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