1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* internal.h: mm/ internal definitions 3 * 4 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved. 5 * Written by David Howells (dhowells@redhat.com) 6 */ 7 #ifndef __MM_INTERNAL_H 8 #define __MM_INTERNAL_H 9 10 #include <linux/fs.h> 11 #include <linux/mm.h> 12 #include <linux/pagemap.h> 13 #include <linux/tracepoint-defs.h> 14 15 /* 16 * The set of flags that only affect watermark checking and reclaim 17 * behaviour. This is used by the MM to obey the caller constraints 18 * about IO, FS and watermark checking while ignoring placement 19 * hints such as HIGHMEM usage. 20 */ 21 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\ 22 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\ 23 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\ 24 __GFP_ATOMIC) 25 26 /* The GFP flags allowed during early boot */ 27 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS)) 28 29 /* Control allocation cpuset and node placement constraints */ 30 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE) 31 32 /* Do not use these with a slab allocator */ 33 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK) 34 35 void page_writeback_init(void); 36 37 vm_fault_t do_swap_page(struct vm_fault *vmf); 38 39 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma, 40 unsigned long floor, unsigned long ceiling); 41 42 static inline bool can_madv_lru_vma(struct vm_area_struct *vma) 43 { 44 return !(vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP)); 45 } 46 47 void unmap_page_range(struct mmu_gather *tlb, 48 struct vm_area_struct *vma, 49 unsigned long addr, unsigned long end, 50 struct zap_details *details); 51 52 void do_page_cache_ra(struct readahead_control *, unsigned long nr_to_read, 53 unsigned long lookahead_size); 54 void force_page_cache_ra(struct readahead_control *, struct file_ra_state *, 55 unsigned long nr); 56 static inline void force_page_cache_readahead(struct address_space *mapping, 57 struct file *file, pgoff_t index, unsigned long nr_to_read) 58 { 59 DEFINE_READAHEAD(ractl, file, mapping, index); 60 force_page_cache_ra(&ractl, &file->f_ra, nr_to_read); 61 } 62 63 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, 64 pgoff_t end, struct pagevec *pvec, pgoff_t *indices); 65 66 /** 67 * page_evictable - test whether a page is evictable 68 * @page: the page to test 69 * 70 * Test whether page is evictable--i.e., should be placed on active/inactive 71 * lists vs unevictable list. 72 * 73 * Reasons page might not be evictable: 74 * (1) page's mapping marked unevictable 75 * (2) page is part of an mlocked VMA 76 * 77 */ 78 static inline bool page_evictable(struct page *page) 79 { 80 bool ret; 81 82 /* Prevent address_space of inode and swap cache from being freed */ 83 rcu_read_lock(); 84 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); 85 rcu_read_unlock(); 86 return ret; 87 } 88 89 /* 90 * Turn a non-refcounted page (->_refcount == 0) into refcounted with 91 * a count of one. 92 */ 93 static inline void set_page_refcounted(struct page *page) 94 { 95 VM_BUG_ON_PAGE(PageTail(page), page); 96 VM_BUG_ON_PAGE(page_ref_count(page), page); 97 set_page_count(page, 1); 98 } 99 100 extern unsigned long highest_memmap_pfn; 101 102 /* 103 * Maximum number of reclaim retries without progress before the OOM 104 * killer is consider the only way forward. 105 */ 106 #define MAX_RECLAIM_RETRIES 16 107 108 /* 109 * in mm/vmscan.c: 110 */ 111 extern int isolate_lru_page(struct page *page); 112 extern void putback_lru_page(struct page *page); 113 114 /* 115 * in mm/rmap.c: 116 */ 117 extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address); 118 119 /* 120 * in mm/page_alloc.c 121 */ 122 123 /* 124 * Structure for holding the mostly immutable allocation parameters passed 125 * between functions involved in allocations, including the alloc_pages* 126 * family of functions. 127 * 128 * nodemask, migratetype and highest_zoneidx are initialized only once in 129 * __alloc_pages_nodemask() and then never change. 130 * 131 * zonelist, preferred_zone and highest_zoneidx are set first in 132 * __alloc_pages_nodemask() for the fast path, and might be later changed 133 * in __alloc_pages_slowpath(). All other functions pass the whole structure 134 * by a const pointer. 135 */ 136 struct alloc_context { 137 struct zonelist *zonelist; 138 nodemask_t *nodemask; 139 struct zoneref *preferred_zoneref; 140 int migratetype; 141 142 /* 143 * highest_zoneidx represents highest usable zone index of 144 * the allocation request. Due to the nature of the zone, 145 * memory on lower zone than the highest_zoneidx will be 146 * protected by lowmem_reserve[highest_zoneidx]. 147 * 148 * highest_zoneidx is also used by reclaim/compaction to limit 149 * the target zone since higher zone than this index cannot be 150 * usable for this allocation request. 151 */ 152 enum zone_type highest_zoneidx; 153 bool spread_dirty_pages; 154 }; 155 156 /* 157 * Locate the struct page for both the matching buddy in our 158 * pair (buddy1) and the combined O(n+1) page they form (page). 159 * 160 * 1) Any buddy B1 will have an order O twin B2 which satisfies 161 * the following equation: 162 * B2 = B1 ^ (1 << O) 163 * For example, if the starting buddy (buddy2) is #8 its order 164 * 1 buddy is #10: 165 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 166 * 167 * 2) Any buddy B will have an order O+1 parent P which 168 * satisfies the following equation: 169 * P = B & ~(1 << O) 170 * 171 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER 172 */ 173 static inline unsigned long 174 __find_buddy_pfn(unsigned long page_pfn, unsigned int order) 175 { 176 return page_pfn ^ (1 << order); 177 } 178 179 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn, 180 unsigned long end_pfn, struct zone *zone); 181 182 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn, 183 unsigned long end_pfn, struct zone *zone) 184 { 185 if (zone->contiguous) 186 return pfn_to_page(start_pfn); 187 188 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone); 189 } 190 191 extern int __isolate_free_page(struct page *page, unsigned int order); 192 extern void __putback_isolated_page(struct page *page, unsigned int order, 193 int mt); 194 extern void memblock_free_pages(struct page *page, unsigned long pfn, 195 unsigned int order); 196 extern void __free_pages_core(struct page *page, unsigned int order); 197 extern void prep_compound_page(struct page *page, unsigned int order); 198 extern void post_alloc_hook(struct page *page, unsigned int order, 199 gfp_t gfp_flags); 200 extern int user_min_free_kbytes; 201 202 extern void free_unref_page(struct page *page); 203 extern void free_unref_page_list(struct list_head *list); 204 205 extern void zone_pcp_update(struct zone *zone); 206 extern void zone_pcp_reset(struct zone *zone); 207 extern void zone_pcp_disable(struct zone *zone); 208 extern void zone_pcp_enable(struct zone *zone); 209 210 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 211 212 /* 213 * in mm/compaction.c 214 */ 215 /* 216 * compact_control is used to track pages being migrated and the free pages 217 * they are being migrated to during memory compaction. The free_pfn starts 218 * at the end of a zone and migrate_pfn begins at the start. Movable pages 219 * are moved to the end of a zone during a compaction run and the run 220 * completes when free_pfn <= migrate_pfn 221 */ 222 struct compact_control { 223 struct list_head freepages; /* List of free pages to migrate to */ 224 struct list_head migratepages; /* List of pages being migrated */ 225 unsigned int nr_freepages; /* Number of isolated free pages */ 226 unsigned int nr_migratepages; /* Number of pages to migrate */ 227 unsigned long free_pfn; /* isolate_freepages search base */ 228 unsigned long migrate_pfn; /* isolate_migratepages search base */ 229 unsigned long fast_start_pfn; /* a pfn to start linear scan from */ 230 struct zone *zone; 231 unsigned long total_migrate_scanned; 232 unsigned long total_free_scanned; 233 unsigned short fast_search_fail;/* failures to use free list searches */ 234 short search_order; /* order to start a fast search at */ 235 const gfp_t gfp_mask; /* gfp mask of a direct compactor */ 236 int order; /* order a direct compactor needs */ 237 int migratetype; /* migratetype of direct compactor */ 238 const unsigned int alloc_flags; /* alloc flags of a direct compactor */ 239 const int highest_zoneidx; /* zone index of a direct compactor */ 240 enum migrate_mode mode; /* Async or sync migration mode */ 241 bool ignore_skip_hint; /* Scan blocks even if marked skip */ 242 bool no_set_skip_hint; /* Don't mark blocks for skipping */ 243 bool ignore_block_suitable; /* Scan blocks considered unsuitable */ 244 bool direct_compaction; /* False from kcompactd or /proc/... */ 245 bool proactive_compaction; /* kcompactd proactive compaction */ 246 bool whole_zone; /* Whole zone should/has been scanned */ 247 bool contended; /* Signal lock or sched contention */ 248 bool rescan; /* Rescanning the same pageblock */ 249 bool alloc_contig; /* alloc_contig_range allocation */ 250 }; 251 252 /* 253 * Used in direct compaction when a page should be taken from the freelists 254 * immediately when one is created during the free path. 255 */ 256 struct capture_control { 257 struct compact_control *cc; 258 struct page *page; 259 }; 260 261 unsigned long 262 isolate_freepages_range(struct compact_control *cc, 263 unsigned long start_pfn, unsigned long end_pfn); 264 unsigned long 265 isolate_migratepages_range(struct compact_control *cc, 266 unsigned long low_pfn, unsigned long end_pfn); 267 int find_suitable_fallback(struct free_area *area, unsigned int order, 268 int migratetype, bool only_stealable, bool *can_steal); 269 270 #endif 271 272 /* 273 * This function returns the order of a free page in the buddy system. In 274 * general, page_zone(page)->lock must be held by the caller to prevent the 275 * page from being allocated in parallel and returning garbage as the order. 276 * If a caller does not hold page_zone(page)->lock, it must guarantee that the 277 * page cannot be allocated or merged in parallel. Alternatively, it must 278 * handle invalid values gracefully, and use buddy_order_unsafe() below. 279 */ 280 static inline unsigned int buddy_order(struct page *page) 281 { 282 /* PageBuddy() must be checked by the caller */ 283 return page_private(page); 284 } 285 286 /* 287 * Like buddy_order(), but for callers who cannot afford to hold the zone lock. 288 * PageBuddy() should be checked first by the caller to minimize race window, 289 * and invalid values must be handled gracefully. 290 * 291 * READ_ONCE is used so that if the caller assigns the result into a local 292 * variable and e.g. tests it for valid range before using, the compiler cannot 293 * decide to remove the variable and inline the page_private(page) multiple 294 * times, potentially observing different values in the tests and the actual 295 * use of the result. 296 */ 297 #define buddy_order_unsafe(page) READ_ONCE(page_private(page)) 298 299 static inline bool is_cow_mapping(vm_flags_t flags) 300 { 301 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 302 } 303 304 /* 305 * These three helpers classifies VMAs for virtual memory accounting. 306 */ 307 308 /* 309 * Executable code area - executable, not writable, not stack 310 */ 311 static inline bool is_exec_mapping(vm_flags_t flags) 312 { 313 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC; 314 } 315 316 /* 317 * Stack area - atomatically grows in one direction 318 * 319 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous: 320 * do_mmap() forbids all other combinations. 321 */ 322 static inline bool is_stack_mapping(vm_flags_t flags) 323 { 324 return (flags & VM_STACK) == VM_STACK; 325 } 326 327 /* 328 * Data area - private, writable, not stack 329 */ 330 static inline bool is_data_mapping(vm_flags_t flags) 331 { 332 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE; 333 } 334 335 /* mm/util.c */ 336 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 337 struct vm_area_struct *prev); 338 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma); 339 340 #ifdef CONFIG_MMU 341 extern long populate_vma_page_range(struct vm_area_struct *vma, 342 unsigned long start, unsigned long end, int *nonblocking); 343 extern void munlock_vma_pages_range(struct vm_area_struct *vma, 344 unsigned long start, unsigned long end); 345 static inline void munlock_vma_pages_all(struct vm_area_struct *vma) 346 { 347 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end); 348 } 349 350 /* 351 * must be called with vma's mmap_lock held for read or write, and page locked. 352 */ 353 extern void mlock_vma_page(struct page *page); 354 extern unsigned int munlock_vma_page(struct page *page); 355 356 /* 357 * Clear the page's PageMlocked(). This can be useful in a situation where 358 * we want to unconditionally remove a page from the pagecache -- e.g., 359 * on truncation or freeing. 360 * 361 * It is legal to call this function for any page, mlocked or not. 362 * If called for a page that is still mapped by mlocked vmas, all we do 363 * is revert to lazy LRU behaviour -- semantics are not broken. 364 */ 365 extern void clear_page_mlock(struct page *page); 366 367 /* 368 * mlock_migrate_page - called only from migrate_misplaced_transhuge_page() 369 * (because that does not go through the full procedure of migration ptes): 370 * to migrate the Mlocked page flag; update statistics. 371 */ 372 static inline void mlock_migrate_page(struct page *newpage, struct page *page) 373 { 374 if (TestClearPageMlocked(page)) { 375 int nr_pages = thp_nr_pages(page); 376 377 /* Holding pmd lock, no change in irq context: __mod is safe */ 378 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); 379 SetPageMlocked(newpage); 380 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages); 381 } 382 } 383 384 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma); 385 386 /* 387 * At what user virtual address is page expected in @vma? 388 */ 389 static inline unsigned long 390 __vma_address(struct page *page, struct vm_area_struct *vma) 391 { 392 pgoff_t pgoff = page_to_pgoff(page); 393 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 394 } 395 396 static inline unsigned long 397 vma_address(struct page *page, struct vm_area_struct *vma) 398 { 399 unsigned long start, end; 400 401 start = __vma_address(page, vma); 402 end = start + thp_size(page) - PAGE_SIZE; 403 404 /* page should be within @vma mapping range */ 405 VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma); 406 407 return max(start, vma->vm_start); 408 } 409 410 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf, 411 struct file *fpin) 412 { 413 int flags = vmf->flags; 414 415 if (fpin) 416 return fpin; 417 418 /* 419 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or 420 * anything, so we only pin the file and drop the mmap_lock if only 421 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt. 422 */ 423 if (fault_flag_allow_retry_first(flags) && 424 !(flags & FAULT_FLAG_RETRY_NOWAIT)) { 425 fpin = get_file(vmf->vma->vm_file); 426 mmap_read_unlock(vmf->vma->vm_mm); 427 } 428 return fpin; 429 } 430 431 #else /* !CONFIG_MMU */ 432 static inline void clear_page_mlock(struct page *page) { } 433 static inline void mlock_vma_page(struct page *page) { } 434 static inline void mlock_migrate_page(struct page *new, struct page *old) { } 435 436 #endif /* !CONFIG_MMU */ 437 438 /* 439 * Return the mem_map entry representing the 'offset' subpage within 440 * the maximally aligned gigantic page 'base'. Handle any discontiguity 441 * in the mem_map at MAX_ORDER_NR_PAGES boundaries. 442 */ 443 static inline struct page *mem_map_offset(struct page *base, int offset) 444 { 445 if (unlikely(offset >= MAX_ORDER_NR_PAGES)) 446 return nth_page(base, offset); 447 return base + offset; 448 } 449 450 /* 451 * Iterator over all subpages within the maximally aligned gigantic 452 * page 'base'. Handle any discontiguity in the mem_map. 453 */ 454 static inline struct page *mem_map_next(struct page *iter, 455 struct page *base, int offset) 456 { 457 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) { 458 unsigned long pfn = page_to_pfn(base) + offset; 459 if (!pfn_valid(pfn)) 460 return NULL; 461 return pfn_to_page(pfn); 462 } 463 return iter + 1; 464 } 465 466 /* Memory initialisation debug and verification */ 467 enum mminit_level { 468 MMINIT_WARNING, 469 MMINIT_VERIFY, 470 MMINIT_TRACE 471 }; 472 473 #ifdef CONFIG_DEBUG_MEMORY_INIT 474 475 extern int mminit_loglevel; 476 477 #define mminit_dprintk(level, prefix, fmt, arg...) \ 478 do { \ 479 if (level < mminit_loglevel) { \ 480 if (level <= MMINIT_WARNING) \ 481 pr_warn("mminit::" prefix " " fmt, ##arg); \ 482 else \ 483 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \ 484 } \ 485 } while (0) 486 487 extern void mminit_verify_pageflags_layout(void); 488 extern void mminit_verify_zonelist(void); 489 #else 490 491 static inline void mminit_dprintk(enum mminit_level level, 492 const char *prefix, const char *fmt, ...) 493 { 494 } 495 496 static inline void mminit_verify_pageflags_layout(void) 497 { 498 } 499 500 static inline void mminit_verify_zonelist(void) 501 { 502 } 503 #endif /* CONFIG_DEBUG_MEMORY_INIT */ 504 505 /* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */ 506 #if defined(CONFIG_SPARSEMEM) 507 extern void mminit_validate_memmodel_limits(unsigned long *start_pfn, 508 unsigned long *end_pfn); 509 #else 510 static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn, 511 unsigned long *end_pfn) 512 { 513 } 514 #endif /* CONFIG_SPARSEMEM */ 515 516 #define NODE_RECLAIM_NOSCAN -2 517 #define NODE_RECLAIM_FULL -1 518 #define NODE_RECLAIM_SOME 0 519 #define NODE_RECLAIM_SUCCESS 1 520 521 #ifdef CONFIG_NUMA 522 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int); 523 #else 524 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask, 525 unsigned int order) 526 { 527 return NODE_RECLAIM_NOSCAN; 528 } 529 #endif 530 531 extern int hwpoison_filter(struct page *p); 532 533 extern u32 hwpoison_filter_dev_major; 534 extern u32 hwpoison_filter_dev_minor; 535 extern u64 hwpoison_filter_flags_mask; 536 extern u64 hwpoison_filter_flags_value; 537 extern u64 hwpoison_filter_memcg; 538 extern u32 hwpoison_filter_enable; 539 540 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long, 541 unsigned long, unsigned long, 542 unsigned long, unsigned long); 543 544 extern void set_pageblock_order(void); 545 unsigned int reclaim_clean_pages_from_list(struct zone *zone, 546 struct list_head *page_list); 547 /* The ALLOC_WMARK bits are used as an index to zone->watermark */ 548 #define ALLOC_WMARK_MIN WMARK_MIN 549 #define ALLOC_WMARK_LOW WMARK_LOW 550 #define ALLOC_WMARK_HIGH WMARK_HIGH 551 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ 552 553 /* Mask to get the watermark bits */ 554 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) 555 556 /* 557 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we 558 * cannot assume a reduced access to memory reserves is sufficient for 559 * !MMU 560 */ 561 #ifdef CONFIG_MMU 562 #define ALLOC_OOM 0x08 563 #else 564 #define ALLOC_OOM ALLOC_NO_WATERMARKS 565 #endif 566 567 #define ALLOC_HARDER 0x10 /* try to alloc harder */ 568 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ 569 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ 570 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */ 571 #ifdef CONFIG_ZONE_DMA32 572 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */ 573 #else 574 #define ALLOC_NOFRAGMENT 0x0 575 #endif 576 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */ 577 578 enum ttu_flags; 579 struct tlbflush_unmap_batch; 580 581 582 /* 583 * only for MM internal work items which do not depend on 584 * any allocations or locks which might depend on allocations 585 */ 586 extern struct workqueue_struct *mm_percpu_wq; 587 588 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 589 void try_to_unmap_flush(void); 590 void try_to_unmap_flush_dirty(void); 591 void flush_tlb_batched_pending(struct mm_struct *mm); 592 #else 593 static inline void try_to_unmap_flush(void) 594 { 595 } 596 static inline void try_to_unmap_flush_dirty(void) 597 { 598 } 599 static inline void flush_tlb_batched_pending(struct mm_struct *mm) 600 { 601 } 602 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ 603 604 extern const struct trace_print_flags pageflag_names[]; 605 extern const struct trace_print_flags vmaflag_names[]; 606 extern const struct trace_print_flags gfpflag_names[]; 607 608 static inline bool is_migrate_highatomic(enum migratetype migratetype) 609 { 610 return migratetype == MIGRATE_HIGHATOMIC; 611 } 612 613 static inline bool is_migrate_highatomic_page(struct page *page) 614 { 615 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC; 616 } 617 618 void setup_zone_pageset(struct zone *zone); 619 620 struct migration_target_control { 621 int nid; /* preferred node id */ 622 nodemask_t *nmask; 623 gfp_t gfp_mask; 624 }; 625 626 #endif /* __MM_INTERNAL_H */ 627