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