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_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
  39 void page_writeback_init(void);
  40
  41 vm_fault_t 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 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);
  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  * 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  */
  92 extern int isolate_lru_page(struct page *page);
  93 extern void putback_lru_page(struct page *page);
  94
  95 /*
  96  * in mm/rmap.c:
  97  */
  98 extern 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  */
 117 struct 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  */
 145 static inline unsigned long
 146 __find_buddy_pfn(unsigned long page_pfn, unsigned int order)
 147 {
 148         return page_pfn ^ (1 << order);
 149 }
 150
 151 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
 152                                 unsigned long end_pfn, struct zone *zone);
 153
 154 static 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
 163 extern int __isolate_free_page(struct page *page, unsigned int order);
 164 extern void memblock_free_pages(struct page *page, unsigned long pfn,
 165                                         unsigned int order);
 166 extern void __free_pages_core(struct page *page, unsigned int order);
 167 extern void prep_compound_page(struct page *page, unsigned int order);
 168 extern void post_alloc_hook(struct page *page, unsigned int order,
 169                                         gfp_t gfp_flags);
 170 extern int user_min_free_kbytes;
 171
 172 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
 173
 174 /*
 175  * in mm/compaction.c
 176  */
 177 /*
 178  * compact_control is used to track pages being migrated and the free pages
 179  * they are being migrated to during memory compaction. The free_pfn starts
 180  * at the end of a zone and migrate_pfn begins at the start. Movable pages
 181  * are moved to the end of a zone during a compaction run and the run
 182  * completes when free_pfn <= migrate_pfn
 183  */
 184 struct compact_control {
 185         struct list_head freepages;     /* List of free pages to migrate to */
 186         struct list_head migratepages;  /* List of pages being migrated */
 187         unsigned int nr_freepages;      /* Number of isolated free pages */
 188         unsigned int nr_migratepages;   /* Number of pages to migrate */
 189         unsigned long free_pfn;         /* isolate_freepages search base */
 190         unsigned long migrate_pfn;      /* isolate_migratepages search base */
 191         unsigned long fast_start_pfn;   /* a pfn to start linear scan from */
 192         struct zone *zone;
 193         unsigned long total_migrate_scanned;
 194         unsigned long total_free_scanned;
 195         unsigned short fast_search_fail;/* failures to use free list searches */
 196         short search_order;             /* order to start a fast search at */
 197         const gfp_t gfp_mask;           /* gfp mask of a direct compactor */
 198         int order;                      /* order a direct compactor needs */
 199         int migratetype;                /* migratetype of direct compactor */
 200         const unsigned int alloc_flags; /* alloc flags of a direct compactor */
 201         const int classzone_idx;        /* zone index of a direct compactor */
 202         enum migrate_mode mode;         /* Async or sync migration mode */
 203         bool ignore_skip_hint;          /* Scan blocks even if marked skip */
 204         bool no_set_skip_hint;          /* Don't mark blocks for skipping */
 205         bool ignore_block_suitable;     /* Scan blocks considered unsuitable */
 206         bool direct_compaction;         /* False from kcompactd or /proc/... */
 207         bool whole_zone;                /* Whole zone should/has been scanned */
 208         bool contended;                 /* Signal lock or sched contention */
 209         bool rescan;                    /* Rescanning the same pageblock */
 210 };
 211
 212 /*
 213  * Used in direct compaction when a page should be taken from the freelists
 214  * immediately when one is created during the free path.
 215  */
 216 struct capture_control {
 217         struct compact_control *cc;
 218         struct page *page;
 219 };
 220
 221 unsigned long
 222 isolate_freepages_range(struct compact_control *cc,
 223                         unsigned long start_pfn, unsigned long end_pfn);
 224 unsigned long
 225 isolate_migratepages_range(struct compact_control *cc,
 226                            unsigned long low_pfn, unsigned long end_pfn);
 227 int find_suitable_fallback(struct free_area *area, unsigned int order,
 228                         int migratetype, bool only_stealable, bool *can_steal);
 229
 230 #endif
 231
 232 /*
 233  * This function returns the order of a free page in the buddy system. In
 234  * general, page_zone(page)->lock must be held by the caller to prevent the
 235  * page from being allocated in parallel and returning garbage as the order.
 236  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
 237  * page cannot be allocated or merged in parallel. Alternatively, it must
 238  * handle invalid values gracefully, and use page_order_unsafe() below.
 239  */
 240 static inline unsigned int page_order(struct page *page)
 241 {
 242         /* PageBuddy() must be checked by the caller */
 243         return page_private(page);
 244 }
 245
 246 /*
 247  * Like page_order(), but for callers who cannot afford to hold the zone lock.
 248  * PageBuddy() should be checked first by the caller to minimize race window,
 249  * and invalid values must be handled gracefully.
 250  *
 251  * READ_ONCE is used so that if the caller assigns the result into a local
 252  * variable and e.g. tests it for valid range before using, the compiler cannot
 253  * decide to remove the variable and inline the page_private(page) multiple
 254  * times, potentially observing different values in the tests and the actual
 255  * use of the result.
 256  */
 257 #define page_order_unsafe(page)         READ_ONCE(page_private(page))
 258
 259 static inline bool is_cow_mapping(vm_flags_t flags)
 260 {
 261         return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
 262 }
 263
 264 /*
 265  * These three helpers classifies VMAs for virtual memory accounting.
 266  */
 267
 268 /*
 269  * Executable code area - executable, not writable, not stack
 270  */
 271 static inline bool is_exec_mapping(vm_flags_t flags)
 272 {
 273         return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
 274 }
 275
 276 /*
 277  * Stack area - atomatically grows in one direction
 278  *
 279  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
 280  * do_mmap() forbids all other combinations.
 281  */
 282 static inline bool is_stack_mapping(vm_flags_t flags)
 283 {
 284         return (flags & VM_STACK) == VM_STACK;
 285 }
 286
 287 /*
 288  * Data area - private, writable, not stack
 289  */
 290 static inline bool is_data_mapping(vm_flags_t flags)
 291 {
 292         return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
 293 }
 294
 295 /* mm/util.c */
 296 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
 297                 struct vm_area_struct *prev, struct rb_node *rb_parent);
 298
 299 #ifdef CONFIG_MMU
 300 extern long populate_vma_page_range(struct vm_area_struct *vma,
 301                 unsigned long start, unsigned long end, int *nonblocking);
 302 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
 303                         unsigned long start, unsigned long end);
 304 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
 305 {
 306         munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
 307 }
 308
 309 /*
 310  * must be called with vma's mmap_sem held for read or write, and page locked.
 311  */
 312 extern void mlock_vma_page(struct page *page);
 313 extern unsigned int munlock_vma_page(struct page *page);
 314
 315 /*
 316  * Clear the page's PageMlocked().  This can be useful in a situation where
 317  * we want to unconditionally remove a page from the pagecache -- e.g.,
 318  * on truncation or freeing.
 319  *
 320  * It is legal to call this function for any page, mlocked or not.
 321  * If called for a page that is still mapped by mlocked vmas, all we do
 322  * is revert to lazy LRU behaviour -- semantics are not broken.
 323  */
 324 extern void clear_page_mlock(struct page *page);
 325
 326 /*
 327  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
 328  * (because that does not go through the full procedure of migration ptes):
 329  * to migrate the Mlocked page flag; update statistics.
 330  */
 331 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
 332 {
 333         if (TestClearPageMlocked(page)) {
 334                 int nr_pages = hpage_nr_pages(page);
 335
 336                 /* Holding pmd lock, no change in irq context: __mod is safe */
 337                 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 338                 SetPageMlocked(newpage);
 339                 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
 340         }
 341 }
 342
 343 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
 344
 345 /*
 346  * At what user virtual address is page expected in @vma?
 347  */
 348 static inline unsigned long
 349 __vma_address(struct page *page, struct vm_area_struct *vma)
 350 {
 351         pgoff_t pgoff = page_to_pgoff(page);
 352         return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 353 }
 354
 355 static inline unsigned long
 356 vma_address(struct page *page, struct vm_area_struct *vma)
 357 {
 358         unsigned long start, end;
 359
 360         start = __vma_address(page, vma);
 361         end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
 362
 363         /* page should be within @vma mapping range */
 364         VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma);
 365
 366         return max(start, vma->vm_start);
 367 }
 368
 369 #else /* !CONFIG_MMU */
 370 static inline void clear_page_mlock(struct page *page) { }
 371 static inline void mlock_vma_page(struct page *page) { }
 372 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
 373
 374 #endif /* !CONFIG_MMU */
 375
 376 /*
 377  * Return the mem_map entry representing the 'offset' subpage within
 378  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
 379  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
 380  */
 381 static inline struct page *mem_map_offset(struct page *base, int offset)
 382 {
 383         if (unlikely(offset >= MAX_ORDER_NR_PAGES))
 384                 return nth_page(base, offset);
 385         return base + offset;
 386 }
 387
 388 /*
 389  * Iterator over all subpages within the maximally aligned gigantic
 390  * page 'base'.  Handle any discontiguity in the mem_map.
 391  */
 392 static inline struct page *mem_map_next(struct page *iter,
 393                                                 struct page *base, int offset)
 394 {
 395         if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
 396                 unsigned long pfn = page_to_pfn(base) + offset;
 397                 if (!pfn_valid(pfn))
 398                         return NULL;
 399                 return pfn_to_page(pfn);
 400         }
 401         return iter + 1;
 402 }
 403
 404 /* Memory initialisation debug and verification */
 405 enum mminit_level {
 406         MMINIT_WARNING,
 407         MMINIT_VERIFY,
 408         MMINIT_TRACE
 409 };
 410
 411 #ifdef CONFIG_DEBUG_MEMORY_INIT
 412
 413 extern int mminit_loglevel;
 414
 415 #define mminit_dprintk(level, prefix, fmt, arg...) \
 416 do { \
 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
 425 extern void mminit_verify_pageflags_layout(void);
 426 extern void mminit_verify_zonelist(void);
 427 #else
 428
 429 static inline void mminit_dprintk(enum mminit_level level,
 430                                 const char *prefix, const char *fmt, ...)
 431 {
 432 }
 433
 434 static inline void mminit_verify_pageflags_layout(void)
 435 {
 436 }
 437
 438 static 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)
 445 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 446                                 unsigned long *end_pfn);
 447 #else
 448 static 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
 459 #ifdef CONFIG_NUMA
 460 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
 461 #else
 462 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
 463                                 unsigned int order)
 464 {
 465         return NODE_RECLAIM_NOSCAN;
 466 }
 467 #endif
 468
 469 extern int hwpoison_filter(struct page *p);
 470
 471 extern u32 hwpoison_filter_dev_major;
 472 extern u32 hwpoison_filter_dev_minor;
 473 extern u64 hwpoison_filter_flags_mask;
 474 extern u64 hwpoison_filter_flags_value;
 475 extern u64 hwpoison_filter_memcg;
 476 extern u32 hwpoison_filter_enable;
 477
 478 extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
 479         unsigned long, unsigned long,
 480         unsigned long, unsigned long);
 481
 482 extern void set_pageblock_order(void);
 483 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
 484                                             struct list_head *page_list);
 485 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
 486 #define ALLOC_WMARK_MIN         WMARK_MIN
 487 #define ALLOC_WMARK_LOW         WMARK_LOW
 488 #define ALLOC_WMARK_HIGH        WMARK_HIGH
 489 #define ALLOC_NO_WATERMARKS     0x04 /* don't check watermarks at all */
 490
 491 /* Mask to get the watermark bits */
 492 #define ALLOC_WMARK_MASK        (ALLOC_NO_WATERMARKS-1)
 493
 494 /*
 495  * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
 496  * cannot assume a reduced access to memory reserves is sufficient for
 497  * !MMU
 498  */
 499 #ifdef CONFIG_MMU
 500 #define ALLOC_OOM               0x08
 501 #else
 502 #define ALLOC_OOM               ALLOC_NO_WATERMARKS
 503 #endif
 504
 505 #define ALLOC_HARDER             0x10 /* try to alloc harder */
 506 #define ALLOC_HIGH               0x20 /* __GFP_HIGH set */
 507 #define ALLOC_CPUSET             0x40 /* check for correct cpuset */
 508 #define ALLOC_CMA                0x80 /* allow allocations from CMA areas */
 509 #ifdef CONFIG_ZONE_DMA32
 510 #define ALLOC_NOFRAGMENT        0x100 /* avoid mixing pageblock types */
 511 #else
 512 #define ALLOC_NOFRAGMENT          0x0
 513 #endif
 514 #define ALLOC_KSWAPD            0x200 /* allow waking of kswapd */
 515
 516 enum ttu_flags;
 517 struct tlbflush_unmap_batch;
 518
 519
 520 /*
 521  * only for MM internal work items which do not depend on
 522  * any allocations or locks which might depend on allocations
 523  */
 524 extern struct workqueue_struct *mm_percpu_wq;
 525
 526 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
 527 void try_to_unmap_flush(void);
 528 void try_to_unmap_flush_dirty(void);
 529 void flush_tlb_batched_pending(struct mm_struct *mm);
 530 #else
 531 static inline void try_to_unmap_flush(void)
 532 {
 533 }
 534 static inline void try_to_unmap_flush_dirty(void)
 535 {
 536 }
 537 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
 538 {
 539 }
 540 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
 541
 542 extern const struct trace_print_flags pageflag_names[];
 543 extern const struct trace_print_flags vmaflag_names[];
 544 extern const struct trace_print_flags gfpflag_names[];
 545
 546 static inline bool is_migrate_highatomic(enum migratetype migratetype)
 547 {
 548         return migratetype == MIGRATE_HIGHATOMIC;
 549 }
 550
 551 static inline bool is_migrate_highatomic_page(struct page *page)
 552 {
 553         return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
 554 }
 555
 556 void setup_zone_pageset(struct zone *zone);
 557 extern struct page *alloc_new_node_page(struct page *page, unsigned long node);
 558 #endif  /* __MM_INTERNAL_H */