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