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/rmap.h> 14 #include <linux/tracepoint-defs.h> 15 16 struct folio_batch; 17 18 /* 19 * The set of flags that only affect watermark checking and reclaim 20 * behaviour. This is used by the MM to obey the caller constraints 21 * about IO, FS and watermark checking while ignoring placement 22 * hints such as HIGHMEM usage. 23 */ 24 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\ 25 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\ 26 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\ 27 __GFP_ATOMIC|__GFP_NOLOCKDEP) 28 29 /* The GFP flags allowed during early boot */ 30 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS)) 31 32 /* Control allocation cpuset and node placement constraints */ 33 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE) 34 35 /* Do not use these with a slab allocator */ 36 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK) 37 38 /* 39 * Different from WARN_ON_ONCE(), no warning will be issued 40 * when we specify __GFP_NOWARN. 41 */ 42 #define WARN_ON_ONCE_GFP(cond, gfp) ({ \ 43 static bool __section(".data.once") __warned; \ 44 int __ret_warn_once = !!(cond); \ 45 \ 46 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \ 47 __warned = true; \ 48 WARN_ON(1); \ 49 } \ 50 unlikely(__ret_warn_once); \ 51 }) 52 53 void page_writeback_init(void); 54 55 static inline void *folio_raw_mapping(struct folio *folio) 56 { 57 unsigned long mapping = (unsigned long)folio->mapping; 58 59 return (void *)(mapping & ~PAGE_MAPPING_FLAGS); 60 } 61 62 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio, 63 int nr_throttled); 64 static inline void acct_reclaim_writeback(struct folio *folio) 65 { 66 pg_data_t *pgdat = folio_pgdat(folio); 67 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled); 68 69 if (nr_throttled) 70 __acct_reclaim_writeback(pgdat, folio, nr_throttled); 71 } 72 73 static inline void wake_throttle_isolated(pg_data_t *pgdat) 74 { 75 wait_queue_head_t *wqh; 76 77 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED]; 78 if (waitqueue_active(wqh)) 79 wake_up(wqh); 80 } 81 82 vm_fault_t do_swap_page(struct vm_fault *vmf); 83 void folio_rotate_reclaimable(struct folio *folio); 84 bool __folio_end_writeback(struct folio *folio); 85 void deactivate_file_folio(struct folio *folio); 86 void folio_activate(struct folio *folio); 87 88 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma, 89 unsigned long floor, unsigned long ceiling); 90 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte); 91 92 struct zap_details; 93 void unmap_page_range(struct mmu_gather *tlb, 94 struct vm_area_struct *vma, 95 unsigned long addr, unsigned long end, 96 struct zap_details *details); 97 98 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *, 99 unsigned int order); 100 void force_page_cache_ra(struct readahead_control *, unsigned long nr); 101 static inline void force_page_cache_readahead(struct address_space *mapping, 102 struct file *file, pgoff_t index, unsigned long nr_to_read) 103 { 104 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index); 105 force_page_cache_ra(&ractl, nr_to_read); 106 } 107 108 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, 109 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 110 unsigned find_get_entries(struct address_space *mapping, pgoff_t start, 111 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 112 void filemap_free_folio(struct address_space *mapping, struct folio *folio); 113 int truncate_inode_folio(struct address_space *mapping, struct folio *folio); 114 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, 115 loff_t end); 116 long invalidate_inode_page(struct page *page); 117 unsigned long invalidate_mapping_pagevec(struct address_space *mapping, 118 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec); 119 120 /** 121 * folio_evictable - Test whether a folio is evictable. 122 * @folio: The folio to test. 123 * 124 * Test whether @folio is evictable -- i.e., should be placed on 125 * active/inactive lists vs unevictable list. 126 * 127 * Reasons folio might not be evictable: 128 * 1. folio's mapping marked unevictable 129 * 2. One of the pages in the folio is part of an mlocked VMA 130 */ 131 static inline bool folio_evictable(struct folio *folio) 132 { 133 bool ret; 134 135 /* Prevent address_space of inode and swap cache from being freed */ 136 rcu_read_lock(); 137 ret = !mapping_unevictable(folio_mapping(folio)) && 138 !folio_test_mlocked(folio); 139 rcu_read_unlock(); 140 return ret; 141 } 142 143 static inline bool page_evictable(struct page *page) 144 { 145 bool ret; 146 147 /* Prevent address_space of inode and swap cache from being freed */ 148 rcu_read_lock(); 149 ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); 150 rcu_read_unlock(); 151 return ret; 152 } 153 154 /* 155 * Turn a non-refcounted page (->_refcount == 0) into refcounted with 156 * a count of one. 157 */ 158 static inline void set_page_refcounted(struct page *page) 159 { 160 VM_BUG_ON_PAGE(PageTail(page), page); 161 VM_BUG_ON_PAGE(page_ref_count(page), page); 162 set_page_count(page, 1); 163 } 164 165 extern unsigned long highest_memmap_pfn; 166 167 /* 168 * Maximum number of reclaim retries without progress before the OOM 169 * killer is consider the only way forward. 170 */ 171 #define MAX_RECLAIM_RETRIES 16 172 173 /* 174 * in mm/early_ioremap.c 175 */ 176 pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr, 177 unsigned long size, pgprot_t prot); 178 179 /* 180 * in mm/vmscan.c: 181 */ 182 int isolate_lru_page(struct page *page); 183 int folio_isolate_lru(struct folio *folio); 184 void putback_lru_page(struct page *page); 185 void folio_putback_lru(struct folio *folio); 186 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason); 187 188 /* 189 * in mm/rmap.c: 190 */ 191 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address); 192 193 /* 194 * in mm/page_alloc.c 195 */ 196 197 /* 198 * Structure for holding the mostly immutable allocation parameters passed 199 * between functions involved in allocations, including the alloc_pages* 200 * family of functions. 201 * 202 * nodemask, migratetype and highest_zoneidx are initialized only once in 203 * __alloc_pages() and then never change. 204 * 205 * zonelist, preferred_zone and highest_zoneidx are set first in 206 * __alloc_pages() for the fast path, and might be later changed 207 * in __alloc_pages_slowpath(). All other functions pass the whole structure 208 * by a const pointer. 209 */ 210 struct alloc_context { 211 struct zonelist *zonelist; 212 nodemask_t *nodemask; 213 struct zoneref *preferred_zoneref; 214 int migratetype; 215 216 /* 217 * highest_zoneidx represents highest usable zone index of 218 * the allocation request. Due to the nature of the zone, 219 * memory on lower zone than the highest_zoneidx will be 220 * protected by lowmem_reserve[highest_zoneidx]. 221 * 222 * highest_zoneidx is also used by reclaim/compaction to limit 223 * the target zone since higher zone than this index cannot be 224 * usable for this allocation request. 225 */ 226 enum zone_type highest_zoneidx; 227 bool spread_dirty_pages; 228 }; 229 230 /* 231 * This function returns the order of a free page in the buddy system. In 232 * general, page_zone(page)->lock must be held by the caller to prevent the 233 * page from being allocated in parallel and returning garbage as the order. 234 * If a caller does not hold page_zone(page)->lock, it must guarantee that the 235 * page cannot be allocated or merged in parallel. Alternatively, it must 236 * handle invalid values gracefully, and use buddy_order_unsafe() below. 237 */ 238 static inline unsigned int buddy_order(struct page *page) 239 { 240 /* PageBuddy() must be checked by the caller */ 241 return page_private(page); 242 } 243 244 /* 245 * Like buddy_order(), but for callers who cannot afford to hold the zone lock. 246 * PageBuddy() should be checked first by the caller to minimize race window, 247 * and invalid values must be handled gracefully. 248 * 249 * READ_ONCE is used so that if the caller assigns the result into a local 250 * variable and e.g. tests it for valid range before using, the compiler cannot 251 * decide to remove the variable and inline the page_private(page) multiple 252 * times, potentially observing different values in the tests and the actual 253 * use of the result. 254 */ 255 #define buddy_order_unsafe(page) READ_ONCE(page_private(page)) 256 257 /* 258 * This function checks whether a page is free && is the buddy 259 * we can coalesce a page and its buddy if 260 * (a) the buddy is not in a hole (check before calling!) && 261 * (b) the buddy is in the buddy system && 262 * (c) a page and its buddy have the same order && 263 * (d) a page and its buddy are in the same zone. 264 * 265 * For recording whether a page is in the buddy system, we set PageBuddy. 266 * Setting, clearing, and testing PageBuddy is serialized by zone->lock. 267 * 268 * For recording page's order, we use page_private(page). 269 */ 270 static inline bool page_is_buddy(struct page *page, struct page *buddy, 271 unsigned int order) 272 { 273 if (!page_is_guard(buddy) && !PageBuddy(buddy)) 274 return false; 275 276 if (buddy_order(buddy) != order) 277 return false; 278 279 /* 280 * zone check is done late to avoid uselessly calculating 281 * zone/node ids for pages that could never merge. 282 */ 283 if (page_zone_id(page) != page_zone_id(buddy)) 284 return false; 285 286 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); 287 288 return true; 289 } 290 291 /* 292 * Locate the struct page for both the matching buddy in our 293 * pair (buddy1) and the combined O(n+1) page they form (page). 294 * 295 * 1) Any buddy B1 will have an order O twin B2 which satisfies 296 * the following equation: 297 * B2 = B1 ^ (1 << O) 298 * For example, if the starting buddy (buddy2) is #8 its order 299 * 1 buddy is #10: 300 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 301 * 302 * 2) Any buddy B will have an order O+1 parent P which 303 * satisfies the following equation: 304 * P = B & ~(1 << O) 305 * 306 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER 307 */ 308 static inline unsigned long 309 __find_buddy_pfn(unsigned long page_pfn, unsigned int order) 310 { 311 return page_pfn ^ (1 << order); 312 } 313 314 /* 315 * Find the buddy of @page and validate it. 316 * @page: The input page 317 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the 318 * function is used in the performance-critical __free_one_page(). 319 * @order: The order of the page 320 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to 321 * page_to_pfn(). 322 * 323 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is 324 * not the same as @page. The validation is necessary before use it. 325 * 326 * Return: the found buddy page or NULL if not found. 327 */ 328 static inline struct page *find_buddy_page_pfn(struct page *page, 329 unsigned long pfn, unsigned int order, unsigned long *buddy_pfn) 330 { 331 unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order); 332 struct page *buddy; 333 334 buddy = page + (__buddy_pfn - pfn); 335 if (buddy_pfn) 336 *buddy_pfn = __buddy_pfn; 337 338 if (page_is_buddy(page, buddy, order)) 339 return buddy; 340 return NULL; 341 } 342 343 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn, 344 unsigned long end_pfn, struct zone *zone); 345 346 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn, 347 unsigned long end_pfn, struct zone *zone) 348 { 349 if (zone->contiguous) 350 return pfn_to_page(start_pfn); 351 352 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone); 353 } 354 355 extern int __isolate_free_page(struct page *page, unsigned int order); 356 extern void __putback_isolated_page(struct page *page, unsigned int order, 357 int mt); 358 extern void memblock_free_pages(struct page *page, unsigned long pfn, 359 unsigned int order); 360 extern void __free_pages_core(struct page *page, unsigned int order); 361 extern void prep_compound_page(struct page *page, unsigned int order); 362 extern void post_alloc_hook(struct page *page, unsigned int order, 363 gfp_t gfp_flags); 364 extern int user_min_free_kbytes; 365 366 extern void free_unref_page(struct page *page, unsigned int order); 367 extern void free_unref_page_list(struct list_head *list); 368 369 extern void zone_pcp_update(struct zone *zone, int cpu_online); 370 extern void zone_pcp_reset(struct zone *zone); 371 extern void zone_pcp_disable(struct zone *zone); 372 extern void zone_pcp_enable(struct zone *zone); 373 374 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align, 375 phys_addr_t min_addr, 376 int nid, bool exact_nid); 377 378 int split_free_page(struct page *free_page, 379 unsigned int order, unsigned long split_pfn_offset); 380 381 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 382 383 /* 384 * in mm/compaction.c 385 */ 386 /* 387 * compact_control is used to track pages being migrated and the free pages 388 * they are being migrated to during memory compaction. The free_pfn starts 389 * at the end of a zone and migrate_pfn begins at the start. Movable pages 390 * are moved to the end of a zone during a compaction run and the run 391 * completes when free_pfn <= migrate_pfn 392 */ 393 struct compact_control { 394 struct list_head freepages; /* List of free pages to migrate to */ 395 struct list_head migratepages; /* List of pages being migrated */ 396 unsigned int nr_freepages; /* Number of isolated free pages */ 397 unsigned int nr_migratepages; /* Number of pages to migrate */ 398 unsigned long free_pfn; /* isolate_freepages search base */ 399 /* 400 * Acts as an in/out parameter to page isolation for migration. 401 * isolate_migratepages uses it as a search base. 402 * isolate_migratepages_block will update the value to the next pfn 403 * after the last isolated one. 404 */ 405 unsigned long migrate_pfn; 406 unsigned long fast_start_pfn; /* a pfn to start linear scan from */ 407 struct zone *zone; 408 unsigned long total_migrate_scanned; 409 unsigned long total_free_scanned; 410 unsigned short fast_search_fail;/* failures to use free list searches */ 411 short search_order; /* order to start a fast search at */ 412 const gfp_t gfp_mask; /* gfp mask of a direct compactor */ 413 int order; /* order a direct compactor needs */ 414 int migratetype; /* migratetype of direct compactor */ 415 const unsigned int alloc_flags; /* alloc flags of a direct compactor */ 416 const int highest_zoneidx; /* zone index of a direct compactor */ 417 enum migrate_mode mode; /* Async or sync migration mode */ 418 bool ignore_skip_hint; /* Scan blocks even if marked skip */ 419 bool no_set_skip_hint; /* Don't mark blocks for skipping */ 420 bool ignore_block_suitable; /* Scan blocks considered unsuitable */ 421 bool direct_compaction; /* False from kcompactd or /proc/... */ 422 bool proactive_compaction; /* kcompactd proactive compaction */ 423 bool whole_zone; /* Whole zone should/has been scanned */ 424 bool contended; /* Signal lock contention */ 425 bool rescan; /* Rescanning the same pageblock */ 426 bool alloc_contig; /* alloc_contig_range allocation */ 427 }; 428 429 /* 430 * Used in direct compaction when a page should be taken from the freelists 431 * immediately when one is created during the free path. 432 */ 433 struct capture_control { 434 struct compact_control *cc; 435 struct page *page; 436 }; 437 438 unsigned long 439 isolate_freepages_range(struct compact_control *cc, 440 unsigned long start_pfn, unsigned long end_pfn); 441 int 442 isolate_migratepages_range(struct compact_control *cc, 443 unsigned long low_pfn, unsigned long end_pfn); 444 445 int __alloc_contig_migrate_range(struct compact_control *cc, 446 unsigned long start, unsigned long end); 447 #endif 448 int find_suitable_fallback(struct free_area *area, unsigned int order, 449 int migratetype, bool only_stealable, bool *can_steal); 450 451 /* 452 * These three helpers classifies VMAs for virtual memory accounting. 453 */ 454 455 /* 456 * Executable code area - executable, not writable, not stack 457 */ 458 static inline bool is_exec_mapping(vm_flags_t flags) 459 { 460 return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC; 461 } 462 463 /* 464 * Stack area - automatically grows in one direction 465 * 466 * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous: 467 * do_mmap() forbids all other combinations. 468 */ 469 static inline bool is_stack_mapping(vm_flags_t flags) 470 { 471 return (flags & VM_STACK) == VM_STACK; 472 } 473 474 /* 475 * Data area - private, writable, not stack 476 */ 477 static inline bool is_data_mapping(vm_flags_t flags) 478 { 479 return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE; 480 } 481 482 /* mm/util.c */ 483 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 484 struct vm_area_struct *prev); 485 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma); 486 struct anon_vma *folio_anon_vma(struct folio *folio); 487 488 #ifdef CONFIG_MMU 489 void unmap_mapping_folio(struct folio *folio); 490 extern long populate_vma_page_range(struct vm_area_struct *vma, 491 unsigned long start, unsigned long end, int *locked); 492 extern long faultin_vma_page_range(struct vm_area_struct *vma, 493 unsigned long start, unsigned long end, 494 bool write, int *locked); 495 extern int mlock_future_check(struct mm_struct *mm, unsigned long flags, 496 unsigned long len); 497 /* 498 * mlock_vma_page() and munlock_vma_page(): 499 * should be called with vma's mmap_lock held for read or write, 500 * under page table lock for the pte/pmd being added or removed. 501 * 502 * mlock is usually called at the end of page_add_*_rmap(), 503 * munlock at the end of page_remove_rmap(); but new anon 504 * pages are managed by lru_cache_add_inactive_or_unevictable() 505 * calling mlock_new_page(). 506 * 507 * @compound is used to include pmd mappings of THPs, but filter out 508 * pte mappings of THPs, which cannot be consistently counted: a pte 509 * mapping of the THP head cannot be distinguished by the page alone. 510 */ 511 void mlock_folio(struct folio *folio); 512 static inline void mlock_vma_folio(struct folio *folio, 513 struct vm_area_struct *vma, bool compound) 514 { 515 /* 516 * The VM_SPECIAL check here serves two purposes. 517 * 1) VM_IO check prevents migration from double-counting during mlock. 518 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED 519 * is never left set on a VM_SPECIAL vma, there is an interval while 520 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may 521 * still be set while VM_SPECIAL bits are added: so ignore it then. 522 */ 523 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) && 524 (compound || !folio_test_large(folio))) 525 mlock_folio(folio); 526 } 527 528 static inline void mlock_vma_page(struct page *page, 529 struct vm_area_struct *vma, bool compound) 530 { 531 mlock_vma_folio(page_folio(page), vma, compound); 532 } 533 534 void munlock_page(struct page *page); 535 static inline void munlock_vma_page(struct page *page, 536 struct vm_area_struct *vma, bool compound) 537 { 538 if (unlikely(vma->vm_flags & VM_LOCKED) && 539 (compound || !PageTransCompound(page))) 540 munlock_page(page); 541 } 542 void mlock_new_page(struct page *page); 543 bool need_mlock_page_drain(int cpu); 544 void mlock_page_drain_local(void); 545 void mlock_page_drain_remote(int cpu); 546 547 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma); 548 549 /* 550 * Return the start of user virtual address at the specific offset within 551 * a vma. 552 */ 553 static inline unsigned long 554 vma_pgoff_address(pgoff_t pgoff, unsigned long nr_pages, 555 struct vm_area_struct *vma) 556 { 557 unsigned long address; 558 559 if (pgoff >= vma->vm_pgoff) { 560 address = vma->vm_start + 561 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 562 /* Check for address beyond vma (or wrapped through 0?) */ 563 if (address < vma->vm_start || address >= vma->vm_end) 564 address = -EFAULT; 565 } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) { 566 /* Test above avoids possibility of wrap to 0 on 32-bit */ 567 address = vma->vm_start; 568 } else { 569 address = -EFAULT; 570 } 571 return address; 572 } 573 574 /* 575 * Return the start of user virtual address of a page within a vma. 576 * Returns -EFAULT if all of the page is outside the range of vma. 577 * If page is a compound head, the entire compound page is considered. 578 */ 579 static inline unsigned long 580 vma_address(struct page *page, struct vm_area_struct *vma) 581 { 582 VM_BUG_ON_PAGE(PageKsm(page), page); /* KSM page->index unusable */ 583 return vma_pgoff_address(page_to_pgoff(page), compound_nr(page), vma); 584 } 585 586 /* 587 * Then at what user virtual address will none of the range be found in vma? 588 * Assumes that vma_address() already returned a good starting address. 589 */ 590 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw) 591 { 592 struct vm_area_struct *vma = pvmw->vma; 593 pgoff_t pgoff; 594 unsigned long address; 595 596 /* Common case, plus ->pgoff is invalid for KSM */ 597 if (pvmw->nr_pages == 1) 598 return pvmw->address + PAGE_SIZE; 599 600 pgoff = pvmw->pgoff + pvmw->nr_pages; 601 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 602 /* Check for address beyond vma (or wrapped through 0?) */ 603 if (address < vma->vm_start || address > vma->vm_end) 604 address = vma->vm_end; 605 return address; 606 } 607 608 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf, 609 struct file *fpin) 610 { 611 int flags = vmf->flags; 612 613 if (fpin) 614 return fpin; 615 616 /* 617 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or 618 * anything, so we only pin the file and drop the mmap_lock if only 619 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt. 620 */ 621 if (fault_flag_allow_retry_first(flags) && 622 !(flags & FAULT_FLAG_RETRY_NOWAIT)) { 623 fpin = get_file(vmf->vma->vm_file); 624 mmap_read_unlock(vmf->vma->vm_mm); 625 } 626 return fpin; 627 } 628 #else /* !CONFIG_MMU */ 629 static inline void unmap_mapping_folio(struct folio *folio) { } 630 static inline void mlock_vma_page(struct page *page, 631 struct vm_area_struct *vma, bool compound) { } 632 static inline void munlock_vma_page(struct page *page, 633 struct vm_area_struct *vma, bool compound) { } 634 static inline void mlock_new_page(struct page *page) { } 635 static inline bool need_mlock_page_drain(int cpu) { return false; } 636 static inline void mlock_page_drain_local(void) { } 637 static inline void mlock_page_drain_remote(int cpu) { } 638 static inline void vunmap_range_noflush(unsigned long start, unsigned long end) 639 { 640 } 641 #endif /* !CONFIG_MMU */ 642 643 /* 644 * Return the mem_map entry representing the 'offset' subpage within 645 * the maximally aligned gigantic page 'base'. Handle any discontiguity 646 * in the mem_map at MAX_ORDER_NR_PAGES boundaries. 647 */ 648 static inline struct page *mem_map_offset(struct page *base, int offset) 649 { 650 if (unlikely(offset >= MAX_ORDER_NR_PAGES)) 651 return nth_page(base, offset); 652 return base + offset; 653 } 654 655 /* 656 * Iterator over all subpages within the maximally aligned gigantic 657 * page 'base'. Handle any discontiguity in the mem_map. 658 */ 659 static inline struct page *mem_map_next(struct page *iter, 660 struct page *base, int offset) 661 { 662 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) { 663 unsigned long pfn = page_to_pfn(base) + offset; 664 if (!pfn_valid(pfn)) 665 return NULL; 666 return pfn_to_page(pfn); 667 } 668 return iter + 1; 669 } 670 671 /* Memory initialisation debug and verification */ 672 enum mminit_level { 673 MMINIT_WARNING, 674 MMINIT_VERIFY, 675 MMINIT_TRACE 676 }; 677 678 #ifdef CONFIG_DEBUG_MEMORY_INIT 679 680 extern int mminit_loglevel; 681 682 #define mminit_dprintk(level, prefix, fmt, arg...) \ 683 do { \ 684 if (level < mminit_loglevel) { \ 685 if (level <= MMINIT_WARNING) \ 686 pr_warn("mminit::" prefix " " fmt, ##arg); \ 687 else \ 688 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \ 689 } \ 690 } while (0) 691 692 extern void mminit_verify_pageflags_layout(void); 693 extern void mminit_verify_zonelist(void); 694 #else 695 696 static inline void mminit_dprintk(enum mminit_level level, 697 const char *prefix, const char *fmt, ...) 698 { 699 } 700 701 static inline void mminit_verify_pageflags_layout(void) 702 { 703 } 704 705 static inline void mminit_verify_zonelist(void) 706 { 707 } 708 #endif /* CONFIG_DEBUG_MEMORY_INIT */ 709 710 #define NODE_RECLAIM_NOSCAN -2 711 #define NODE_RECLAIM_FULL -1 712 #define NODE_RECLAIM_SOME 0 713 #define NODE_RECLAIM_SUCCESS 1 714 715 #ifdef CONFIG_NUMA 716 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int); 717 extern int find_next_best_node(int node, nodemask_t *used_node_mask); 718 #else 719 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask, 720 unsigned int order) 721 { 722 return NODE_RECLAIM_NOSCAN; 723 } 724 static inline int find_next_best_node(int node, nodemask_t *used_node_mask) 725 { 726 return NUMA_NO_NODE; 727 } 728 #endif 729 730 /* 731 * mm/memory-failure.c 732 */ 733 extern int hwpoison_filter(struct page *p); 734 735 extern u32 hwpoison_filter_dev_major; 736 extern u32 hwpoison_filter_dev_minor; 737 extern u64 hwpoison_filter_flags_mask; 738 extern u64 hwpoison_filter_flags_value; 739 extern u64 hwpoison_filter_memcg; 740 extern u32 hwpoison_filter_enable; 741 742 #ifdef CONFIG_MEMORY_FAILURE 743 void clear_hwpoisoned_pages(struct page *memmap, int nr_pages); 744 #else 745 static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) 746 { 747 } 748 #endif 749 750 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long, 751 unsigned long, unsigned long, 752 unsigned long, unsigned long); 753 754 extern void set_pageblock_order(void); 755 unsigned int reclaim_clean_pages_from_list(struct zone *zone, 756 struct list_head *page_list); 757 /* The ALLOC_WMARK bits are used as an index to zone->watermark */ 758 #define ALLOC_WMARK_MIN WMARK_MIN 759 #define ALLOC_WMARK_LOW WMARK_LOW 760 #define ALLOC_WMARK_HIGH WMARK_HIGH 761 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ 762 763 /* Mask to get the watermark bits */ 764 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) 765 766 /* 767 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we 768 * cannot assume a reduced access to memory reserves is sufficient for 769 * !MMU 770 */ 771 #ifdef CONFIG_MMU 772 #define ALLOC_OOM 0x08 773 #else 774 #define ALLOC_OOM ALLOC_NO_WATERMARKS 775 #endif 776 777 #define ALLOC_HARDER 0x10 /* try to alloc harder */ 778 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ 779 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ 780 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */ 781 #ifdef CONFIG_ZONE_DMA32 782 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */ 783 #else 784 #define ALLOC_NOFRAGMENT 0x0 785 #endif 786 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */ 787 788 enum ttu_flags; 789 struct tlbflush_unmap_batch; 790 791 792 /* 793 * only for MM internal work items which do not depend on 794 * any allocations or locks which might depend on allocations 795 */ 796 extern struct workqueue_struct *mm_percpu_wq; 797 798 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 799 void try_to_unmap_flush(void); 800 void try_to_unmap_flush_dirty(void); 801 void flush_tlb_batched_pending(struct mm_struct *mm); 802 #else 803 static inline void try_to_unmap_flush(void) 804 { 805 } 806 static inline void try_to_unmap_flush_dirty(void) 807 { 808 } 809 static inline void flush_tlb_batched_pending(struct mm_struct *mm) 810 { 811 } 812 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ 813 814 extern const struct trace_print_flags pageflag_names[]; 815 extern const struct trace_print_flags vmaflag_names[]; 816 extern const struct trace_print_flags gfpflag_names[]; 817 818 static inline bool is_migrate_highatomic(enum migratetype migratetype) 819 { 820 return migratetype == MIGRATE_HIGHATOMIC; 821 } 822 823 static inline bool is_migrate_highatomic_page(struct page *page) 824 { 825 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC; 826 } 827 828 void setup_zone_pageset(struct zone *zone); 829 830 struct migration_target_control { 831 int nid; /* preferred node id */ 832 nodemask_t *nmask; 833 gfp_t gfp_mask; 834 }; 835 836 /* 837 * mm/vmalloc.c 838 */ 839 #ifdef CONFIG_MMU 840 int vmap_pages_range_noflush(unsigned long addr, unsigned long end, 841 pgprot_t prot, struct page **pages, unsigned int page_shift); 842 #else 843 static inline 844 int vmap_pages_range_noflush(unsigned long addr, unsigned long end, 845 pgprot_t prot, struct page **pages, unsigned int page_shift) 846 { 847 return -EINVAL; 848 } 849 #endif 850 851 void vunmap_range_noflush(unsigned long start, unsigned long end); 852 853 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 854 unsigned long addr, int page_nid, int *flags); 855 856 void free_zone_device_page(struct page *page); 857 int migrate_device_coherent_page(struct page *page); 858 859 /* 860 * mm/gup.c 861 */ 862 struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags); 863 864 DECLARE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); 865 866 extern bool mirrored_kernelcore; 867 868 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma) 869 { 870 /* 871 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty 872 * enablements, because when without soft-dirty being compiled in, 873 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY) 874 * will be constantly true. 875 */ 876 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 877 return false; 878 879 /* 880 * Soft-dirty is kind of special: its tracking is enabled when the 881 * vma flags not set. 882 */ 883 return !(vma->vm_flags & VM_SOFTDIRTY); 884 } 885 886 #endif /* __MM_INTERNAL_H */ 887