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/khugepaged.h> 12 #include <linux/mm.h> 13 #include <linux/mm_inline.h> 14 #include <linux/pagemap.h> 15 #include <linux/rmap.h> 16 #include <linux/swap.h> 17 #include <linux/swapops.h> 18 #include <linux/swap_cgroup.h> 19 #include <linux/tracepoint-defs.h> 20 21 /* Internal core VMA manipulation functions. */ 22 #include "vma.h" 23 24 struct folio_batch; 25 26 /* 27 * The set of flags that only affect watermark checking and reclaim 28 * behaviour. This is used by the MM to obey the caller constraints 29 * about IO, FS and watermark checking while ignoring placement 30 * hints such as HIGHMEM usage. 31 */ 32 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\ 33 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\ 34 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\ 35 __GFP_NOLOCKDEP) 36 37 /* The GFP flags allowed during early boot */ 38 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS)) 39 40 /* Control allocation cpuset and node placement constraints */ 41 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE) 42 43 /* Do not use these with a slab allocator */ 44 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK) 45 46 /* 47 * Different from WARN_ON_ONCE(), no warning will be issued 48 * when we specify __GFP_NOWARN. 49 */ 50 #define WARN_ON_ONCE_GFP(cond, gfp) ({ \ 51 static bool __section(".data.once") __warned; \ 52 int __ret_warn_once = !!(cond); \ 53 \ 54 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \ 55 __warned = true; \ 56 WARN_ON(1); \ 57 } \ 58 unlikely(__ret_warn_once); \ 59 }) 60 61 void page_writeback_init(void); 62 63 /* 64 * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages, 65 * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit 66 * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE). Hugetlb currently 67 * leaves nr_pages_mapped at 0, but avoid surprise if it participates later. 68 */ 69 #define ENTIRELY_MAPPED 0x800000 70 #define FOLIO_PAGES_MAPPED (ENTIRELY_MAPPED - 1) 71 72 /* 73 * Flags passed to __show_mem() and show_free_areas() to suppress output in 74 * various contexts. 75 */ 76 #define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */ 77 78 /* 79 * How many individual pages have an elevated _mapcount. Excludes 80 * the folio's entire_mapcount. 81 * 82 * Don't use this function outside of debugging code. 83 */ 84 static inline int folio_nr_pages_mapped(const struct folio *folio) 85 { 86 return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED; 87 } 88 89 /* 90 * Retrieve the first entry of a folio based on a provided entry within the 91 * folio. We cannot rely on folio->swap as there is no guarantee that it has 92 * been initialized. Used for calling arch_swap_restore() 93 */ 94 static inline swp_entry_t folio_swap(swp_entry_t entry, 95 const struct folio *folio) 96 { 97 swp_entry_t swap = { 98 .val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)), 99 }; 100 101 return swap; 102 } 103 104 static inline void *folio_raw_mapping(const struct folio *folio) 105 { 106 unsigned long mapping = (unsigned long)folio->mapping; 107 108 return (void *)(mapping & ~PAGE_MAPPING_FLAGS); 109 } 110 111 /* 112 * This is a file-backed mapping, and is about to be memory mapped - invoke its 113 * mmap hook and safely handle error conditions. On error, VMA hooks will be 114 * mutated. 115 * 116 * @file: File which backs the mapping. 117 * @vma: VMA which we are mapping. 118 * 119 * Returns: 0 if success, error otherwise. 120 */ 121 static inline int mmap_file(struct file *file, struct vm_area_struct *vma) 122 { 123 int err = call_mmap(file, vma); 124 125 if (likely(!err)) 126 return 0; 127 128 /* 129 * OK, we tried to call the file hook for mmap(), but an error 130 * arose. The mapping is in an inconsistent state and we most not invoke 131 * any further hooks on it. 132 */ 133 vma->vm_ops = &vma_dummy_vm_ops; 134 135 return err; 136 } 137 138 /* 139 * If the VMA has a close hook then close it, and since closing it might leave 140 * it in an inconsistent state which makes the use of any hooks suspect, clear 141 * them down by installing dummy empty hooks. 142 */ 143 static inline void vma_close(struct vm_area_struct *vma) 144 { 145 if (vma->vm_ops && vma->vm_ops->close) { 146 vma->vm_ops->close(vma); 147 148 /* 149 * The mapping is in an inconsistent state, and no further hooks 150 * may be invoked upon it. 151 */ 152 vma->vm_ops = &vma_dummy_vm_ops; 153 } 154 } 155 156 #ifdef CONFIG_MMU 157 158 /* Flags for folio_pte_batch(). */ 159 typedef int __bitwise fpb_t; 160 161 /* Compare PTEs after pte_mkclean(), ignoring the dirty bit. */ 162 #define FPB_IGNORE_DIRTY ((__force fpb_t)BIT(0)) 163 164 /* Compare PTEs after pte_clear_soft_dirty(), ignoring the soft-dirty bit. */ 165 #define FPB_IGNORE_SOFT_DIRTY ((__force fpb_t)BIT(1)) 166 167 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags) 168 { 169 if (flags & FPB_IGNORE_DIRTY) 170 pte = pte_mkclean(pte); 171 if (likely(flags & FPB_IGNORE_SOFT_DIRTY)) 172 pte = pte_clear_soft_dirty(pte); 173 return pte_wrprotect(pte_mkold(pte)); 174 } 175 176 /** 177 * folio_pte_batch - detect a PTE batch for a large folio 178 * @folio: The large folio to detect a PTE batch for. 179 * @addr: The user virtual address the first page is mapped at. 180 * @start_ptep: Page table pointer for the first entry. 181 * @pte: Page table entry for the first page. 182 * @max_nr: The maximum number of table entries to consider. 183 * @flags: Flags to modify the PTE batch semantics. 184 * @any_writable: Optional pointer to indicate whether any entry except the 185 * first one is writable. 186 * @any_young: Optional pointer to indicate whether any entry except the 187 * first one is young. 188 * @any_dirty: Optional pointer to indicate whether any entry except the 189 * first one is dirty. 190 * 191 * Detect a PTE batch: consecutive (present) PTEs that map consecutive 192 * pages of the same large folio. 193 * 194 * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN, 195 * the accessed bit, writable bit, dirty bit (with FPB_IGNORE_DIRTY) and 196 * soft-dirty bit (with FPB_IGNORE_SOFT_DIRTY). 197 * 198 * start_ptep must map any page of the folio. max_nr must be at least one and 199 * must be limited by the caller so scanning cannot exceed a single page table. 200 * 201 * Return: the number of table entries in the batch. 202 */ 203 static inline int folio_pte_batch(struct folio *folio, unsigned long addr, 204 pte_t *start_ptep, pte_t pte, int max_nr, fpb_t flags, 205 bool *any_writable, bool *any_young, bool *any_dirty) 206 { 207 unsigned long folio_end_pfn = folio_pfn(folio) + folio_nr_pages(folio); 208 const pte_t *end_ptep = start_ptep + max_nr; 209 pte_t expected_pte, *ptep; 210 bool writable, young, dirty; 211 int nr; 212 213 if (any_writable) 214 *any_writable = false; 215 if (any_young) 216 *any_young = false; 217 if (any_dirty) 218 *any_dirty = false; 219 220 VM_WARN_ON_FOLIO(!pte_present(pte), folio); 221 VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio); 222 VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio); 223 224 nr = pte_batch_hint(start_ptep, pte); 225 expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags); 226 ptep = start_ptep + nr; 227 228 while (ptep < end_ptep) { 229 pte = ptep_get(ptep); 230 if (any_writable) 231 writable = !!pte_write(pte); 232 if (any_young) 233 young = !!pte_young(pte); 234 if (any_dirty) 235 dirty = !!pte_dirty(pte); 236 pte = __pte_batch_clear_ignored(pte, flags); 237 238 if (!pte_same(pte, expected_pte)) 239 break; 240 241 /* 242 * Stop immediately once we reached the end of the folio. In 243 * corner cases the next PFN might fall into a different 244 * folio. 245 */ 246 if (pte_pfn(pte) >= folio_end_pfn) 247 break; 248 249 if (any_writable) 250 *any_writable |= writable; 251 if (any_young) 252 *any_young |= young; 253 if (any_dirty) 254 *any_dirty |= dirty; 255 256 nr = pte_batch_hint(ptep, pte); 257 expected_pte = pte_advance_pfn(expected_pte, nr); 258 ptep += nr; 259 } 260 261 return min(ptep - start_ptep, max_nr); 262 } 263 264 /** 265 * pte_move_swp_offset - Move the swap entry offset field of a swap pte 266 * forward or backward by delta 267 * @pte: The initial pte state; is_swap_pte(pte) must be true and 268 * non_swap_entry() must be false. 269 * @delta: The direction and the offset we are moving; forward if delta 270 * is positive; backward if delta is negative 271 * 272 * Moves the swap offset, while maintaining all other fields, including 273 * swap type, and any swp pte bits. The resulting pte is returned. 274 */ 275 static inline pte_t pte_move_swp_offset(pte_t pte, long delta) 276 { 277 swp_entry_t entry = pte_to_swp_entry(pte); 278 pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry), 279 (swp_offset(entry) + delta))); 280 281 if (pte_swp_soft_dirty(pte)) 282 new = pte_swp_mksoft_dirty(new); 283 if (pte_swp_exclusive(pte)) 284 new = pte_swp_mkexclusive(new); 285 if (pte_swp_uffd_wp(pte)) 286 new = pte_swp_mkuffd_wp(new); 287 288 return new; 289 } 290 291 292 /** 293 * pte_next_swp_offset - Increment the swap entry offset field of a swap pte. 294 * @pte: The initial pte state; is_swap_pte(pte) must be true and 295 * non_swap_entry() must be false. 296 * 297 * Increments the swap offset, while maintaining all other fields, including 298 * swap type, and any swp pte bits. The resulting pte is returned. 299 */ 300 static inline pte_t pte_next_swp_offset(pte_t pte) 301 { 302 return pte_move_swp_offset(pte, 1); 303 } 304 305 /** 306 * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries 307 * @start_ptep: Page table pointer for the first entry. 308 * @max_nr: The maximum number of table entries to consider. 309 * @pte: Page table entry for the first entry. 310 * 311 * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs 312 * containing swap entries all with consecutive offsets and targeting the same 313 * swap type, all with matching swp pte bits. 314 * 315 * max_nr must be at least one and must be limited by the caller so scanning 316 * cannot exceed a single page table. 317 * 318 * Return: the number of table entries in the batch. 319 */ 320 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte) 321 { 322 pte_t expected_pte = pte_next_swp_offset(pte); 323 const pte_t *end_ptep = start_ptep + max_nr; 324 swp_entry_t entry = pte_to_swp_entry(pte); 325 pte_t *ptep = start_ptep + 1; 326 unsigned short cgroup_id; 327 328 VM_WARN_ON(max_nr < 1); 329 VM_WARN_ON(!is_swap_pte(pte)); 330 VM_WARN_ON(non_swap_entry(entry)); 331 332 cgroup_id = lookup_swap_cgroup_id(entry); 333 while (ptep < end_ptep) { 334 pte = ptep_get(ptep); 335 336 if (!pte_same(pte, expected_pte)) 337 break; 338 if (lookup_swap_cgroup_id(pte_to_swp_entry(pte)) != cgroup_id) 339 break; 340 expected_pte = pte_next_swp_offset(expected_pte); 341 ptep++; 342 } 343 344 return ptep - start_ptep; 345 } 346 #endif /* CONFIG_MMU */ 347 348 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio, 349 int nr_throttled); 350 static inline void acct_reclaim_writeback(struct folio *folio) 351 { 352 pg_data_t *pgdat = folio_pgdat(folio); 353 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled); 354 355 if (nr_throttled) 356 __acct_reclaim_writeback(pgdat, folio, nr_throttled); 357 } 358 359 static inline void wake_throttle_isolated(pg_data_t *pgdat) 360 { 361 wait_queue_head_t *wqh; 362 363 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED]; 364 if (waitqueue_active(wqh)) 365 wake_up(wqh); 366 } 367 368 vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf); 369 static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf) 370 { 371 vm_fault_t ret = __vmf_anon_prepare(vmf); 372 373 if (unlikely(ret & VM_FAULT_RETRY)) 374 vma_end_read(vmf->vma); 375 return ret; 376 } 377 378 vm_fault_t do_swap_page(struct vm_fault *vmf); 379 void folio_rotate_reclaimable(struct folio *folio); 380 bool __folio_end_writeback(struct folio *folio); 381 void deactivate_file_folio(struct folio *folio); 382 void folio_activate(struct folio *folio); 383 384 void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas, 385 struct vm_area_struct *start_vma, unsigned long floor, 386 unsigned long ceiling, bool mm_wr_locked); 387 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte); 388 389 struct zap_details; 390 void unmap_page_range(struct mmu_gather *tlb, 391 struct vm_area_struct *vma, 392 unsigned long addr, unsigned long end, 393 struct zap_details *details); 394 395 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *, 396 unsigned int order); 397 void force_page_cache_ra(struct readahead_control *, unsigned long nr); 398 static inline void force_page_cache_readahead(struct address_space *mapping, 399 struct file *file, pgoff_t index, unsigned long nr_to_read) 400 { 401 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index); 402 force_page_cache_ra(&ractl, nr_to_read); 403 } 404 405 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start, 406 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 407 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start, 408 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 409 void filemap_free_folio(struct address_space *mapping, struct folio *folio); 410 int truncate_inode_folio(struct address_space *mapping, struct folio *folio); 411 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, 412 loff_t end); 413 long mapping_evict_folio(struct address_space *mapping, struct folio *folio); 414 unsigned long mapping_try_invalidate(struct address_space *mapping, 415 pgoff_t start, pgoff_t end, unsigned long *nr_failed); 416 417 /** 418 * folio_evictable - Test whether a folio is evictable. 419 * @folio: The folio to test. 420 * 421 * Test whether @folio is evictable -- i.e., should be placed on 422 * active/inactive lists vs unevictable list. 423 * 424 * Reasons folio might not be evictable: 425 * 1. folio's mapping marked unevictable 426 * 2. One of the pages in the folio is part of an mlocked VMA 427 */ 428 static inline bool folio_evictable(struct folio *folio) 429 { 430 bool ret; 431 432 /* Prevent address_space of inode and swap cache from being freed */ 433 rcu_read_lock(); 434 ret = !mapping_unevictable(folio_mapping(folio)) && 435 !folio_test_mlocked(folio); 436 rcu_read_unlock(); 437 return ret; 438 } 439 440 /* 441 * Turn a non-refcounted page (->_refcount == 0) into refcounted with 442 * a count of one. 443 */ 444 static inline void set_page_refcounted(struct page *page) 445 { 446 VM_BUG_ON_PAGE(PageTail(page), page); 447 VM_BUG_ON_PAGE(page_ref_count(page), page); 448 set_page_count(page, 1); 449 } 450 451 /* 452 * Return true if a folio needs ->release_folio() calling upon it. 453 */ 454 static inline bool folio_needs_release(struct folio *folio) 455 { 456 struct address_space *mapping = folio_mapping(folio); 457 458 return folio_has_private(folio) || 459 (mapping && mapping_release_always(mapping)); 460 } 461 462 extern unsigned long highest_memmap_pfn; 463 464 /* 465 * Maximum number of reclaim retries without progress before the OOM 466 * killer is consider the only way forward. 467 */ 468 #define MAX_RECLAIM_RETRIES 16 469 470 /* 471 * in mm/vmscan.c: 472 */ 473 bool folio_isolate_lru(struct folio *folio); 474 void folio_putback_lru(struct folio *folio); 475 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason); 476 477 /* 478 * in mm/rmap.c: 479 */ 480 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address); 481 482 /* 483 * in mm/page_alloc.c 484 */ 485 #define K(x) ((x) << (PAGE_SHIFT-10)) 486 487 extern char * const zone_names[MAX_NR_ZONES]; 488 489 /* perform sanity checks on struct pages being allocated or freed */ 490 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled); 491 492 extern int min_free_kbytes; 493 494 void setup_per_zone_wmarks(void); 495 void calculate_min_free_kbytes(void); 496 int __meminit init_per_zone_wmark_min(void); 497 void page_alloc_sysctl_init(void); 498 499 /* 500 * Structure for holding the mostly immutable allocation parameters passed 501 * between functions involved in allocations, including the alloc_pages* 502 * family of functions. 503 * 504 * nodemask, migratetype and highest_zoneidx are initialized only once in 505 * __alloc_pages() and then never change. 506 * 507 * zonelist, preferred_zone and highest_zoneidx are set first in 508 * __alloc_pages() for the fast path, and might be later changed 509 * in __alloc_pages_slowpath(). All other functions pass the whole structure 510 * by a const pointer. 511 */ 512 struct alloc_context { 513 struct zonelist *zonelist; 514 nodemask_t *nodemask; 515 struct zoneref *preferred_zoneref; 516 int migratetype; 517 518 /* 519 * highest_zoneidx represents highest usable zone index of 520 * the allocation request. Due to the nature of the zone, 521 * memory on lower zone than the highest_zoneidx will be 522 * protected by lowmem_reserve[highest_zoneidx]. 523 * 524 * highest_zoneidx is also used by reclaim/compaction to limit 525 * the target zone since higher zone than this index cannot be 526 * usable for this allocation request. 527 */ 528 enum zone_type highest_zoneidx; 529 bool spread_dirty_pages; 530 }; 531 532 /* 533 * This function returns the order of a free page in the buddy system. In 534 * general, page_zone(page)->lock must be held by the caller to prevent the 535 * page from being allocated in parallel and returning garbage as the order. 536 * If a caller does not hold page_zone(page)->lock, it must guarantee that the 537 * page cannot be allocated or merged in parallel. Alternatively, it must 538 * handle invalid values gracefully, and use buddy_order_unsafe() below. 539 */ 540 static inline unsigned int buddy_order(struct page *page) 541 { 542 /* PageBuddy() must be checked by the caller */ 543 return page_private(page); 544 } 545 546 /* 547 * Like buddy_order(), but for callers who cannot afford to hold the zone lock. 548 * PageBuddy() should be checked first by the caller to minimize race window, 549 * and invalid values must be handled gracefully. 550 * 551 * READ_ONCE is used so that if the caller assigns the result into a local 552 * variable and e.g. tests it for valid range before using, the compiler cannot 553 * decide to remove the variable and inline the page_private(page) multiple 554 * times, potentially observing different values in the tests and the actual 555 * use of the result. 556 */ 557 #define buddy_order_unsafe(page) READ_ONCE(page_private(page)) 558 559 /* 560 * This function checks whether a page is free && is the buddy 561 * we can coalesce a page and its buddy if 562 * (a) the buddy is not in a hole (check before calling!) && 563 * (b) the buddy is in the buddy system && 564 * (c) a page and its buddy have the same order && 565 * (d) a page and its buddy are in the same zone. 566 * 567 * For recording whether a page is in the buddy system, we set PageBuddy. 568 * Setting, clearing, and testing PageBuddy is serialized by zone->lock. 569 * 570 * For recording page's order, we use page_private(page). 571 */ 572 static inline bool page_is_buddy(struct page *page, struct page *buddy, 573 unsigned int order) 574 { 575 if (!page_is_guard(buddy) && !PageBuddy(buddy)) 576 return false; 577 578 if (buddy_order(buddy) != order) 579 return false; 580 581 /* 582 * zone check is done late to avoid uselessly calculating 583 * zone/node ids for pages that could never merge. 584 */ 585 if (page_zone_id(page) != page_zone_id(buddy)) 586 return false; 587 588 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); 589 590 return true; 591 } 592 593 /* 594 * Locate the struct page for both the matching buddy in our 595 * pair (buddy1) and the combined O(n+1) page they form (page). 596 * 597 * 1) Any buddy B1 will have an order O twin B2 which satisfies 598 * the following equation: 599 * B2 = B1 ^ (1 << O) 600 * For example, if the starting buddy (buddy2) is #8 its order 601 * 1 buddy is #10: 602 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 603 * 604 * 2) Any buddy B will have an order O+1 parent P which 605 * satisfies the following equation: 606 * P = B & ~(1 << O) 607 * 608 * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER 609 */ 610 static inline unsigned long 611 __find_buddy_pfn(unsigned long page_pfn, unsigned int order) 612 { 613 return page_pfn ^ (1 << order); 614 } 615 616 /* 617 * Find the buddy of @page and validate it. 618 * @page: The input page 619 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the 620 * function is used in the performance-critical __free_one_page(). 621 * @order: The order of the page 622 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to 623 * page_to_pfn(). 624 * 625 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is 626 * not the same as @page. The validation is necessary before use it. 627 * 628 * Return: the found buddy page or NULL if not found. 629 */ 630 static inline struct page *find_buddy_page_pfn(struct page *page, 631 unsigned long pfn, unsigned int order, unsigned long *buddy_pfn) 632 { 633 unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order); 634 struct page *buddy; 635 636 buddy = page + (__buddy_pfn - pfn); 637 if (buddy_pfn) 638 *buddy_pfn = __buddy_pfn; 639 640 if (page_is_buddy(page, buddy, order)) 641 return buddy; 642 return NULL; 643 } 644 645 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn, 646 unsigned long end_pfn, struct zone *zone); 647 648 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn, 649 unsigned long end_pfn, struct zone *zone) 650 { 651 if (zone->contiguous) 652 return pfn_to_page(start_pfn); 653 654 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone); 655 } 656 657 void set_zone_contiguous(struct zone *zone); 658 659 static inline void clear_zone_contiguous(struct zone *zone) 660 { 661 zone->contiguous = false; 662 } 663 664 extern int __isolate_free_page(struct page *page, unsigned int order); 665 extern void __putback_isolated_page(struct page *page, unsigned int order, 666 int mt); 667 extern void memblock_free_pages(struct page *page, unsigned long pfn, 668 unsigned int order); 669 extern void __free_pages_core(struct page *page, unsigned int order, 670 enum meminit_context context); 671 672 /* 673 * This will have no effect, other than possibly generating a warning, if the 674 * caller passes in a non-large folio. 675 */ 676 static inline void folio_set_order(struct folio *folio, unsigned int order) 677 { 678 if (WARN_ON_ONCE(!order || !folio_test_large(folio))) 679 return; 680 681 folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order; 682 #ifdef CONFIG_64BIT 683 folio->_folio_nr_pages = 1U << order; 684 #endif 685 } 686 687 bool __folio_unqueue_deferred_split(struct folio *folio); 688 static inline bool folio_unqueue_deferred_split(struct folio *folio) 689 { 690 if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio)) 691 return false; 692 693 /* 694 * At this point, there is no one trying to add the folio to 695 * deferred_list. If folio is not in deferred_list, it's safe 696 * to check without acquiring the split_queue_lock. 697 */ 698 if (data_race(list_empty(&folio->_deferred_list))) 699 return false; 700 701 return __folio_unqueue_deferred_split(folio); 702 } 703 704 static inline struct folio *page_rmappable_folio(struct page *page) 705 { 706 struct folio *folio = (struct folio *)page; 707 708 if (folio && folio_test_large(folio)) 709 folio_set_large_rmappable(folio); 710 return folio; 711 } 712 713 static inline void prep_compound_head(struct page *page, unsigned int order) 714 { 715 struct folio *folio = (struct folio *)page; 716 717 folio_set_order(folio, order); 718 atomic_set(&folio->_large_mapcount, -1); 719 atomic_set(&folio->_entire_mapcount, -1); 720 atomic_set(&folio->_nr_pages_mapped, 0); 721 atomic_set(&folio->_pincount, 0); 722 if (order > 1) 723 INIT_LIST_HEAD(&folio->_deferred_list); 724 } 725 726 static inline void prep_compound_tail(struct page *head, int tail_idx) 727 { 728 struct page *p = head + tail_idx; 729 730 p->mapping = TAIL_MAPPING; 731 set_compound_head(p, head); 732 set_page_private(p, 0); 733 } 734 735 extern void prep_compound_page(struct page *page, unsigned int order); 736 737 extern void post_alloc_hook(struct page *page, unsigned int order, 738 gfp_t gfp_flags); 739 extern bool free_pages_prepare(struct page *page, unsigned int order); 740 741 extern int user_min_free_kbytes; 742 743 void free_unref_page(struct page *page, unsigned int order); 744 void free_unref_folios(struct folio_batch *fbatch); 745 746 extern void zone_pcp_reset(struct zone *zone); 747 extern void zone_pcp_disable(struct zone *zone); 748 extern void zone_pcp_enable(struct zone *zone); 749 extern void zone_pcp_init(struct zone *zone); 750 751 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align, 752 phys_addr_t min_addr, 753 int nid, bool exact_nid); 754 755 void memmap_init_range(unsigned long, int, unsigned long, unsigned long, 756 unsigned long, enum meminit_context, struct vmem_altmap *, int); 757 758 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 759 760 /* 761 * in mm/compaction.c 762 */ 763 /* 764 * compact_control is used to track pages being migrated and the free pages 765 * they are being migrated to during memory compaction. The free_pfn starts 766 * at the end of a zone and migrate_pfn begins at the start. Movable pages 767 * are moved to the end of a zone during a compaction run and the run 768 * completes when free_pfn <= migrate_pfn 769 */ 770 struct compact_control { 771 struct list_head freepages[NR_PAGE_ORDERS]; /* List of free pages to migrate to */ 772 struct list_head migratepages; /* List of pages being migrated */ 773 unsigned int nr_freepages; /* Number of isolated free pages */ 774 unsigned int nr_migratepages; /* Number of pages to migrate */ 775 unsigned long free_pfn; /* isolate_freepages search base */ 776 /* 777 * Acts as an in/out parameter to page isolation for migration. 778 * isolate_migratepages uses it as a search base. 779 * isolate_migratepages_block will update the value to the next pfn 780 * after the last isolated one. 781 */ 782 unsigned long migrate_pfn; 783 unsigned long fast_start_pfn; /* a pfn to start linear scan from */ 784 struct zone *zone; 785 unsigned long total_migrate_scanned; 786 unsigned long total_free_scanned; 787 unsigned short fast_search_fail;/* failures to use free list searches */ 788 short search_order; /* order to start a fast search at */ 789 const gfp_t gfp_mask; /* gfp mask of a direct compactor */ 790 int order; /* order a direct compactor needs */ 791 int migratetype; /* migratetype of direct compactor */ 792 const unsigned int alloc_flags; /* alloc flags of a direct compactor */ 793 const int highest_zoneidx; /* zone index of a direct compactor */ 794 enum migrate_mode mode; /* Async or sync migration mode */ 795 bool ignore_skip_hint; /* Scan blocks even if marked skip */ 796 bool no_set_skip_hint; /* Don't mark blocks for skipping */ 797 bool ignore_block_suitable; /* Scan blocks considered unsuitable */ 798 bool direct_compaction; /* False from kcompactd or /proc/... */ 799 bool proactive_compaction; /* kcompactd proactive compaction */ 800 bool whole_zone; /* Whole zone should/has been scanned */ 801 bool contended; /* Signal lock contention */ 802 bool finish_pageblock; /* Scan the remainder of a pageblock. Used 803 * when there are potentially transient 804 * isolation or migration failures to 805 * ensure forward progress. 806 */ 807 bool alloc_contig; /* alloc_contig_range allocation */ 808 }; 809 810 /* 811 * Used in direct compaction when a page should be taken from the freelists 812 * immediately when one is created during the free path. 813 */ 814 struct capture_control { 815 struct compact_control *cc; 816 struct page *page; 817 }; 818 819 unsigned long 820 isolate_freepages_range(struct compact_control *cc, 821 unsigned long start_pfn, unsigned long end_pfn); 822 int 823 isolate_migratepages_range(struct compact_control *cc, 824 unsigned long low_pfn, unsigned long end_pfn); 825 826 int __alloc_contig_migrate_range(struct compact_control *cc, 827 unsigned long start, unsigned long end, 828 int migratetype); 829 830 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ 831 void init_cma_reserved_pageblock(struct page *page); 832 833 #endif /* CONFIG_COMPACTION || CONFIG_CMA */ 834 835 int find_suitable_fallback(struct free_area *area, unsigned int order, 836 int migratetype, bool only_stealable, bool *can_steal); 837 838 static inline bool free_area_empty(struct free_area *area, int migratetype) 839 { 840 return list_empty(&area->free_list[migratetype]); 841 } 842 843 /* mm/util.c */ 844 struct anon_vma *folio_anon_vma(struct folio *folio); 845 846 #ifdef CONFIG_MMU 847 void unmap_mapping_folio(struct folio *folio); 848 extern long populate_vma_page_range(struct vm_area_struct *vma, 849 unsigned long start, unsigned long end, int *locked); 850 extern long faultin_page_range(struct mm_struct *mm, unsigned long start, 851 unsigned long end, bool write, int *locked); 852 extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 853 unsigned long bytes); 854 855 /* 856 * NOTE: This function can't tell whether the folio is "fully mapped" in the 857 * range. 858 * "fully mapped" means all the pages of folio is associated with the page 859 * table of range while this function just check whether the folio range is 860 * within the range [start, end). Function caller needs to do page table 861 * check if it cares about the page table association. 862 * 863 * Typical usage (like mlock or madvise) is: 864 * Caller knows at least 1 page of folio is associated with page table of VMA 865 * and the range [start, end) is intersect with the VMA range. Caller wants 866 * to know whether the folio is fully associated with the range. It calls 867 * this function to check whether the folio is in the range first. Then checks 868 * the page table to know whether the folio is fully mapped to the range. 869 */ 870 static inline bool 871 folio_within_range(struct folio *folio, struct vm_area_struct *vma, 872 unsigned long start, unsigned long end) 873 { 874 pgoff_t pgoff, addr; 875 unsigned long vma_pglen = vma_pages(vma); 876 877 VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio); 878 if (start > end) 879 return false; 880 881 if (start < vma->vm_start) 882 start = vma->vm_start; 883 884 if (end > vma->vm_end) 885 end = vma->vm_end; 886 887 pgoff = folio_pgoff(folio); 888 889 /* if folio start address is not in vma range */ 890 if (!in_range(pgoff, vma->vm_pgoff, vma_pglen)) 891 return false; 892 893 addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 894 895 return !(addr < start || end - addr < folio_size(folio)); 896 } 897 898 static inline bool 899 folio_within_vma(struct folio *folio, struct vm_area_struct *vma) 900 { 901 return folio_within_range(folio, vma, vma->vm_start, vma->vm_end); 902 } 903 904 /* 905 * mlock_vma_folio() and munlock_vma_folio(): 906 * should be called with vma's mmap_lock held for read or write, 907 * under page table lock for the pte/pmd being added or removed. 908 * 909 * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at 910 * the end of folio_remove_rmap_*(); but new anon folios are managed by 911 * folio_add_lru_vma() calling mlock_new_folio(). 912 */ 913 void mlock_folio(struct folio *folio); 914 static inline void mlock_vma_folio(struct folio *folio, 915 struct vm_area_struct *vma) 916 { 917 /* 918 * The VM_SPECIAL check here serves two purposes. 919 * 1) VM_IO check prevents migration from double-counting during mlock. 920 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED 921 * is never left set on a VM_SPECIAL vma, there is an interval while 922 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may 923 * still be set while VM_SPECIAL bits are added: so ignore it then. 924 */ 925 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED)) 926 mlock_folio(folio); 927 } 928 929 void munlock_folio(struct folio *folio); 930 static inline void munlock_vma_folio(struct folio *folio, 931 struct vm_area_struct *vma) 932 { 933 /* 934 * munlock if the function is called. Ideally, we should only 935 * do munlock if any page of folio is unmapped from VMA and 936 * cause folio not fully mapped to VMA. 937 * 938 * But it's not easy to confirm that's the situation. So we 939 * always munlock the folio and page reclaim will correct it 940 * if it's wrong. 941 */ 942 if (unlikely(vma->vm_flags & VM_LOCKED)) 943 munlock_folio(folio); 944 } 945 946 void mlock_new_folio(struct folio *folio); 947 bool need_mlock_drain(int cpu); 948 void mlock_drain_local(void); 949 void mlock_drain_remote(int cpu); 950 951 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma); 952 953 /** 954 * vma_address - Find the virtual address a page range is mapped at 955 * @vma: The vma which maps this object. 956 * @pgoff: The page offset within its object. 957 * @nr_pages: The number of pages to consider. 958 * 959 * If any page in this range is mapped by this VMA, return the first address 960 * where any of these pages appear. Otherwise, return -EFAULT. 961 */ 962 static inline unsigned long vma_address(struct vm_area_struct *vma, 963 pgoff_t pgoff, unsigned long nr_pages) 964 { 965 unsigned long address; 966 967 if (pgoff >= vma->vm_pgoff) { 968 address = vma->vm_start + 969 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 970 /* Check for address beyond vma (or wrapped through 0?) */ 971 if (address < vma->vm_start || address >= vma->vm_end) 972 address = -EFAULT; 973 } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) { 974 /* Test above avoids possibility of wrap to 0 on 32-bit */ 975 address = vma->vm_start; 976 } else { 977 address = -EFAULT; 978 } 979 return address; 980 } 981 982 /* 983 * Then at what user virtual address will none of the range be found in vma? 984 * Assumes that vma_address() already returned a good starting address. 985 */ 986 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw) 987 { 988 struct vm_area_struct *vma = pvmw->vma; 989 pgoff_t pgoff; 990 unsigned long address; 991 992 /* Common case, plus ->pgoff is invalid for KSM */ 993 if (pvmw->nr_pages == 1) 994 return pvmw->address + PAGE_SIZE; 995 996 pgoff = pvmw->pgoff + pvmw->nr_pages; 997 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 998 /* Check for address beyond vma (or wrapped through 0?) */ 999 if (address < vma->vm_start || address > vma->vm_end) 1000 address = vma->vm_end; 1001 return address; 1002 } 1003 1004 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf, 1005 struct file *fpin) 1006 { 1007 int flags = vmf->flags; 1008 1009 if (fpin) 1010 return fpin; 1011 1012 /* 1013 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or 1014 * anything, so we only pin the file and drop the mmap_lock if only 1015 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt. 1016 */ 1017 if (fault_flag_allow_retry_first(flags) && 1018 !(flags & FAULT_FLAG_RETRY_NOWAIT)) { 1019 fpin = get_file(vmf->vma->vm_file); 1020 release_fault_lock(vmf); 1021 } 1022 return fpin; 1023 } 1024 #else /* !CONFIG_MMU */ 1025 static inline void unmap_mapping_folio(struct folio *folio) { } 1026 static inline void mlock_new_folio(struct folio *folio) { } 1027 static inline bool need_mlock_drain(int cpu) { return false; } 1028 static inline void mlock_drain_local(void) { } 1029 static inline void mlock_drain_remote(int cpu) { } 1030 static inline void vunmap_range_noflush(unsigned long start, unsigned long end) 1031 { 1032 } 1033 #endif /* !CONFIG_MMU */ 1034 1035 /* Memory initialisation debug and verification */ 1036 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1037 DECLARE_STATIC_KEY_TRUE(deferred_pages); 1038 1039 bool __init deferred_grow_zone(struct zone *zone, unsigned int order); 1040 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1041 1042 enum mminit_level { 1043 MMINIT_WARNING, 1044 MMINIT_VERIFY, 1045 MMINIT_TRACE 1046 }; 1047 1048 #ifdef CONFIG_DEBUG_MEMORY_INIT 1049 1050 extern int mminit_loglevel; 1051 1052 #define mminit_dprintk(level, prefix, fmt, arg...) \ 1053 do { \ 1054 if (level < mminit_loglevel) { \ 1055 if (level <= MMINIT_WARNING) \ 1056 pr_warn("mminit::" prefix " " fmt, ##arg); \ 1057 else \ 1058 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \ 1059 } \ 1060 } while (0) 1061 1062 extern void mminit_verify_pageflags_layout(void); 1063 extern void mminit_verify_zonelist(void); 1064 #else 1065 1066 static inline void mminit_dprintk(enum mminit_level level, 1067 const char *prefix, const char *fmt, ...) 1068 { 1069 } 1070 1071 static inline void mminit_verify_pageflags_layout(void) 1072 { 1073 } 1074 1075 static inline void mminit_verify_zonelist(void) 1076 { 1077 } 1078 #endif /* CONFIG_DEBUG_MEMORY_INIT */ 1079 1080 #define NODE_RECLAIM_NOSCAN -2 1081 #define NODE_RECLAIM_FULL -1 1082 #define NODE_RECLAIM_SOME 0 1083 #define NODE_RECLAIM_SUCCESS 1 1084 1085 #ifdef CONFIG_NUMA 1086 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int); 1087 extern int find_next_best_node(int node, nodemask_t *used_node_mask); 1088 #else 1089 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask, 1090 unsigned int order) 1091 { 1092 return NODE_RECLAIM_NOSCAN; 1093 } 1094 static inline int find_next_best_node(int node, nodemask_t *used_node_mask) 1095 { 1096 return NUMA_NO_NODE; 1097 } 1098 #endif 1099 1100 /* 1101 * mm/memory-failure.c 1102 */ 1103 #ifdef CONFIG_MEMORY_FAILURE 1104 void unmap_poisoned_folio(struct folio *folio, enum ttu_flags ttu); 1105 void shake_folio(struct folio *folio); 1106 extern int hwpoison_filter(struct page *p); 1107 1108 extern u32 hwpoison_filter_dev_major; 1109 extern u32 hwpoison_filter_dev_minor; 1110 extern u64 hwpoison_filter_flags_mask; 1111 extern u64 hwpoison_filter_flags_value; 1112 extern u64 hwpoison_filter_memcg; 1113 extern u32 hwpoison_filter_enable; 1114 #define MAGIC_HWPOISON 0x48575053U /* HWPS */ 1115 void SetPageHWPoisonTakenOff(struct page *page); 1116 void ClearPageHWPoisonTakenOff(struct page *page); 1117 bool take_page_off_buddy(struct page *page); 1118 bool put_page_back_buddy(struct page *page); 1119 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early); 1120 void add_to_kill_ksm(struct task_struct *tsk, struct page *p, 1121 struct vm_area_struct *vma, struct list_head *to_kill, 1122 unsigned long ksm_addr); 1123 unsigned long page_mapped_in_vma(struct page *page, struct vm_area_struct *vma); 1124 1125 #else 1126 static inline void unmap_poisoned_folio(struct folio *folio, enum ttu_flags ttu) 1127 { 1128 } 1129 #endif 1130 1131 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long, 1132 unsigned long, unsigned long, 1133 unsigned long, unsigned long); 1134 1135 extern void set_pageblock_order(void); 1136 struct folio *alloc_migrate_folio(struct folio *src, unsigned long private); 1137 unsigned long reclaim_pages(struct list_head *folio_list); 1138 unsigned int reclaim_clean_pages_from_list(struct zone *zone, 1139 struct list_head *folio_list); 1140 /* The ALLOC_WMARK bits are used as an index to zone->watermark */ 1141 #define ALLOC_WMARK_MIN WMARK_MIN 1142 #define ALLOC_WMARK_LOW WMARK_LOW 1143 #define ALLOC_WMARK_HIGH WMARK_HIGH 1144 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ 1145 1146 /* Mask to get the watermark bits */ 1147 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) 1148 1149 /* 1150 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we 1151 * cannot assume a reduced access to memory reserves is sufficient for 1152 * !MMU 1153 */ 1154 #ifdef CONFIG_MMU 1155 #define ALLOC_OOM 0x08 1156 #else 1157 #define ALLOC_OOM ALLOC_NO_WATERMARKS 1158 #endif 1159 1160 #define ALLOC_NON_BLOCK 0x10 /* Caller cannot block. Allow access 1161 * to 25% of the min watermark or 1162 * 62.5% if __GFP_HIGH is set. 1163 */ 1164 #define ALLOC_MIN_RESERVE 0x20 /* __GFP_HIGH set. Allow access to 50% 1165 * of the min watermark. 1166 */ 1167 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ 1168 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */ 1169 #ifdef CONFIG_ZONE_DMA32 1170 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */ 1171 #else 1172 #define ALLOC_NOFRAGMENT 0x0 1173 #endif 1174 #define ALLOC_HIGHATOMIC 0x200 /* Allows access to MIGRATE_HIGHATOMIC */ 1175 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */ 1176 1177 /* Flags that allow allocations below the min watermark. */ 1178 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM) 1179 1180 enum ttu_flags; 1181 struct tlbflush_unmap_batch; 1182 1183 1184 /* 1185 * only for MM internal work items which do not depend on 1186 * any allocations or locks which might depend on allocations 1187 */ 1188 extern struct workqueue_struct *mm_percpu_wq; 1189 1190 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 1191 void try_to_unmap_flush(void); 1192 void try_to_unmap_flush_dirty(void); 1193 void flush_tlb_batched_pending(struct mm_struct *mm); 1194 #else 1195 static inline void try_to_unmap_flush(void) 1196 { 1197 } 1198 static inline void try_to_unmap_flush_dirty(void) 1199 { 1200 } 1201 static inline void flush_tlb_batched_pending(struct mm_struct *mm) 1202 { 1203 } 1204 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ 1205 1206 extern const struct trace_print_flags pageflag_names[]; 1207 extern const struct trace_print_flags vmaflag_names[]; 1208 extern const struct trace_print_flags gfpflag_names[]; 1209 1210 static inline bool is_migrate_highatomic(enum migratetype migratetype) 1211 { 1212 return migratetype == MIGRATE_HIGHATOMIC; 1213 } 1214 1215 void setup_zone_pageset(struct zone *zone); 1216 1217 struct migration_target_control { 1218 int nid; /* preferred node id */ 1219 nodemask_t *nmask; 1220 gfp_t gfp_mask; 1221 enum migrate_reason reason; 1222 }; 1223 1224 /* 1225 * mm/filemap.c 1226 */ 1227 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe, 1228 struct folio *folio, loff_t fpos, size_t size); 1229 1230 /* 1231 * mm/vmalloc.c 1232 */ 1233 #ifdef CONFIG_MMU 1234 void __init vmalloc_init(void); 1235 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1236 pgprot_t prot, struct page **pages, unsigned int page_shift); 1237 #else 1238 static inline void vmalloc_init(void) 1239 { 1240 } 1241 1242 static inline 1243 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1244 pgprot_t prot, struct page **pages, unsigned int page_shift) 1245 { 1246 return -EINVAL; 1247 } 1248 #endif 1249 1250 int __must_check __vmap_pages_range_noflush(unsigned long addr, 1251 unsigned long end, pgprot_t prot, 1252 struct page **pages, unsigned int page_shift); 1253 1254 void vunmap_range_noflush(unsigned long start, unsigned long end); 1255 1256 void __vunmap_range_noflush(unsigned long start, unsigned long end); 1257 1258 int numa_migrate_check(struct folio *folio, struct vm_fault *vmf, 1259 unsigned long addr, int *flags, bool writable, 1260 int *last_cpupid); 1261 1262 void free_zone_device_folio(struct folio *folio); 1263 int migrate_device_coherent_folio(struct folio *folio); 1264 1265 /* 1266 * mm/gup.c 1267 */ 1268 int __must_check try_grab_folio(struct folio *folio, int refs, 1269 unsigned int flags); 1270 1271 /* 1272 * mm/huge_memory.c 1273 */ 1274 void touch_pud(struct vm_area_struct *vma, unsigned long addr, 1275 pud_t *pud, bool write); 1276 void touch_pmd(struct vm_area_struct *vma, unsigned long addr, 1277 pmd_t *pmd, bool write); 1278 1279 enum { 1280 /* mark page accessed */ 1281 FOLL_TOUCH = 1 << 16, 1282 /* a retry, previous pass started an IO */ 1283 FOLL_TRIED = 1 << 17, 1284 /* we are working on non-current tsk/mm */ 1285 FOLL_REMOTE = 1 << 18, 1286 /* pages must be released via unpin_user_page */ 1287 FOLL_PIN = 1 << 19, 1288 /* gup_fast: prevent fall-back to slow gup */ 1289 FOLL_FAST_ONLY = 1 << 20, 1290 /* allow unlocking the mmap lock */ 1291 FOLL_UNLOCKABLE = 1 << 21, 1292 /* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */ 1293 FOLL_MADV_POPULATE = 1 << 22, 1294 }; 1295 1296 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \ 1297 FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \ 1298 FOLL_MADV_POPULATE) 1299 1300 /* 1301 * Indicates for which pages that are write-protected in the page table, 1302 * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the 1303 * GUP pin will remain consistent with the pages mapped into the page tables 1304 * of the MM. 1305 * 1306 * Temporary unmapping of PageAnonExclusive() pages or clearing of 1307 * PageAnonExclusive() has to protect against concurrent GUP: 1308 * * Ordinary GUP: Using the PT lock 1309 * * GUP-fast and fork(): mm->write_protect_seq 1310 * * GUP-fast and KSM or temporary unmapping (swap, migration): see 1311 * folio_try_share_anon_rmap_*() 1312 * 1313 * Must be called with the (sub)page that's actually referenced via the 1314 * page table entry, which might not necessarily be the head page for a 1315 * PTE-mapped THP. 1316 * 1317 * If the vma is NULL, we're coming from the GUP-fast path and might have 1318 * to fallback to the slow path just to lookup the vma. 1319 */ 1320 static inline bool gup_must_unshare(struct vm_area_struct *vma, 1321 unsigned int flags, struct page *page) 1322 { 1323 /* 1324 * FOLL_WRITE is implicitly handled correctly as the page table entry 1325 * has to be writable -- and if it references (part of) an anonymous 1326 * folio, that part is required to be marked exclusive. 1327 */ 1328 if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN) 1329 return false; 1330 /* 1331 * Note: PageAnon(page) is stable until the page is actually getting 1332 * freed. 1333 */ 1334 if (!PageAnon(page)) { 1335 /* 1336 * We only care about R/O long-term pining: R/O short-term 1337 * pinning does not have the semantics to observe successive 1338 * changes through the process page tables. 1339 */ 1340 if (!(flags & FOLL_LONGTERM)) 1341 return false; 1342 1343 /* We really need the vma ... */ 1344 if (!vma) 1345 return true; 1346 1347 /* 1348 * ... because we only care about writable private ("COW") 1349 * mappings where we have to break COW early. 1350 */ 1351 return is_cow_mapping(vma->vm_flags); 1352 } 1353 1354 /* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */ 1355 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) 1356 smp_rmb(); 1357 1358 /* 1359 * Note that PageKsm() pages cannot be exclusive, and consequently, 1360 * cannot get pinned. 1361 */ 1362 return !PageAnonExclusive(page); 1363 } 1364 1365 extern bool mirrored_kernelcore; 1366 extern bool memblock_has_mirror(void); 1367 1368 static __always_inline void vma_set_range(struct vm_area_struct *vma, 1369 unsigned long start, unsigned long end, 1370 pgoff_t pgoff) 1371 { 1372 vma->vm_start = start; 1373 vma->vm_end = end; 1374 vma->vm_pgoff = pgoff; 1375 } 1376 1377 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma) 1378 { 1379 /* 1380 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty 1381 * enablements, because when without soft-dirty being compiled in, 1382 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY) 1383 * will be constantly true. 1384 */ 1385 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 1386 return false; 1387 1388 /* 1389 * Soft-dirty is kind of special: its tracking is enabled when the 1390 * vma flags not set. 1391 */ 1392 return !(vma->vm_flags & VM_SOFTDIRTY); 1393 } 1394 1395 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd) 1396 { 1397 return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd); 1398 } 1399 1400 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte) 1401 { 1402 return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte); 1403 } 1404 1405 void __meminit __init_single_page(struct page *page, unsigned long pfn, 1406 unsigned long zone, int nid); 1407 1408 /* shrinker related functions */ 1409 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg, 1410 int priority); 1411 1412 #ifdef CONFIG_64BIT 1413 static inline int can_do_mseal(unsigned long flags) 1414 { 1415 if (flags) 1416 return -EINVAL; 1417 1418 return 0; 1419 } 1420 1421 #else 1422 static inline int can_do_mseal(unsigned long flags) 1423 { 1424 return -EPERM; 1425 } 1426 #endif 1427 1428 #ifdef CONFIG_SHRINKER_DEBUG 1429 static inline __printf(2, 0) int shrinker_debugfs_name_alloc( 1430 struct shrinker *shrinker, const char *fmt, va_list ap) 1431 { 1432 shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap); 1433 1434 return shrinker->name ? 0 : -ENOMEM; 1435 } 1436 1437 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1438 { 1439 kfree_const(shrinker->name); 1440 shrinker->name = NULL; 1441 } 1442 1443 extern int shrinker_debugfs_add(struct shrinker *shrinker); 1444 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1445 int *debugfs_id); 1446 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1447 int debugfs_id); 1448 #else /* CONFIG_SHRINKER_DEBUG */ 1449 static inline int shrinker_debugfs_add(struct shrinker *shrinker) 1450 { 1451 return 0; 1452 } 1453 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker, 1454 const char *fmt, va_list ap) 1455 { 1456 return 0; 1457 } 1458 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1459 { 1460 } 1461 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1462 int *debugfs_id) 1463 { 1464 *debugfs_id = -1; 1465 return NULL; 1466 } 1467 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1468 int debugfs_id) 1469 { 1470 } 1471 #endif /* CONFIG_SHRINKER_DEBUG */ 1472 1473 /* Only track the nodes of mappings with shadow entries */ 1474 void workingset_update_node(struct xa_node *node); 1475 extern struct list_lru shadow_nodes; 1476 1477 /* mremap.c */ 1478 unsigned long move_page_tables(struct vm_area_struct *vma, 1479 unsigned long old_addr, struct vm_area_struct *new_vma, 1480 unsigned long new_addr, unsigned long len, 1481 bool need_rmap_locks, bool for_stack); 1482 1483 #ifdef CONFIG_UNACCEPTED_MEMORY 1484 void accept_page(struct page *page); 1485 #else /* CONFIG_UNACCEPTED_MEMORY */ 1486 static inline void accept_page(struct page *page) 1487 { 1488 } 1489 #endif /* CONFIG_UNACCEPTED_MEMORY */ 1490 1491 #endif /* __MM_INTERNAL_H */ 1492