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