Loading...
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_RETRY_MAYFAIL|__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
39void page_writeback_init(void);
40
41int do_swap_page(struct vm_fault *vmf);
42
43void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
44 unsigned long floor, unsigned long ceiling);
45
46static 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
51void 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
56extern 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 */
63static 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 */
74static 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
81extern 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 */
92extern int isolate_lru_page(struct page *page);
93extern void putback_lru_page(struct page *page);
94
95/*
96 * in mm/rmap.c:
97 */
98extern 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 */
117struct 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 */
145static inline unsigned long
146__find_buddy_pfn(unsigned long page_pfn, unsigned int order)
147{
148 return page_pfn ^ (1 << order);
149}
150
151extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
152 unsigned long end_pfn, struct zone *zone);
153
154static 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
163extern int __isolate_free_page(struct page *page, unsigned int order);
164extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
165 unsigned int order);
166extern void prep_compound_page(struct page *page, unsigned int order);
167extern void post_alloc_hook(struct page *page, unsigned int order,
168 gfp_t gfp_flags);
169extern 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 */
183struct 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 struct zone *zone;
187 unsigned long nr_freepages; /* Number of isolated free pages */
188 unsigned long nr_migratepages; /* Number of pages to migrate */
189 unsigned long total_migrate_scanned;
190 unsigned long total_free_scanned;
191 unsigned long free_pfn; /* isolate_freepages search base */
192 unsigned long migrate_pfn; /* isolate_migratepages search base */
193 unsigned long last_migrated_pfn;/* Not yet flushed page being freed */
194 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
195 int order; /* order a direct compactor needs */
196 int migratetype; /* migratetype of direct compactor */
197 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
198 const int classzone_idx; /* zone index of a direct compactor */
199 enum migrate_mode mode; /* Async or sync migration mode */
200 bool ignore_skip_hint; /* Scan blocks even if marked skip */
201 bool no_set_skip_hint; /* Don't mark blocks for skipping */
202 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
203 bool direct_compaction; /* False from kcompactd or /proc/... */
204 bool whole_zone; /* Whole zone should/has been scanned */
205 bool contended; /* Signal lock or sched contention */
206 bool finishing_block; /* Finishing current pageblock */
207};
208
209unsigned long
210isolate_freepages_range(struct compact_control *cc,
211 unsigned long start_pfn, unsigned long end_pfn);
212unsigned long
213isolate_migratepages_range(struct compact_control *cc,
214 unsigned long low_pfn, unsigned long end_pfn);
215int find_suitable_fallback(struct free_area *area, unsigned int order,
216 int migratetype, bool only_stealable, bool *can_steal);
217
218#endif
219
220/*
221 * This function returns the order of a free page in the buddy system. In
222 * general, page_zone(page)->lock must be held by the caller to prevent the
223 * page from being allocated in parallel and returning garbage as the order.
224 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
225 * page cannot be allocated or merged in parallel. Alternatively, it must
226 * handle invalid values gracefully, and use page_order_unsafe() below.
227 */
228static inline unsigned int page_order(struct page *page)
229{
230 /* PageBuddy() must be checked by the caller */
231 return page_private(page);
232}
233
234/*
235 * Like page_order(), but for callers who cannot afford to hold the zone lock.
236 * PageBuddy() should be checked first by the caller to minimize race window,
237 * and invalid values must be handled gracefully.
238 *
239 * READ_ONCE is used so that if the caller assigns the result into a local
240 * variable and e.g. tests it for valid range before using, the compiler cannot
241 * decide to remove the variable and inline the page_private(page) multiple
242 * times, potentially observing different values in the tests and the actual
243 * use of the result.
244 */
245#define page_order_unsafe(page) READ_ONCE(page_private(page))
246
247static inline bool is_cow_mapping(vm_flags_t flags)
248{
249 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
250}
251
252/*
253 * These three helpers classifies VMAs for virtual memory accounting.
254 */
255
256/*
257 * Executable code area - executable, not writable, not stack
258 */
259static inline bool is_exec_mapping(vm_flags_t flags)
260{
261 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
262}
263
264/*
265 * Stack area - atomatically grows in one direction
266 *
267 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
268 * do_mmap() forbids all other combinations.
269 */
270static inline bool is_stack_mapping(vm_flags_t flags)
271{
272 return (flags & VM_STACK) == VM_STACK;
273}
274
275/*
276 * Data area - private, writable, not stack
277 */
278static inline bool is_data_mapping(vm_flags_t flags)
279{
280 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
281}
282
283/* mm/util.c */
284void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
285 struct vm_area_struct *prev, struct rb_node *rb_parent);
286
287#ifdef CONFIG_MMU
288extern long populate_vma_page_range(struct vm_area_struct *vma,
289 unsigned long start, unsigned long end, int *nonblocking);
290extern void munlock_vma_pages_range(struct vm_area_struct *vma,
291 unsigned long start, unsigned long end);
292static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
293{
294 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
295}
296
297/*
298 * must be called with vma's mmap_sem held for read or write, and page locked.
299 */
300extern void mlock_vma_page(struct page *page);
301extern unsigned int munlock_vma_page(struct page *page);
302
303/*
304 * Clear the page's PageMlocked(). This can be useful in a situation where
305 * we want to unconditionally remove a page from the pagecache -- e.g.,
306 * on truncation or freeing.
307 *
308 * It is legal to call this function for any page, mlocked or not.
309 * If called for a page that is still mapped by mlocked vmas, all we do
310 * is revert to lazy LRU behaviour -- semantics are not broken.
311 */
312extern void clear_page_mlock(struct page *page);
313
314/*
315 * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
316 * (because that does not go through the full procedure of migration ptes):
317 * to migrate the Mlocked page flag; update statistics.
318 */
319static inline void mlock_migrate_page(struct page *newpage, struct page *page)
320{
321 if (TestClearPageMlocked(page)) {
322 int nr_pages = hpage_nr_pages(page);
323
324 /* Holding pmd lock, no change in irq context: __mod is safe */
325 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
326 SetPageMlocked(newpage);
327 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
328 }
329}
330
331extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
332
333/*
334 * At what user virtual address is page expected in @vma?
335 */
336static inline unsigned long
337__vma_address(struct page *page, struct vm_area_struct *vma)
338{
339 pgoff_t pgoff = page_to_pgoff(page);
340 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
341}
342
343static inline unsigned long
344vma_address(struct page *page, struct vm_area_struct *vma)
345{
346 unsigned long start, end;
347
348 start = __vma_address(page, vma);
349 end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
350
351 /* page should be within @vma mapping range */
352 VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma);
353
354 return max(start, vma->vm_start);
355}
356
357#else /* !CONFIG_MMU */
358static inline void clear_page_mlock(struct page *page) { }
359static inline void mlock_vma_page(struct page *page) { }
360static inline void mlock_migrate_page(struct page *new, struct page *old) { }
361
362#endif /* !CONFIG_MMU */
363
364/*
365 * Return the mem_map entry representing the 'offset' subpage within
366 * the maximally aligned gigantic page 'base'. Handle any discontiguity
367 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
368 */
369static inline struct page *mem_map_offset(struct page *base, int offset)
370{
371 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
372 return nth_page(base, offset);
373 return base + offset;
374}
375
376/*
377 * Iterator over all subpages within the maximally aligned gigantic
378 * page 'base'. Handle any discontiguity in the mem_map.
379 */
380static inline struct page *mem_map_next(struct page *iter,
381 struct page *base, int offset)
382{
383 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
384 unsigned long pfn = page_to_pfn(base) + offset;
385 if (!pfn_valid(pfn))
386 return NULL;
387 return pfn_to_page(pfn);
388 }
389 return iter + 1;
390}
391
392/*
393 * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
394 * so all functions starting at paging_init should be marked __init
395 * in those cases. SPARSEMEM, however, allows for memory hotplug,
396 * and alloc_bootmem_node is not used.
397 */
398#ifdef CONFIG_SPARSEMEM
399#define __paginginit __meminit
400#else
401#define __paginginit __init
402#endif
403
404/* Memory initialisation debug and verification */
405enum mminit_level {
406 MMINIT_WARNING,
407 MMINIT_VERIFY,
408 MMINIT_TRACE
409};
410
411#ifdef CONFIG_DEBUG_MEMORY_INIT
412
413extern int mminit_loglevel;
414
415#define mminit_dprintk(level, prefix, fmt, arg...) \
416do { \
417 if (level < mminit_loglevel) { \
418 if (level <= MMINIT_WARNING) \
419 pr_warn("mminit::" prefix " " fmt, ##arg); \
420 else \
421 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
422 } \
423} while (0)
424
425extern void mminit_verify_pageflags_layout(void);
426extern void mminit_verify_zonelist(void);
427#else
428
429static inline void mminit_dprintk(enum mminit_level level,
430 const char *prefix, const char *fmt, ...)
431{
432}
433
434static inline void mminit_verify_pageflags_layout(void)
435{
436}
437
438static inline void mminit_verify_zonelist(void)
439{
440}
441#endif /* CONFIG_DEBUG_MEMORY_INIT */
442
443/* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
444#if defined(CONFIG_SPARSEMEM)
445extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
446 unsigned long *end_pfn);
447#else
448static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
449 unsigned long *end_pfn)
450{
451}
452#endif /* CONFIG_SPARSEMEM */
453
454#define NODE_RECLAIM_NOSCAN -2
455#define NODE_RECLAIM_FULL -1
456#define NODE_RECLAIM_SOME 0
457#define NODE_RECLAIM_SUCCESS 1
458
459extern int hwpoison_filter(struct page *p);
460
461extern u32 hwpoison_filter_dev_major;
462extern u32 hwpoison_filter_dev_minor;
463extern u64 hwpoison_filter_flags_mask;
464extern u64 hwpoison_filter_flags_value;
465extern u64 hwpoison_filter_memcg;
466extern u32 hwpoison_filter_enable;
467
468extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
469 unsigned long, unsigned long,
470 unsigned long, unsigned long);
471
472extern void set_pageblock_order(void);
473unsigned long reclaim_clean_pages_from_list(struct zone *zone,
474 struct list_head *page_list);
475/* The ALLOC_WMARK bits are used as an index to zone->watermark */
476#define ALLOC_WMARK_MIN WMARK_MIN
477#define ALLOC_WMARK_LOW WMARK_LOW
478#define ALLOC_WMARK_HIGH WMARK_HIGH
479#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
480
481/* Mask to get the watermark bits */
482#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
483
484/*
485 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
486 * cannot assume a reduced access to memory reserves is sufficient for
487 * !MMU
488 */
489#ifdef CONFIG_MMU
490#define ALLOC_OOM 0x08
491#else
492#define ALLOC_OOM ALLOC_NO_WATERMARKS
493#endif
494
495#define ALLOC_HARDER 0x10 /* try to alloc harder */
496#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
497#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
498#define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
499
500enum ttu_flags;
501struct tlbflush_unmap_batch;
502
503
504/*
505 * only for MM internal work items which do not depend on
506 * any allocations or locks which might depend on allocations
507 */
508extern struct workqueue_struct *mm_percpu_wq;
509
510#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
511void try_to_unmap_flush(void);
512void try_to_unmap_flush_dirty(void);
513void flush_tlb_batched_pending(struct mm_struct *mm);
514#else
515static inline void try_to_unmap_flush(void)
516{
517}
518static inline void try_to_unmap_flush_dirty(void)
519{
520}
521static inline void flush_tlb_batched_pending(struct mm_struct *mm)
522{
523}
524#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
525
526extern const struct trace_print_flags pageflag_names[];
527extern const struct trace_print_flags vmaflag_names[];
528extern const struct trace_print_flags gfpflag_names[];
529
530static inline bool is_migrate_highatomic(enum migratetype migratetype)
531{
532 return migratetype == MIGRATE_HIGHATOMIC;
533}
534
535static inline bool is_migrate_highatomic_page(struct page *page)
536{
537 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
538}
539
540void setup_zone_pageset(struct zone *zone);
541extern struct page *alloc_new_node_page(struct page *page, unsigned long node);
542#endif /* __MM_INTERNAL_H */
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
35void page_writeback_init(void);
36
37vm_fault_t do_swap_page(struct vm_fault *vmf);
38
39void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
40 unsigned long floor, unsigned long ceiling);
41
42static 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
47void 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
52void do_page_cache_ra(struct readahead_control *, unsigned long nr_to_read,
53 unsigned long lookahead_size);
54void force_page_cache_ra(struct readahead_control *, unsigned long nr);
55static inline void force_page_cache_readahead(struct address_space *mapping,
56 struct file *file, pgoff_t index, unsigned long nr_to_read)
57{
58 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
59 force_page_cache_ra(&ractl, nr_to_read);
60}
61
62unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
63 pgoff_t end, struct pagevec *pvec, pgoff_t *indices);
64
65/**
66 * page_evictable - test whether a page is evictable
67 * @page: the page to test
68 *
69 * Test whether page is evictable--i.e., should be placed on active/inactive
70 * lists vs unevictable list.
71 *
72 * Reasons page might not be evictable:
73 * (1) page's mapping marked unevictable
74 * (2) page is part of an mlocked VMA
75 *
76 */
77static inline bool page_evictable(struct page *page)
78{
79 bool ret;
80
81 /* Prevent address_space of inode and swap cache from being freed */
82 rcu_read_lock();
83 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
84 rcu_read_unlock();
85 return ret;
86}
87
88/*
89 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
90 * a count of one.
91 */
92static inline void set_page_refcounted(struct page *page)
93{
94 VM_BUG_ON_PAGE(PageTail(page), page);
95 VM_BUG_ON_PAGE(page_ref_count(page), page);
96 set_page_count(page, 1);
97}
98
99extern unsigned long highest_memmap_pfn;
100
101/*
102 * Maximum number of reclaim retries without progress before the OOM
103 * killer is consider the only way forward.
104 */
105#define MAX_RECLAIM_RETRIES 16
106
107/*
108 * in mm/vmscan.c:
109 */
110extern int isolate_lru_page(struct page *page);
111extern void putback_lru_page(struct page *page);
112
113/*
114 * in mm/rmap.c:
115 */
116extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
117
118/*
119 * in mm/memcontrol.c:
120 */
121extern bool cgroup_memory_nokmem;
122
123/*
124 * in mm/page_alloc.c
125 */
126
127/*
128 * Structure for holding the mostly immutable allocation parameters passed
129 * between functions involved in allocations, including the alloc_pages*
130 * family of functions.
131 *
132 * nodemask, migratetype and highest_zoneidx are initialized only once in
133 * __alloc_pages() and then never change.
134 *
135 * zonelist, preferred_zone and highest_zoneidx are set first in
136 * __alloc_pages() for the fast path, and might be later changed
137 * in __alloc_pages_slowpath(). All other functions pass the whole structure
138 * by a const pointer.
139 */
140struct alloc_context {
141 struct zonelist *zonelist;
142 nodemask_t *nodemask;
143 struct zoneref *preferred_zoneref;
144 int migratetype;
145
146 /*
147 * highest_zoneidx represents highest usable zone index of
148 * the allocation request. Due to the nature of the zone,
149 * memory on lower zone than the highest_zoneidx will be
150 * protected by lowmem_reserve[highest_zoneidx].
151 *
152 * highest_zoneidx is also used by reclaim/compaction to limit
153 * the target zone since higher zone than this index cannot be
154 * usable for this allocation request.
155 */
156 enum zone_type highest_zoneidx;
157 bool spread_dirty_pages;
158};
159
160/*
161 * Locate the struct page for both the matching buddy in our
162 * pair (buddy1) and the combined O(n+1) page they form (page).
163 *
164 * 1) Any buddy B1 will have an order O twin B2 which satisfies
165 * the following equation:
166 * B2 = B1 ^ (1 << O)
167 * For example, if the starting buddy (buddy2) is #8 its order
168 * 1 buddy is #10:
169 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
170 *
171 * 2) Any buddy B will have an order O+1 parent P which
172 * satisfies the following equation:
173 * P = B & ~(1 << O)
174 *
175 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
176 */
177static inline unsigned long
178__find_buddy_pfn(unsigned long page_pfn, unsigned int order)
179{
180 return page_pfn ^ (1 << order);
181}
182
183extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
184 unsigned long end_pfn, struct zone *zone);
185
186static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
187 unsigned long end_pfn, struct zone *zone)
188{
189 if (zone->contiguous)
190 return pfn_to_page(start_pfn);
191
192 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
193}
194
195extern int __isolate_free_page(struct page *page, unsigned int order);
196extern void __putback_isolated_page(struct page *page, unsigned int order,
197 int mt);
198extern void memblock_free_pages(struct page *page, unsigned long pfn,
199 unsigned int order);
200extern void __free_pages_core(struct page *page, unsigned int order);
201extern void prep_compound_page(struct page *page, unsigned int order);
202extern void post_alloc_hook(struct page *page, unsigned int order,
203 gfp_t gfp_flags);
204extern int user_min_free_kbytes;
205
206extern void free_unref_page(struct page *page, unsigned int order);
207extern void free_unref_page_list(struct list_head *list);
208
209extern void zone_pcp_update(struct zone *zone, int cpu_online);
210extern void zone_pcp_reset(struct zone *zone);
211extern void zone_pcp_disable(struct zone *zone);
212extern void zone_pcp_enable(struct zone *zone);
213
214#if defined CONFIG_COMPACTION || defined CONFIG_CMA
215
216/*
217 * in mm/compaction.c
218 */
219/*
220 * compact_control is used to track pages being migrated and the free pages
221 * they are being migrated to during memory compaction. The free_pfn starts
222 * at the end of a zone and migrate_pfn begins at the start. Movable pages
223 * are moved to the end of a zone during a compaction run and the run
224 * completes when free_pfn <= migrate_pfn
225 */
226struct compact_control {
227 struct list_head freepages; /* List of free pages to migrate to */
228 struct list_head migratepages; /* List of pages being migrated */
229 unsigned int nr_freepages; /* Number of isolated free pages */
230 unsigned int nr_migratepages; /* Number of pages to migrate */
231 unsigned long free_pfn; /* isolate_freepages search base */
232 /*
233 * Acts as an in/out parameter to page isolation for migration.
234 * isolate_migratepages uses it as a search base.
235 * isolate_migratepages_block will update the value to the next pfn
236 * after the last isolated one.
237 */
238 unsigned long migrate_pfn;
239 unsigned long fast_start_pfn; /* a pfn to start linear scan from */
240 struct zone *zone;
241 unsigned long total_migrate_scanned;
242 unsigned long total_free_scanned;
243 unsigned short fast_search_fail;/* failures to use free list searches */
244 short search_order; /* order to start a fast search at */
245 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
246 int order; /* order a direct compactor needs */
247 int migratetype; /* migratetype of direct compactor */
248 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
249 const int highest_zoneidx; /* zone index of a direct compactor */
250 enum migrate_mode mode; /* Async or sync migration mode */
251 bool ignore_skip_hint; /* Scan blocks even if marked skip */
252 bool no_set_skip_hint; /* Don't mark blocks for skipping */
253 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
254 bool direct_compaction; /* False from kcompactd or /proc/... */
255 bool proactive_compaction; /* kcompactd proactive compaction */
256 bool whole_zone; /* Whole zone should/has been scanned */
257 bool contended; /* Signal lock or sched contention */
258 bool rescan; /* Rescanning the same pageblock */
259 bool alloc_contig; /* alloc_contig_range allocation */
260};
261
262/*
263 * Used in direct compaction when a page should be taken from the freelists
264 * immediately when one is created during the free path.
265 */
266struct capture_control {
267 struct compact_control *cc;
268 struct page *page;
269};
270
271unsigned long
272isolate_freepages_range(struct compact_control *cc,
273 unsigned long start_pfn, unsigned long end_pfn);
274int
275isolate_migratepages_range(struct compact_control *cc,
276 unsigned long low_pfn, unsigned long end_pfn);
277#endif
278int find_suitable_fallback(struct free_area *area, unsigned int order,
279 int migratetype, bool only_stealable, bool *can_steal);
280
281/*
282 * This function returns the order of a free page in the buddy system. In
283 * general, page_zone(page)->lock must be held by the caller to prevent the
284 * page from being allocated in parallel and returning garbage as the order.
285 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
286 * page cannot be allocated or merged in parallel. Alternatively, it must
287 * handle invalid values gracefully, and use buddy_order_unsafe() below.
288 */
289static inline unsigned int buddy_order(struct page *page)
290{
291 /* PageBuddy() must be checked by the caller */
292 return page_private(page);
293}
294
295/*
296 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
297 * PageBuddy() should be checked first by the caller to minimize race window,
298 * and invalid values must be handled gracefully.
299 *
300 * READ_ONCE is used so that if the caller assigns the result into a local
301 * variable and e.g. tests it for valid range before using, the compiler cannot
302 * decide to remove the variable and inline the page_private(page) multiple
303 * times, potentially observing different values in the tests and the actual
304 * use of the result.
305 */
306#define buddy_order_unsafe(page) READ_ONCE(page_private(page))
307
308/*
309 * These three helpers classifies VMAs for virtual memory accounting.
310 */
311
312/*
313 * Executable code area - executable, not writable, not stack
314 */
315static inline bool is_exec_mapping(vm_flags_t flags)
316{
317 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
318}
319
320/*
321 * Stack area - automatically grows in one direction
322 *
323 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
324 * do_mmap() forbids all other combinations.
325 */
326static inline bool is_stack_mapping(vm_flags_t flags)
327{
328 return (flags & VM_STACK) == VM_STACK;
329}
330
331/*
332 * Data area - private, writable, not stack
333 */
334static inline bool is_data_mapping(vm_flags_t flags)
335{
336 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
337}
338
339/* mm/util.c */
340void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
341 struct vm_area_struct *prev);
342void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma);
343
344#ifdef CONFIG_MMU
345extern long populate_vma_page_range(struct vm_area_struct *vma,
346 unsigned long start, unsigned long end, int *locked);
347extern long faultin_vma_page_range(struct vm_area_struct *vma,
348 unsigned long start, unsigned long end,
349 bool write, int *locked);
350extern void munlock_vma_pages_range(struct vm_area_struct *vma,
351 unsigned long start, unsigned long end);
352static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
353{
354 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
355}
356
357/*
358 * must be called with vma's mmap_lock held for read or write, and page locked.
359 */
360extern void mlock_vma_page(struct page *page);
361extern unsigned int munlock_vma_page(struct page *page);
362
363extern int mlock_future_check(struct mm_struct *mm, unsigned long flags,
364 unsigned long len);
365
366/*
367 * Clear the page's PageMlocked(). This can be useful in a situation where
368 * we want to unconditionally remove a page from the pagecache -- e.g.,
369 * on truncation or freeing.
370 *
371 * It is legal to call this function for any page, mlocked or not.
372 * If called for a page that is still mapped by mlocked vmas, all we do
373 * is revert to lazy LRU behaviour -- semantics are not broken.
374 */
375extern void clear_page_mlock(struct page *page);
376
377extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
378
379/*
380 * At what user virtual address is page expected in vma?
381 * Returns -EFAULT if all of the page is outside the range of vma.
382 * If page is a compound head, the entire compound page is considered.
383 */
384static inline unsigned long
385vma_address(struct page *page, struct vm_area_struct *vma)
386{
387 pgoff_t pgoff;
388 unsigned long address;
389
390 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
391 pgoff = page_to_pgoff(page);
392 if (pgoff >= vma->vm_pgoff) {
393 address = vma->vm_start +
394 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
395 /* Check for address beyond vma (or wrapped through 0?) */
396 if (address < vma->vm_start || address >= vma->vm_end)
397 address = -EFAULT;
398 } else if (PageHead(page) &&
399 pgoff + compound_nr(page) - 1 >= vma->vm_pgoff) {
400 /* Test above avoids possibility of wrap to 0 on 32-bit */
401 address = vma->vm_start;
402 } else {
403 address = -EFAULT;
404 }
405 return address;
406}
407
408/*
409 * Then at what user virtual address will none of the page be found in vma?
410 * Assumes that vma_address() already returned a good starting address.
411 * If page is a compound head, the entire compound page is considered.
412 */
413static inline unsigned long
414vma_address_end(struct page *page, struct vm_area_struct *vma)
415{
416 pgoff_t pgoff;
417 unsigned long address;
418
419 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */
420 pgoff = page_to_pgoff(page) + compound_nr(page);
421 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
422 /* Check for address beyond vma (or wrapped through 0?) */
423 if (address < vma->vm_start || address > vma->vm_end)
424 address = vma->vm_end;
425 return address;
426}
427
428static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
429 struct file *fpin)
430{
431 int flags = vmf->flags;
432
433 if (fpin)
434 return fpin;
435
436 /*
437 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
438 * anything, so we only pin the file and drop the mmap_lock if only
439 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
440 */
441 if (fault_flag_allow_retry_first(flags) &&
442 !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
443 fpin = get_file(vmf->vma->vm_file);
444 mmap_read_unlock(vmf->vma->vm_mm);
445 }
446 return fpin;
447}
448
449#else /* !CONFIG_MMU */
450static inline void clear_page_mlock(struct page *page) { }
451static inline void mlock_vma_page(struct page *page) { }
452static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
453{
454}
455#endif /* !CONFIG_MMU */
456
457/*
458 * Return the mem_map entry representing the 'offset' subpage within
459 * the maximally aligned gigantic page 'base'. Handle any discontiguity
460 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
461 */
462static inline struct page *mem_map_offset(struct page *base, int offset)
463{
464 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
465 return nth_page(base, offset);
466 return base + offset;
467}
468
469/*
470 * Iterator over all subpages within the maximally aligned gigantic
471 * page 'base'. Handle any discontiguity in the mem_map.
472 */
473static inline struct page *mem_map_next(struct page *iter,
474 struct page *base, int offset)
475{
476 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
477 unsigned long pfn = page_to_pfn(base) + offset;
478 if (!pfn_valid(pfn))
479 return NULL;
480 return pfn_to_page(pfn);
481 }
482 return iter + 1;
483}
484
485/* Memory initialisation debug and verification */
486enum mminit_level {
487 MMINIT_WARNING,
488 MMINIT_VERIFY,
489 MMINIT_TRACE
490};
491
492#ifdef CONFIG_DEBUG_MEMORY_INIT
493
494extern int mminit_loglevel;
495
496#define mminit_dprintk(level, prefix, fmt, arg...) \
497do { \
498 if (level < mminit_loglevel) { \
499 if (level <= MMINIT_WARNING) \
500 pr_warn("mminit::" prefix " " fmt, ##arg); \
501 else \
502 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
503 } \
504} while (0)
505
506extern void mminit_verify_pageflags_layout(void);
507extern void mminit_verify_zonelist(void);
508#else
509
510static inline void mminit_dprintk(enum mminit_level level,
511 const char *prefix, const char *fmt, ...)
512{
513}
514
515static inline void mminit_verify_pageflags_layout(void)
516{
517}
518
519static inline void mminit_verify_zonelist(void)
520{
521}
522#endif /* CONFIG_DEBUG_MEMORY_INIT */
523
524/* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
525#if defined(CONFIG_SPARSEMEM)
526extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
527 unsigned long *end_pfn);
528#else
529static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
530 unsigned long *end_pfn)
531{
532}
533#endif /* CONFIG_SPARSEMEM */
534
535#define NODE_RECLAIM_NOSCAN -2
536#define NODE_RECLAIM_FULL -1
537#define NODE_RECLAIM_SOME 0
538#define NODE_RECLAIM_SUCCESS 1
539
540#ifdef CONFIG_NUMA
541extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
542#else
543static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
544 unsigned int order)
545{
546 return NODE_RECLAIM_NOSCAN;
547}
548#endif
549
550extern int hwpoison_filter(struct page *p);
551
552extern u32 hwpoison_filter_dev_major;
553extern u32 hwpoison_filter_dev_minor;
554extern u64 hwpoison_filter_flags_mask;
555extern u64 hwpoison_filter_flags_value;
556extern u64 hwpoison_filter_memcg;
557extern u32 hwpoison_filter_enable;
558
559extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
560 unsigned long, unsigned long,
561 unsigned long, unsigned long);
562
563extern void set_pageblock_order(void);
564unsigned int reclaim_clean_pages_from_list(struct zone *zone,
565 struct list_head *page_list);
566/* The ALLOC_WMARK bits are used as an index to zone->watermark */
567#define ALLOC_WMARK_MIN WMARK_MIN
568#define ALLOC_WMARK_LOW WMARK_LOW
569#define ALLOC_WMARK_HIGH WMARK_HIGH
570#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
571
572/* Mask to get the watermark bits */
573#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
574
575/*
576 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
577 * cannot assume a reduced access to memory reserves is sufficient for
578 * !MMU
579 */
580#ifdef CONFIG_MMU
581#define ALLOC_OOM 0x08
582#else
583#define ALLOC_OOM ALLOC_NO_WATERMARKS
584#endif
585
586#define ALLOC_HARDER 0x10 /* try to alloc harder */
587#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
588#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
589#define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
590#ifdef CONFIG_ZONE_DMA32
591#define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */
592#else
593#define ALLOC_NOFRAGMENT 0x0
594#endif
595#define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
596
597enum ttu_flags;
598struct tlbflush_unmap_batch;
599
600
601/*
602 * only for MM internal work items which do not depend on
603 * any allocations or locks which might depend on allocations
604 */
605extern struct workqueue_struct *mm_percpu_wq;
606
607#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
608void try_to_unmap_flush(void);
609void try_to_unmap_flush_dirty(void);
610void flush_tlb_batched_pending(struct mm_struct *mm);
611#else
612static inline void try_to_unmap_flush(void)
613{
614}
615static inline void try_to_unmap_flush_dirty(void)
616{
617}
618static inline void flush_tlb_batched_pending(struct mm_struct *mm)
619{
620}
621#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
622
623extern const struct trace_print_flags pageflag_names[];
624extern const struct trace_print_flags vmaflag_names[];
625extern const struct trace_print_flags gfpflag_names[];
626
627static inline bool is_migrate_highatomic(enum migratetype migratetype)
628{
629 return migratetype == MIGRATE_HIGHATOMIC;
630}
631
632static inline bool is_migrate_highatomic_page(struct page *page)
633{
634 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
635}
636
637void setup_zone_pageset(struct zone *zone);
638
639struct migration_target_control {
640 int nid; /* preferred node id */
641 nodemask_t *nmask;
642 gfp_t gfp_mask;
643};
644
645/*
646 * mm/vmalloc.c
647 */
648#ifdef CONFIG_MMU
649int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
650 pgprot_t prot, struct page **pages, unsigned int page_shift);
651#else
652static inline
653int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
654 pgprot_t prot, struct page **pages, unsigned int page_shift)
655{
656 return -EINVAL;
657}
658#endif
659
660void vunmap_range_noflush(unsigned long start, unsigned long end);
661
662int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
663 unsigned long addr, int page_nid, int *flags);
664
665#endif /* __MM_INTERNAL_H */