIG_MMU */
391 static inline void clear_page_mlock(struct page *page) { }
392 static inline void mlock_vma_page(struct page *page) { }
393 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
394 
395 #endif /* !CONFIG_MMU */
396 
397 /*
398  * Return the mem_map entry representing the 'offset' subpage within
399  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
400  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
401  */
402 static inline struct page *mem_map_offset(struct page *base, int offset)
403 {
404 	if (unlikely(offset >= MAX_ORDER_NR_PAGES))
405 		return nth_page(base, offset);
406 	return base + offset;
407 }
408 
409 /*
410  * Iterator over all subpages within the maximally aligned gigantic
411  * page 'base'.  Handle any discontiguity in the mem_map.
412  */
413 static inline struct page *mem_map_next(struct page *iter,
414 						struct page *base, int offset)
415 {
416 	if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
417 		unsigned long pfn = page_to_pfn(base) + offset;
418 		if (!pfn_valid(pfn))
419 			return NULL;
420 		return pfn_to_page(pfn);
421 	}
422 	return iter + 1;
423 }
424 
425 /* Memory initialisation debug and verification */
426 enum mminit_level {
427 	MMINIT_WARNING,
428 	MMINIT_VERIFY,
429 	MMINIT_TRACE
430 };
431 
432 #ifdef CONFIG_DEBUG_MEMORY_INIT
433 
434 extern int mminit_loglevel;
435 
436 #define mminit_dprintk(level, prefix, fmt, arg...) \
437 do { \
438 	if (level < mminit_loglevel) { \
439 		if (level <= MMINIT_WARNING) \
440 			pr_warn("mminit::" prefix " " fmt, ##arg);	\
441 		else \
442 			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
443 	} \
444 } while (0)
445 
446 extern void mminit_verify_pageflags_layout(void);
447 extern void mminit_verify_zonelist(void);
448 #else
449 
450 static inline void mminit_dprintk(enum mminit_level level,
451 				const char *prefix, const char *fmt, ...)
452 {
453 }
454 
455 static inline void mminit_verify_pageflags_layout(void)
456 {
457 }
458 
459 static inline void mminit_verify_zonelist(void)
460 {
461 }
462 #endif /* CONFIG_DEBUG_MEMORY_INIT */
463 
464 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
465 #if defined(CONFIG_SPARSEMEM)
466 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
467 				unsigned long *end_pfn);
468 #else
469 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
470 				unsigned long *end_pfn)
471 {
472 }
473 #endif /* CONFIG_SPARSEMEM */
474 
475 #define NODE_RECLAIM_NOSCAN	-2
476 #define NODE_RECLAIM_FULL	-1
477 #define NODE_RECLAIM_SOME	0
478 #define NODE_RECLAIM_SUCCESS	1
479 
480 #ifdef CONFIG_NUMA
481 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
482 #else
483 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
484 				unsigned int order)
485 {
486 	return NODE_RECLAIM_NOSCAN;
487 }
488 #endif
489 
490 extern int hwpoison_filter(struct page *p);
491 
492 extern u32 hwpoison_filter_dev_major;
493 extern u32 hwpoison_filter_dev_minor;
494 extern u64 hwpoison_filter_flags_mask;
495 extern u64 hwpoison_filter_flags_value;
496 extern u64 hwpoison_filter_memcg;
497 extern u32 hwpoison_filter_enable;
498 
499 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
500         unsigned long, unsigned long,
501         unsigned long, unsigned long);
502 
503 extern void set_pageblock_order(void);
504 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
505 					    struct list_head *page_list);
506 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
507 #define ALLOC_WMARK_MIN		WMARK_MIN
508 #define ALLOC_WMARK_LOW		WMARK_LOW
509 #define ALLOC_WMARK_HIGH	WMARK_HIGH
510 #define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
511 
512 /* Mask to get the watermark bits */
513 #define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
514 
515 /*
516  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
517  * cannot assume a reduced access to memory reserves is sufficient for
518  * !MMU
519  */
520 #ifdef CONFIG_MMU
521 #define ALLOC_OOM		0x08
522 #else
523 #define ALLOC_OOM		ALLOC_NO_WATERMARKS
524 #endif
525 
526 #define ALLOC_HARDER		 0x10 /* try to alloc harder */
527 #define ALLOC_HIGH		 0x20 /* __GFP_HIGH set */
528 #define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
529 #define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
530 #ifdef CONFIG_ZONE_DMA32
531 #define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
532 #else
533 #define ALLOC_NOFRAGMENT	  0x0
534 #endif
535 #define ALLOC_KSWAPD		0x200 /* allow waking of kswapd */
536 
537 enum ttu_flags;
538 struct tlbflush_unmap_batch;
539 
540 
541 /*
542  * only for MM internal work items which do not depend on
543  * any allocations or locks which might depend on allocations
544  */
545 extern struct workqueue_struct *mm_percpu_wq;
546 
547 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
548 void try_to_unmap_flush(void);
549 void try_to_unmap_flush_dirty(void);
550 void flush_tlb_batched_pending(struct mm_struct *mm);
551 #else
552 static inline void try_to_unmap_flush(void)
553 {
554 }
555 static inline void try_to_unmap_flush_dirty(void)
556 {
557 }
558 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
559 {
560 }
561 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
562 
563 extern const struct trace_print_flags pageflag_names[];
564 extern const struct trace_print_flags vmaflag_names[];
565 extern const struct trace_print_flags gfpflag_names[];
566 
567 static inline bool is_migrate_highatomic(enum migratetype migratetype)
568 {
569 	return migratetype == MIGRATE_HIGHATOMIC;
570 }
571 
572 static inline bool is_migrate_highatomic_page(struct page *page)
573 {
574 	return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
575 }
576 
577 void setup_zone_pageset(struct zone *zone);
578 extern struct page *alloc_new_node_page(struct page *page, unsigned long node);
579 #endif	/* __MM_INTERNAL_H */
580