1 /* internal.h: mm/ internal definitions
3 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
11 #ifndef __MM_INTERNAL_H
12 #define __MM_INTERNAL_H
16 #include <linux/pagemap.h>
17 #include <linux/tracepoint-defs.h>
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.
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|\
30 /* The GFP flags allowed during early boot */
31 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
33 /* Control allocation cpuset and node placement constraints */
34 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
36 /* Do not use these with a slab allocator */
37 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
39 void page_writeback_init(void);
41 vm_fault_t do_swap_page(struct vm_fault *vmf);
43 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
44 unsigned long floor, unsigned long ceiling);
46 static inline bool can_madv_dontneed_vma(struct vm_area_struct *vma)
48 return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP));
51 void 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);
56 extern unsigned 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);
61 * Submit IO for the read-ahead request in file_ra_state.
63 static inline unsigned long ra_submit(struct file_ra_state *ra,
64 struct address_space *mapping, struct file *filp)
66 return __do_page_cache_readahead(mapping, filp,
67 ra->start, ra->size, ra->async_size);
71 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
74 static inline void set_page_refcounted(struct page *page)
76 VM_BUG_ON_PAGE(PageTail(page), page);
77 VM_BUG_ON_PAGE(page_ref_count(page), page);
78 set_page_count(page, 1);
81 extern unsigned long highest_memmap_pfn;
84 * Maximum number of reclaim retries without progress before the OOM
85 * killer is consider the only way forward.
87 #define MAX_RECLAIM_RETRIES 16
92 extern int isolate_lru_page(struct page *page);
93 extern void putback_lru_page(struct page *page);
98 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
105 * Structure for holding the mostly immutable allocation parameters passed
106 * between functions involved in allocations, including the alloc_pages*
107 * family of functions.
109 * nodemask, migratetype and high_zoneidx are initialized only once in
110 * __alloc_pages_nodemask() and then never change.
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.
117 struct alloc_context {
118 struct zonelist *zonelist;
119 nodemask_t *nodemask;
120 struct zoneref *preferred_zoneref;
122 enum zone_type high_zoneidx;
123 bool spread_dirty_pages;
126 #define ac_classzone_idx(ac) zonelist_zone_idx(ac->preferred_zoneref)
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).
132 * 1) Any buddy B1 will have an order O twin B2 which satisfies
133 * the following equation:
135 * For example, if the starting buddy (buddy2) is #8 its order
137 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
139 * 2) Any buddy B will have an order O+1 parent P which
140 * satisfies the following equation:
143 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
145 static inline unsigned long
146 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
148 return page_pfn ^ (1 << order);
151 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
152 unsigned long end_pfn, struct zone *zone);
154 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
155 unsigned long end_pfn, struct zone *zone)
157 if (zone->contiguous)
158 return pfn_to_page(start_pfn);
160 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
163 extern int __isolate_free_page(struct page *page, unsigned int order);
164 extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
166 extern void prep_compound_page(struct page *page, unsigned int order);
167 extern void post_alloc_hook(struct page *page, unsigned int order,
169 extern int user_min_free_kbytes;
171 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
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
183 struct compact_control {
184 struct list_head freepages; /* List of free pages to migrate to */
185 struct list_head migratepages; /* List of pages being migrated */
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 */
210 isolate_freepages_range(struct compact_control *cc,
211 unsigned long start_pfn, unsigned long end_pfn);
213 isolate_migratepages_range(struct compact_control *cc,
214 unsigned long low_pfn, unsigned long end_pfn);
215 int find_suitable_fallback(struct free_area *area, unsigned int order,
216 int migratetype, bool only_stealable, bool *can_steal);
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.
228 static inline unsigned int page_order(struct page *page)
230 /* PageBuddy() must be checked by the caller */
231 return page_private(page);
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.
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
245 #define page_order_unsafe(page) READ_ONCE(page_private(page))
247 static inline bool is_cow_mapping(vm_flags_t flags)
249 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
253 * These three helpers classifies VMAs for virtual memory accounting.
257 * Executable code area - executable, not writable, not stack
259 static inline bool is_exec_mapping(vm_flags_t flags)
261 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
265 * Stack area - atomatically grows in one direction
267 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
268 * do_mmap() forbids all other combinations.
270 static inline bool is_stack_mapping(vm_flags_t flags)
272 return (flags & VM_STACK) == VM_STACK;
276 * Data area - private, writable, not stack
278 static inline bool is_data_mapping(vm_flags_t flags)
280 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
284 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
285 struct vm_area_struct *prev, struct rb_node *rb_parent);
288 extern long populate_vma_page_range(struct vm_area_struct *vma,
289 unsigned long start, unsigned long end, int *nonblocking);
290 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
291 unsigned long start, unsigned long end);
292 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
294 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
298 * must be called with vma's mmap_sem held for read or write, and page locked.
300 extern void mlock_vma_page(struct page *page);
301 extern unsigned int munlock_vma_page(struct page *page);
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.
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.
312 extern void clear_page_mlock(struct page *page);
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.
319 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
321 if (TestClearPageMlocked(page)) {
322 int nr_pages = hpage_nr_pages(page);
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);
331 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
334 * At what user virtual address is page expected in @vma?
336 static inline unsigned long
337 __vma_address(struct page *page, struct vm_area_struct *vma)
339 pgoff_t pgoff = page_to_pgoff(page);
340 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
343 static inline unsigned long
344 vma_address(struct page *page, struct vm_area_struct *vma)
346 unsigned long start, end;
348 start = __vma_address(page, vma);
349 end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
351 /* page should be within @vma mapping range */
352 VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma);
354 return max(start, vma->vm_start);
357 #else /* !CONFIG_MMU */
358 static inline void clear_page_mlock(struct page *page) { }
359 static inline void mlock_vma_page(struct page *page) { }
360 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
362 #endif /* !CONFIG_MMU */
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.
369 static inline struct page *mem_map_offset(struct page *base, int offset)
371 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
372 return nth_page(base, offset);
373 return base + offset;
377 * Iterator over all subpages within the maximally aligned gigantic
378 * page 'base'. Handle any discontiguity in the mem_map.
380 static inline struct page *mem_map_next(struct page *iter,
381 struct page *base, int offset)
383 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
384 unsigned long pfn = page_to_pfn(base) + offset;
387 return pfn_to_page(pfn);
392 /* Memory initialisation debug and verification */
399 #ifdef CONFIG_DEBUG_MEMORY_INIT
401 extern int mminit_loglevel;
403 #define mminit_dprintk(level, prefix, fmt, arg...) \
405 if (level < mminit_loglevel) { \
406 if (level <= MMINIT_WARNING) \
407 pr_warn("mminit::" prefix " " fmt, ##arg); \
409 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
413 extern void mminit_verify_pageflags_layout(void);
414 extern void mminit_verify_zonelist(void);
417 static inline void mminit_dprintk(enum mminit_level level,
418 const char *prefix, const char *fmt, ...)
422 static inline void mminit_verify_pageflags_layout(void)
426 static inline void mminit_verify_zonelist(void)
429 #endif /* CONFIG_DEBUG_MEMORY_INIT */
431 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
432 #if defined(CONFIG_SPARSEMEM)
433 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
434 unsigned long *end_pfn);
436 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
437 unsigned long *end_pfn)
440 #endif /* CONFIG_SPARSEMEM */
442 #define NODE_RECLAIM_NOSCAN -2
443 #define NODE_RECLAIM_FULL -1
444 #define NODE_RECLAIM_SOME 0
445 #define NODE_RECLAIM_SUCCESS 1
447 extern int hwpoison_filter(struct page *p);
449 extern u32 hwpoison_filter_dev_major;
450 extern u32 hwpoison_filter_dev_minor;
451 extern u64 hwpoison_filter_flags_mask;
452 extern u64 hwpoison_filter_flags_value;
453 extern u64 hwpoison_filter_memcg;
454 extern u32 hwpoison_filter_enable;
456 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long,
457 unsigned long, unsigned long,
458 unsigned long, unsigned long);
460 extern void set_pageblock_order(void);
461 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
462 struct list_head *page_list);
463 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
464 #define ALLOC_WMARK_MIN WMARK_MIN
465 #define ALLOC_WMARK_LOW WMARK_LOW
466 #define ALLOC_WMARK_HIGH WMARK_HIGH
467 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
469 /* Mask to get the watermark bits */
470 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
473 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
474 * cannot assume a reduced access to memory reserves is sufficient for
478 #define ALLOC_OOM 0x08
480 #define ALLOC_OOM ALLOC_NO_WATERMARKS
483 #define ALLOC_HARDER 0x10 /* try to alloc harder */
484 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
485 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
486 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
489 struct tlbflush_unmap_batch;
493 * only for MM internal work items which do not depend on
494 * any allocations or locks which might depend on allocations
496 extern struct workqueue_struct *mm_percpu_wq;
498 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
499 void try_to_unmap_flush(void);
500 void try_to_unmap_flush_dirty(void);
501 void flush_tlb_batched_pending(struct mm_struct *mm);
503 static inline void try_to_unmap_flush(void)
506 static inline void try_to_unmap_flush_dirty(void)
509 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
512 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
514 extern const struct trace_print_flags pageflag_names[];
515 extern const struct trace_print_flags vmaflag_names[];
516 extern const struct trace_print_flags gfpflag_names[];
518 static inline bool is_migrate_highatomic(enum migratetype migratetype)
520 return migratetype == MIGRATE_HIGHATOMIC;
523 static inline bool is_migrate_highatomic_page(struct page *page)
525 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
528 void setup_zone_pageset(struct zone *zone);
529 extern struct page *alloc_new_node_page(struct page *page, unsigned long node);
530 #endif /* __MM_INTERNAL_H */