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
  17 /*
  18  * The set of flags that only affect watermark checking and reclaim
  19  * behaviour. This is used by the MM to obey the caller constraints
  20  * about IO, FS and watermark checking while ignoring placement
  21  * hints such as HIGHMEM usage.
  22  */
  23 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
  24                         __GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
  25                         __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
  26                         __GFP_ATOMIC)
  27
  28 /* The GFP flags allowed during early boot */
  29 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
  30
  31 /* Control allocation cpuset and node placement constraints */
  32 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
  33
  34 /* Do not use these with a slab allocator */
  35 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
  36
  37 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
  38                 unsigned long floor, unsigned long ceiling);
  39
  40 static inline void set_page_count(struct page *page, int v)
  41 {
  42         atomic_set(&page->_count, v);
  43 }
  44
  45 extern int __do_page_cache_readahead(struct address_space *mapping,
  46                 struct file *filp, pgoff_t offset, unsigned long nr_to_read,
  47                 unsigned long lookahead_size);
  48
  49 /*
  50  * Submit IO for the read-ahead request in file_ra_state.
  51  */
  52 static inline unsigned long ra_submit(struct file_ra_state *ra,
  53                 struct address_space *mapping, struct file *filp)
  54 {
  55         return __do_page_cache_readahead(mapping, filp,
  56                                         ra->start, ra->size, ra->async_size);
  57 }
  58
  59 /*
  60  * Turn a non-refcounted page (->_count == 0) into refcounted with
  61  * a count of one.
  62  */
  63 static inline void set_page_refcounted(struct page *page)
  64 {
  65         VM_BUG_ON_PAGE(PageTail(page), page);
  66         VM_BUG_ON_PAGE(atomic_read(&page->_count), page);
  67         set_page_count(page, 1);
  68 }
  69
  70 static inline void __get_page_tail_foll(struct page *page,
  71                                         bool get_page_head)
  72 {
  73         /*
  74          * If we're getting a tail page, the elevated page->_count is
  75          * required only in the head page and we will elevate the head
  76          * page->_count and tail page->_mapcount.
  77          *
  78          * We elevate page_tail->_mapcount for tail pages to force
  79          * page_tail->_count to be zero at all times to avoid getting
  80          * false positives from get_page_unless_zero() with
  81          * speculative page access (like in
  82          * page_cache_get_speculative()) on tail pages.
  83          */
  84         VM_BUG_ON_PAGE(atomic_read(&compound_head(page)->_count) <= 0, page);
  85         if (get_page_head)
  86                 atomic_inc(&compound_head(page)->_count);
  87         get_huge_page_tail(page);
  88 }
  89
  90 /*
  91  * This is meant to be called as the FOLL_GET operation of
  92  * follow_page() and it must be called while holding the proper PT
  93  * lock while the pte (or pmd_trans_huge) is still mapping the page.
  94  */
  95 static inline void get_page_foll(struct page *page)
  96 {
  97         if (unlikely(PageTail(page)))
  98                 /*
  99                  * This is safe only because
 100                  * __split_huge_page_refcount() can't run under
 101                  * get_page_foll() because we hold the proper PT lock.
 102                  */
 103                 __get_page_tail_foll(page, true);
 104         else {
 105                 /*
 106                  * Getting a normal page or the head of a compound page
 107                  * requires to already have an elevated page->_count.
 108                  */
 109                 VM_BUG_ON_PAGE(atomic_read(&page->_count) <= 0, page);
 110                 atomic_inc(&page->_count);
 111         }
 112 }
 113
 114 extern unsigned long highest_memmap_pfn;
 115
 116 /*
 117  * in mm/vmscan.c:
 118  */
 119 extern int isolate_lru_page(struct page *page);
 120 extern void putback_lru_page(struct page *page);
 121 extern bool zone_reclaimable(struct zone *zone);
 122
 123 /*
 124  * in mm/rmap.c:
 125  */
 126 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
 127
 128 /*
 129  * in mm/page_alloc.c
 130  */
 131
 132 /*
 133  * Structure for holding the mostly immutable allocation parameters passed
 134  * between functions involved in allocations, including the alloc_pages*
 135  * family of functions.
 136  *
 137  * nodemask, migratetype and high_zoneidx are initialized only once in
 138  * __alloc_pages_nodemask() and then never change.
 139  *
 140  * zonelist, preferred_zone and classzone_idx are set first in
 141  * __alloc_pages_nodemask() for the fast path, and might be later changed
 142  * in __alloc_pages_slowpath(). All other functions pass the whole strucure
 143  * by a const pointer.
 144  */
 145 struct alloc_context {
 146         struct zonelist *zonelist;
 147         nodemask_t *nodemask;
 148         struct zone *preferred_zone;
 149         int classzone_idx;
 150         int migratetype;
 151         enum zone_type high_zoneidx;
 152         bool spread_dirty_pages;
 153 };
 154
 155 /*
 156  * Locate the struct page for both the matching buddy in our
 157  * pair (buddy1) and the combined O(n+1) page they form (page).
 158  *
 159  * 1) Any buddy B1 will have an order O twin B2 which satisfies
 160  * the following equation:
 161  *     B2 = B1 ^ (1 << O)
 162  * For example, if the starting buddy (buddy2) is #8 its order
 163  * 1 buddy is #10:
 164  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
 165  *
 166  * 2) Any buddy B will have an order O+1 parent P which
 167  * satisfies the following equation:
 168  *     P = B & ~(1 << O)
 169  *
 170  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
 171  */
 172 static inline unsigned long
 173 __find_buddy_index(unsigned long page_idx, unsigned int order)
 174 {
 175         return page_idx ^ (1 << order);
 176 }
 177
 178 extern int __isolate_free_page(struct page *page, unsigned int order);
 179 extern void __free_pages_bootmem(struct page *page, unsigned long pfn,
 180                                         unsigned int order);
 181 extern void prep_compound_page(struct page *page, unsigned int order);
 182 #ifdef CONFIG_MEMORY_FAILURE
 183 extern bool is_free_buddy_page(struct page *page);
 184 #endif
 185 extern int user_min_free_kbytes;
 186
 187 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
 188
 189 /*
 190  * in mm/compaction.c
 191  */
 192 /*
 193  * compact_control is used to track pages being migrated and the free pages
 194  * they are being migrated to during memory compaction. The free_pfn starts
 195  * at the end of a zone and migrate_pfn begins at the start. Movable pages
 196  * are moved to the end of a zone during a compaction run and the run
 197  * completes when free_pfn <= migrate_pfn
 198  */
 199 struct compact_control {
 200         struct list_head freepages;     /* List of free pages to migrate to */
 201         struct list_head migratepages;  /* List of pages being migrated */
 202         unsigned long nr_freepages;     /* Number of isolated free pages */
 203         unsigned long nr_migratepages;  /* Number of pages to migrate */
 204         unsigned long free_pfn;         /* isolate_freepages search base */
 205         unsigned long migrate_pfn;      /* isolate_migratepages search base */
 206         unsigned long last_migrated_pfn;/* Not yet flushed page being freed */
 207         enum migrate_mode mode;         /* Async or sync migration mode */
 208         bool ignore_skip_hint;          /* Scan blocks even if marked skip */
 209         int order;                      /* order a direct compactor needs */
 210         const gfp_t gfp_mask;           /* gfp mask of a direct compactor */
 211         const int alloc_flags;          /* alloc flags of a direct compactor */
 212         const int classzone_idx;        /* zone index of a direct compactor */
 213         struct zone *zone;
 214         int contended;                  /* Signal need_sched() or lock
 215                                          * contention detected during
 216                                          * compaction
 217                                          */
 218 };
 219
 220 unsigned long
 221 isolate_freepages_range(struct compact_control *cc,
 222                         unsigned long start_pfn, unsigned long end_pfn);
 223 unsigned long
 224 isolate_migratepages_range(struct compact_control *cc,
 225                            unsigned long low_pfn, unsigned long end_pfn);
 226 int find_suitable_fallback(struct free_area *area, unsigned int order,
 227                         int migratetype, bool only_stealable, bool *can_steal);
 228
 229 #endif
 230
 231 /*
 232  * This function returns the order of a free page in the buddy system. In
 233  * general, page_zone(page)->lock must be held by the caller to prevent the
 234  * page from being allocated in parallel and returning garbage as the order.
 235  * If a caller does not hold page_zone(page)->lock, it must guarantee that the
 236  * page cannot be allocated or merged in parallel. Alternatively, it must
 237  * handle invalid values gracefully, and use page_order_unsafe() below.
 238  */
 239 static inline unsigned int page_order(struct page *page)
 240 {
 241         /* PageBuddy() must be checked by the caller */
 242         return page_private(page);
 243 }
 244
 245 /*
 246  * Like page_order(), but for callers who cannot afford to hold the zone lock.
 247  * PageBuddy() should be checked first by the caller to minimize race window,
 248  * and invalid values must be handled gracefully.
 249  *
 250  * READ_ONCE is used so that if the caller assigns the result into a local
 251  * variable and e.g. tests it for valid range before using, the compiler cannot
 252  * decide to remove the variable and inline the page_private(page) multiple
 253  * times, potentially observing different values in the tests and the actual
 254  * use of the result.
 255  */
 256 #define page_order_unsafe(page)         READ_ONCE(page_private(page))
 257
 258 static inline bool is_cow_mapping(vm_flags_t flags)
 259 {
 260         return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
 261 }
 262
 263 /* mm/util.c */
 264 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
 265                 struct vm_area_struct *prev, struct rb_node *rb_parent);
 266
 267 #ifdef CONFIG_MMU
 268 extern long populate_vma_page_range(struct vm_area_struct *vma,
 269                 unsigned long start, unsigned long end, int *nonblocking);
 270 extern void munlock_vma_pages_range(struct vm_area_struct *vma,
 271                         unsigned long start, unsigned long end);
 272 static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
 273 {
 274         munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
 275 }
 276
 277 /*
 278  * must be called with vma's mmap_sem held for read or write, and page locked.
 279  */
 280 extern void mlock_vma_page(struct page *page);
 281 extern unsigned int munlock_vma_page(struct page *page);
 282
 283 /*
 284  * Clear the page's PageMlocked().  This can be useful in a situation where
 285  * we want to unconditionally remove a page from the pagecache -- e.g.,
 286  * on truncation or freeing.
 287  *
 288  * It is legal to call this function for any page, mlocked or not.
 289  * If called for a page that is still mapped by mlocked vmas, all we do
 290  * is revert to lazy LRU behaviour -- semantics are not broken.
 291  */
 292 extern void clear_page_mlock(struct page *page);
 293
 294 /*
 295  * mlock_migrate_page - called only from migrate_misplaced_transhuge_page()
 296  * (because that does not go through the full procedure of migration ptes):
 297  * to migrate the Mlocked page flag; update statistics.
 298  */
 299 static inline void mlock_migrate_page(struct page *newpage, struct page *page)
 300 {
 301         if (TestClearPageMlocked(page)) {
 302                 int nr_pages = hpage_nr_pages(page);
 303
 304                 /* Holding pmd lock, no change in irq context: __mod is safe */
 305                 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 306                 SetPageMlocked(newpage);
 307                 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
 308         }
 309 }
 310
 311 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
 312
 313 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 314 extern unsigned long vma_address(struct page *page,
 315                                  struct vm_area_struct *vma);
 316 #endif
 317 #else /* !CONFIG_MMU */
 318 static inline void clear_page_mlock(struct page *page) { }
 319 static inline void mlock_vma_page(struct page *page) { }
 320 static inline void mlock_migrate_page(struct page *new, struct page *old) { }
 321
 322 #endif /* !CONFIG_MMU */
 323
 324 /*
 325  * Return the mem_map entry representing the 'offset' subpage within
 326  * the maximally aligned gigantic page 'base'.  Handle any discontiguity
 327  * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
 328  */
 329 static inline struct page *mem_map_offset(struct page *base, int offset)
 330 {
 331         if (unlikely(offset >= MAX_ORDER_NR_PAGES))
 332                 return nth_page(base, offset);
 333         return base + offset;
 334 }
 335
 336 /*
 337  * Iterator over all subpages within the maximally aligned gigantic
 338  * page 'base'.  Handle any discontiguity in the mem_map.
 339  */
 340 static inline struct page *mem_map_next(struct page *iter,
 341                                                 struct page *base, int offset)
 342 {
 343         if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
 344                 unsigned long pfn = page_to_pfn(base) + offset;
 345                 if (!pfn_valid(pfn))
 346                         return NULL;
 347                 return pfn_to_page(pfn);
 348         }
 349         return iter + 1;
 350 }
 351
 352 /*
 353  * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
 354  * so all functions starting at paging_init should be marked __init
 355  * in those cases. SPARSEMEM, however, allows for memory hotplug,
 356  * and alloc_bootmem_node is not used.
 357  */
 358 #ifdef CONFIG_SPARSEMEM
 359 #define __paginginit __meminit
 360 #else
 361 #define __paginginit __init
 362 #endif
 363
 364 /* Memory initialisation debug and verification */
 365 enum mminit_level {
 366         MMINIT_WARNING,
 367         MMINIT_VERIFY,
 368         MMINIT_TRACE
 369 };
 370
 371 #ifdef CONFIG_DEBUG_MEMORY_INIT
 372
 373 extern int mminit_loglevel;
 374
 375 #define mminit_dprintk(level, prefix, fmt, arg...) \
 376 do { \
 377         if (level < mminit_loglevel) { \
 378                 if (level <= MMINIT_WARNING) \
 379                         printk(KERN_WARNING "mminit::" prefix " " fmt, ##arg); \
 380                 else \
 381                         printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
 382         } \
 383 } while (0)
 384
 385 extern void mminit_verify_pageflags_layout(void);
 386 extern void mminit_verify_zonelist(void);
 387 #else
 388
 389 static inline void mminit_dprintk(enum mminit_level level,
 390                                 const char *prefix, const char *fmt, ...)
 391 {
 392 }
 393
 394 static inline void mminit_verify_pageflags_layout(void)
 395 {
 396 }
 397
 398 static inline void mminit_verify_zonelist(void)
 399 {
 400 }
 401 #endif /* CONFIG_DEBUG_MEMORY_INIT */
 402
 403 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
 404 #if defined(CONFIG_SPARSEMEM)
 405 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 406                                 unsigned long *end_pfn);
 407 #else
 408 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
 409                                 unsigned long *end_pfn)
 410 {
 411 }
 412 #endif /* CONFIG_SPARSEMEM */
 413
 414 #define ZONE_RECLAIM_NOSCAN     -2
 415 #define ZONE_RECLAIM_FULL       -1
 416 #define ZONE_RECLAIM_SOME       0
 417 #define ZONE_RECLAIM_SUCCESS    1
 418
 419 extern int hwpoison_filter(struct page *p);
 420
 421 extern u32 hwpoison_filter_dev_major;
 422 extern u32 hwpoison_filter_dev_minor;
 423 extern u64 hwpoison_filter_flags_mask;
 424 extern u64 hwpoison_filter_flags_value;
 425 extern u64 hwpoison_filter_memcg;
 426 extern u32 hwpoison_filter_enable;
 427
 428 extern unsigned long vm_mmap_pgoff(struct file *, unsigned long,
 429         unsigned long, unsigned long,
 430         unsigned long, unsigned long);
 431
 432 extern void set_pageblock_order(void);
 433 unsigned long reclaim_clean_pages_from_list(struct zone *zone,
 434                                             struct list_head *page_list);
 435 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
 436 #define ALLOC_WMARK_MIN         WMARK_MIN
 437 #define ALLOC_WMARK_LOW         WMARK_LOW
 438 #define ALLOC_WMARK_HIGH        WMARK_HIGH
 439 #define ALLOC_NO_WATERMARKS     0x04 /* don't check watermarks at all */
 440
 441 /* Mask to get the watermark bits */
 442 #define ALLOC_WMARK_MASK        (ALLOC_NO_WATERMARKS-1)
 443
 444 #define ALLOC_HARDER            0x10 /* try to alloc harder */
 445 #define ALLOC_HIGH              0x20 /* __GFP_HIGH set */
 446 #define ALLOC_CPUSET            0x40 /* check for correct cpuset */
 447 #define ALLOC_CMA               0x80 /* allow allocations from CMA areas */
 448 #define ALLOC_FAIR              0x100 /* fair zone allocation */
 449
 450 enum ttu_flags;
 451 struct tlbflush_unmap_batch;
 452
 453 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
 454 void try_to_unmap_flush(void);
 455 void try_to_unmap_flush_dirty(void);
 456 void flush_tlb_batched_pending(struct mm_struct *mm);
 457 #else
 458 static inline void try_to_unmap_flush(void)
 459 {
 460 }
 461 static inline void try_to_unmap_flush_dirty(void)
 462 {
 463 }
 464 static inline void flush_tlb_batched_pending(struct mm_struct *mm)
 465 {
 466 }
 467 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
 468 #endif  /* __MM_INTERNAL_H */