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/pagewalk.h> 16 #include <linux/rmap.h> 17 #include <linux/swap.h> 18 #include <linux/leafops.h> 19 #include <linux/swap_cgroup.h> 20 #include <linux/tracepoint-defs.h> 21 22 /* Internal core VMA manipulation functions. */ 23 #include "vma.h" 24 25 struct folio_batch; 26 27 /* 28 * Maintains state across a page table move. The operation assumes both source 29 * and destination VMAs already exist and are specified by the user. 30 * 31 * Partial moves are permitted, but the old and new ranges must both reside 32 * within a VMA. 33 * 34 * mmap lock must be held in write and VMA write locks must be held on any VMA 35 * that is visible. 36 * 37 * Use the PAGETABLE_MOVE() macro to initialise this struct. 38 * 39 * The old_addr and new_addr fields are updated as the page table move is 40 * executed. 41 * 42 * NOTE: The page table move is affected by reading from [old_addr, old_end), 43 * and old_addr may be updated for better page table alignment, so len_in 44 * represents the length of the range being copied as specified by the user. 45 */ 46 struct pagetable_move_control { 47 struct vm_area_struct *old; /* Source VMA. */ 48 struct vm_area_struct *new; /* Destination VMA. */ 49 unsigned long old_addr; /* Address from which the move begins. */ 50 unsigned long old_end; /* Exclusive address at which old range ends. */ 51 unsigned long new_addr; /* Address to move page tables to. */ 52 unsigned long len_in; /* Bytes to remap specified by user. */ 53 54 bool need_rmap_locks; /* Do rmap locks need to be taken? */ 55 bool for_stack; /* Is this an early temp stack being moved? */ 56 }; 57 58 #define PAGETABLE_MOVE(name, old_, new_, old_addr_, new_addr_, len_) \ 59 struct pagetable_move_control name = { \ 60 .old = old_, \ 61 .new = new_, \ 62 .old_addr = old_addr_, \ 63 .old_end = (old_addr_) + (len_), \ 64 .new_addr = new_addr_, \ 65 .len_in = len_, \ 66 } 67 68 /* 69 * The set of flags that only affect watermark checking and reclaim 70 * behaviour. This is used by the MM to obey the caller constraints 71 * about IO, FS and watermark checking while ignoring placement 72 * hints such as HIGHMEM usage. 73 */ 74 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\ 75 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\ 76 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\ 77 __GFP_NOLOCKDEP) 78 79 /* The GFP flags allowed during early boot */ 80 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS)) 81 82 /* Control allocation cpuset and node placement constraints */ 83 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE) 84 85 /* Do not use these with a slab allocator */ 86 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK) 87 88 /* 89 * Different from WARN_ON_ONCE(), no warning will be issued 90 * when we specify __GFP_NOWARN. 91 */ 92 #define WARN_ON_ONCE_GFP(cond, gfp) ({ \ 93 static bool __section(".data..once") __warned; \ 94 int __ret_warn_once = !!(cond); \ 95 \ 96 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \ 97 __warned = true; \ 98 WARN_ON(1); \ 99 } \ 100 unlikely(__ret_warn_once); \ 101 }) 102 103 void page_writeback_init(void); 104 105 /* 106 * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages, 107 * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit 108 * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE). Hugetlb currently 109 * leaves nr_pages_mapped at 0, but avoid surprise if it participates later. 110 */ 111 #define ENTIRELY_MAPPED 0x800000 112 #define FOLIO_PAGES_MAPPED (ENTIRELY_MAPPED - 1) 113 114 /* 115 * Flags passed to __show_mem() and show_free_areas() to suppress output in 116 * various contexts. 117 */ 118 #define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */ 119 120 /* 121 * How many individual pages have an elevated _mapcount. Excludes 122 * the folio's entire_mapcount. 123 * 124 * Don't use this function outside of debugging code. 125 */ 126 static inline int folio_nr_pages_mapped(const struct folio *folio) 127 { 128 if (IS_ENABLED(CONFIG_NO_PAGE_MAPCOUNT)) 129 return -1; 130 return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED; 131 } 132 133 /* 134 * Retrieve the first entry of a folio based on a provided entry within the 135 * folio. We cannot rely on folio->swap as there is no guarantee that it has 136 * been initialized. Used for calling arch_swap_restore() 137 */ 138 static inline swp_entry_t folio_swap(swp_entry_t entry, 139 const struct folio *folio) 140 { 141 swp_entry_t swap = { 142 .val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)), 143 }; 144 145 return swap; 146 } 147 148 static inline void *folio_raw_mapping(const struct folio *folio) 149 { 150 unsigned long mapping = (unsigned long)folio->mapping; 151 152 return (void *)(mapping & ~FOLIO_MAPPING_FLAGS); 153 } 154 155 /* 156 * This is a file-backed mapping, and is about to be memory mapped - invoke its 157 * mmap hook and safely handle error conditions. On error, VMA hooks will be 158 * mutated. 159 * 160 * @file: File which backs the mapping. 161 * @vma: VMA which we are mapping. 162 * 163 * Returns: 0 if success, error otherwise. 164 */ 165 static inline int mmap_file(struct file *file, struct vm_area_struct *vma) 166 { 167 int err = vfs_mmap(file, vma); 168 169 if (likely(!err)) 170 return 0; 171 172 /* 173 * OK, we tried to call the file hook for mmap(), but an error 174 * arose. The mapping is in an inconsistent state and we must not invoke 175 * any further hooks on it. 176 */ 177 vma->vm_ops = &vma_dummy_vm_ops; 178 179 return err; 180 } 181 182 /* 183 * If the VMA has a close hook then close it, and since closing it might leave 184 * it in an inconsistent state which makes the use of any hooks suspect, clear 185 * them down by installing dummy empty hooks. 186 */ 187 static inline void vma_close(struct vm_area_struct *vma) 188 { 189 if (vma->vm_ops && vma->vm_ops->close) { 190 vma->vm_ops->close(vma); 191 192 /* 193 * The mapping is in an inconsistent state, and no further hooks 194 * may be invoked upon it. 195 */ 196 vma->vm_ops = &vma_dummy_vm_ops; 197 } 198 } 199 200 /* unmap_vmas is in mm/memory.c */ 201 void unmap_vmas(struct mmu_gather *tlb, struct unmap_desc *unmap); 202 203 #ifdef CONFIG_MMU 204 205 static inline void get_anon_vma(struct anon_vma *anon_vma) 206 { 207 atomic_inc(&anon_vma->refcount); 208 } 209 210 void __put_anon_vma(struct anon_vma *anon_vma); 211 212 static inline void put_anon_vma(struct anon_vma *anon_vma) 213 { 214 if (atomic_dec_and_test(&anon_vma->refcount)) 215 __put_anon_vma(anon_vma); 216 } 217 218 static inline void anon_vma_lock_write(struct anon_vma *anon_vma) 219 { 220 down_write(&anon_vma->root->rwsem); 221 } 222 223 static inline int anon_vma_trylock_write(struct anon_vma *anon_vma) 224 { 225 return down_write_trylock(&anon_vma->root->rwsem); 226 } 227 228 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma) 229 { 230 up_write(&anon_vma->root->rwsem); 231 } 232 233 static inline void anon_vma_lock_read(struct anon_vma *anon_vma) 234 { 235 down_read(&anon_vma->root->rwsem); 236 } 237 238 static inline int anon_vma_trylock_read(struct anon_vma *anon_vma) 239 { 240 return down_read_trylock(&anon_vma->root->rwsem); 241 } 242 243 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma) 244 { 245 up_read(&anon_vma->root->rwsem); 246 } 247 248 struct anon_vma *folio_get_anon_vma(const struct folio *folio); 249 250 /* Operations which modify VMAs. */ 251 enum vma_operation { 252 VMA_OP_SPLIT, 253 VMA_OP_MERGE_UNFAULTED, 254 VMA_OP_REMAP, 255 VMA_OP_FORK, 256 }; 257 258 int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src, 259 enum vma_operation operation); 260 int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma); 261 int __anon_vma_prepare(struct vm_area_struct *vma); 262 void unlink_anon_vmas(struct vm_area_struct *vma); 263 264 static inline int anon_vma_prepare(struct vm_area_struct *vma) 265 { 266 if (likely(vma->anon_vma)) 267 return 0; 268 269 return __anon_vma_prepare(vma); 270 } 271 272 /* Flags for folio_pte_batch(). */ 273 typedef int __bitwise fpb_t; 274 275 /* Compare PTEs respecting the dirty bit. */ 276 #define FPB_RESPECT_DIRTY ((__force fpb_t)BIT(0)) 277 278 /* Compare PTEs respecting the soft-dirty bit. */ 279 #define FPB_RESPECT_SOFT_DIRTY ((__force fpb_t)BIT(1)) 280 281 /* Compare PTEs respecting the writable bit. */ 282 #define FPB_RESPECT_WRITE ((__force fpb_t)BIT(2)) 283 284 /* 285 * Merge PTE write bits: if any PTE in the batch is writable, modify the 286 * PTE at @ptentp to be writable. 287 */ 288 #define FPB_MERGE_WRITE ((__force fpb_t)BIT(3)) 289 290 /* 291 * Merge PTE young and dirty bits: if any PTE in the batch is young or dirty, 292 * modify the PTE at @ptentp to be young or dirty, respectively. 293 */ 294 #define FPB_MERGE_YOUNG_DIRTY ((__force fpb_t)BIT(4)) 295 296 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags) 297 { 298 if (!(flags & FPB_RESPECT_DIRTY)) 299 pte = pte_mkclean(pte); 300 if (likely(!(flags & FPB_RESPECT_SOFT_DIRTY))) 301 pte = pte_clear_soft_dirty(pte); 302 if (likely(!(flags & FPB_RESPECT_WRITE))) 303 pte = pte_wrprotect(pte); 304 return pte_mkold(pte); 305 } 306 307 /** 308 * folio_pte_batch_flags - detect a PTE batch for a large folio 309 * @folio: The large folio to detect a PTE batch for. 310 * @vma: The VMA. Only relevant with FPB_MERGE_WRITE, otherwise can be NULL. 311 * @ptep: Page table pointer for the first entry. 312 * @ptentp: Pointer to a COPY of the first page table entry whose flags this 313 * function updates based on @flags if appropriate. 314 * @max_nr: The maximum number of table entries to consider. 315 * @flags: Flags to modify the PTE batch semantics. 316 * 317 * Detect a PTE batch: consecutive (present) PTEs that map consecutive 318 * pages of the same large folio in a single VMA and a single page table. 319 * 320 * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN, 321 * the accessed bit, writable bit, dirty bit (unless FPB_RESPECT_DIRTY is set) 322 * and soft-dirty bit (unless FPB_RESPECT_SOFT_DIRTY is set). 323 * 324 * @ptep must map any page of the folio. max_nr must be at least one and 325 * must be limited by the caller so scanning cannot exceed a single VMA and 326 * a single page table. 327 * 328 * Depending on the FPB_MERGE_* flags, the pte stored at @ptentp will 329 * be updated: it's crucial that a pointer to a COPY of the first 330 * page table entry, obtained through ptep_get(), is provided as @ptentp. 331 * 332 * This function will be inlined to optimize based on the input parameters; 333 * consider using folio_pte_batch() instead if applicable. 334 * 335 * Return: the number of table entries in the batch. 336 */ 337 static inline unsigned int folio_pte_batch_flags(struct folio *folio, 338 struct vm_area_struct *vma, pte_t *ptep, pte_t *ptentp, 339 unsigned int max_nr, fpb_t flags) 340 { 341 bool any_writable = false, any_young = false, any_dirty = false; 342 pte_t expected_pte, pte = *ptentp; 343 unsigned int nr, cur_nr; 344 345 VM_WARN_ON_FOLIO(!pte_present(pte), folio); 346 VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio); 347 VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio); 348 /* 349 * Ensure this is a pointer to a copy not a pointer into a page table. 350 * If this is a stack value, it won't be a valid virtual address, but 351 * that's fine because it also cannot be pointing into the page table. 352 */ 353 VM_WARN_ON(virt_addr_valid(ptentp) && PageTable(virt_to_page(ptentp))); 354 355 /* Limit max_nr to the actual remaining PFNs in the folio we could batch. */ 356 max_nr = min_t(unsigned long, max_nr, 357 folio_pfn(folio) + folio_nr_pages(folio) - pte_pfn(pte)); 358 359 nr = pte_batch_hint(ptep, pte); 360 expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags); 361 ptep = ptep + nr; 362 363 while (nr < max_nr) { 364 pte = ptep_get(ptep); 365 366 if (!pte_same(__pte_batch_clear_ignored(pte, flags), expected_pte)) 367 break; 368 369 if (flags & FPB_MERGE_WRITE) 370 any_writable |= pte_write(pte); 371 if (flags & FPB_MERGE_YOUNG_DIRTY) { 372 any_young |= pte_young(pte); 373 any_dirty |= pte_dirty(pte); 374 } 375 376 cur_nr = pte_batch_hint(ptep, pte); 377 expected_pte = pte_advance_pfn(expected_pte, cur_nr); 378 ptep += cur_nr; 379 nr += cur_nr; 380 } 381 382 if (any_writable) 383 *ptentp = pte_mkwrite(*ptentp, vma); 384 if (any_young) 385 *ptentp = pte_mkyoung(*ptentp); 386 if (any_dirty) 387 *ptentp = pte_mkdirty(*ptentp); 388 389 return min(nr, max_nr); 390 } 391 392 unsigned int folio_pte_batch(struct folio *folio, pte_t *ptep, pte_t pte, 393 unsigned int max_nr); 394 395 /** 396 * pte_move_swp_offset - Move the swap entry offset field of a swap pte 397 * forward or backward by delta 398 * @pte: The initial pte state; must be a swap entry 399 * @delta: The direction and the offset we are moving; forward if delta 400 * is positive; backward if delta is negative 401 * 402 * Moves the swap offset, while maintaining all other fields, including 403 * swap type, and any swp pte bits. The resulting pte is returned. 404 */ 405 static inline pte_t pte_move_swp_offset(pte_t pte, long delta) 406 { 407 const softleaf_t entry = softleaf_from_pte(pte); 408 pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry), 409 (swp_offset(entry) + delta))); 410 411 if (pte_swp_soft_dirty(pte)) 412 new = pte_swp_mksoft_dirty(new); 413 if (pte_swp_exclusive(pte)) 414 new = pte_swp_mkexclusive(new); 415 if (pte_swp_uffd_wp(pte)) 416 new = pte_swp_mkuffd_wp(new); 417 418 return new; 419 } 420 421 422 /** 423 * pte_next_swp_offset - Increment the swap entry offset field of a swap pte. 424 * @pte: The initial pte state; must be a swap entry. 425 * 426 * Increments the swap offset, while maintaining all other fields, including 427 * swap type, and any swp pte bits. The resulting pte is returned. 428 */ 429 static inline pte_t pte_next_swp_offset(pte_t pte) 430 { 431 return pte_move_swp_offset(pte, 1); 432 } 433 434 /** 435 * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries 436 * @start_ptep: Page table pointer for the first entry. 437 * @max_nr: The maximum number of table entries to consider. 438 * @pte: Page table entry for the first entry. 439 * 440 * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs 441 * containing swap entries all with consecutive offsets and targeting the same 442 * swap type, all with matching swp pte bits. 443 * 444 * max_nr must be at least one and must be limited by the caller so scanning 445 * cannot exceed a single page table. 446 * 447 * Return: the number of table entries in the batch. 448 */ 449 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte) 450 { 451 pte_t expected_pte = pte_next_swp_offset(pte); 452 const pte_t *end_ptep = start_ptep + max_nr; 453 const softleaf_t entry = softleaf_from_pte(pte); 454 pte_t *ptep = start_ptep + 1; 455 unsigned short cgroup_id; 456 457 VM_WARN_ON(max_nr < 1); 458 VM_WARN_ON(!softleaf_is_swap(entry)); 459 460 cgroup_id = lookup_swap_cgroup_id(entry); 461 while (ptep < end_ptep) { 462 softleaf_t entry; 463 464 pte = ptep_get(ptep); 465 466 if (!pte_same(pte, expected_pte)) 467 break; 468 entry = softleaf_from_pte(pte); 469 if (lookup_swap_cgroup_id(entry) != cgroup_id) 470 break; 471 expected_pte = pte_next_swp_offset(expected_pte); 472 ptep++; 473 } 474 475 return ptep - start_ptep; 476 } 477 #endif /* CONFIG_MMU */ 478 479 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio, 480 int nr_throttled); 481 static inline void acct_reclaim_writeback(struct folio *folio) 482 { 483 pg_data_t *pgdat = folio_pgdat(folio); 484 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled); 485 486 if (nr_throttled) 487 __acct_reclaim_writeback(pgdat, folio, nr_throttled); 488 } 489 490 static inline void wake_throttle_isolated(pg_data_t *pgdat) 491 { 492 wait_queue_head_t *wqh; 493 494 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED]; 495 if (waitqueue_active(wqh)) 496 wake_up(wqh); 497 } 498 499 vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf); 500 static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf) 501 { 502 vm_fault_t ret = __vmf_anon_prepare(vmf); 503 504 if (unlikely(ret & VM_FAULT_RETRY)) 505 vma_end_read(vmf->vma); 506 return ret; 507 } 508 509 vm_fault_t do_swap_page(struct vm_fault *vmf); 510 void folio_rotate_reclaimable(struct folio *folio); 511 bool __folio_end_writeback(struct folio *folio); 512 void deactivate_file_folio(struct folio *folio); 513 void folio_activate(struct folio *folio); 514 515 void free_pgtables(struct mmu_gather *tlb, struct unmap_desc *desc); 516 517 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte); 518 519 /** 520 * sync_with_folio_pmd_zap - sync with concurrent zapping of a folio PMD 521 * @mm: The mm_struct. 522 * @pmdp: Pointer to the pmd that was found to be pmd_none(). 523 * 524 * When we find a pmd_none() while unmapping a folio without holding the PTL, 525 * zap_huge_pmd() may have cleared the PMD but not yet modified the folio to 526 * indicate that it's unmapped. Skipping the PMD without synchronization could 527 * make folio unmapping code assume that unmapping failed. 528 * 529 * Wait for concurrent zapping to complete by grabbing the PTL. 530 */ 531 static inline void sync_with_folio_pmd_zap(struct mm_struct *mm, pmd_t *pmdp) 532 { 533 spinlock_t *ptl = pmd_lock(mm, pmdp); 534 535 spin_unlock(ptl); 536 } 537 538 struct zap_details; 539 void unmap_page_range(struct mmu_gather *tlb, 540 struct vm_area_struct *vma, 541 unsigned long addr, unsigned long end, 542 struct zap_details *details); 543 void zap_page_range_single_batched(struct mmu_gather *tlb, 544 struct vm_area_struct *vma, unsigned long addr, 545 unsigned long size, struct zap_details *details); 546 int folio_unmap_invalidate(struct address_space *mapping, struct folio *folio, 547 gfp_t gfp); 548 549 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *); 550 void force_page_cache_ra(struct readahead_control *, unsigned long nr); 551 static inline void force_page_cache_readahead(struct address_space *mapping, 552 struct file *file, pgoff_t index, unsigned long nr_to_read) 553 { 554 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index); 555 force_page_cache_ra(&ractl, nr_to_read); 556 } 557 558 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start, 559 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 560 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start, 561 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 562 void filemap_free_folio(struct address_space *mapping, struct folio *folio); 563 int truncate_inode_folio(struct address_space *mapping, struct folio *folio); 564 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, 565 loff_t end); 566 long mapping_evict_folio(struct address_space *mapping, struct folio *folio); 567 unsigned long mapping_try_invalidate(struct address_space *mapping, 568 pgoff_t start, pgoff_t end, unsigned long *nr_failed); 569 570 /** 571 * folio_evictable - Test whether a folio is evictable. 572 * @folio: The folio to test. 573 * 574 * Test whether @folio is evictable -- i.e., should be placed on 575 * active/inactive lists vs unevictable list. 576 * 577 * Reasons folio might not be evictable: 578 * 1. folio's mapping marked unevictable 579 * 2. One of the pages in the folio is part of an mlocked VMA 580 */ 581 static inline bool folio_evictable(struct folio *folio) 582 { 583 bool ret; 584 585 /* Prevent address_space of inode and swap cache from being freed */ 586 rcu_read_lock(); 587 ret = !mapping_unevictable(folio_mapping(folio)) && 588 !folio_test_mlocked(folio); 589 rcu_read_unlock(); 590 return ret; 591 } 592 593 /* 594 * Turn a non-refcounted page (->_refcount == 0) into refcounted with 595 * a count of one. 596 */ 597 static inline void set_page_refcounted(struct page *page) 598 { 599 VM_BUG_ON_PAGE(PageTail(page), page); 600 VM_BUG_ON_PAGE(page_ref_count(page), page); 601 set_page_count(page, 1); 602 } 603 604 static inline void set_pages_refcounted(struct page *page, unsigned long nr_pages) 605 { 606 unsigned long pfn = page_to_pfn(page); 607 608 for (; nr_pages--; pfn++) 609 set_page_refcounted(pfn_to_page(pfn)); 610 } 611 612 /* 613 * Return true if a folio needs ->release_folio() calling upon it. 614 */ 615 static inline bool folio_needs_release(struct folio *folio) 616 { 617 struct address_space *mapping = folio_mapping(folio); 618 619 return folio_has_private(folio) || 620 (mapping && mapping_release_always(mapping)); 621 } 622 623 extern unsigned long highest_memmap_pfn; 624 625 /* 626 * Maximum number of reclaim retries without progress before the OOM 627 * killer is consider the only way forward. 628 */ 629 #define MAX_RECLAIM_RETRIES 16 630 631 /* 632 * in mm/vmscan.c: 633 */ 634 bool folio_isolate_lru(struct folio *folio); 635 void folio_putback_lru(struct folio *folio); 636 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason); 637 int user_proactive_reclaim(char *buf, 638 struct mem_cgroup *memcg, pg_data_t *pgdat); 639 640 /* 641 * in mm/rmap.c: 642 */ 643 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address); 644 645 /* 646 * in mm/page_alloc.c 647 */ 648 #define K(x) ((x) << (PAGE_SHIFT-10)) 649 650 extern char * const zone_names[MAX_NR_ZONES]; 651 652 /* perform sanity checks on struct pages being allocated or freed */ 653 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled); 654 655 extern int min_free_kbytes; 656 extern int defrag_mode; 657 658 void setup_per_zone_wmarks(void); 659 void calculate_min_free_kbytes(void); 660 int __meminit init_per_zone_wmark_min(void); 661 void page_alloc_sysctl_init(void); 662 663 /* 664 * Structure for holding the mostly immutable allocation parameters passed 665 * between functions involved in allocations, including the alloc_pages* 666 * family of functions. 667 * 668 * nodemask, migratetype and highest_zoneidx are initialized only once in 669 * __alloc_pages() and then never change. 670 * 671 * zonelist, preferred_zone and highest_zoneidx are set first in 672 * __alloc_pages() for the fast path, and might be later changed 673 * in __alloc_pages_slowpath(). All other functions pass the whole structure 674 * by a const pointer. 675 */ 676 struct alloc_context { 677 struct zonelist *zonelist; 678 nodemask_t *nodemask; 679 struct zoneref *preferred_zoneref; 680 int migratetype; 681 682 /* 683 * highest_zoneidx represents highest usable zone index of 684 * the allocation request. Due to the nature of the zone, 685 * memory on lower zone than the highest_zoneidx will be 686 * protected by lowmem_reserve[highest_zoneidx]. 687 * 688 * highest_zoneidx is also used by reclaim/compaction to limit 689 * the target zone since higher zone than this index cannot be 690 * usable for this allocation request. 691 */ 692 enum zone_type highest_zoneidx; 693 bool spread_dirty_pages; 694 }; 695 696 /* 697 * This function returns the order of a free page in the buddy system. In 698 * general, page_zone(page)->lock must be held by the caller to prevent the 699 * page from being allocated in parallel and returning garbage as the order. 700 * If a caller does not hold page_zone(page)->lock, it must guarantee that the 701 * page cannot be allocated or merged in parallel. Alternatively, it must 702 * handle invalid values gracefully, and use buddy_order_unsafe() below. 703 */ 704 static inline unsigned int buddy_order(struct page *page) 705 { 706 /* PageBuddy() must be checked by the caller */ 707 return page_private(page); 708 } 709 710 /* 711 * Like buddy_order(), but for callers who cannot afford to hold the zone lock. 712 * PageBuddy() should be checked first by the caller to minimize race window, 713 * and invalid values must be handled gracefully. 714 * 715 * READ_ONCE is used so that if the caller assigns the result into a local 716 * variable and e.g. tests it for valid range before using, the compiler cannot 717 * decide to remove the variable and inline the page_private(page) multiple 718 * times, potentially observing different values in the tests and the actual 719 * use of the result. 720 */ 721 #define buddy_order_unsafe(page) READ_ONCE(page_private(page)) 722 723 /* 724 * This function checks whether a page is free && is the buddy 725 * we can coalesce a page and its buddy if 726 * (a) the buddy is not in a hole (check before calling!) && 727 * (b) the buddy is in the buddy system && 728 * (c) a page and its buddy have the same order && 729 * (d) a page and its buddy are in the same zone. 730 * 731 * For recording whether a page is in the buddy system, we set PageBuddy. 732 * Setting, clearing, and testing PageBuddy is serialized by zone->lock. 733 * 734 * For recording page's order, we use page_private(page). 735 */ 736 static inline bool page_is_buddy(struct page *page, struct page *buddy, 737 unsigned int order) 738 { 739 if (!page_is_guard(buddy) && !PageBuddy(buddy)) 740 return false; 741 742 if (buddy_order(buddy) != order) 743 return false; 744 745 /* 746 * zone check is done late to avoid uselessly calculating 747 * zone/node ids for pages that could never merge. 748 */ 749 if (page_zone_id(page) != page_zone_id(buddy)) 750 return false; 751 752 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); 753 754 return true; 755 } 756 757 /* 758 * Locate the struct page for both the matching buddy in our 759 * pair (buddy1) and the combined O(n+1) page they form (page). 760 * 761 * 1) Any buddy B1 will have an order O twin B2 which satisfies 762 * the following equation: 763 * B2 = B1 ^ (1 << O) 764 * For example, if the starting buddy (buddy2) is #8 its order 765 * 1 buddy is #10: 766 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 767 * 768 * 2) Any buddy B will have an order O+1 parent P which 769 * satisfies the following equation: 770 * P = B & ~(1 << O) 771 * 772 * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER 773 */ 774 static inline unsigned long 775 __find_buddy_pfn(unsigned long page_pfn, unsigned int order) 776 { 777 return page_pfn ^ (1 << order); 778 } 779 780 /* 781 * Find the buddy of @page and validate it. 782 * @page: The input page 783 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the 784 * function is used in the performance-critical __free_one_page(). 785 * @order: The order of the page 786 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to 787 * page_to_pfn(). 788 * 789 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is 790 * not the same as @page. The validation is necessary before use it. 791 * 792 * Return: the found buddy page or NULL if not found. 793 */ 794 static inline struct page *find_buddy_page_pfn(struct page *page, 795 unsigned long pfn, unsigned int order, unsigned long *buddy_pfn) 796 { 797 unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order); 798 struct page *buddy; 799 800 buddy = page + (__buddy_pfn - pfn); 801 if (buddy_pfn) 802 *buddy_pfn = __buddy_pfn; 803 804 if (page_is_buddy(page, buddy, order)) 805 return buddy; 806 return NULL; 807 } 808 809 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn, 810 unsigned long end_pfn, struct zone *zone); 811 812 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn, 813 unsigned long end_pfn, struct zone *zone) 814 { 815 if (zone->contiguous) 816 return pfn_to_page(start_pfn); 817 818 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone); 819 } 820 821 void set_zone_contiguous(struct zone *zone); 822 bool pfn_range_intersects_zones(int nid, unsigned long start_pfn, 823 unsigned long nr_pages); 824 825 static inline void clear_zone_contiguous(struct zone *zone) 826 { 827 zone->contiguous = false; 828 } 829 830 extern int __isolate_free_page(struct page *page, unsigned int order); 831 extern void __putback_isolated_page(struct page *page, unsigned int order, 832 int mt); 833 extern void memblock_free_pages(unsigned long pfn, unsigned int order); 834 extern void __free_pages_core(struct page *page, unsigned int order, 835 enum meminit_context context); 836 837 /* 838 * This will have no effect, other than possibly generating a warning, if the 839 * caller passes in a non-large folio. 840 */ 841 static inline void folio_set_order(struct folio *folio, unsigned int order) 842 { 843 if (WARN_ON_ONCE(!order || !folio_test_large(folio))) 844 return; 845 VM_WARN_ON_ONCE(order > MAX_FOLIO_ORDER); 846 847 folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order; 848 #ifdef NR_PAGES_IN_LARGE_FOLIO 849 folio->_nr_pages = 1U << order; 850 #endif 851 } 852 853 bool __folio_unqueue_deferred_split(struct folio *folio); 854 static inline bool folio_unqueue_deferred_split(struct folio *folio) 855 { 856 if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio)) 857 return false; 858 859 /* 860 * At this point, there is no one trying to add the folio to 861 * deferred_list. If folio is not in deferred_list, it's safe 862 * to check without acquiring the split_queue_lock. 863 */ 864 if (data_race(list_empty(&folio->_deferred_list))) 865 return false; 866 867 return __folio_unqueue_deferred_split(folio); 868 } 869 870 static inline struct folio *page_rmappable_folio(struct page *page) 871 { 872 struct folio *folio = (struct folio *)page; 873 874 if (folio && folio_test_large(folio)) 875 folio_set_large_rmappable(folio); 876 return folio; 877 } 878 879 static inline void prep_compound_head(struct page *page, unsigned int order) 880 { 881 struct folio *folio = (struct folio *)page; 882 883 folio_set_order(folio, order); 884 atomic_set(&folio->_large_mapcount, -1); 885 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT)) 886 atomic_set(&folio->_nr_pages_mapped, 0); 887 if (IS_ENABLED(CONFIG_MM_ID)) { 888 folio->_mm_ids = 0; 889 folio->_mm_id_mapcount[0] = -1; 890 folio->_mm_id_mapcount[1] = -1; 891 } 892 if (IS_ENABLED(CONFIG_64BIT) || order > 1) { 893 atomic_set(&folio->_pincount, 0); 894 atomic_set(&folio->_entire_mapcount, -1); 895 } 896 if (order > 1) 897 INIT_LIST_HEAD(&folio->_deferred_list); 898 } 899 900 static inline void prep_compound_tail(struct page *head, int tail_idx) 901 { 902 struct page *p = head + tail_idx; 903 904 p->mapping = TAIL_MAPPING; 905 set_compound_head(p, head); 906 set_page_private(p, 0); 907 } 908 909 void post_alloc_hook(struct page *page, unsigned int order, gfp_t gfp_flags); 910 extern bool free_pages_prepare(struct page *page, unsigned int order); 911 912 extern int user_min_free_kbytes; 913 914 struct page *__alloc_frozen_pages_noprof(gfp_t, unsigned int order, int nid, 915 nodemask_t *); 916 #define __alloc_frozen_pages(...) \ 917 alloc_hooks(__alloc_frozen_pages_noprof(__VA_ARGS__)) 918 void free_frozen_pages(struct page *page, unsigned int order); 919 void free_unref_folios(struct folio_batch *fbatch); 920 921 #ifdef CONFIG_NUMA 922 struct page *alloc_frozen_pages_noprof(gfp_t, unsigned int order); 923 #else 924 static inline struct page *alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order) 925 { 926 return __alloc_frozen_pages_noprof(gfp, order, numa_node_id(), NULL); 927 } 928 #endif 929 930 #define alloc_frozen_pages(...) \ 931 alloc_hooks(alloc_frozen_pages_noprof(__VA_ARGS__)) 932 933 struct page *alloc_frozen_pages_nolock_noprof(gfp_t gfp_flags, int nid, unsigned int order); 934 #define alloc_frozen_pages_nolock(...) \ 935 alloc_hooks(alloc_frozen_pages_nolock_noprof(__VA_ARGS__)) 936 void free_frozen_pages_nolock(struct page *page, unsigned int order); 937 938 extern void zone_pcp_reset(struct zone *zone); 939 extern void zone_pcp_disable(struct zone *zone); 940 extern void zone_pcp_enable(struct zone *zone); 941 extern void zone_pcp_init(struct zone *zone); 942 943 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align, 944 phys_addr_t min_addr, 945 int nid, bool exact_nid); 946 947 void memmap_init_range(unsigned long, int, unsigned long, unsigned long, 948 unsigned long, enum meminit_context, struct vmem_altmap *, int, 949 bool); 950 951 #ifdef CONFIG_SPARSEMEM 952 void sparse_init(void); 953 #else 954 static inline void sparse_init(void) {} 955 #endif /* CONFIG_SPARSEMEM */ 956 957 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 958 959 /* 960 * in mm/compaction.c 961 */ 962 /* 963 * compact_control is used to track pages being migrated and the free pages 964 * they are being migrated to during memory compaction. The free_pfn starts 965 * at the end of a zone and migrate_pfn begins at the start. Movable pages 966 * are moved to the end of a zone during a compaction run and the run 967 * completes when free_pfn <= migrate_pfn 968 */ 969 struct compact_control { 970 struct list_head freepages[NR_PAGE_ORDERS]; /* List of free pages to migrate to */ 971 struct list_head migratepages; /* List of pages being migrated */ 972 unsigned int nr_freepages; /* Number of isolated free pages */ 973 unsigned int nr_migratepages; /* Number of pages to migrate */ 974 unsigned long free_pfn; /* isolate_freepages search base */ 975 /* 976 * Acts as an in/out parameter to page isolation for migration. 977 * isolate_migratepages uses it as a search base. 978 * isolate_migratepages_block will update the value to the next pfn 979 * after the last isolated one. 980 */ 981 unsigned long migrate_pfn; 982 unsigned long fast_start_pfn; /* a pfn to start linear scan from */ 983 struct zone *zone; 984 unsigned long total_migrate_scanned; 985 unsigned long total_free_scanned; 986 unsigned short fast_search_fail;/* failures to use free list searches */ 987 short search_order; /* order to start a fast search at */ 988 const gfp_t gfp_mask; /* gfp mask of a direct compactor */ 989 int order; /* order a direct compactor needs */ 990 int migratetype; /* migratetype of direct compactor */ 991 const unsigned int alloc_flags; /* alloc flags of a direct compactor */ 992 const int highest_zoneidx; /* zone index of a direct compactor */ 993 enum migrate_mode mode; /* Async or sync migration mode */ 994 bool ignore_skip_hint; /* Scan blocks even if marked skip */ 995 bool no_set_skip_hint; /* Don't mark blocks for skipping */ 996 bool ignore_block_suitable; /* Scan blocks considered unsuitable */ 997 bool direct_compaction; /* False from kcompactd or /proc/... */ 998 bool proactive_compaction; /* kcompactd proactive compaction */ 999 bool whole_zone; /* Whole zone should/has been scanned */ 1000 bool contended; /* Signal lock contention */ 1001 bool finish_pageblock; /* Scan the remainder of a pageblock. Used 1002 * when there are potentially transient 1003 * isolation or migration failures to 1004 * ensure forward progress. 1005 */ 1006 bool alloc_contig; /* alloc_contig_range allocation */ 1007 }; 1008 1009 /* 1010 * Used in direct compaction when a page should be taken from the freelists 1011 * immediately when one is created during the free path. 1012 */ 1013 struct capture_control { 1014 struct compact_control *cc; 1015 struct page *page; 1016 }; 1017 1018 unsigned long 1019 isolate_freepages_range(struct compact_control *cc, 1020 unsigned long start_pfn, unsigned long end_pfn); 1021 int 1022 isolate_migratepages_range(struct compact_control *cc, 1023 unsigned long low_pfn, unsigned long end_pfn); 1024 1025 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ 1026 void init_cma_reserved_pageblock(struct page *page); 1027 1028 #endif /* CONFIG_COMPACTION || CONFIG_CMA */ 1029 1030 struct cma; 1031 1032 #ifdef CONFIG_CMA 1033 bool cma_validate_zones(struct cma *cma); 1034 void *cma_reserve_early(struct cma *cma, unsigned long size); 1035 void init_cma_pageblock(struct page *page); 1036 #else 1037 static inline bool cma_validate_zones(struct cma *cma) 1038 { 1039 return false; 1040 } 1041 static inline void *cma_reserve_early(struct cma *cma, unsigned long size) 1042 { 1043 return NULL; 1044 } 1045 static inline void init_cma_pageblock(struct page *page) 1046 { 1047 } 1048 #endif 1049 1050 1051 int find_suitable_fallback(struct free_area *area, unsigned int order, 1052 int migratetype, bool claimable); 1053 1054 static inline bool free_area_empty(struct free_area *area, int migratetype) 1055 { 1056 return list_empty(&area->free_list[migratetype]); 1057 } 1058 1059 /* mm/util.c */ 1060 struct anon_vma *folio_anon_vma(const struct folio *folio); 1061 1062 #ifdef CONFIG_MMU 1063 void unmap_mapping_folio(struct folio *folio); 1064 extern long populate_vma_page_range(struct vm_area_struct *vma, 1065 unsigned long start, unsigned long end, int *locked); 1066 extern long faultin_page_range(struct mm_struct *mm, unsigned long start, 1067 unsigned long end, bool write, int *locked); 1068 bool mlock_future_ok(const struct mm_struct *mm, bool is_vma_locked, 1069 unsigned long bytes); 1070 1071 /* 1072 * NOTE: This function can't tell whether the folio is "fully mapped" in the 1073 * range. 1074 * "fully mapped" means all the pages of folio is associated with the page 1075 * table of range while this function just check whether the folio range is 1076 * within the range [start, end). Function caller needs to do page table 1077 * check if it cares about the page table association. 1078 * 1079 * Typical usage (like mlock or madvise) is: 1080 * Caller knows at least 1 page of folio is associated with page table of VMA 1081 * and the range [start, end) is intersect with the VMA range. Caller wants 1082 * to know whether the folio is fully associated with the range. It calls 1083 * this function to check whether the folio is in the range first. Then checks 1084 * the page table to know whether the folio is fully mapped to the range. 1085 */ 1086 static inline bool 1087 folio_within_range(struct folio *folio, struct vm_area_struct *vma, 1088 unsigned long start, unsigned long end) 1089 { 1090 pgoff_t pgoff, addr; 1091 unsigned long vma_pglen = vma_pages(vma); 1092 1093 VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio); 1094 if (start > end) 1095 return false; 1096 1097 if (start < vma->vm_start) 1098 start = vma->vm_start; 1099 1100 if (end > vma->vm_end) 1101 end = vma->vm_end; 1102 1103 pgoff = folio_pgoff(folio); 1104 1105 /* if folio start address is not in vma range */ 1106 if (!in_range(pgoff, vma->vm_pgoff, vma_pglen)) 1107 return false; 1108 1109 addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 1110 1111 return !(addr < start || end - addr < folio_size(folio)); 1112 } 1113 1114 static inline bool 1115 folio_within_vma(struct folio *folio, struct vm_area_struct *vma) 1116 { 1117 return folio_within_range(folio, vma, vma->vm_start, vma->vm_end); 1118 } 1119 1120 /* 1121 * mlock_vma_folio() and munlock_vma_folio(): 1122 * should be called with vma's mmap_lock held for read or write, 1123 * under page table lock for the pte/pmd being added or removed. 1124 * 1125 * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at 1126 * the end of folio_remove_rmap_*(); but new anon folios are managed by 1127 * folio_add_lru_vma() calling mlock_new_folio(). 1128 */ 1129 void mlock_folio(struct folio *folio); 1130 static inline void mlock_vma_folio(struct folio *folio, 1131 struct vm_area_struct *vma) 1132 { 1133 /* 1134 * The VM_SPECIAL check here serves two purposes. 1135 * 1) VM_IO check prevents migration from double-counting during mlock. 1136 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED 1137 * is never left set on a VM_SPECIAL vma, there is an interval while 1138 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may 1139 * still be set while VM_SPECIAL bits are added: so ignore it then. 1140 */ 1141 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED)) 1142 mlock_folio(folio); 1143 } 1144 1145 void munlock_folio(struct folio *folio); 1146 static inline void munlock_vma_folio(struct folio *folio, 1147 struct vm_area_struct *vma) 1148 { 1149 /* 1150 * munlock if the function is called. Ideally, we should only 1151 * do munlock if any page of folio is unmapped from VMA and 1152 * cause folio not fully mapped to VMA. 1153 * 1154 * But it's not easy to confirm that's the situation. So we 1155 * always munlock the folio and page reclaim will correct it 1156 * if it's wrong. 1157 */ 1158 if (unlikely(vma->vm_flags & VM_LOCKED)) 1159 munlock_folio(folio); 1160 } 1161 1162 void mlock_new_folio(struct folio *folio); 1163 bool need_mlock_drain(int cpu); 1164 void mlock_drain_local(void); 1165 void mlock_drain_remote(int cpu); 1166 1167 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma); 1168 1169 /** 1170 * vma_address - Find the virtual address a page range is mapped at 1171 * @vma: The vma which maps this object. 1172 * @pgoff: The page offset within its object. 1173 * @nr_pages: The number of pages to consider. 1174 * 1175 * If any page in this range is mapped by this VMA, return the first address 1176 * where any of these pages appear. Otherwise, return -EFAULT. 1177 */ 1178 static inline unsigned long vma_address(const struct vm_area_struct *vma, 1179 pgoff_t pgoff, unsigned long nr_pages) 1180 { 1181 unsigned long address; 1182 1183 if (pgoff >= vma->vm_pgoff) { 1184 address = vma->vm_start + 1185 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 1186 /* Check for address beyond vma (or wrapped through 0?) */ 1187 if (address < vma->vm_start || address >= vma->vm_end) 1188 address = -EFAULT; 1189 } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) { 1190 /* Test above avoids possibility of wrap to 0 on 32-bit */ 1191 address = vma->vm_start; 1192 } else { 1193 address = -EFAULT; 1194 } 1195 return address; 1196 } 1197 1198 /* 1199 * Then at what user virtual address will none of the range be found in vma? 1200 * Assumes that vma_address() already returned a good starting address. 1201 */ 1202 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw) 1203 { 1204 struct vm_area_struct *vma = pvmw->vma; 1205 pgoff_t pgoff; 1206 unsigned long address; 1207 1208 /* Common case, plus ->pgoff is invalid for KSM */ 1209 if (pvmw->nr_pages == 1) 1210 return pvmw->address + PAGE_SIZE; 1211 1212 pgoff = pvmw->pgoff + pvmw->nr_pages; 1213 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 1214 /* Check for address beyond vma (or wrapped through 0?) */ 1215 if (address < vma->vm_start || address > vma->vm_end) 1216 address = vma->vm_end; 1217 return address; 1218 } 1219 1220 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf, 1221 struct file *fpin) 1222 { 1223 int flags = vmf->flags; 1224 1225 if (fpin) 1226 return fpin; 1227 1228 /* 1229 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or 1230 * anything, so we only pin the file and drop the mmap_lock if only 1231 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt. 1232 */ 1233 if (fault_flag_allow_retry_first(flags) && 1234 !(flags & FAULT_FLAG_RETRY_NOWAIT)) { 1235 fpin = get_file(vmf->vma->vm_file); 1236 release_fault_lock(vmf); 1237 } 1238 return fpin; 1239 } 1240 #else /* !CONFIG_MMU */ 1241 static inline void unmap_mapping_folio(struct folio *folio) { } 1242 static inline void mlock_new_folio(struct folio *folio) { } 1243 static inline bool need_mlock_drain(int cpu) { return false; } 1244 static inline void mlock_drain_local(void) { } 1245 static inline void mlock_drain_remote(int cpu) { } 1246 static inline void vunmap_range_noflush(unsigned long start, unsigned long end) 1247 { 1248 } 1249 #endif /* !CONFIG_MMU */ 1250 1251 /* Memory initialisation debug and verification */ 1252 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1253 DECLARE_STATIC_KEY_TRUE(deferred_pages); 1254 1255 bool __init deferred_grow_zone(struct zone *zone, unsigned int order); 1256 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1257 1258 void init_deferred_page(unsigned long pfn, int nid); 1259 1260 enum mminit_level { 1261 MMINIT_WARNING, 1262 MMINIT_VERIFY, 1263 MMINIT_TRACE 1264 }; 1265 1266 #ifdef CONFIG_DEBUG_MEMORY_INIT 1267 1268 extern int mminit_loglevel; 1269 1270 #define mminit_dprintk(level, prefix, fmt, arg...) \ 1271 do { \ 1272 if (level < mminit_loglevel) { \ 1273 if (level <= MMINIT_WARNING) \ 1274 pr_warn("mminit::" prefix " " fmt, ##arg); \ 1275 else \ 1276 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \ 1277 } \ 1278 } while (0) 1279 1280 extern void mminit_verify_pageflags_layout(void); 1281 extern void mminit_verify_zonelist(void); 1282 #else 1283 1284 static inline void mminit_dprintk(enum mminit_level level, 1285 const char *prefix, const char *fmt, ...) 1286 { 1287 } 1288 1289 static inline void mminit_verify_pageflags_layout(void) 1290 { 1291 } 1292 1293 static inline void mminit_verify_zonelist(void) 1294 { 1295 } 1296 #endif /* CONFIG_DEBUG_MEMORY_INIT */ 1297 1298 #define NODE_RECLAIM_NOSCAN -2 1299 #define NODE_RECLAIM_FULL -1 1300 #define NODE_RECLAIM_SOME 0 1301 #define NODE_RECLAIM_SUCCESS 1 1302 1303 #ifdef CONFIG_NUMA 1304 extern int node_reclaim_mode; 1305 1306 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int); 1307 extern int find_next_best_node(int node, nodemask_t *used_node_mask); 1308 #else 1309 #define node_reclaim_mode 0 1310 1311 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask, 1312 unsigned int order) 1313 { 1314 return NODE_RECLAIM_NOSCAN; 1315 } 1316 static inline int find_next_best_node(int node, nodemask_t *used_node_mask) 1317 { 1318 return NUMA_NO_NODE; 1319 } 1320 #endif 1321 1322 static inline bool node_reclaim_enabled(void) 1323 { 1324 /* Is any node_reclaim_mode bit set? */ 1325 return node_reclaim_mode & (RECLAIM_ZONE|RECLAIM_WRITE|RECLAIM_UNMAP); 1326 } 1327 1328 /* 1329 * mm/memory-failure.c 1330 */ 1331 #ifdef CONFIG_MEMORY_FAILURE 1332 int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill); 1333 void shake_folio(struct folio *folio); 1334 typedef int hwpoison_filter_func_t(struct page *p); 1335 void hwpoison_filter_register(hwpoison_filter_func_t *filter); 1336 void hwpoison_filter_unregister(void); 1337 1338 #define MAGIC_HWPOISON 0x48575053U /* HWPS */ 1339 void SetPageHWPoisonTakenOff(struct page *page); 1340 void ClearPageHWPoisonTakenOff(struct page *page); 1341 bool take_page_off_buddy(struct page *page); 1342 bool put_page_back_buddy(struct page *page); 1343 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early); 1344 void add_to_kill_ksm(struct task_struct *tsk, const struct page *p, 1345 struct vm_area_struct *vma, struct list_head *to_kill, 1346 unsigned long ksm_addr); 1347 unsigned long page_mapped_in_vma(const struct page *page, 1348 struct vm_area_struct *vma); 1349 1350 #else 1351 static inline int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill) 1352 { 1353 return -EBUSY; 1354 } 1355 #endif 1356 1357 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long, 1358 unsigned long, unsigned long, 1359 unsigned long, unsigned long); 1360 1361 extern void set_pageblock_order(void); 1362 unsigned long reclaim_pages(struct list_head *folio_list); 1363 unsigned int reclaim_clean_pages_from_list(struct zone *zone, 1364 struct list_head *folio_list); 1365 /* The ALLOC_WMARK bits are used as an index to zone->watermark */ 1366 #define ALLOC_WMARK_MIN WMARK_MIN 1367 #define ALLOC_WMARK_LOW WMARK_LOW 1368 #define ALLOC_WMARK_HIGH WMARK_HIGH 1369 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ 1370 1371 /* Mask to get the watermark bits */ 1372 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) 1373 1374 /* 1375 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we 1376 * cannot assume a reduced access to memory reserves is sufficient for 1377 * !MMU 1378 */ 1379 #ifdef CONFIG_MMU 1380 #define ALLOC_OOM 0x08 1381 #else 1382 #define ALLOC_OOM ALLOC_NO_WATERMARKS 1383 #endif 1384 1385 #define ALLOC_NON_BLOCK 0x10 /* Caller cannot block. Allow access 1386 * to 25% of the min watermark or 1387 * 62.5% if __GFP_HIGH is set. 1388 */ 1389 #define ALLOC_MIN_RESERVE 0x20 /* __GFP_HIGH set. Allow access to 50% 1390 * of the min watermark. 1391 */ 1392 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ 1393 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */ 1394 #ifdef CONFIG_ZONE_DMA32 1395 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */ 1396 #else 1397 #define ALLOC_NOFRAGMENT 0x0 1398 #endif 1399 #define ALLOC_HIGHATOMIC 0x200 /* Allows access to MIGRATE_HIGHATOMIC */ 1400 #define ALLOC_TRYLOCK 0x400 /* Only use spin_trylock in allocation path */ 1401 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */ 1402 1403 /* Flags that allow allocations below the min watermark. */ 1404 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM) 1405 1406 enum ttu_flags; 1407 struct tlbflush_unmap_batch; 1408 1409 1410 /* 1411 * only for MM internal work items which do not depend on 1412 * any allocations or locks which might depend on allocations 1413 */ 1414 extern struct workqueue_struct *mm_percpu_wq; 1415 1416 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 1417 void try_to_unmap_flush(void); 1418 void try_to_unmap_flush_dirty(void); 1419 void flush_tlb_batched_pending(struct mm_struct *mm); 1420 #else 1421 static inline void try_to_unmap_flush(void) 1422 { 1423 } 1424 static inline void try_to_unmap_flush_dirty(void) 1425 { 1426 } 1427 static inline void flush_tlb_batched_pending(struct mm_struct *mm) 1428 { 1429 } 1430 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ 1431 1432 extern const struct trace_print_flags pageflag_names[]; 1433 extern const struct trace_print_flags vmaflag_names[]; 1434 extern const struct trace_print_flags gfpflag_names[]; 1435 1436 void setup_zone_pageset(struct zone *zone); 1437 1438 struct migration_target_control { 1439 int nid; /* preferred node id */ 1440 nodemask_t *nmask; 1441 gfp_t gfp_mask; 1442 enum migrate_reason reason; 1443 }; 1444 1445 /* 1446 * mm/filemap.c 1447 */ 1448 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe, 1449 struct folio *folio, loff_t fpos, size_t size); 1450 1451 /* 1452 * mm/vmalloc.c 1453 */ 1454 #ifdef CONFIG_MMU 1455 void __init vmalloc_init(void); 1456 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1457 pgprot_t prot, struct page **pages, unsigned int page_shift, gfp_t gfp_mask); 1458 unsigned int get_vm_area_page_order(struct vm_struct *vm); 1459 #else 1460 static inline void vmalloc_init(void) 1461 { 1462 } 1463 1464 static inline 1465 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1466 pgprot_t prot, struct page **pages, unsigned int page_shift, gfp_t gfp_mask) 1467 { 1468 return -EINVAL; 1469 } 1470 #endif 1471 1472 void clear_vm_uninitialized_flag(struct vm_struct *vm); 1473 1474 int __must_check __vmap_pages_range_noflush(unsigned long addr, 1475 unsigned long end, pgprot_t prot, 1476 struct page **pages, unsigned int page_shift); 1477 1478 void vunmap_range_noflush(unsigned long start, unsigned long end); 1479 1480 void __vunmap_range_noflush(unsigned long start, unsigned long end); 1481 1482 static inline bool vma_is_single_threaded_private(struct vm_area_struct *vma) 1483 { 1484 if (vma->vm_flags & VM_SHARED) 1485 return false; 1486 1487 return atomic_read(&vma->vm_mm->mm_users) == 1; 1488 } 1489 1490 #ifdef CONFIG_NUMA_BALANCING 1491 bool folio_can_map_prot_numa(struct folio *folio, struct vm_area_struct *vma, 1492 bool is_private_single_threaded); 1493 1494 #else 1495 static inline bool folio_can_map_prot_numa(struct folio *folio, 1496 struct vm_area_struct *vma, bool is_private_single_threaded) 1497 { 1498 return false; 1499 } 1500 #endif 1501 1502 int numa_migrate_check(struct folio *folio, struct vm_fault *vmf, 1503 unsigned long addr, int *flags, bool writable, 1504 int *last_cpupid); 1505 1506 void free_zone_device_folio(struct folio *folio); 1507 int migrate_device_coherent_folio(struct folio *folio); 1508 1509 struct vm_struct *__get_vm_area_node(unsigned long size, 1510 unsigned long align, unsigned long shift, 1511 unsigned long vm_flags, unsigned long start, 1512 unsigned long end, int node, gfp_t gfp_mask, 1513 const void *caller); 1514 1515 /* 1516 * mm/gup.c 1517 */ 1518 int __must_check try_grab_folio(struct folio *folio, int refs, 1519 unsigned int flags); 1520 1521 /* 1522 * mm/huge_memory.c 1523 */ 1524 void touch_pud(struct vm_area_struct *vma, unsigned long addr, 1525 pud_t *pud, bool write); 1526 bool touch_pmd(struct vm_area_struct *vma, unsigned long addr, 1527 pmd_t *pmd, bool write); 1528 1529 /* 1530 * Parses a string with mem suffixes into its order. Useful to parse kernel 1531 * parameters. 1532 */ 1533 static inline int get_order_from_str(const char *size_str, 1534 unsigned long valid_orders) 1535 { 1536 unsigned long size; 1537 char *endptr; 1538 int order; 1539 1540 size = memparse(size_str, &endptr); 1541 1542 if (!is_power_of_2(size)) 1543 return -EINVAL; 1544 order = get_order(size); 1545 if (BIT(order) & ~valid_orders) 1546 return -EINVAL; 1547 1548 return order; 1549 } 1550 1551 enum { 1552 /* mark page accessed */ 1553 FOLL_TOUCH = 1 << 16, 1554 /* a retry, previous pass started an IO */ 1555 FOLL_TRIED = 1 << 17, 1556 /* we are working on non-current tsk/mm */ 1557 FOLL_REMOTE = 1 << 18, 1558 /* pages must be released via unpin_user_page */ 1559 FOLL_PIN = 1 << 19, 1560 /* gup_fast: prevent fall-back to slow gup */ 1561 FOLL_FAST_ONLY = 1 << 20, 1562 /* allow unlocking the mmap lock */ 1563 FOLL_UNLOCKABLE = 1 << 21, 1564 /* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */ 1565 FOLL_MADV_POPULATE = 1 << 22, 1566 }; 1567 1568 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \ 1569 FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \ 1570 FOLL_MADV_POPULATE) 1571 1572 /* 1573 * Indicates for which pages that are write-protected in the page table, 1574 * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the 1575 * GUP pin will remain consistent with the pages mapped into the page tables 1576 * of the MM. 1577 * 1578 * Temporary unmapping of PageAnonExclusive() pages or clearing of 1579 * PageAnonExclusive() has to protect against concurrent GUP: 1580 * * Ordinary GUP: Using the PT lock 1581 * * GUP-fast and fork(): mm->write_protect_seq 1582 * * GUP-fast and KSM or temporary unmapping (swap, migration): see 1583 * folio_try_share_anon_rmap_*() 1584 * 1585 * Must be called with the (sub)page that's actually referenced via the 1586 * page table entry, which might not necessarily be the head page for a 1587 * PTE-mapped THP. 1588 * 1589 * If the vma is NULL, we're coming from the GUP-fast path and might have 1590 * to fallback to the slow path just to lookup the vma. 1591 */ 1592 static inline bool gup_must_unshare(struct vm_area_struct *vma, 1593 unsigned int flags, struct page *page) 1594 { 1595 /* 1596 * FOLL_WRITE is implicitly handled correctly as the page table entry 1597 * has to be writable -- and if it references (part of) an anonymous 1598 * folio, that part is required to be marked exclusive. 1599 */ 1600 if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN) 1601 return false; 1602 /* 1603 * Note: PageAnon(page) is stable until the page is actually getting 1604 * freed. 1605 */ 1606 if (!PageAnon(page)) { 1607 /* 1608 * We only care about R/O long-term pining: R/O short-term 1609 * pinning does not have the semantics to observe successive 1610 * changes through the process page tables. 1611 */ 1612 if (!(flags & FOLL_LONGTERM)) 1613 return false; 1614 1615 /* We really need the vma ... */ 1616 if (!vma) 1617 return true; 1618 1619 /* 1620 * ... because we only care about writable private ("COW") 1621 * mappings where we have to break COW early. 1622 */ 1623 return is_cow_mapping(vma->vm_flags); 1624 } 1625 1626 /* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */ 1627 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) 1628 smp_rmb(); 1629 1630 /* 1631 * Note that KSM pages cannot be exclusive, and consequently, 1632 * cannot get pinned. 1633 */ 1634 return !PageAnonExclusive(page); 1635 } 1636 1637 extern bool mirrored_kernelcore; 1638 bool memblock_has_mirror(void); 1639 void memblock_free_all(void); 1640 1641 static __always_inline void vma_set_range(struct vm_area_struct *vma, 1642 unsigned long start, unsigned long end, 1643 pgoff_t pgoff) 1644 { 1645 vma->vm_start = start; 1646 vma->vm_end = end; 1647 vma->vm_pgoff = pgoff; 1648 } 1649 1650 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma) 1651 { 1652 /* 1653 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty 1654 * enablements, because when without soft-dirty being compiled in, 1655 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY) 1656 * will be constantly true. 1657 */ 1658 if (!pgtable_supports_soft_dirty()) 1659 return false; 1660 1661 /* 1662 * Soft-dirty is kind of special: its tracking is enabled when the 1663 * vma flags not set. 1664 */ 1665 return !(vma->vm_flags & VM_SOFTDIRTY); 1666 } 1667 1668 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd) 1669 { 1670 return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd); 1671 } 1672 1673 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte) 1674 { 1675 return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte); 1676 } 1677 1678 void __meminit __init_single_page(struct page *page, unsigned long pfn, 1679 unsigned long zone, int nid); 1680 void __meminit __init_page_from_nid(unsigned long pfn, int nid); 1681 1682 /* shrinker related functions */ 1683 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg, 1684 int priority); 1685 1686 int shmem_add_to_page_cache(struct folio *folio, 1687 struct address_space *mapping, 1688 pgoff_t index, void *expected, gfp_t gfp); 1689 int shmem_inode_acct_blocks(struct inode *inode, long pages); 1690 bool shmem_recalc_inode(struct inode *inode, long alloced, long swapped); 1691 1692 #ifdef CONFIG_SHRINKER_DEBUG 1693 static inline __printf(2, 0) int shrinker_debugfs_name_alloc( 1694 struct shrinker *shrinker, const char *fmt, va_list ap) 1695 { 1696 shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap); 1697 1698 return shrinker->name ? 0 : -ENOMEM; 1699 } 1700 1701 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1702 { 1703 kfree_const(shrinker->name); 1704 shrinker->name = NULL; 1705 } 1706 1707 extern int shrinker_debugfs_add(struct shrinker *shrinker); 1708 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1709 int *debugfs_id); 1710 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1711 int debugfs_id); 1712 #else /* CONFIG_SHRINKER_DEBUG */ 1713 static inline int shrinker_debugfs_add(struct shrinker *shrinker) 1714 { 1715 return 0; 1716 } 1717 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker, 1718 const char *fmt, va_list ap) 1719 { 1720 return 0; 1721 } 1722 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1723 { 1724 } 1725 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1726 int *debugfs_id) 1727 { 1728 *debugfs_id = -1; 1729 return NULL; 1730 } 1731 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1732 int debugfs_id) 1733 { 1734 } 1735 #endif /* CONFIG_SHRINKER_DEBUG */ 1736 1737 /* Only track the nodes of mappings with shadow entries */ 1738 void workingset_update_node(struct xa_node *node); 1739 extern struct list_lru shadow_nodes; 1740 #define mapping_set_update(xas, mapping) do { \ 1741 if (!dax_mapping(mapping) && !shmem_mapping(mapping)) { \ 1742 xas_set_update(xas, workingset_update_node); \ 1743 xas_set_lru(xas, &shadow_nodes); \ 1744 } \ 1745 } while (0) 1746 1747 /* mremap.c */ 1748 unsigned long move_page_tables(struct pagetable_move_control *pmc); 1749 1750 #ifdef CONFIG_UNACCEPTED_MEMORY 1751 void accept_page(struct page *page); 1752 #else /* CONFIG_UNACCEPTED_MEMORY */ 1753 static inline void accept_page(struct page *page) 1754 { 1755 } 1756 #endif /* CONFIG_UNACCEPTED_MEMORY */ 1757 1758 /* pagewalk.c */ 1759 int walk_page_range_mm_unsafe(struct mm_struct *mm, unsigned long start, 1760 unsigned long end, const struct mm_walk_ops *ops, 1761 void *private); 1762 int walk_page_range_vma_unsafe(struct vm_area_struct *vma, unsigned long start, 1763 unsigned long end, const struct mm_walk_ops *ops, 1764 void *private); 1765 int walk_page_range_debug(struct mm_struct *mm, unsigned long start, 1766 unsigned long end, const struct mm_walk_ops *ops, 1767 pgd_t *pgd, void *private); 1768 1769 void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm); 1770 int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm); 1771 1772 void remap_pfn_range_prepare(struct vm_area_desc *desc, unsigned long pfn); 1773 int remap_pfn_range_complete(struct vm_area_struct *vma, unsigned long addr, 1774 unsigned long pfn, unsigned long size, pgprot_t pgprot); 1775 1776 static inline void io_remap_pfn_range_prepare(struct vm_area_desc *desc, 1777 unsigned long orig_pfn, unsigned long size) 1778 { 1779 const unsigned long pfn = io_remap_pfn_range_pfn(orig_pfn, size); 1780 1781 return remap_pfn_range_prepare(desc, pfn); 1782 } 1783 1784 static inline int io_remap_pfn_range_complete(struct vm_area_struct *vma, 1785 unsigned long addr, unsigned long orig_pfn, unsigned long size, 1786 pgprot_t orig_prot) 1787 { 1788 const unsigned long pfn = io_remap_pfn_range_pfn(orig_pfn, size); 1789 const pgprot_t prot = pgprot_decrypted(orig_prot); 1790 1791 return remap_pfn_range_complete(vma, addr, pfn, size, prot); 1792 } 1793 1794 #endif /* __MM_INTERNAL_H */ 1795