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 int do_swap_page(struct fault_env *fe, pte_t orig_pte);
  40
  41 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
  42                 unsigned long floor, unsigned long ceiling);
  43
  44 void unmap_page_range(struct mmu_gather *tlb,
  45                              struct vm_area_struct *vma,
  46                              unsigned long addr, unsigned long end,
  47                              struct zap_details *details);
  48
  49 extern int __do_page_cache_readahead(struct address_space *mapping,
  50                 struct file *filp, pgoff_t offset, unsigned long nr_to_read,
  51                 unsigned long lookahead_size);
  52
  53 /*
  54  * Submit IO for the read-ahead request in file_ra_state.
  55  */
  56 static inline unsigned long ra_submit(struct file_ra_state *ra,
  57                 struct address_space *mapping, struct file *filp)
  58 {
  59         return __do_page_cache_readahead(mapping, filp,
  60                                         ra->start, ra->size, ra->async_size);
  61 }
  62
  63 /*
  64  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
  65  * a count of one.
  66  */
  67 static inline void set_page_refcounted(struct page *page)
  68 {
  69         VM_BUG_ON_PAGE(PageTail(page), page);
  70         VM_BUG_ON_PAGE(page_ref_count(page), page);
  71         set_page_count(page, 1);
  72 }
  73
  74 extern unsigned long highest_memmap_pfn;
  75
  76 /*
  77  * in mm/vmscan.c:
  78  */
  79 extern int isolate_lru_page(struct page *page);
  80 extern void putback_lru_page(struct page *page);
  81 extern bool pgdat_reclaimable(struct pglist_data *pgdat);
  82
  83 /*
  84  * in mm/rmap.c:
  85  */
  86 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
  87
  88 /*
  89  * in mm/page_alloc.c
  90  */
  91
  92 /*
  93  * Structure for holding the mostly immutable allocation parameters passed
  94  * between functions involved in allocations, including the alloc_pages*
  95  * family of functions.
  96  *
  97  * nodemask, migratetype and high_zoneidx are initialized only once in
  98  * __alloc_pages_nodemask() and then never change.
  99  *
 100  * zonelist, preferred_zone and classzone_idx are set first in
 101  * __alloc_pages_nodemask() for the fast path, and might be later changed
 102  * in __alloc_pages_slowpath(). All other functions pass the whole strucure
 103  * by a const pointer.
 104  */
 105 struct alloc_context {
 106         struct zonelist *zonelist;
 107         nodemask_t *nodemask;
 108         struct zoneref *preferred_zoneref;
 109         int migratetype;
 110         enum zone_type high_zoneidx;
 111         bool spread_dirty_pages;
 112 };
 113
 114 #define ac_classzone_idx(ac) zonelist_zone_idx(ac->preferred_zoneref)
 115
 116 /*
 117  * Locate the struct page for both the matching buddy in our
 118  * pair (buddy1) and the combined O(n+1) page they form (page).
 119  *
 120  * 1) Any buddy B1 will have an order O twin B2 which satisfies
 121  * the following equation:
 122  *     B2 = B1 ^ (1 << O)
 123  * For example, if the starting buddy (buddy2) is #8 its order
 124  * 1 buddy is #10:
 125  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
 126  *
 127  * 2) Any buddy B will have an order O+1 parent P which
 128  * satisfies the following equation:
 129  *     P = B & ~(1 << O)
 130  *
 131  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
 132  */
 133 static inline unsigned long
 134 __find_buddy_index(unsigned long page_idx, unsigned int order)
 135 {
 136         return page_idx ^ (1 << order);
 137 }
 138
 139 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
 140                                 unsigned long end_pfn, struct zone *zone);
 141
 142 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
 143                                 unsigned long end_pfn, struct zone *zone)
 144 {
 145         if (zone->contiguous)
 146                 return pfn_to_page(start_pfn);
 147
 148         return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
 149 }
 150
 151 extern int __isolate_free_page(struct page *page, unsigned int order);
 152 extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
 153                                         unsigned int order);
 154 extern void prep_compound_page(struct page *page, unsigned int order);
 155 extern void post_alloc_hook(struct page *page, unsigned int order,
 156                                         gfp_t gfp_flags);
 157 extern int user_min_free_kbytes;
 158
 159 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
 160
 161 /*
 162  * in mm/compaction.c
 163  */
 164 /*
 165  * compact_control is used to track pages being migrated and the free pages
 166  * they are being migrated to during memory compaction. The free_pfn starts
 167  * at the end of a zone and migrate_pfn begins at the start. Movable pages
 168  * are moved to the end of a zone during a compaction run and the run
 169  * completes when free_pfn <= migrate_pfn
 170  */
 171 struct compact_control {
 172         struct list_head freepages;     /* List of free pages to migrate to */
 173         struct list_head migratepages;  /* List of pages being migrated */
 174         unsigned long nr_freepages;     /* Number of isolated free pages */
 175         unsigned long nr_migratepages;  /* Number of pages to migrate */
 176         unsigned long free_pfn;         /* isolate_freepages search base */
 177         unsigned long migrate_pfn;      /* isolate_migratepages search base */
 178         unsigned long last_migrated_pfn;/* Not yet flushed page being freed */
 179         enum migrate_mode mode;         /* Async or sync migration mode */
 180         bool ignore_skip_hint;          /* Scan blocks even if marked skip */
 181         bool ignore_block_suitable;     /* Scan blocks considered unsuitable */
 182         bool direct_compaction;         /* False from kcompactd or /proc/... */
 183         bool whole_zone;                /* Whole zone should/has been scanned */
 184         int order;                      /* order a direct compactor needs */
 185         const gfp_t gfp_mask;           /* gfp mask of a direct compactor */
 186         const unsigned int alloc_flags; /* alloc flags of a direct compactor */
 187         const int classzone_idx;        /* zone index of a direct compactor */
 188         struct zone *zone;
 189         bool contended;                 /* Signal lock or sched contention */
 190 };
 191
 192 unsigned long
 193 isolate_freepages_range(struct compact_control *cc,
 194                         unsigned long start_pfn, unsigned long end_pfn);
 195 unsigned long
 196 isolate_migratepages_range(struct compact_control *cc,
 197                            unsigned long low_pfn, unsigned long end_pfn);
 198 int find_suitable_fallback(struct free_area *area, unsigned int order,
 199                         int migratetype, bool only_stealable, bool *can_steal);
 200
 201 #endif
 202
 203 /*
 204  * This function returns the order of a free page in the buddy system. In
 205  * general, page_zone(page)->lock must be held by the caller to prevent the
 206  * page from being allocated in parallel and returning garbage as the order.
 207  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
 208  * page cannot be allocated or merged in parallel. Alternatively, it must
 209  * handle invalid values gracefully, and use page_order_unsafe() below.
 210  */
 211 static inline unsigned int page_order(struct page *page)
 212 {
 213         /* PageBuddy() must be checked by the caller */
 214         return page_private(page);
 215 }
 216
 217 /*
 218  * Like page_order(), but for callers who cannot afford to hold the zone lock.
 219  * PageBuddy() should be checked first by the caller to minimize race window,
 220  * and invalid values must be handled gracefully.
 221  *
 222  * READ_ONCE is used so that if the caller assigns the result into a local
 223  * variable and e.g. tests it for valid range before using, the compiler cannot
 224  * decide to remove the variable and inline the page_private(page) multiple
 225  * times, potentially observing different values in the tests and the actual
 226  * use of the result.
 227  */
 228 #define page_order_unsafe(page)         READ_ONCE(page_private(page))
 229
 230 static inline bool is_cow_mapping(vm_flags_t flags)
 231 {
 232         return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
 233 }
 234
 235 /*
 236  * These three helpers classifies VMAs for virtual memory accounting.
 237  */
 238
 239 /*
 240  * Executable code area - executable, not writable, not stack
 241  */
 242 static inline bool is_exec_mapping(vm_flags_t flags)
 243 {
 244         return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
 245 }
 246
 247 /*
 248  * Stack area - atomatically grows in one direction
 249  *
 250  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
 251  * do_mmap() forbids all other combinations.
 252  */
 253 static inline bool is_stack_mapping(vm_flags_t flags)
 254 {
 255         return (flags & VM_STACK) == VM_STACK;
 256 }
 257
 258 /*
 259  * Data area - private, writable, not stack
 260  */
 261 static inline bool is_data_mapping(vm_flags_t flags)
 262 {
 263         return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
 264 }
 265
 266 /* mm/util.c */
 267 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
 268                 struct vm_area_struct *prev, struct rb_node *rb_parent);
 269
 270 #ifdef CONFIG_MMU
 271 extern long populate_vma_page_range(struct vm_area_struct *vma,
 272                 unsigned long start, unsigned long end, int *nonblocking);
 273 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
 274                         unsigned long start, unsigned long end);
 275 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
 276 {
 277         munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
 278 }
 279
 280 /*
 281  * must be called with vma's mmap_sem held for read or write, and page locked.
 282  */
 283 extern void mlock_vma_page(struct page *page);
 284 extern unsigned int munlock_vma_page(struct page *page);
 285
 286 /*
 287  * Clear the page's PageMlocked().  This can be useful in a situation where
 288  * we want to unconditionally remove a page from the pagecache -- e.g.,
 289  * on truncation or freeing.
 290  *
 291  * It is legal to call this function for any page, mlocked or not.
 292  * If called for a page that is still mapped by mlocked vmas, all we do
 293  * is revert to lazy LRU behaviour -- semantics are not broken.
 294  */
 295 extern void clear_page_mlock(struct page *page);
 296
 297 /*
 298  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
 299  * (because that does not go through the full procedure of migration ptes):
 300  * to migrate the Mlocked page flag; update statistics.
 301  */
 302 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
 303 {
 304         if (TestClearPageMlocked(page)) {
 305                 int nr_pages = hpage_nr_pages(page);
 306
 307                 /* Holding pmd lock, no change in irq context: __mod is safe */
 308                 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 309                 SetPageMlocked(newpage);
 310                 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
 311         }
 312 }
 313
 314 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
 315
 316 /*
 317  * At what user virtual address is page expected in @vma?
 318  */
 319 static inline unsigned long
 320 __vma_address(struct page *page, struct vm_area_struct *vma)
 321 {
 322         pgoff_t pgoff = page_to_pgoff(page);
 323         return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 324 }
 325
 326 static inline unsigned long
 327 vma_address(struct page *page, struct vm_area_struct *vma)
 328 {
 329         unsigned long address = __vma_address(page, vma);
 330
 331         /* page should be within @vma mapping range */
 332         VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
 333
 334         return address;
 335 }
 336
 337 #else /* !CONFIG_MMU */
 338 static inline void clear_page_mlock(struct page *page) { }
 339 static inline void mlock_vma_page(struct page *page) { }
 340 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
 341
 342 #endif /* !CONFIG_MMU */
 343
 344 /*
 345  * Return the mem_map entry representing the 'offset' subpage within
 346  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
 347  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
 348  */
 349 static inline struct page *mem_map_offset(struct page *base, int offset)
 350 {
 351         if (unlikely(offset >= MAX_ORDER_NR_PAGES))
 352                 return nth_page(base, offset);
 353         return base + offset;
 354 }
 355
 356 /*
 357  * Iterator over all subpages within the maximally aligned gigantic
 358  * page 'base'.  Handle any discontiguity in the mem_map.
 359  */
 360 static inline struct page *mem_map_next(struct page *iter,
 361                                                 struct page *base, int offset)
 362 {
 363         if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
 364                 unsigned long pfn = page_to_pfn(base) + offset;
 365                 if (!pfn_valid(pfn))
 366                         return NULL;
 367                 return pfn_to_page(pfn);
 368         }
 369         return iter + 1;
 370 }
 371
 372 /*
 373  * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
 374  * so all functions starting at paging_init should be marked __init
 375  * in those cases. SPARSEMEM, however, allows for memory hotplug,
 376  * and alloc_bootmem_node is not used.
 377  */
 378 #ifdef CONFIG_SPARSEMEM
 379 #define __paginginit __meminit
 380 #else
 381 #define __paginginit __init
 382 #endif
 383
 384 /* Memory initialisation debug and verification */
 385 enum mminit_level {
 386         MMINIT_WARNING,
 387         MMINIT_VERIFY,
 388         MMINIT_TRACE
 389 };
 390
 391 #ifdef CONFIG_DEBUG_MEMORY_INIT
 392
 393 extern int mminit_loglevel;
 394
 395 #define mminit_dprintk(level, prefix, fmt, arg...) \
 396 do { \
 397         if (level < mminit_loglevel) { \
 398                 if (level <= MMINIT_WARNING) \
 399                         pr_warn("mminit::" prefix " " fmt, ##arg);      \
 400                 else \
 401                         printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
 402         } \
 403 } while (0)
 404
 405 extern void mminit_verify_pageflags_layout(void);
 406 extern void mminit_verify_zonelist(void);
 407 #else
 408
 409 static inline void mminit_dprintk(enum mminit_level level,
 410                                 const char *prefix, const char *fmt, ...)
 411 {
 412 }
 413
 414 static inline void mminit_verify_pageflags_layout(void)
 415 {
 416 }
 417
 418 static inline void mminit_verify_zonelist(void)
 419 {
 420 }
 421 #endif /* CONFIG_DEBUG_MEMORY_INIT */
 422
 423 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
 424 #if defined(CONFIG_SPARSEMEM)
 425 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 426                                 unsigned long *end_pfn);
 427 #else
 428 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 429                                 unsigned long *end_pfn)
 430 {
 431 }
 432 #endif /* CONFIG_SPARSEMEM */
 433
 434 #define NODE_RECLAIM_NOSCAN     -2
 435 #define NODE_RECLAIM_FULL       -1
 436 #define NODE_RECLAIM_SOME       0
 437 #define NODE_RECLAIM_SUCCESS    1
 438
 439 extern int hwpoison_filter(struct page *p);
 440
 441 extern u32 hwpoison_filter_dev_major;
 442 extern u32 hwpoison_filter_dev_minor;
 443 extern u64 hwpoison_filter_flags_mask;
 444 extern u64 hwpoison_filter_flags_value;
 445 extern u64 hwpoison_filter_memcg;
 446 extern u32 hwpoison_filter_enable;
 447
 448 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
 449         unsigned long, unsigned long,
 450         unsigned long, unsigned long);
 451
 452 extern void set_pageblock_order(void);
 453 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
 454                                             struct list_head *page_list);
 455 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
 456 #define ALLOC_WMARK_MIN         WMARK_MIN
 457 #define ALLOC_WMARK_LOW         WMARK_LOW
 458 #define ALLOC_WMARK_HIGH        WMARK_HIGH
 459 #define ALLOC_NO_WATERMARKS     0x04 /* don't check watermarks at all */
 460
 461 /* Mask to get the watermark bits */
 462 #define ALLOC_WMARK_MASK        (ALLOC_NO_WATERMARKS-1)
 463
 464 #define ALLOC_HARDER            0x10 /* try to alloc harder */
 465 #define ALLOC_HIGH              0x20 /* __GFP_HIGH set */
 466 #define ALLOC_CPUSET            0x40 /* check for correct cpuset */
 467 #define ALLOC_CMA               0x80 /* allow allocations from CMA areas */
 468
 469 enum ttu_flags;
 470 struct tlbflush_unmap_batch;
 471
 472 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
 473 void try_to_unmap_flush(void);
 474 void try_to_unmap_flush_dirty(void);
 475 #else
 476 static inline void try_to_unmap_flush(void)
 477 {
 478 }
 479 static inline void try_to_unmap_flush_dirty(void)
 480 {
 481 }
 482
 483 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
 484
 485 extern const struct trace_print_flags pageflag_names[];
 486 extern const struct trace_print_flags vmaflag_names[];
 487 extern const struct trace_print_flags gfpflag_names[];
 488
 489 #endif  /* __MM_INTERNAL_H */