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 void unmap_page_range(struct mmu_gather *tlb,
  47                              struct vm_area_struct *vma,
  48                              unsigned long addr, unsigned long end,
  49                              struct zap_details *details);
  50
  51 extern int __do_page_cache_readahead(struct address_space *mapping,
  52                 struct file *filp, pgoff_t offset, unsigned long nr_to_read,
  53                 unsigned long lookahead_size);
  54
  55 /*
  56  * Submit IO for the read-ahead request in file_ra_state.
  57  */
  58 static inline unsigned long ra_submit(struct file_ra_state *ra,
  59                 struct address_space *mapping, struct file *filp)
  60 {
  61         return __do_page_cache_readahead(mapping, filp,
  62                                         ra->start, ra->size, ra->async_size);
  63 }
  64
  65 /*
  66  * Turn a non-refcounted page (->_refcount == 0) into refcounted with
  67  * a count of one.
  68  */
  69 static inline void set_page_refcounted(struct page *page)
  70 {
  71         VM_BUG_ON_PAGE(PageTail(page), page);
  72         VM_BUG_ON_PAGE(page_ref_count(page), page);
  73         set_page_count(page, 1);
  74 }
  75
  76 extern unsigned long highest_memmap_pfn;
  77
  78 /*
  79  * in mm/vmscan.c:
  80  */
  81 extern int isolate_lru_page(struct page *page);
  82 extern void putback_lru_page(struct page *page);
  83 extern bool pgdat_reclaimable(struct pglist_data *pgdat);
  84
  85 /*
  86  * in mm/rmap.c:
  87  */
  88 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
  89
  90 /*
  91  * in mm/page_alloc.c
  92  */
  93
  94 /*
  95  * Structure for holding the mostly immutable allocation parameters passed
  96  * between functions involved in allocations, including the alloc_pages*
  97  * family of functions.
  98  *
  99  * nodemask, migratetype and high_zoneidx are initialized only once in
 100  * __alloc_pages_nodemask() and then never change.
 101  *
 102  * zonelist, preferred_zone and classzone_idx are set first in
 103  * __alloc_pages_nodemask() for the fast path, and might be later changed
 104  * in __alloc_pages_slowpath(). All other functions pass the whole strucure
 105  * by a const pointer.
 106  */
 107 struct alloc_context {
 108         struct zonelist *zonelist;
 109         nodemask_t *nodemask;
 110         struct zoneref *preferred_zoneref;
 111         int migratetype;
 112         enum zone_type high_zoneidx;
 113         bool spread_dirty_pages;
 114 };
 115
 116 #define ac_classzone_idx(ac) zonelist_zone_idx(ac->preferred_zoneref)
 117
 118 /*
 119  * Locate the struct page for both the matching buddy in our
 120  * pair (buddy1) and the combined O(n+1) page they form (page).
 121  *
 122  * 1) Any buddy B1 will have an order O twin B2 which satisfies
 123  * the following equation:
 124  *     B2 = B1 ^ (1 << O)
 125  * For example, if the starting buddy (buddy2) is #8 its order
 126  * 1 buddy is #10:
 127  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
 128  *
 129  * 2) Any buddy B will have an order O+1 parent P which
 130  * satisfies the following equation:
 131  *     P = B & ~(1 << O)
 132  *
 133  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
 134  */
 135 static inline unsigned long
 136 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
 137 {
 138         return page_pfn ^ (1 << order);
 139 }
 140
 141 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
 142                                 unsigned long end_pfn, struct zone *zone);
 143
 144 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
 145                                 unsigned long end_pfn, struct zone *zone)
 146 {
 147         if (zone->contiguous)
 148                 return pfn_to_page(start_pfn);
 149
 150         return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
 151 }
 152
 153 extern int __isolate_free_page(struct page *page, unsigned int order);
 154 extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
 155                                         unsigned int order);
 156 extern void prep_compound_page(struct page *page, unsigned int order);
 157 extern void post_alloc_hook(struct page *page, unsigned int order,
 158                                         gfp_t gfp_flags);
 159 extern int user_min_free_kbytes;
 160
 161 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
 162
 163 /*
 164  * in mm/compaction.c
 165  */
 166 /*
 167  * compact_control is used to track pages being migrated and the free pages
 168  * they are being migrated to during memory compaction. The free_pfn starts
 169  * at the end of a zone and migrate_pfn begins at the start. Movable pages
 170  * are moved to the end of a zone during a compaction run and the run
 171  * completes when free_pfn <= migrate_pfn
 172  */
 173 struct compact_control {
 174         struct list_head freepages;     /* List of free pages to migrate to */
 175         struct list_head migratepages;  /* List of pages being migrated */
 176         unsigned long nr_freepages;     /* Number of isolated free pages */
 177         unsigned long nr_migratepages;  /* Number of pages to migrate */
 178         unsigned long total_migrate_scanned;
 179         unsigned long total_free_scanned;
 180         unsigned long free_pfn;         /* isolate_freepages search base */
 181         unsigned long migrate_pfn;      /* isolate_migratepages search base */
 182         unsigned long last_migrated_pfn;/* Not yet flushed page being freed */
 183         enum migrate_mode mode;         /* Async or sync migration mode */
 184         bool ignore_skip_hint;          /* Scan blocks even if marked skip */
 185         bool ignore_block_suitable;     /* Scan blocks considered unsuitable */
 186         bool direct_compaction;         /* False from kcompactd or /proc/... */
 187         bool whole_zone;                /* Whole zone should/has been scanned */
 188         int order;                      /* order a direct compactor needs */
 189         const gfp_t gfp_mask;           /* gfp mask of a direct compactor */
 190         const unsigned int alloc_flags; /* alloc flags of a direct compactor */
 191         const int classzone_idx;        /* zone index of a direct compactor */
 192         struct zone *zone;
 193         bool contended;                 /* Signal lock or sched contention */
 194 };
 195
 196 unsigned long
 197 isolate_freepages_range(struct compact_control *cc,
 198                         unsigned long start_pfn, unsigned long end_pfn);
 199 unsigned long
 200 isolate_migratepages_range(struct compact_control *cc,
 201                            unsigned long low_pfn, unsigned long end_pfn);
 202 int find_suitable_fallback(struct free_area *area, unsigned int order,
 203                         int migratetype, bool only_stealable, bool *can_steal);
 204
 205 #endif
 206
 207 /*
 208  * This function returns the order of a free page in the buddy system. In
 209  * general, page_zone(page)->lock must be held by the caller to prevent the
 210  * page from being allocated in parallel and returning garbage as the order.
 211  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
 212  * page cannot be allocated or merged in parallel. Alternatively, it must
 213  * handle invalid values gracefully, and use page_order_unsafe() below.
 214  */
 215 static inline unsigned int page_order(struct page *page)
 216 {
 217         /* PageBuddy() must be checked by the caller */
 218         return page_private(page);
 219 }
 220
 221 /*
 222  * Like page_order(), but for callers who cannot afford to hold the zone lock.
 223  * PageBuddy() should be checked first by the caller to minimize race window,
 224  * and invalid values must be handled gracefully.
 225  *
 226  * READ_ONCE is used so that if the caller assigns the result into a local
 227  * variable and e.g. tests it for valid range before using, the compiler cannot
 228  * decide to remove the variable and inline the page_private(page) multiple
 229  * times, potentially observing different values in the tests and the actual
 230  * use of the result.
 231  */
 232 #define page_order_unsafe(page)         READ_ONCE(page_private(page))
 233
 234 static inline bool is_cow_mapping(vm_flags_t flags)
 235 {
 236         return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
 237 }
 238
 239 /*
 240  * These three helpers classifies VMAs for virtual memory accounting.
 241  */
 242
 243 /*
 244  * Executable code area - executable, not writable, not stack
 245  */
 246 static inline bool is_exec_mapping(vm_flags_t flags)
 247 {
 248         return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
 249 }
 250
 251 /*
 252  * Stack area - atomatically grows in one direction
 253  *
 254  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
 255  * do_mmap() forbids all other combinations.
 256  */
 257 static inline bool is_stack_mapping(vm_flags_t flags)
 258 {
 259         return (flags & VM_STACK) == VM_STACK;
 260 }
 261
 262 /*
 263  * Data area - private, writable, not stack
 264  */
 265 static inline bool is_data_mapping(vm_flags_t flags)
 266 {
 267         return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
 268 }
 269
 270 /* mm/util.c */
 271 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
 272                 struct vm_area_struct *prev, struct rb_node *rb_parent);
 273
 274 #ifdef CONFIG_MMU
 275 extern long populate_vma_page_range(struct vm_area_struct *vma,
 276                 unsigned long start, unsigned long end, int *nonblocking);
 277 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
 278                         unsigned long start, unsigned long end);
 279 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
 280 {
 281         munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
 282 }
 283
 284 /*
 285  * must be called with vma's mmap_sem held for read or write, and page locked.
 286  */
 287 extern void mlock_vma_page(struct page *page);
 288 extern unsigned int munlock_vma_page(struct page *page);
 289
 290 /*
 291  * Clear the page's PageMlocked().  This can be useful in a situation where
 292  * we want to unconditionally remove a page from the pagecache -- e.g.,
 293  * on truncation or freeing.
 294  *
 295  * It is legal to call this function for any page, mlocked or not.
 296  * If called for a page that is still mapped by mlocked vmas, all we do
 297  * is revert to lazy LRU behaviour -- semantics are not broken.
 298  */
 299 extern void clear_page_mlock(struct page *page);
 300
 301 /*
 302  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
 303  * (because that does not go through the full procedure of migration ptes):
 304  * to migrate the Mlocked page flag; update statistics.
 305  */
 306 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
 307 {
 308         if (TestClearPageMlocked(page)) {
 309                 int nr_pages = hpage_nr_pages(page);
 310
 311                 /* Holding pmd lock, no change in irq context: __mod is safe */
 312                 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 313                 SetPageMlocked(newpage);
 314                 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
 315         }
 316 }
 317
 318 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
 319
 320 /*
 321  * At what user virtual address is page expected in @vma?
 322  */
 323 static inline unsigned long
 324 __vma_address(struct page *page, struct vm_area_struct *vma)
 325 {
 326         pgoff_t pgoff = page_to_pgoff(page);
 327         return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 328 }
 329
 330 static inline unsigned long
 331 vma_address(struct page *page, struct vm_area_struct *vma)
 332 {
 333         unsigned long address = __vma_address(page, vma);
 334
 335         /* page should be within @vma mapping range */
 336         VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
 337
 338         return address;
 339 }
 340
 341 #else /* !CONFIG_MMU */
 342 static inline void clear_page_mlock(struct page *page) { }
 343 static inline void mlock_vma_page(struct page *page) { }
 344 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
 345
 346 #endif /* !CONFIG_MMU */
 347
 348 /*
 349  * Return the mem_map entry representing the 'offset' subpage within
 350  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
 351  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
 352  */
 353 static inline struct page *mem_map_offset(struct page *base, int offset)
 354 {
 355         if (unlikely(offset >= MAX_ORDER_NR_PAGES))
 356                 return nth_page(base, offset);
 357         return base + offset;
 358 }
 359
 360 /*
 361  * Iterator over all subpages within the maximally aligned gigantic
 362  * page 'base'.  Handle any discontiguity in the mem_map.
 363  */
 364 static inline struct page *mem_map_next(struct page *iter,
 365                                                 struct page *base, int offset)
 366 {
 367         if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
 368                 unsigned long pfn = page_to_pfn(base) + offset;
 369                 if (!pfn_valid(pfn))
 370                         return NULL;
 371                 return pfn_to_page(pfn);
 372         }
 373         return iter + 1;
 374 }
 375
 376 /*
 377  * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
 378  * so all functions starting at paging_init should be marked __init
 379  * in those cases. SPARSEMEM, however, allows for memory hotplug,
 380  * and alloc_bootmem_node is not used.
 381  */
 382 #ifdef CONFIG_SPARSEMEM
 383 #define __paginginit __meminit
 384 #else
 385 #define __paginginit __init
 386 #endif
 387
 388 /* Memory initialisation debug and verification */
 389 enum mminit_level {
 390         MMINIT_WARNING,
 391         MMINIT_VERIFY,
 392         MMINIT_TRACE
 393 };
 394
 395 #ifdef CONFIG_DEBUG_MEMORY_INIT
 396
 397 extern int mminit_loglevel;
 398
 399 #define mminit_dprintk(level, prefix, fmt, arg...) \
 400 do { \
 401         if (level < mminit_loglevel) { \
 402                 if (level <= MMINIT_WARNING) \
 403                         pr_warn("mminit::" prefix " " fmt, ##arg);      \
 404                 else \
 405                         printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
 406         } \
 407 } while (0)
 408
 409 extern void mminit_verify_pageflags_layout(void);
 410 extern void mminit_verify_zonelist(void);
 411 #else
 412
 413 static inline void mminit_dprintk(enum mminit_level level,
 414                                 const char *prefix, const char *fmt, ...)
 415 {
 416 }
 417
 418 static inline void mminit_verify_pageflags_layout(void)
 419 {
 420 }
 421
 422 static inline void mminit_verify_zonelist(void)
 423 {
 424 }
 425 #endif /* CONFIG_DEBUG_MEMORY_INIT */
 426
 427 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
 428 #if defined(CONFIG_SPARSEMEM)
 429 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 430                                 unsigned long *end_pfn);
 431 #else
 432 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 433                                 unsigned long *end_pfn)
 434 {
 435 }
 436 #endif /* CONFIG_SPARSEMEM */
 437
 438 #define NODE_RECLAIM_NOSCAN     -2
 439 #define NODE_RECLAIM_FULL       -1
 440 #define NODE_RECLAIM_SOME       0
 441 #define NODE_RECLAIM_SUCCESS    1
 442
 443 extern int hwpoison_filter(struct page *p);
 444
 445 extern u32 hwpoison_filter_dev_major;
 446 extern u32 hwpoison_filter_dev_minor;
 447 extern u64 hwpoison_filter_flags_mask;
 448 extern u64 hwpoison_filter_flags_value;
 449 extern u64 hwpoison_filter_memcg;
 450 extern u32 hwpoison_filter_enable;
 451
 452 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
 453         unsigned long, unsigned long,
 454         unsigned long, unsigned long);
 455
 456 extern void set_pageblock_order(void);
 457 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
 458                                             struct list_head *page_list);
 459 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
 460 #define ALLOC_WMARK_MIN         WMARK_MIN
 461 #define ALLOC_WMARK_LOW         WMARK_LOW
 462 #define ALLOC_WMARK_HIGH        WMARK_HIGH
 463 #define ALLOC_NO_WATERMARKS     0x04 /* don't check watermarks at all */
 464
 465 /* Mask to get the watermark bits */
 466 #define ALLOC_WMARK_MASK        (ALLOC_NO_WATERMARKS-1)
 467
 468 #define ALLOC_HARDER            0x10 /* try to alloc harder */
 469 #define ALLOC_HIGH              0x20 /* __GFP_HIGH set */
 470 #define ALLOC_CPUSET            0x40 /* check for correct cpuset */
 471 #define ALLOC_CMA               0x80 /* allow allocations from CMA areas */
 472
 473 enum ttu_flags;
 474 struct tlbflush_unmap_batch;
 475
 476 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
 477 void try_to_unmap_flush(void);
 478 void try_to_unmap_flush_dirty(void);
 479 #else
 480 static inline void try_to_unmap_flush(void)
 481 {
 482 }
 483 static inline void try_to_unmap_flush_dirty(void)
 484 {
 485 }
 486
 487 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
 488
 489 extern const struct trace_print_flags pageflag_names[];
 490 extern const struct trace_print_flags vmaflag_names[];
 491 extern const struct trace_print_flags gfpflag_names[];
 492
 493 #endif  /* __MM_INTERNAL_H */