mm/internal.h

   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_REPEAT|__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
  39 void page_writeback_init(void);
  40
  41 int do_swap_page(struct vm_fault *vmf);
  42
  43 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
  44                 unsigned long floor, unsigned long ceiling);
  45
  46 static 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
  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);
  55
  56 extern 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  */
  63 static 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  */
  74 static 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
  81 extern unsigned long highest_memmap_pfn;
  82
  83 /*
  84  * in mm/vmscan.c:
  85  */
  86 extern int isolate_lru_page(struct page *page);
  87 extern void putback_lru_page(struct page *page);
  88 extern bool pgdat_reclaimable(struct pglist_data *pgdat);
  89
  90 /*
  91  * in mm/rmap.c:
  92  */
  93 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
  94
  95 /*
  96  * in mm/page_alloc.c
  97  */
  98
  99 /*
 100  * Structure for holding the mostly immutable allocation parameters passed
 101  * between functions involved in allocations, including the alloc_pages*
 102  * family of functions.
 103  *
 104  * nodemask, migratetype and high_zoneidx are initialized only once in
 105  * __alloc_pages_nodemask() and then never change.
 106  *
 107  * zonelist, preferred_zone and classzone_idx are set first in
 108  * __alloc_pages_nodemask() for the fast path, and might be later changed
 109  * in __alloc_pages_slowpath(). All other functions pass the whole strucure
 110  * by a const pointer.
 111  */
 112 struct alloc_context {
 113         struct zonelist *zonelist;
 114         nodemask_t *nodemask;
 115         struct zoneref *preferred_zoneref;
 116         int migratetype;
 117         enum zone_type high_zoneidx;
 118         bool spread_dirty_pages;
 119 };
 120
 121 #define ac_classzone_idx(ac) zonelist_zone_idx(ac->preferred_zoneref)
 122
 123 /*
 124  * Locate the struct page for both the matching buddy in our
 125  * pair (buddy1) and the combined O(n+1) page they form (page).
 126  *
 127  * 1) Any buddy B1 will have an order O twin B2 which satisfies
 128  * the following equation:
 129  *     B2 = B1 ^ (1 << O)
 130  * For example, if the starting buddy (buddy2) is #8 its order
 131  * 1 buddy is #10:
 132  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
 133  *
 134  * 2) Any buddy B will have an order O+1 parent P which
 135  * satisfies the following equation:
 136  *     P = B & ~(1 << O)
 137  *
 138  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
 139  */
 140 static inline unsigned long
 141 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
 142 {
 143         return page_pfn ^ (1 << order);
 144 }
 145
 146 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
 147                                 unsigned long end_pfn, struct zone *zone);
 148
 149 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
 150                                 unsigned long end_pfn, struct zone *zone)
 151 {
 152         if (zone->contiguous)
 153                 return pfn_to_page(start_pfn);
 154
 155         return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
 156 }
 157
 158 extern int __isolate_free_page(struct page *page, unsigned int order);
 159 extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
 160                                         unsigned int order);
 161 extern void prep_compound_page(struct page *page, unsigned int order);
 162 extern void post_alloc_hook(struct page *page, unsigned int order,
 163                                         gfp_t gfp_flags);
 164 extern int user_min_free_kbytes;
 165
 166 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
 167
 168 /*
 169  * in mm/compaction.c
 170  */
 171 /*
 172  * compact_control is used to track pages being migrated and the free pages
 173  * they are being migrated to during memory compaction. The free_pfn starts
 174  * at the end of a zone and migrate_pfn begins at the start. Movable pages
 175  * are moved to the end of a zone during a compaction run and the run
 176  * completes when free_pfn <= migrate_pfn
 177  */
 178 struct compact_control {
 179         struct list_head freepages;     /* List of free pages to migrate to */
 180         struct list_head migratepages;  /* List of pages being migrated */
 181         unsigned long nr_freepages;     /* Number of isolated free pages */
 182         unsigned long nr_migratepages;  /* Number of pages to migrate */
 183         unsigned long total_migrate_scanned;
 184         unsigned long total_free_scanned;
 185         unsigned long free_pfn;         /* isolate_freepages search base */
 186         unsigned long migrate_pfn;      /* isolate_migratepages search base */
 187         unsigned long last_migrated_pfn;/* Not yet flushed page being freed */
 188         enum migrate_mode mode;         /* Async or sync migration mode */
 189         bool ignore_skip_hint;          /* Scan blocks even if marked skip */
 190         bool ignore_block_suitable;     /* Scan blocks considered unsuitable */
 191         bool direct_compaction;         /* False from kcompactd or /proc/... */
 192         bool whole_zone;                /* Whole zone should/has been scanned */
 193         int order;                      /* order a direct compactor needs */
 194         const gfp_t gfp_mask;           /* gfp mask of a direct compactor */
 195         const unsigned int alloc_flags; /* alloc flags of a direct compactor */
 196         const int classzone_idx;        /* zone index of a direct compactor */
 197         struct zone *zone;
 198         bool contended;                 /* Signal lock or sched contention */
 199 };
 200
 201 unsigned long
 202 isolate_freepages_range(struct compact_control *cc,
 203                         unsigned long start_pfn, unsigned long end_pfn);
 204 unsigned long
 205 isolate_migratepages_range(struct compact_control *cc,
 206                            unsigned long low_pfn, unsigned long end_pfn);
 207 int find_suitable_fallback(struct free_area *area, unsigned int order,
 208                         int migratetype, bool only_stealable, bool *can_steal);
 209
 210 #endif
 211
 212 /*
 213  * This function returns the order of a free page in the buddy system. In
 214  * general, page_zone(page)->lock must be held by the caller to prevent the
 215  * page from being allocated in parallel and returning garbage as the order.
 216  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
 217  * page cannot be allocated or merged in parallel. Alternatively, it must
 218  * handle invalid values gracefully, and use page_order_unsafe() below.
 219  */
 220 static inline unsigned int page_order(struct page *page)
 221 {
 222         /* PageBuddy() must be checked by the caller */
 223         return page_private(page);
 224 }
 225
 226 /*
 227  * Like page_order(), but for callers who cannot afford to hold the zone lock.
 228  * PageBuddy() should be checked first by the caller to minimize race window,
 229  * and invalid values must be handled gracefully.
 230  *
 231  * READ_ONCE is used so that if the caller assigns the result into a local
 232  * variable and e.g. tests it for valid range before using, the compiler cannot
 233  * decide to remove the variable and inline the page_private(page) multiple
 234  * times, potentially observing different values in the tests and the actual
 235  * use of the result.
 236  */
 237 #define page_order_unsafe(page)         READ_ONCE(page_private(page))
 238
 239 static inline bool is_cow_mapping(vm_flags_t flags)
 240 {
 241         return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
 242 }
 243
 244 /*
 245  * These three helpers classifies VMAs for virtual memory accounting.
 246  */
 247
 248 /*
 249  * Executable code area - executable, not writable, not stack
 250  */
 251 static inline bool is_exec_mapping(vm_flags_t flags)
 252 {
 253         return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
 254 }
 255
 256 /*
 257  * Stack area - atomatically grows in one direction
 258  *
 259  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
 260  * do_mmap() forbids all other combinations.
 261  */
 262 static inline bool is_stack_mapping(vm_flags_t flags)
 263 {
 264         return (flags & VM_STACK) == VM_STACK;
 265 }
 266
 267 /*
 268  * Data area - private, writable, not stack
 269  */
 270 static inline bool is_data_mapping(vm_flags_t flags)
 271 {
 272         return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
 273 }
 274
 275 /* mm/util.c */
 276 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
 277                 struct vm_area_struct *prev, struct rb_node *rb_parent);
 278
 279 #ifdef CONFIG_MMU
 280 extern long populate_vma_page_range(struct vm_area_struct *vma,
 281                 unsigned long start, unsigned long end, int *nonblocking);
 282 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
 283                         unsigned long start, unsigned long end);
 284 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
 285 {
 286         munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
 287 }
 288
 289 /*
 290  * must be called with vma's mmap_sem held for read or write, and page locked.
 291  */
 292 extern void mlock_vma_page(struct page *page);
 293 extern unsigned int munlock_vma_page(struct page *page);
 294
 295 /*
 296  * Clear the page's PageMlocked().  This can be useful in a situation where
 297  * we want to unconditionally remove a page from the pagecache -- e.g.,
 298  * on truncation or freeing.
 299  *
 300  * It is legal to call this function for any page, mlocked or not.
 301  * If called for a page that is still mapped by mlocked vmas, all we do
 302  * is revert to lazy LRU behaviour -- semantics are not broken.
 303  */
 304 extern void clear_page_mlock(struct page *page);
 305
 306 /*
 307  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
 308  * (because that does not go through the full procedure of migration ptes):
 309  * to migrate the Mlocked page flag; update statistics.
 310  */
 311 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
 312 {
 313         if (TestClearPageMlocked(page)) {
 314                 int nr_pages = hpage_nr_pages(page);
 315
 316                 /* Holding pmd lock, no change in irq context: __mod is safe */
 317                 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 318                 SetPageMlocked(newpage);
 319                 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
 320         }
 321 }
 322
 323 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
 324
 325 /*
 326  * At what user virtual address is page expected in @vma?
 327  */
 328 static inline unsigned long
 329 __vma_address(struct page *page, struct vm_area_struct *vma)
 330 {
 331         pgoff_t pgoff = page_to_pgoff(page);
 332         return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 333 }
 334
 335 static inline unsigned long
 336 vma_address(struct page *page, struct vm_area_struct *vma)
 337 {
 338         unsigned long start, end;
 339
 340         start = __vma_address(page, vma);
 341         end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
 342
 343         /* page should be within @vma mapping range */
 344         VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma);
 345
 346         return max(start, vma->vm_start);
 347 }
 348
 349 #else /* !CONFIG_MMU */
 350 static inline void clear_page_mlock(struct page *page) { }
 351 static inline void mlock_vma_page(struct page *page) { }
 352 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
 353
 354 #endif /* !CONFIG_MMU */
 355
 356 /*
 357  * Return the mem_map entry representing the 'offset' subpage within
 358  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
 359  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
 360  */
 361 static inline struct page *mem_map_offset(struct page *base, int offset)
 362 {
 363         if (unlikely(offset >= MAX_ORDER_NR_PAGES))
 364                 return nth_page(base, offset);
 365         return base + offset;
 366 }
 367
 368 /*
 369  * Iterator over all subpages within the maximally aligned gigantic
 370  * page 'base'.  Handle any discontiguity in the mem_map.
 371  */
 372 static inline struct page *mem_map_next(struct page *iter,
 373                                                 struct page *base, int offset)
 374 {
 375         if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
 376                 unsigned long pfn = page_to_pfn(base) + offset;
 377                 if (!pfn_valid(pfn))
 378                         return NULL;
 379                 return pfn_to_page(pfn);
 380         }
 381         return iter + 1;
 382 }
 383
 384 /*
 385  * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
 386  * so all functions starting at paging_init should be marked __init
 387  * in those cases. SPARSEMEM, however, allows for memory hotplug,
 388  * and alloc_bootmem_node is not used.
 389  */
 390 #ifdef CONFIG_SPARSEMEM
 391 #define __paginginit __meminit
 392 #else
 393 #define __paginginit __init
 394 #endif
 395
 396 /* Memory initialisation debug and verification */
 397 enum mminit_level {
 398         MMINIT_WARNING,
 399         MMINIT_VERIFY,
 400         MMINIT_TRACE
 401 };
 402
 403 #ifdef CONFIG_DEBUG_MEMORY_INIT
 404
 405 extern int mminit_loglevel;
 406
 407 #define mminit_dprintk(level, prefix, fmt, arg...) \
 408 do { \
 409         if (level < mminit_loglevel) { \
 410                 if (level <= MMINIT_WARNING) \
 411                         pr_warn("mminit::" prefix " " fmt, ##arg);      \
 412                 else \
 413                         printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
 414         } \
 415 } while (0)
 416
 417 extern void mminit_verify_pageflags_layout(void);
 418 extern void mminit_verify_zonelist(void);
 419 #else
 420
 421 static inline void mminit_dprintk(enum mminit_level level,
 422                                 const char *prefix, const char *fmt, ...)
 423 {
 424 }
 425
 426 static inline void mminit_verify_pageflags_layout(void)
 427 {
 428 }
 429
 430 static inline void mminit_verify_zonelist(void)
 431 {
 432 }
 433 #endif /* CONFIG_DEBUG_MEMORY_INIT */
 434
 435 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
 436 #if defined(CONFIG_SPARSEMEM)
 437 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 438                                 unsigned long *end_pfn);
 439 #else
 440 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 441                                 unsigned long *end_pfn)
 442 {
 443 }
 444 #endif /* CONFIG_SPARSEMEM */
 445
 446 #define NODE_RECLAIM_NOSCAN     -2
 447 #define NODE_RECLAIM_FULL       -1
 448 #define NODE_RECLAIM_SOME       0
 449 #define NODE_RECLAIM_SUCCESS    1
 450
 451 extern int hwpoison_filter(struct page *p);
 452
 453 extern u32 hwpoison_filter_dev_major;
 454 extern u32 hwpoison_filter_dev_minor;
 455 extern u64 hwpoison_filter_flags_mask;
 456 extern u64 hwpoison_filter_flags_value;
 457 extern u64 hwpoison_filter_memcg;
 458 extern u32 hwpoison_filter_enable;
 459
 460 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
 461         unsigned long, unsigned long,
 462         unsigned long, unsigned long);
 463
 464 extern void set_pageblock_order(void);
 465 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
 466                                             struct list_head *page_list);
 467 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
 468 #define ALLOC_WMARK_MIN         WMARK_MIN
 469 #define ALLOC_WMARK_LOW         WMARK_LOW
 470 #define ALLOC_WMARK_HIGH        WMARK_HIGH
 471 #define ALLOC_NO_WATERMARKS     0x04 /* don't check watermarks at all */
 472
 473 /* Mask to get the watermark bits */
 474 #define ALLOC_WMARK_MASK        (ALLOC_NO_WATERMARKS-1)
 475
 476 #define ALLOC_HARDER            0x10 /* try to alloc harder */
 477 #define ALLOC_HIGH              0x20 /* __GFP_HIGH set */
 478 #define ALLOC_CPUSET            0x40 /* check for correct cpuset */
 479 #define ALLOC_CMA               0x80 /* allow allocations from CMA areas */
 480
 481 enum ttu_flags;
 482 struct tlbflush_unmap_batch;
 483
 484 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
 485 void try_to_unmap_flush(void);
 486 void try_to_unmap_flush_dirty(void);
 487 #else
 488 static inline void try_to_unmap_flush(void)
 489 {
 490 }
 491 static inline void try_to_unmap_flush_dirty(void)
 492 {
 493 }
 494
 495 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
 496
 497 extern const struct trace_print_flags pageflag_names[];
 498 extern const struct trace_print_flags vmaflag_names[];
 499 extern const struct trace_print_flags gfpflag_names[];
 500
 501 #endif  /* __MM_INTERNAL_H */