blob: 36b23f1e2ca62612e6e1d1b2b9d74c3cd7e87db7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* 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 */
Nick Piggin0f8053a2006-03-22 00:08:33 -080011#ifndef __MM_INTERNAL_H
12#define __MM_INTERNAL_H
13
Fabian Frederick29f175d2014-04-07 15:37:55 -070014#include <linux/fs.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080015#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016
Jan Beulich42b77722008-07-23 21:27:10 -070017void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
18 unsigned long floor, unsigned long ceiling);
19
Nick Piggin7835e982006-03-22 00:08:40 -080020static inline void set_page_count(struct page *page, int v)
Nick Piggin77a8a782006-01-06 00:10:57 -080021{
Nick Piggin7835e982006-03-22 00:08:40 -080022 atomic_set(&page->_count, v);
23}
24
Fabian Frederick29f175d2014-04-07 15:37:55 -070025extern int __do_page_cache_readahead(struct address_space *mapping,
26 struct file *filp, pgoff_t offset, unsigned long nr_to_read,
27 unsigned long lookahead_size);
28
29/*
30 * Submit IO for the read-ahead request in file_ra_state.
31 */
32static inline unsigned long ra_submit(struct file_ra_state *ra,
33 struct address_space *mapping, struct file *filp)
34{
35 return __do_page_cache_readahead(mapping, filp,
36 ra->start, ra->size, ra->async_size);
37}
38
Nick Piggin7835e982006-03-22 00:08:40 -080039/*
40 * Turn a non-refcounted page (->_count == 0) into refcounted with
41 * a count of one.
42 */
43static inline void set_page_refcounted(struct page *page)
44{
Sasha Levin309381fea2014-01-23 15:52:54 -080045 VM_BUG_ON_PAGE(PageTail(page), page);
46 VM_BUG_ON_PAGE(atomic_read(&page->_count), page);
Nick Piggin77a8a782006-01-06 00:10:57 -080047 set_page_count(page, 1);
Nick Piggin77a8a782006-01-06 00:10:57 -080048}
49
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070050static inline void __get_page_tail_foll(struct page *page,
51 bool get_page_head)
52{
53 /*
54 * If we're getting a tail page, the elevated page->_count is
55 * required only in the head page and we will elevate the head
56 * page->_count and tail page->_mapcount.
57 *
58 * We elevate page_tail->_mapcount for tail pages to force
59 * page_tail->_count to be zero at all times to avoid getting
60 * false positives from get_page_unless_zero() with
61 * speculative page access (like in
62 * page_cache_get_speculative()) on tail pages.
63 */
Sasha Levin309381fea2014-01-23 15:52:54 -080064 VM_BUG_ON_PAGE(atomic_read(&page->first_page->_count) <= 0, page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070065 if (get_page_head)
66 atomic_inc(&page->first_page->_count);
Oleg Nesterovc7288522014-01-21 15:49:02 -080067 get_huge_page_tail(page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070068}
69
70/*
71 * This is meant to be called as the FOLL_GET operation of
72 * follow_page() and it must be called while holding the proper PT
73 * lock while the pte (or pmd_trans_huge) is still mapping the page.
74 */
75static inline void get_page_foll(struct page *page)
76{
77 if (unlikely(PageTail(page)))
78 /*
79 * This is safe only because
80 * __split_huge_page_refcount() can't run under
81 * get_page_foll() because we hold the proper PT lock.
82 */
83 __get_page_tail_foll(page, true);
84 else {
85 /*
86 * Getting a normal page or the head of a compound page
87 * requires to already have an elevated page->_count.
88 */
Sasha Levin309381fea2014-01-23 15:52:54 -080089 VM_BUG_ON_PAGE(atomic_read(&page->_count) <= 0, page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070090 atomic_inc(&page->_count);
91 }
92}
93
Hugh Dickins03f64622009-09-21 17:03:35 -070094extern unsigned long highest_memmap_pfn;
95
Lee Schermerhorn894bc312008-10-18 20:26:39 -070096/*
97 * in mm/vmscan.c:
98 */
Nick Piggin62695a82008-10-18 20:26:09 -070099extern int isolate_lru_page(struct page *page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700100extern void putback_lru_page(struct page *page);
Lisa Du6e543d52013-09-11 14:22:36 -0700101extern bool zone_reclaimable(struct zone *zone);
Nick Piggin62695a82008-10-18 20:26:09 -0700102
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700103/*
Bob Liu62190492012-12-11 16:00:37 -0800104 * in mm/rmap.c:
105 */
106extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
107
108/*
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700109 * in mm/page_alloc.c
110 */
Joonsoo Kim3c605092014-11-13 15:19:21 -0800111
112/*
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -0800113 * Structure for holding the mostly immutable allocation parameters passed
114 * between functions involved in allocations, including the alloc_pages*
115 * family of functions.
116 *
117 * nodemask, migratetype and high_zoneidx are initialized only once in
118 * __alloc_pages_nodemask() and then never change.
119 *
120 * zonelist, preferred_zone and classzone_idx are set first in
121 * __alloc_pages_nodemask() for the fast path, and might be later changed
122 * in __alloc_pages_slowpath(). All other functions pass the whole strucure
123 * by a const pointer.
124 */
125struct alloc_context {
126 struct zonelist *zonelist;
127 nodemask_t *nodemask;
128 struct zone *preferred_zone;
129 int classzone_idx;
130 int migratetype;
131 enum zone_type high_zoneidx;
132};
133
134/*
Joonsoo Kim3c605092014-11-13 15:19:21 -0800135 * Locate the struct page for both the matching buddy in our
136 * pair (buddy1) and the combined O(n+1) page they form (page).
137 *
138 * 1) Any buddy B1 will have an order O twin B2 which satisfies
139 * the following equation:
140 * B2 = B1 ^ (1 << O)
141 * For example, if the starting buddy (buddy2) is #8 its order
142 * 1 buddy is #10:
143 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
144 *
145 * 2) Any buddy B will have an order O+1 parent P which
146 * satisfies the following equation:
147 * P = B & ~(1 << O)
148 *
149 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
150 */
151static inline unsigned long
152__find_buddy_index(unsigned long page_idx, unsigned int order)
153{
154 return page_idx ^ (1 << order);
155}
156
157extern int __isolate_free_page(struct page *page, unsigned int order);
Mel Gormand70ddd72015-06-30 14:56:52 -0700158extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
159 unsigned int order);
Wu Fengguang20a03072009-06-16 15:32:22 -0700160extern void prep_compound_page(struct page *page, unsigned long order);
Wu Fengguang8d22ba12009-12-16 12:19:58 +0100161#ifdef CONFIG_MEMORY_FAILURE
162extern bool is_free_buddy_page(struct page *page);
163#endif
Han Pingtian42aa83c2014-01-23 15:53:28 -0800164extern int user_min_free_kbytes;
Wu Fengguang20a03072009-06-16 15:32:22 -0700165
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100166#if defined CONFIG_COMPACTION || defined CONFIG_CMA
167
168/*
169 * in mm/compaction.c
170 */
171/*
172 * compact_control is used to track pages being migrated and the free pages
173 * they are being migrated to during memory compaction. The free_pfn starts
174 * at the end of a zone and migrate_pfn begins at the start. Movable pages
175 * are moved to the end of a zone during a compaction run and the run
176 * completes when free_pfn <= migrate_pfn
177 */
178struct compact_control {
179 struct list_head freepages; /* List of free pages to migrate to */
180 struct list_head migratepages; /* List of pages being migrated */
181 unsigned long nr_freepages; /* Number of isolated free pages */
182 unsigned long nr_migratepages; /* Number of pages to migrate */
183 unsigned long free_pfn; /* isolate_freepages search base */
184 unsigned long migrate_pfn; /* isolate_migratepages search base */
David Rientjese0b9dae2014-06-04 16:08:28 -0700185 enum migrate_mode mode; /* Async or sync migration mode */
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700186 bool ignore_skip_hint; /* Scan blocks even if marked skip */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100187 int order; /* order a direct compactor needs */
David Rientjes6d7ce552014-10-09 15:27:27 -0700188 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
Vlastimil Babkaebff3982014-12-10 15:43:22 -0800189 const int alloc_flags; /* alloc flags of a direct compactor */
190 const int classzone_idx; /* zone index of a direct compactor */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100191 struct zone *zone;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -0700192 int contended; /* Signal need_sched() or lock
193 * contention detected during
Vlastimil Babkabe976572014-06-04 16:10:41 -0700194 * compaction
195 */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100196};
197
198unsigned long
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700199isolate_freepages_range(struct compact_control *cc,
200 unsigned long start_pfn, unsigned long end_pfn);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100201unsigned long
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700202isolate_migratepages_range(struct compact_control *cc,
203 unsigned long low_pfn, unsigned long end_pfn);
Joonsoo Kim2149cda2015-04-14 15:45:21 -0700204int find_suitable_fallback(struct free_area *area, unsigned int order,
205 int migratetype, bool only_stealable, bool *can_steal);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100206
207#endif
Nick Piggin0f8053a2006-03-22 00:08:33 -0800208
Mel Gorman48f13bf2007-10-16 01:26:10 -0700209/*
Mel Gorman6c144662014-01-23 15:53:38 -0800210 * This function returns the order of a free page in the buddy system. In
211 * general, page_zone(page)->lock must be held by the caller to prevent the
212 * page from being allocated in parallel and returning garbage as the order.
213 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700214 * page cannot be allocated or merged in parallel. Alternatively, it must
215 * handle invalid values gracefully, and use page_order_unsafe() below.
Mel Gorman48f13bf2007-10-16 01:26:10 -0700216 */
217static inline unsigned long page_order(struct page *page)
218{
KAMEZAWA Hiroyuki572438f2010-10-26 14:22:08 -0700219 /* PageBuddy() must be checked by the caller */
Mel Gorman48f13bf2007-10-16 01:26:10 -0700220 return page_private(page);
221}
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -0800222
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700223/*
224 * Like page_order(), but for callers who cannot afford to hold the zone lock.
225 * PageBuddy() should be checked first by the caller to minimize race window,
226 * and invalid values must be handled gracefully.
227 *
Jason Low4db0c3c2015-04-15 16:14:08 -0700228 * READ_ONCE is used so that if the caller assigns the result into a local
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700229 * variable and e.g. tests it for valid range before using, the compiler cannot
230 * decide to remove the variable and inline the page_private(page) multiple
231 * times, potentially observing different values in the tests and the actual
232 * use of the result.
233 */
Jason Low4db0c3c2015-04-15 16:14:08 -0700234#define page_order_unsafe(page) READ_ONCE(page_private(page))
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700235
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700236static inline bool is_cow_mapping(vm_flags_t flags)
237{
238 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
239}
240
Namhyung Kim6038def2011-05-24 17:11:22 -0700241/* mm/util.c */
242void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
243 struct vm_area_struct *prev, struct rb_node *rb_parent);
244
Hugh Dickinsaf8e3352009-12-14 17:58:59 -0800245#ifdef CONFIG_MMU
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700246extern long populate_vma_page_range(struct vm_area_struct *vma,
Michel Lespinassecea10a12013-02-22 16:32:44 -0800247 unsigned long start, unsigned long end, int *nonblocking);
Hugh Dickinsaf8e3352009-12-14 17:58:59 -0800248extern void munlock_vma_pages_range(struct vm_area_struct *vma,
249 unsigned long start, unsigned long end);
250static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
251{
252 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
253}
254
Nick Pigginb291f002008-10-18 20:26:44 -0700255/*
Hugh Dickins73848b42009-12-14 17:59:22 -0800256 * must be called with vma's mmap_sem held for read or write, and page locked.
Nick Pigginb291f002008-10-18 20:26:44 -0700257 */
258extern void mlock_vma_page(struct page *page);
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800259extern unsigned int munlock_vma_page(struct page *page);
Nick Pigginb291f002008-10-18 20:26:44 -0700260
261/*
262 * Clear the page's PageMlocked(). This can be useful in a situation where
263 * we want to unconditionally remove a page from the pagecache -- e.g.,
264 * on truncation or freeing.
265 *
266 * It is legal to call this function for any page, mlocked or not.
267 * If called for a page that is still mapped by mlocked vmas, all we do
268 * is revert to lazy LRU behaviour -- semantics are not broken.
269 */
Hugh Dickinse6c509f2012-10-08 16:33:19 -0700270extern void clear_page_mlock(struct page *page);
Nick Pigginb291f002008-10-18 20:26:44 -0700271
272/*
273 * mlock_migrate_page - called only from migrate_page_copy() to
Nick Piggin5344b7e2008-10-18 20:26:51 -0700274 * migrate the Mlocked page flag; update statistics.
Nick Pigginb291f002008-10-18 20:26:44 -0700275 */
276static inline void mlock_migrate_page(struct page *newpage, struct page *page)
277{
Nick Piggin5344b7e2008-10-18 20:26:51 -0700278 if (TestClearPageMlocked(page)) {
279 unsigned long flags;
Mel Gormanb32967f2012-11-19 12:35:47 +0000280 int nr_pages = hpage_nr_pages(page);
Nick Piggin5344b7e2008-10-18 20:26:51 -0700281
282 local_irq_save(flags);
Mel Gormanb32967f2012-11-19 12:35:47 +0000283 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
Nick Pigginb291f002008-10-18 20:26:44 -0700284 SetPageMlocked(newpage);
Mel Gormanb32967f2012-11-19 12:35:47 +0000285 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
Nick Piggin5344b7e2008-10-18 20:26:51 -0700286 local_irq_restore(flags);
287 }
Nick Pigginb291f002008-10-18 20:26:44 -0700288}
289
Mel Gormanb32967f2012-11-19 12:35:47 +0000290extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
291
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800292#ifdef CONFIG_TRANSPARENT_HUGEPAGE
293extern unsigned long vma_address(struct page *page,
294 struct vm_area_struct *vma);
295#endif
Hugh Dickinsaf8e3352009-12-14 17:58:59 -0800296#else /* !CONFIG_MMU */
Nick Pigginb291f002008-10-18 20:26:44 -0700297static inline void clear_page_mlock(struct page *page) { }
298static inline void mlock_vma_page(struct page *page) { }
299static inline void mlock_migrate_page(struct page *new, struct page *old) { }
300
Hugh Dickinsaf8e3352009-12-14 17:58:59 -0800301#endif /* !CONFIG_MMU */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700302
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -0800303/*
Andy Whitcroft69d177c2008-11-06 12:53:26 -0800304 * Return the mem_map entry representing the 'offset' subpage within
305 * the maximally aligned gigantic page 'base'. Handle any discontiguity
306 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
307 */
308static inline struct page *mem_map_offset(struct page *base, int offset)
309{
310 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
Fabian Frederickbc7f84c2014-08-06 16:05:17 -0700311 return nth_page(base, offset);
Andy Whitcroft69d177c2008-11-06 12:53:26 -0800312 return base + offset;
313}
314
315/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300316 * Iterator over all subpages within the maximally aligned gigantic
Andy Whitcroft69d177c2008-11-06 12:53:26 -0800317 * page 'base'. Handle any discontiguity in the mem_map.
318 */
319static inline struct page *mem_map_next(struct page *iter,
320 struct page *base, int offset)
321{
322 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
323 unsigned long pfn = page_to_pfn(base) + offset;
324 if (!pfn_valid(pfn))
325 return NULL;
326 return pfn_to_page(pfn);
327 }
328 return iter + 1;
329}
330
331/*
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -0800332 * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
333 * so all functions starting at paging_init should be marked __init
334 * in those cases. SPARSEMEM, however, allows for memory hotplug,
335 * and alloc_bootmem_node is not used.
336 */
337#ifdef CONFIG_SPARSEMEM
338#define __paginginit __meminit
339#else
340#define __paginginit __init
341#endif
342
Mel Gorman6b74ab92008-07-23 21:26:49 -0700343/* Memory initialisation debug and verification */
344enum mminit_level {
345 MMINIT_WARNING,
346 MMINIT_VERIFY,
347 MMINIT_TRACE
348};
349
350#ifdef CONFIG_DEBUG_MEMORY_INIT
351
352extern int mminit_loglevel;
353
354#define mminit_dprintk(level, prefix, fmt, arg...) \
355do { \
356 if (level < mminit_loglevel) { \
Rasmus Villemoesfc5199d2015-02-12 15:00:02 -0800357 if (level <= MMINIT_WARNING) \
358 printk(KERN_WARNING "mminit::" prefix " " fmt, ##arg); \
359 else \
360 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
Mel Gorman6b74ab92008-07-23 21:26:49 -0700361 } \
362} while (0)
363
Mel Gorman708614e2008-07-23 21:26:51 -0700364extern void mminit_verify_pageflags_layout(void);
Mel Gorman68ad8df2008-07-23 21:26:52 -0700365extern void mminit_verify_zonelist(void);
Mel Gorman6b74ab92008-07-23 21:26:49 -0700366#else
367
368static inline void mminit_dprintk(enum mminit_level level,
369 const char *prefix, const char *fmt, ...)
370{
371}
372
Mel Gorman708614e2008-07-23 21:26:51 -0700373static inline void mminit_verify_pageflags_layout(void)
374{
375}
376
Mel Gorman68ad8df2008-07-23 21:26:52 -0700377static inline void mminit_verify_zonelist(void)
378{
379}
Mel Gorman6b74ab92008-07-23 21:26:49 -0700380#endif /* CONFIG_DEBUG_MEMORY_INIT */
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700381
382/* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
383#if defined(CONFIG_SPARSEMEM)
384extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
385 unsigned long *end_pfn);
386#else
387static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
388 unsigned long *end_pfn)
389{
390}
391#endif /* CONFIG_SPARSEMEM */
392
Mel Gormanfa5e0842009-06-16 15:33:22 -0700393#define ZONE_RECLAIM_NOSCAN -2
394#define ZONE_RECLAIM_FULL -1
395#define ZONE_RECLAIM_SOME 0
396#define ZONE_RECLAIM_SUCCESS 1
Wu Fengguang7c116f22009-12-16 12:19:59 +0100397
Wu Fengguang31d3d342009-12-16 12:19:59 +0100398extern int hwpoison_filter(struct page *p);
399
Wu Fengguang7c116f22009-12-16 12:19:59 +0100400extern u32 hwpoison_filter_dev_major;
401extern u32 hwpoison_filter_dev_minor;
Wu Fengguang478c5ff2009-12-16 12:19:59 +0100402extern u64 hwpoison_filter_flags_mask;
403extern u64 hwpoison_filter_flags_value;
Andi Kleen4fd466e2009-12-16 12:19:59 +0100404extern u64 hwpoison_filter_memcg;
Haicheng Li1bfe5fe2009-12-16 12:19:59 +0100405extern u32 hwpoison_filter_enable;
Al Viroeb36c582012-05-30 20:17:35 -0400406
407extern unsigned long vm_mmap_pgoff(struct file *, unsigned long,
408 unsigned long, unsigned long,
409 unsigned long, unsigned long);
Xishi Qiuca57df72012-07-31 16:43:19 -0700410
411extern void set_pageblock_order(void);
Minchan Kim02c6de82012-10-08 16:31:55 -0700412unsigned long reclaim_clean_pages_from_list(struct zone *zone,
413 struct list_head *page_list);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -0700414/* The ALLOC_WMARK bits are used as an index to zone->watermark */
415#define ALLOC_WMARK_MIN WMARK_MIN
416#define ALLOC_WMARK_LOW WMARK_LOW
417#define ALLOC_WMARK_HIGH WMARK_HIGH
418#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
419
420/* Mask to get the watermark bits */
421#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
422
423#define ALLOC_HARDER 0x10 /* try to alloc harder */
424#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
425#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
426#define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
Johannes Weiner3a025762014-04-07 15:37:48 -0700427#define ALLOC_FAIR 0x100 /* fair zone allocation */
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -0700428
Michel Lespinassedb971412012-10-08 16:29:34 -0700429#endif /* __MM_INTERNAL_H */