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/swapops.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 * The set of flags that only affect watermark checking and reclaim 29 * behaviour. This is used by the MM to obey the caller constraints 30 * about IO, FS and watermark checking while ignoring placement 31 * hints such as HIGHMEM usage. 32 */ 33 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\ 34 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\ 35 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\ 36 __GFP_NOLOCKDEP) 37 38 /* The GFP flags allowed during early boot */ 39 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS)) 40 41 /* Control allocation cpuset and node placement constraints */ 42 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE) 43 44 /* Do not use these with a slab allocator */ 45 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK) 46 47 /* 48 * Different from WARN_ON_ONCE(), no warning will be issued 49 * when we specify __GFP_NOWARN. 50 */ 51 #define WARN_ON_ONCE_GFP(cond, gfp) ({ \ 52 static bool __section(".data..once") __warned; \ 53 int __ret_warn_once = !!(cond); \ 54 \ 55 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \ 56 __warned = true; \ 57 WARN_ON(1); \ 58 } \ 59 unlikely(__ret_warn_once); \ 60 }) 61 62 void page_writeback_init(void); 63 64 /* 65 * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages, 66 * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit 67 * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE). Hugetlb currently 68 * leaves nr_pages_mapped at 0, but avoid surprise if it participates later. 69 */ 70 #define ENTIRELY_MAPPED 0x800000 71 #define FOLIO_PAGES_MAPPED (ENTIRELY_MAPPED - 1) 72 73 /* 74 * Flags passed to __show_mem() and show_free_areas() to suppress output in 75 * various contexts. 76 */ 77 #define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */ 78 79 /* 80 * How many individual pages have an elevated _mapcount. Excludes 81 * the folio's entire_mapcount. 82 * 83 * Don't use this function outside of debugging code. 84 */ 85 static inline int folio_nr_pages_mapped(const struct folio *folio) 86 { 87 return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED; 88 } 89 90 /* 91 * Retrieve the first entry of a folio based on a provided entry within the 92 * folio. We cannot rely on folio->swap as there is no guarantee that it has 93 * been initialized. Used for calling arch_swap_restore() 94 */ 95 static inline swp_entry_t folio_swap(swp_entry_t entry, 96 const struct folio *folio) 97 { 98 swp_entry_t swap = { 99 .val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)), 100 }; 101 102 return swap; 103 } 104 105 static inline void *folio_raw_mapping(const struct folio *folio) 106 { 107 unsigned long mapping = (unsigned long)folio->mapping; 108 109 return (void *)(mapping & ~PAGE_MAPPING_FLAGS); 110 } 111 112 /* 113 * This is a file-backed mapping, and is about to be memory mapped - invoke its 114 * mmap hook and safely handle error conditions. On error, VMA hooks will be 115 * mutated. 116 * 117 * @file: File which backs the mapping. 118 * @vma: VMA which we are mapping. 119 * 120 * Returns: 0 if success, error otherwise. 121 */ 122 static inline int mmap_file(struct file *file, struct vm_area_struct *vma) 123 { 124 int err = call_mmap(file, vma); 125 126 if (likely(!err)) 127 return 0; 128 129 /* 130 * OK, we tried to call the file hook for mmap(), but an error 131 * arose. The mapping is in an inconsistent state and we most not invoke 132 * any further hooks on it. 133 */ 134 vma->vm_ops = &vma_dummy_vm_ops; 135 136 return err; 137 } 138 139 /* 140 * If the VMA has a close hook then close it, and since closing it might leave 141 * it in an inconsistent state which makes the use of any hooks suspect, clear 142 * them down by installing dummy empty hooks. 143 */ 144 static inline void vma_close(struct vm_area_struct *vma) 145 { 146 if (vma->vm_ops && vma->vm_ops->close) { 147 vma->vm_ops->close(vma); 148 149 /* 150 * The mapping is in an inconsistent state, and no further hooks 151 * may be invoked upon it. 152 */ 153 vma->vm_ops = &vma_dummy_vm_ops; 154 } 155 } 156 157 #ifdef CONFIG_MMU 158 159 /* Flags for folio_pte_batch(). */ 160 typedef int __bitwise fpb_t; 161 162 /* Compare PTEs after pte_mkclean(), ignoring the dirty bit. */ 163 #define FPB_IGNORE_DIRTY ((__force fpb_t)BIT(0)) 164 165 /* Compare PTEs after pte_clear_soft_dirty(), ignoring the soft-dirty bit. */ 166 #define FPB_IGNORE_SOFT_DIRTY ((__force fpb_t)BIT(1)) 167 168 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags) 169 { 170 if (flags & FPB_IGNORE_DIRTY) 171 pte = pte_mkclean(pte); 172 if (likely(flags & FPB_IGNORE_SOFT_DIRTY)) 173 pte = pte_clear_soft_dirty(pte); 174 return pte_wrprotect(pte_mkold(pte)); 175 } 176 177 /** 178 * folio_pte_batch - detect a PTE batch for a large folio 179 * @folio: The large folio to detect a PTE batch for. 180 * @addr: The user virtual address the first page is mapped at. 181 * @start_ptep: Page table pointer for the first entry. 182 * @pte: Page table entry for the first page. 183 * @max_nr: The maximum number of table entries to consider. 184 * @flags: Flags to modify the PTE batch semantics. 185 * @any_writable: Optional pointer to indicate whether any entry except the 186 * first one is writable. 187 * @any_young: Optional pointer to indicate whether any entry except the 188 * first one is young. 189 * @any_dirty: Optional pointer to indicate whether any entry except the 190 * first one is dirty. 191 * 192 * Detect a PTE batch: consecutive (present) PTEs that map consecutive 193 * pages of the same large folio. 194 * 195 * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN, 196 * the accessed bit, writable bit, dirty bit (with FPB_IGNORE_DIRTY) and 197 * soft-dirty bit (with FPB_IGNORE_SOFT_DIRTY). 198 * 199 * start_ptep must map any page of the folio. max_nr must be at least one and 200 * must be limited by the caller so scanning cannot exceed a single page table. 201 * 202 * Return: the number of table entries in the batch. 203 */ 204 static inline int folio_pte_batch(struct folio *folio, unsigned long addr, 205 pte_t *start_ptep, pte_t pte, int max_nr, fpb_t flags, 206 bool *any_writable, bool *any_young, bool *any_dirty) 207 { 208 unsigned long folio_end_pfn = folio_pfn(folio) + folio_nr_pages(folio); 209 const pte_t *end_ptep = start_ptep + max_nr; 210 pte_t expected_pte, *ptep; 211 bool writable, young, dirty; 212 int nr; 213 214 if (any_writable) 215 *any_writable = false; 216 if (any_young) 217 *any_young = false; 218 if (any_dirty) 219 *any_dirty = false; 220 221 VM_WARN_ON_FOLIO(!pte_present(pte), folio); 222 VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio); 223 VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio); 224 225 nr = pte_batch_hint(start_ptep, pte); 226 expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags); 227 ptep = start_ptep + nr; 228 229 while (ptep < end_ptep) { 230 pte = ptep_get(ptep); 231 if (any_writable) 232 writable = !!pte_write(pte); 233 if (any_young) 234 young = !!pte_young(pte); 235 if (any_dirty) 236 dirty = !!pte_dirty(pte); 237 pte = __pte_batch_clear_ignored(pte, flags); 238 239 if (!pte_same(pte, expected_pte)) 240 break; 241 242 /* 243 * Stop immediately once we reached the end of the folio. In 244 * corner cases the next PFN might fall into a different 245 * folio. 246 */ 247 if (pte_pfn(pte) >= folio_end_pfn) 248 break; 249 250 if (any_writable) 251 *any_writable |= writable; 252 if (any_young) 253 *any_young |= young; 254 if (any_dirty) 255 *any_dirty |= dirty; 256 257 nr = pte_batch_hint(ptep, pte); 258 expected_pte = pte_advance_pfn(expected_pte, nr); 259 ptep += nr; 260 } 261 262 return min(ptep - start_ptep, max_nr); 263 } 264 265 /** 266 * pte_move_swp_offset - Move the swap entry offset field of a swap pte 267 * forward or backward by delta 268 * @pte: The initial pte state; is_swap_pte(pte) must be true and 269 * non_swap_entry() must be false. 270 * @delta: The direction and the offset we are moving; forward if delta 271 * is positive; backward if delta is negative 272 * 273 * Moves the swap offset, while maintaining all other fields, including 274 * swap type, and any swp pte bits. The resulting pte is returned. 275 */ 276 static inline pte_t pte_move_swp_offset(pte_t pte, long delta) 277 { 278 swp_entry_t entry = pte_to_swp_entry(pte); 279 pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry), 280 (swp_offset(entry) + delta))); 281 282 if (pte_swp_soft_dirty(pte)) 283 new = pte_swp_mksoft_dirty(new); 284 if (pte_swp_exclusive(pte)) 285 new = pte_swp_mkexclusive(new); 286 if (pte_swp_uffd_wp(pte)) 287 new = pte_swp_mkuffd_wp(new); 288 289 return new; 290 } 291 292 293 /** 294 * pte_next_swp_offset - Increment the swap entry offset field of a swap pte. 295 * @pte: The initial pte state; is_swap_pte(pte) must be true and 296 * non_swap_entry() must be false. 297 * 298 * Increments the swap offset, while maintaining all other fields, including 299 * swap type, and any swp pte bits. The resulting pte is returned. 300 */ 301 static inline pte_t pte_next_swp_offset(pte_t pte) 302 { 303 return pte_move_swp_offset(pte, 1); 304 } 305 306 /** 307 * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries 308 * @start_ptep: Page table pointer for the first entry. 309 * @max_nr: The maximum number of table entries to consider. 310 * @pte: Page table entry for the first entry. 311 * 312 * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs 313 * containing swap entries all with consecutive offsets and targeting the same 314 * swap type, all with matching swp pte bits. 315 * 316 * max_nr must be at least one and must be limited by the caller so scanning 317 * cannot exceed a single page table. 318 * 319 * Return: the number of table entries in the batch. 320 */ 321 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte) 322 { 323 pte_t expected_pte = pte_next_swp_offset(pte); 324 const pte_t *end_ptep = start_ptep + max_nr; 325 swp_entry_t entry = pte_to_swp_entry(pte); 326 pte_t *ptep = start_ptep + 1; 327 unsigned short cgroup_id; 328 329 VM_WARN_ON(max_nr < 1); 330 VM_WARN_ON(!is_swap_pte(pte)); 331 VM_WARN_ON(non_swap_entry(entry)); 332 333 cgroup_id = lookup_swap_cgroup_id(entry); 334 while (ptep < end_ptep) { 335 pte = ptep_get(ptep); 336 337 if (!pte_same(pte, expected_pte)) 338 break; 339 if (lookup_swap_cgroup_id(pte_to_swp_entry(pte)) != cgroup_id) 340 break; 341 expected_pte = pte_next_swp_offset(expected_pte); 342 ptep++; 343 } 344 345 return ptep - start_ptep; 346 } 347 #endif /* CONFIG_MMU */ 348 349 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio, 350 int nr_throttled); 351 static inline void acct_reclaim_writeback(struct folio *folio) 352 { 353 pg_data_t *pgdat = folio_pgdat(folio); 354 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled); 355 356 if (nr_throttled) 357 __acct_reclaim_writeback(pgdat, folio, nr_throttled); 358 } 359 360 static inline void wake_throttle_isolated(pg_data_t *pgdat) 361 { 362 wait_queue_head_t *wqh; 363 364 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED]; 365 if (waitqueue_active(wqh)) 366 wake_up(wqh); 367 } 368 369 vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf); 370 static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf) 371 { 372 vm_fault_t ret = __vmf_anon_prepare(vmf); 373 374 if (unlikely(ret & VM_FAULT_RETRY)) 375 vma_end_read(vmf->vma); 376 return ret; 377 } 378 379 vm_fault_t do_swap_page(struct vm_fault *vmf); 380 void folio_rotate_reclaimable(struct folio *folio); 381 bool __folio_end_writeback(struct folio *folio); 382 void deactivate_file_folio(struct folio *folio); 383 void folio_activate(struct folio *folio); 384 385 void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas, 386 struct vm_area_struct *start_vma, unsigned long floor, 387 unsigned long ceiling, bool mm_wr_locked); 388 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte); 389 390 struct zap_details; 391 void unmap_page_range(struct mmu_gather *tlb, 392 struct vm_area_struct *vma, 393 unsigned long addr, unsigned long end, 394 struct zap_details *details); 395 int folio_unmap_invalidate(struct address_space *mapping, struct folio *folio, 396 gfp_t gfp); 397 398 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *, 399 unsigned int order); 400 void force_page_cache_ra(struct readahead_control *, unsigned long nr); 401 static inline void force_page_cache_readahead(struct address_space *mapping, 402 struct file *file, pgoff_t index, unsigned long nr_to_read) 403 { 404 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index); 405 force_page_cache_ra(&ractl, nr_to_read); 406 } 407 408 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start, 409 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 410 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start, 411 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 412 void filemap_free_folio(struct address_space *mapping, struct folio *folio); 413 int truncate_inode_folio(struct address_space *mapping, struct folio *folio); 414 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, 415 loff_t end); 416 long mapping_evict_folio(struct address_space *mapping, struct folio *folio); 417 unsigned long mapping_try_invalidate(struct address_space *mapping, 418 pgoff_t start, pgoff_t end, unsigned long *nr_failed); 419 420 /** 421 * folio_evictable - Test whether a folio is evictable. 422 * @folio: The folio to test. 423 * 424 * Test whether @folio is evictable -- i.e., should be placed on 425 * active/inactive lists vs unevictable list. 426 * 427 * Reasons folio might not be evictable: 428 * 1. folio's mapping marked unevictable 429 * 2. One of the pages in the folio is part of an mlocked VMA 430 */ 431 static inline bool folio_evictable(struct folio *folio) 432 { 433 bool ret; 434 435 /* Prevent address_space of inode and swap cache from being freed */ 436 rcu_read_lock(); 437 ret = !mapping_unevictable(folio_mapping(folio)) && 438 !folio_test_mlocked(folio); 439 rcu_read_unlock(); 440 return ret; 441 } 442 443 /* 444 * Turn a non-refcounted page (->_refcount == 0) into refcounted with 445 * a count of one. 446 */ 447 static inline void set_page_refcounted(struct page *page) 448 { 449 VM_BUG_ON_PAGE(PageTail(page), page); 450 VM_BUG_ON_PAGE(page_ref_count(page), page); 451 set_page_count(page, 1); 452 } 453 454 /* 455 * Return true if a folio needs ->release_folio() calling upon it. 456 */ 457 static inline bool folio_needs_release(struct folio *folio) 458 { 459 struct address_space *mapping = folio_mapping(folio); 460 461 return folio_has_private(folio) || 462 (mapping && mapping_release_always(mapping)); 463 } 464 465 extern unsigned long highest_memmap_pfn; 466 467 /* 468 * Maximum number of reclaim retries without progress before the OOM 469 * killer is consider the only way forward. 470 */ 471 #define MAX_RECLAIM_RETRIES 16 472 473 /* 474 * in mm/vmscan.c: 475 */ 476 bool folio_isolate_lru(struct folio *folio); 477 void folio_putback_lru(struct folio *folio); 478 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason); 479 480 /* 481 * in mm/rmap.c: 482 */ 483 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address); 484 485 /* 486 * in mm/page_alloc.c 487 */ 488 #define K(x) ((x) << (PAGE_SHIFT-10)) 489 490 extern char * const zone_names[MAX_NR_ZONES]; 491 492 /* perform sanity checks on struct pages being allocated or freed */ 493 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled); 494 495 extern int min_free_kbytes; 496 497 void setup_per_zone_wmarks(void); 498 void calculate_min_free_kbytes(void); 499 int __meminit init_per_zone_wmark_min(void); 500 void page_alloc_sysctl_init(void); 501 502 /* 503 * Structure for holding the mostly immutable allocation parameters passed 504 * between functions involved in allocations, including the alloc_pages* 505 * family of functions. 506 * 507 * nodemask, migratetype and highest_zoneidx are initialized only once in 508 * __alloc_pages() and then never change. 509 * 510 * zonelist, preferred_zone and highest_zoneidx are set first in 511 * __alloc_pages() for the fast path, and might be later changed 512 * in __alloc_pages_slowpath(). All other functions pass the whole structure 513 * by a const pointer. 514 */ 515 struct alloc_context { 516 struct zonelist *zonelist; 517 nodemask_t *nodemask; 518 struct zoneref *preferred_zoneref; 519 int migratetype; 520 521 /* 522 * highest_zoneidx represents highest usable zone index of 523 * the allocation request. Due to the nature of the zone, 524 * memory on lower zone than the highest_zoneidx will be 525 * protected by lowmem_reserve[highest_zoneidx]. 526 * 527 * highest_zoneidx is also used by reclaim/compaction to limit 528 * the target zone since higher zone than this index cannot be 529 * usable for this allocation request. 530 */ 531 enum zone_type highest_zoneidx; 532 bool spread_dirty_pages; 533 }; 534 535 /* 536 * This function returns the order of a free page in the buddy system. In 537 * general, page_zone(page)->lock must be held by the caller to prevent the 538 * page from being allocated in parallel and returning garbage as the order. 539 * If a caller does not hold page_zone(page)->lock, it must guarantee that the 540 * page cannot be allocated or merged in parallel. Alternatively, it must 541 * handle invalid values gracefully, and use buddy_order_unsafe() below. 542 */ 543 static inline unsigned int buddy_order(struct page *page) 544 { 545 /* PageBuddy() must be checked by the caller */ 546 return page_private(page); 547 } 548 549 /* 550 * Like buddy_order(), but for callers who cannot afford to hold the zone lock. 551 * PageBuddy() should be checked first by the caller to minimize race window, 552 * and invalid values must be handled gracefully. 553 * 554 * READ_ONCE is used so that if the caller assigns the result into a local 555 * variable and e.g. tests it for valid range before using, the compiler cannot 556 * decide to remove the variable and inline the page_private(page) multiple 557 * times, potentially observing different values in the tests and the actual 558 * use of the result. 559 */ 560 #define buddy_order_unsafe(page) READ_ONCE(page_private(page)) 561 562 /* 563 * This function checks whether a page is free && is the buddy 564 * we can coalesce a page and its buddy if 565 * (a) the buddy is not in a hole (check before calling!) && 566 * (b) the buddy is in the buddy system && 567 * (c) a page and its buddy have the same order && 568 * (d) a page and its buddy are in the same zone. 569 * 570 * For recording whether a page is in the buddy system, we set PageBuddy. 571 * Setting, clearing, and testing PageBuddy is serialized by zone->lock. 572 * 573 * For recording page's order, we use page_private(page). 574 */ 575 static inline bool page_is_buddy(struct page *page, struct page *buddy, 576 unsigned int order) 577 { 578 if (!page_is_guard(buddy) && !PageBuddy(buddy)) 579 return false; 580 581 if (buddy_order(buddy) != order) 582 return false; 583 584 /* 585 * zone check is done late to avoid uselessly calculating 586 * zone/node ids for pages that could never merge. 587 */ 588 if (page_zone_id(page) != page_zone_id(buddy)) 589 return false; 590 591 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); 592 593 return true; 594 } 595 596 /* 597 * Locate the struct page for both the matching buddy in our 598 * pair (buddy1) and the combined O(n+1) page they form (page). 599 * 600 * 1) Any buddy B1 will have an order O twin B2 which satisfies 601 * the following equation: 602 * B2 = B1 ^ (1 << O) 603 * For example, if the starting buddy (buddy2) is #8 its order 604 * 1 buddy is #10: 605 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 606 * 607 * 2) Any buddy B will have an order O+1 parent P which 608 * satisfies the following equation: 609 * P = B & ~(1 << O) 610 * 611 * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER 612 */ 613 static inline unsigned long 614 __find_buddy_pfn(unsigned long page_pfn, unsigned int order) 615 { 616 return page_pfn ^ (1 << order); 617 } 618 619 /* 620 * Find the buddy of @page and validate it. 621 * @page: The input page 622 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the 623 * function is used in the performance-critical __free_one_page(). 624 * @order: The order of the page 625 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to 626 * page_to_pfn(). 627 * 628 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is 629 * not the same as @page. The validation is necessary before use it. 630 * 631 * Return: the found buddy page or NULL if not found. 632 */ 633 static inline struct page *find_buddy_page_pfn(struct page *page, 634 unsigned long pfn, unsigned int order, unsigned long *buddy_pfn) 635 { 636 unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order); 637 struct page *buddy; 638 639 buddy = page + (__buddy_pfn - pfn); 640 if (buddy_pfn) 641 *buddy_pfn = __buddy_pfn; 642 643 if (page_is_buddy(page, buddy, order)) 644 return buddy; 645 return NULL; 646 } 647 648 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn, 649 unsigned long end_pfn, struct zone *zone); 650 651 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn, 652 unsigned long end_pfn, struct zone *zone) 653 { 654 if (zone->contiguous) 655 return pfn_to_page(start_pfn); 656 657 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone); 658 } 659 660 void set_zone_contiguous(struct zone *zone); 661 662 static inline void clear_zone_contiguous(struct zone *zone) 663 { 664 zone->contiguous = false; 665 } 666 667 extern int __isolate_free_page(struct page *page, unsigned int order); 668 extern void __putback_isolated_page(struct page *page, unsigned int order, 669 int mt); 670 extern void memblock_free_pages(struct page *page, unsigned long pfn, 671 unsigned int order); 672 extern void __free_pages_core(struct page *page, unsigned int order, 673 enum meminit_context context); 674 675 /* 676 * This will have no effect, other than possibly generating a warning, if the 677 * caller passes in a non-large folio. 678 */ 679 static inline void folio_set_order(struct folio *folio, unsigned int order) 680 { 681 if (WARN_ON_ONCE(!order || !folio_test_large(folio))) 682 return; 683 684 folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order; 685 #ifdef CONFIG_64BIT 686 folio->_folio_nr_pages = 1U << order; 687 #endif 688 } 689 690 bool __folio_unqueue_deferred_split(struct folio *folio); 691 static inline bool folio_unqueue_deferred_split(struct folio *folio) 692 { 693 if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio)) 694 return false; 695 696 /* 697 * At this point, there is no one trying to add the folio to 698 * deferred_list. If folio is not in deferred_list, it's safe 699 * to check without acquiring the split_queue_lock. 700 */ 701 if (data_race(list_empty(&folio->_deferred_list))) 702 return false; 703 704 return __folio_unqueue_deferred_split(folio); 705 } 706 707 static inline struct folio *page_rmappable_folio(struct page *page) 708 { 709 struct folio *folio = (struct folio *)page; 710 711 if (folio && folio_test_large(folio)) 712 folio_set_large_rmappable(folio); 713 return folio; 714 } 715 716 static inline void prep_compound_head(struct page *page, unsigned int order) 717 { 718 struct folio *folio = (struct folio *)page; 719 720 folio_set_order(folio, order); 721 atomic_set(&folio->_large_mapcount, -1); 722 atomic_set(&folio->_entire_mapcount, -1); 723 atomic_set(&folio->_nr_pages_mapped, 0); 724 atomic_set(&folio->_pincount, 0); 725 if (order > 1) 726 INIT_LIST_HEAD(&folio->_deferred_list); 727 } 728 729 static inline void prep_compound_tail(struct page *head, int tail_idx) 730 { 731 struct page *p = head + tail_idx; 732 733 p->mapping = TAIL_MAPPING; 734 set_compound_head(p, head); 735 set_page_private(p, 0); 736 } 737 738 extern void prep_compound_page(struct page *page, unsigned int order); 739 740 void post_alloc_hook(struct page *page, unsigned int order, gfp_t gfp_flags); 741 extern bool free_pages_prepare(struct page *page, unsigned int order); 742 743 extern int user_min_free_kbytes; 744 745 struct page *__alloc_frozen_pages_noprof(gfp_t, unsigned int order, int nid, 746 nodemask_t *); 747 #define __alloc_frozen_pages(...) \ 748 alloc_hooks(__alloc_frozen_pages_noprof(__VA_ARGS__)) 749 void free_frozen_pages(struct page *page, unsigned int order); 750 void free_unref_folios(struct folio_batch *fbatch); 751 752 #ifdef CONFIG_NUMA 753 struct page *alloc_frozen_pages_noprof(gfp_t, unsigned int order); 754 #else 755 static inline struct page *alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order) 756 { 757 return __alloc_frozen_pages_noprof(gfp, order, numa_node_id(), NULL); 758 } 759 #endif 760 761 #define alloc_frozen_pages(...) \ 762 alloc_hooks(alloc_frozen_pages_noprof(__VA_ARGS__)) 763 764 extern void zone_pcp_reset(struct zone *zone); 765 extern void zone_pcp_disable(struct zone *zone); 766 extern void zone_pcp_enable(struct zone *zone); 767 extern void zone_pcp_init(struct zone *zone); 768 769 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align, 770 phys_addr_t min_addr, 771 int nid, bool exact_nid); 772 773 void memmap_init_range(unsigned long, int, unsigned long, unsigned long, 774 unsigned long, enum meminit_context, struct vmem_altmap *, int); 775 776 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 777 778 /* 779 * in mm/compaction.c 780 */ 781 /* 782 * compact_control is used to track pages being migrated and the free pages 783 * they are being migrated to during memory compaction. The free_pfn starts 784 * at the end of a zone and migrate_pfn begins at the start. Movable pages 785 * are moved to the end of a zone during a compaction run and the run 786 * completes when free_pfn <= migrate_pfn 787 */ 788 struct compact_control { 789 struct list_head freepages[NR_PAGE_ORDERS]; /* List of free pages to migrate to */ 790 struct list_head migratepages; /* List of pages being migrated */ 791 unsigned int nr_freepages; /* Number of isolated free pages */ 792 unsigned int nr_migratepages; /* Number of pages to migrate */ 793 unsigned long free_pfn; /* isolate_freepages search base */ 794 /* 795 * Acts as an in/out parameter to page isolation for migration. 796 * isolate_migratepages uses it as a search base. 797 * isolate_migratepages_block will update the value to the next pfn 798 * after the last isolated one. 799 */ 800 unsigned long migrate_pfn; 801 unsigned long fast_start_pfn; /* a pfn to start linear scan from */ 802 struct zone *zone; 803 unsigned long total_migrate_scanned; 804 unsigned long total_free_scanned; 805 unsigned short fast_search_fail;/* failures to use free list searches */ 806 short search_order; /* order to start a fast search at */ 807 const gfp_t gfp_mask; /* gfp mask of a direct compactor */ 808 int order; /* order a direct compactor needs */ 809 int migratetype; /* migratetype of direct compactor */ 810 const unsigned int alloc_flags; /* alloc flags of a direct compactor */ 811 const int highest_zoneidx; /* zone index of a direct compactor */ 812 enum migrate_mode mode; /* Async or sync migration mode */ 813 bool ignore_skip_hint; /* Scan blocks even if marked skip */ 814 bool no_set_skip_hint; /* Don't mark blocks for skipping */ 815 bool ignore_block_suitable; /* Scan blocks considered unsuitable */ 816 bool direct_compaction; /* False from kcompactd or /proc/... */ 817 bool proactive_compaction; /* kcompactd proactive compaction */ 818 bool whole_zone; /* Whole zone should/has been scanned */ 819 bool contended; /* Signal lock contention */ 820 bool finish_pageblock; /* Scan the remainder of a pageblock. Used 821 * when there are potentially transient 822 * isolation or migration failures to 823 * ensure forward progress. 824 */ 825 bool alloc_contig; /* alloc_contig_range allocation */ 826 }; 827 828 /* 829 * Used in direct compaction when a page should be taken from the freelists 830 * immediately when one is created during the free path. 831 */ 832 struct capture_control { 833 struct compact_control *cc; 834 struct page *page; 835 }; 836 837 unsigned long 838 isolate_freepages_range(struct compact_control *cc, 839 unsigned long start_pfn, unsigned long end_pfn); 840 int 841 isolate_migratepages_range(struct compact_control *cc, 842 unsigned long low_pfn, unsigned long end_pfn); 843 844 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ 845 void init_cma_reserved_pageblock(struct page *page); 846 847 #endif /* CONFIG_COMPACTION || CONFIG_CMA */ 848 849 int find_suitable_fallback(struct free_area *area, unsigned int order, 850 int migratetype, bool only_stealable, bool *can_steal); 851 852 static inline bool free_area_empty(struct free_area *area, int migratetype) 853 { 854 return list_empty(&area->free_list[migratetype]); 855 } 856 857 /* mm/util.c */ 858 struct anon_vma *folio_anon_vma(const struct folio *folio); 859 860 #ifdef CONFIG_MMU 861 void unmap_mapping_folio(struct folio *folio); 862 extern long populate_vma_page_range(struct vm_area_struct *vma, 863 unsigned long start, unsigned long end, int *locked); 864 extern long faultin_page_range(struct mm_struct *mm, unsigned long start, 865 unsigned long end, bool write, int *locked); 866 extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 867 unsigned long bytes); 868 869 /* 870 * NOTE: This function can't tell whether the folio is "fully mapped" in the 871 * range. 872 * "fully mapped" means all the pages of folio is associated with the page 873 * table of range while this function just check whether the folio range is 874 * within the range [start, end). Function caller needs to do page table 875 * check if it cares about the page table association. 876 * 877 * Typical usage (like mlock or madvise) is: 878 * Caller knows at least 1 page of folio is associated with page table of VMA 879 * and the range [start, end) is intersect with the VMA range. Caller wants 880 * to know whether the folio is fully associated with the range. It calls 881 * this function to check whether the folio is in the range first. Then checks 882 * the page table to know whether the folio is fully mapped to the range. 883 */ 884 static inline bool 885 folio_within_range(struct folio *folio, struct vm_area_struct *vma, 886 unsigned long start, unsigned long end) 887 { 888 pgoff_t pgoff, addr; 889 unsigned long vma_pglen = vma_pages(vma); 890 891 VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio); 892 if (start > end) 893 return false; 894 895 if (start < vma->vm_start) 896 start = vma->vm_start; 897 898 if (end > vma->vm_end) 899 end = vma->vm_end; 900 901 pgoff = folio_pgoff(folio); 902 903 /* if folio start address is not in vma range */ 904 if (!in_range(pgoff, vma->vm_pgoff, vma_pglen)) 905 return false; 906 907 addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 908 909 return !(addr < start || end - addr < folio_size(folio)); 910 } 911 912 static inline bool 913 folio_within_vma(struct folio *folio, struct vm_area_struct *vma) 914 { 915 return folio_within_range(folio, vma, vma->vm_start, vma->vm_end); 916 } 917 918 /* 919 * mlock_vma_folio() and munlock_vma_folio(): 920 * should be called with vma's mmap_lock held for read or write, 921 * under page table lock for the pte/pmd being added or removed. 922 * 923 * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at 924 * the end of folio_remove_rmap_*(); but new anon folios are managed by 925 * folio_add_lru_vma() calling mlock_new_folio(). 926 */ 927 void mlock_folio(struct folio *folio); 928 static inline void mlock_vma_folio(struct folio *folio, 929 struct vm_area_struct *vma) 930 { 931 /* 932 * The VM_SPECIAL check here serves two purposes. 933 * 1) VM_IO check prevents migration from double-counting during mlock. 934 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED 935 * is never left set on a VM_SPECIAL vma, there is an interval while 936 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may 937 * still be set while VM_SPECIAL bits are added: so ignore it then. 938 */ 939 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED)) 940 mlock_folio(folio); 941 } 942 943 void munlock_folio(struct folio *folio); 944 static inline void munlock_vma_folio(struct folio *folio, 945 struct vm_area_struct *vma) 946 { 947 /* 948 * munlock if the function is called. Ideally, we should only 949 * do munlock if any page of folio is unmapped from VMA and 950 * cause folio not fully mapped to VMA. 951 * 952 * But it's not easy to confirm that's the situation. So we 953 * always munlock the folio and page reclaim will correct it 954 * if it's wrong. 955 */ 956 if (unlikely(vma->vm_flags & VM_LOCKED)) 957 munlock_folio(folio); 958 } 959 960 void mlock_new_folio(struct folio *folio); 961 bool need_mlock_drain(int cpu); 962 void mlock_drain_local(void); 963 void mlock_drain_remote(int cpu); 964 965 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma); 966 967 /** 968 * vma_address - Find the virtual address a page range is mapped at 969 * @vma: The vma which maps this object. 970 * @pgoff: The page offset within its object. 971 * @nr_pages: The number of pages to consider. 972 * 973 * If any page in this range is mapped by this VMA, return the first address 974 * where any of these pages appear. Otherwise, return -EFAULT. 975 */ 976 static inline unsigned long vma_address(const struct vm_area_struct *vma, 977 pgoff_t pgoff, unsigned long nr_pages) 978 { 979 unsigned long address; 980 981 if (pgoff >= vma->vm_pgoff) { 982 address = vma->vm_start + 983 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 984 /* Check for address beyond vma (or wrapped through 0?) */ 985 if (address < vma->vm_start || address >= vma->vm_end) 986 address = -EFAULT; 987 } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) { 988 /* Test above avoids possibility of wrap to 0 on 32-bit */ 989 address = vma->vm_start; 990 } else { 991 address = -EFAULT; 992 } 993 return address; 994 } 995 996 /* 997 * Then at what user virtual address will none of the range be found in vma? 998 * Assumes that vma_address() already returned a good starting address. 999 */ 1000 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw) 1001 { 1002 struct vm_area_struct *vma = pvmw->vma; 1003 pgoff_t pgoff; 1004 unsigned long address; 1005 1006 /* Common case, plus ->pgoff is invalid for KSM */ 1007 if (pvmw->nr_pages == 1) 1008 return pvmw->address + PAGE_SIZE; 1009 1010 pgoff = pvmw->pgoff + pvmw->nr_pages; 1011 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 1012 /* Check for address beyond vma (or wrapped through 0?) */ 1013 if (address < vma->vm_start || address > vma->vm_end) 1014 address = vma->vm_end; 1015 return address; 1016 } 1017 1018 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf, 1019 struct file *fpin) 1020 { 1021 int flags = vmf->flags; 1022 1023 if (fpin) 1024 return fpin; 1025 1026 /* 1027 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or 1028 * anything, so we only pin the file and drop the mmap_lock if only 1029 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt. 1030 */ 1031 if (fault_flag_allow_retry_first(flags) && 1032 !(flags & FAULT_FLAG_RETRY_NOWAIT)) { 1033 fpin = get_file(vmf->vma->vm_file); 1034 release_fault_lock(vmf); 1035 } 1036 return fpin; 1037 } 1038 #else /* !CONFIG_MMU */ 1039 static inline void unmap_mapping_folio(struct folio *folio) { } 1040 static inline void mlock_new_folio(struct folio *folio) { } 1041 static inline bool need_mlock_drain(int cpu) { return false; } 1042 static inline void mlock_drain_local(void) { } 1043 static inline void mlock_drain_remote(int cpu) { } 1044 static inline void vunmap_range_noflush(unsigned long start, unsigned long end) 1045 { 1046 } 1047 #endif /* !CONFIG_MMU */ 1048 1049 /* Memory initialisation debug and verification */ 1050 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1051 DECLARE_STATIC_KEY_TRUE(deferred_pages); 1052 1053 bool __init deferred_grow_zone(struct zone *zone, unsigned int order); 1054 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1055 1056 enum mminit_level { 1057 MMINIT_WARNING, 1058 MMINIT_VERIFY, 1059 MMINIT_TRACE 1060 }; 1061 1062 #ifdef CONFIG_DEBUG_MEMORY_INIT 1063 1064 extern int mminit_loglevel; 1065 1066 #define mminit_dprintk(level, prefix, fmt, arg...) \ 1067 do { \ 1068 if (level < mminit_loglevel) { \ 1069 if (level <= MMINIT_WARNING) \ 1070 pr_warn("mminit::" prefix " " fmt, ##arg); \ 1071 else \ 1072 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \ 1073 } \ 1074 } while (0) 1075 1076 extern void mminit_verify_pageflags_layout(void); 1077 extern void mminit_verify_zonelist(void); 1078 #else 1079 1080 static inline void mminit_dprintk(enum mminit_level level, 1081 const char *prefix, const char *fmt, ...) 1082 { 1083 } 1084 1085 static inline void mminit_verify_pageflags_layout(void) 1086 { 1087 } 1088 1089 static inline void mminit_verify_zonelist(void) 1090 { 1091 } 1092 #endif /* CONFIG_DEBUG_MEMORY_INIT */ 1093 1094 #define NODE_RECLAIM_NOSCAN -2 1095 #define NODE_RECLAIM_FULL -1 1096 #define NODE_RECLAIM_SOME 0 1097 #define NODE_RECLAIM_SUCCESS 1 1098 1099 #ifdef CONFIG_NUMA 1100 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int); 1101 extern int find_next_best_node(int node, nodemask_t *used_node_mask); 1102 #else 1103 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask, 1104 unsigned int order) 1105 { 1106 return NODE_RECLAIM_NOSCAN; 1107 } 1108 static inline int find_next_best_node(int node, nodemask_t *used_node_mask) 1109 { 1110 return NUMA_NO_NODE; 1111 } 1112 #endif 1113 1114 /* 1115 * mm/memory-failure.c 1116 */ 1117 #ifdef CONFIG_MEMORY_FAILURE 1118 void unmap_poisoned_folio(struct folio *folio, enum ttu_flags ttu); 1119 void shake_folio(struct folio *folio); 1120 extern int hwpoison_filter(struct page *p); 1121 1122 extern u32 hwpoison_filter_dev_major; 1123 extern u32 hwpoison_filter_dev_minor; 1124 extern u64 hwpoison_filter_flags_mask; 1125 extern u64 hwpoison_filter_flags_value; 1126 extern u64 hwpoison_filter_memcg; 1127 extern u32 hwpoison_filter_enable; 1128 #define MAGIC_HWPOISON 0x48575053U /* HWPS */ 1129 void SetPageHWPoisonTakenOff(struct page *page); 1130 void ClearPageHWPoisonTakenOff(struct page *page); 1131 bool take_page_off_buddy(struct page *page); 1132 bool put_page_back_buddy(struct page *page); 1133 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early); 1134 void add_to_kill_ksm(struct task_struct *tsk, const struct page *p, 1135 struct vm_area_struct *vma, struct list_head *to_kill, 1136 unsigned long ksm_addr); 1137 unsigned long page_mapped_in_vma(const struct page *page, 1138 struct vm_area_struct *vma); 1139 1140 #else 1141 static inline void unmap_poisoned_folio(struct folio *folio, enum ttu_flags ttu) 1142 { 1143 } 1144 #endif 1145 1146 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long, 1147 unsigned long, unsigned long, 1148 unsigned long, unsigned long); 1149 1150 extern void set_pageblock_order(void); 1151 struct folio *alloc_migrate_folio(struct folio *src, unsigned long private); 1152 unsigned long reclaim_pages(struct list_head *folio_list); 1153 unsigned int reclaim_clean_pages_from_list(struct zone *zone, 1154 struct list_head *folio_list); 1155 /* The ALLOC_WMARK bits are used as an index to zone->watermark */ 1156 #define ALLOC_WMARK_MIN WMARK_MIN 1157 #define ALLOC_WMARK_LOW WMARK_LOW 1158 #define ALLOC_WMARK_HIGH WMARK_HIGH 1159 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ 1160 1161 /* Mask to get the watermark bits */ 1162 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) 1163 1164 /* 1165 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we 1166 * cannot assume a reduced access to memory reserves is sufficient for 1167 * !MMU 1168 */ 1169 #ifdef CONFIG_MMU 1170 #define ALLOC_OOM 0x08 1171 #else 1172 #define ALLOC_OOM ALLOC_NO_WATERMARKS 1173 #endif 1174 1175 #define ALLOC_NON_BLOCK 0x10 /* Caller cannot block. Allow access 1176 * to 25% of the min watermark or 1177 * 62.5% if __GFP_HIGH is set. 1178 */ 1179 #define ALLOC_MIN_RESERVE 0x20 /* __GFP_HIGH set. Allow access to 50% 1180 * of the min watermark. 1181 */ 1182 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ 1183 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */ 1184 #ifdef CONFIG_ZONE_DMA32 1185 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */ 1186 #else 1187 #define ALLOC_NOFRAGMENT 0x0 1188 #endif 1189 #define ALLOC_HIGHATOMIC 0x200 /* Allows access to MIGRATE_HIGHATOMIC */ 1190 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */ 1191 1192 /* Flags that allow allocations below the min watermark. */ 1193 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM) 1194 1195 enum ttu_flags; 1196 struct tlbflush_unmap_batch; 1197 1198 1199 /* 1200 * only for MM internal work items which do not depend on 1201 * any allocations or locks which might depend on allocations 1202 */ 1203 extern struct workqueue_struct *mm_percpu_wq; 1204 1205 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 1206 void try_to_unmap_flush(void); 1207 void try_to_unmap_flush_dirty(void); 1208 void flush_tlb_batched_pending(struct mm_struct *mm); 1209 #else 1210 static inline void try_to_unmap_flush(void) 1211 { 1212 } 1213 static inline void try_to_unmap_flush_dirty(void) 1214 { 1215 } 1216 static inline void flush_tlb_batched_pending(struct mm_struct *mm) 1217 { 1218 } 1219 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ 1220 1221 extern const struct trace_print_flags pageflag_names[]; 1222 extern const struct trace_print_flags vmaflag_names[]; 1223 extern const struct trace_print_flags gfpflag_names[]; 1224 1225 static inline bool is_migrate_highatomic(enum migratetype migratetype) 1226 { 1227 return migratetype == MIGRATE_HIGHATOMIC; 1228 } 1229 1230 void setup_zone_pageset(struct zone *zone); 1231 1232 struct migration_target_control { 1233 int nid; /* preferred node id */ 1234 nodemask_t *nmask; 1235 gfp_t gfp_mask; 1236 enum migrate_reason reason; 1237 }; 1238 1239 /* 1240 * mm/filemap.c 1241 */ 1242 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe, 1243 struct folio *folio, loff_t fpos, size_t size); 1244 1245 /* 1246 * mm/vmalloc.c 1247 */ 1248 #ifdef CONFIG_MMU 1249 void __init vmalloc_init(void); 1250 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1251 pgprot_t prot, struct page **pages, unsigned int page_shift); 1252 unsigned int get_vm_area_page_order(struct vm_struct *vm); 1253 #else 1254 static inline void vmalloc_init(void) 1255 { 1256 } 1257 1258 static inline 1259 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1260 pgprot_t prot, struct page **pages, unsigned int page_shift) 1261 { 1262 return -EINVAL; 1263 } 1264 #endif 1265 1266 int __must_check __vmap_pages_range_noflush(unsigned long addr, 1267 unsigned long end, pgprot_t prot, 1268 struct page **pages, unsigned int page_shift); 1269 1270 void vunmap_range_noflush(unsigned long start, unsigned long end); 1271 1272 void __vunmap_range_noflush(unsigned long start, unsigned long end); 1273 1274 int numa_migrate_check(struct folio *folio, struct vm_fault *vmf, 1275 unsigned long addr, int *flags, bool writable, 1276 int *last_cpupid); 1277 1278 void free_zone_device_folio(struct folio *folio); 1279 int migrate_device_coherent_folio(struct folio *folio); 1280 1281 struct vm_struct *__get_vm_area_node(unsigned long size, 1282 unsigned long align, unsigned long shift, 1283 unsigned long flags, unsigned long start, 1284 unsigned long end, int node, gfp_t gfp_mask, 1285 const void *caller); 1286 1287 /* 1288 * mm/gup.c 1289 */ 1290 int __must_check try_grab_folio(struct folio *folio, int refs, 1291 unsigned int flags); 1292 1293 /* 1294 * mm/huge_memory.c 1295 */ 1296 void touch_pud(struct vm_area_struct *vma, unsigned long addr, 1297 pud_t *pud, bool write); 1298 void touch_pmd(struct vm_area_struct *vma, unsigned long addr, 1299 pmd_t *pmd, bool write); 1300 1301 /* 1302 * Parses a string with mem suffixes into its order. Useful to parse kernel 1303 * parameters. 1304 */ 1305 static inline int get_order_from_str(const char *size_str, 1306 unsigned long valid_orders) 1307 { 1308 unsigned long size; 1309 char *endptr; 1310 int order; 1311 1312 size = memparse(size_str, &endptr); 1313 1314 if (!is_power_of_2(size)) 1315 return -EINVAL; 1316 order = get_order(size); 1317 if (BIT(order) & ~valid_orders) 1318 return -EINVAL; 1319 1320 return order; 1321 } 1322 1323 enum { 1324 /* mark page accessed */ 1325 FOLL_TOUCH = 1 << 16, 1326 /* a retry, previous pass started an IO */ 1327 FOLL_TRIED = 1 << 17, 1328 /* we are working on non-current tsk/mm */ 1329 FOLL_REMOTE = 1 << 18, 1330 /* pages must be released via unpin_user_page */ 1331 FOLL_PIN = 1 << 19, 1332 /* gup_fast: prevent fall-back to slow gup */ 1333 FOLL_FAST_ONLY = 1 << 20, 1334 /* allow unlocking the mmap lock */ 1335 FOLL_UNLOCKABLE = 1 << 21, 1336 /* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */ 1337 FOLL_MADV_POPULATE = 1 << 22, 1338 }; 1339 1340 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \ 1341 FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \ 1342 FOLL_MADV_POPULATE) 1343 1344 /* 1345 * Indicates for which pages that are write-protected in the page table, 1346 * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the 1347 * GUP pin will remain consistent with the pages mapped into the page tables 1348 * of the MM. 1349 * 1350 * Temporary unmapping of PageAnonExclusive() pages or clearing of 1351 * PageAnonExclusive() has to protect against concurrent GUP: 1352 * * Ordinary GUP: Using the PT lock 1353 * * GUP-fast and fork(): mm->write_protect_seq 1354 * * GUP-fast and KSM or temporary unmapping (swap, migration): see 1355 * folio_try_share_anon_rmap_*() 1356 * 1357 * Must be called with the (sub)page that's actually referenced via the 1358 * page table entry, which might not necessarily be the head page for a 1359 * PTE-mapped THP. 1360 * 1361 * If the vma is NULL, we're coming from the GUP-fast path and might have 1362 * to fallback to the slow path just to lookup the vma. 1363 */ 1364 static inline bool gup_must_unshare(struct vm_area_struct *vma, 1365 unsigned int flags, struct page *page) 1366 { 1367 /* 1368 * FOLL_WRITE is implicitly handled correctly as the page table entry 1369 * has to be writable -- and if it references (part of) an anonymous 1370 * folio, that part is required to be marked exclusive. 1371 */ 1372 if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN) 1373 return false; 1374 /* 1375 * Note: PageAnon(page) is stable until the page is actually getting 1376 * freed. 1377 */ 1378 if (!PageAnon(page)) { 1379 /* 1380 * We only care about R/O long-term pining: R/O short-term 1381 * pinning does not have the semantics to observe successive 1382 * changes through the process page tables. 1383 */ 1384 if (!(flags & FOLL_LONGTERM)) 1385 return false; 1386 1387 /* We really need the vma ... */ 1388 if (!vma) 1389 return true; 1390 1391 /* 1392 * ... because we only care about writable private ("COW") 1393 * mappings where we have to break COW early. 1394 */ 1395 return is_cow_mapping(vma->vm_flags); 1396 } 1397 1398 /* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */ 1399 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) 1400 smp_rmb(); 1401 1402 /* 1403 * Note that KSM pages cannot be exclusive, and consequently, 1404 * cannot get pinned. 1405 */ 1406 return !PageAnonExclusive(page); 1407 } 1408 1409 extern bool mirrored_kernelcore; 1410 extern bool memblock_has_mirror(void); 1411 1412 static __always_inline void vma_set_range(struct vm_area_struct *vma, 1413 unsigned long start, unsigned long end, 1414 pgoff_t pgoff) 1415 { 1416 vma->vm_start = start; 1417 vma->vm_end = end; 1418 vma->vm_pgoff = pgoff; 1419 } 1420 1421 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma) 1422 { 1423 /* 1424 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty 1425 * enablements, because when without soft-dirty being compiled in, 1426 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY) 1427 * will be constantly true. 1428 */ 1429 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 1430 return false; 1431 1432 /* 1433 * Soft-dirty is kind of special: its tracking is enabled when the 1434 * vma flags not set. 1435 */ 1436 return !(vma->vm_flags & VM_SOFTDIRTY); 1437 } 1438 1439 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd) 1440 { 1441 return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd); 1442 } 1443 1444 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte) 1445 { 1446 return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte); 1447 } 1448 1449 void __meminit __init_single_page(struct page *page, unsigned long pfn, 1450 unsigned long zone, int nid); 1451 1452 /* shrinker related functions */ 1453 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg, 1454 int priority); 1455 1456 #ifdef CONFIG_SHRINKER_DEBUG 1457 static inline __printf(2, 0) int shrinker_debugfs_name_alloc( 1458 struct shrinker *shrinker, const char *fmt, va_list ap) 1459 { 1460 shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap); 1461 1462 return shrinker->name ? 0 : -ENOMEM; 1463 } 1464 1465 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1466 { 1467 kfree_const(shrinker->name); 1468 shrinker->name = NULL; 1469 } 1470 1471 extern int shrinker_debugfs_add(struct shrinker *shrinker); 1472 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1473 int *debugfs_id); 1474 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1475 int debugfs_id); 1476 #else /* CONFIG_SHRINKER_DEBUG */ 1477 static inline int shrinker_debugfs_add(struct shrinker *shrinker) 1478 { 1479 return 0; 1480 } 1481 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker, 1482 const char *fmt, va_list ap) 1483 { 1484 return 0; 1485 } 1486 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1487 { 1488 } 1489 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1490 int *debugfs_id) 1491 { 1492 *debugfs_id = -1; 1493 return NULL; 1494 } 1495 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1496 int debugfs_id) 1497 { 1498 } 1499 #endif /* CONFIG_SHRINKER_DEBUG */ 1500 1501 /* Only track the nodes of mappings with shadow entries */ 1502 void workingset_update_node(struct xa_node *node); 1503 extern struct list_lru shadow_nodes; 1504 #define mapping_set_update(xas, mapping) do { \ 1505 if (!dax_mapping(mapping) && !shmem_mapping(mapping)) { \ 1506 xas_set_update(xas, workingset_update_node); \ 1507 xas_set_lru(xas, &shadow_nodes); \ 1508 } \ 1509 } while (0) 1510 1511 /* mremap.c */ 1512 unsigned long move_page_tables(struct vm_area_struct *vma, 1513 unsigned long old_addr, struct vm_area_struct *new_vma, 1514 unsigned long new_addr, unsigned long len, 1515 bool need_rmap_locks, bool for_stack); 1516 1517 #ifdef CONFIG_UNACCEPTED_MEMORY 1518 void accept_page(struct page *page); 1519 #else /* CONFIG_UNACCEPTED_MEMORY */ 1520 static inline void accept_page(struct page *page) 1521 { 1522 } 1523 #endif /* CONFIG_UNACCEPTED_MEMORY */ 1524 1525 /* pagewalk.c */ 1526 int walk_page_range_mm(struct mm_struct *mm, unsigned long start, 1527 unsigned long end, const struct mm_walk_ops *ops, 1528 void *private); 1529 1530 /* pt_reclaim.c */ 1531 bool try_get_and_clear_pmd(struct mm_struct *mm, pmd_t *pmd, pmd_t *pmdval); 1532 void free_pte(struct mm_struct *mm, unsigned long addr, struct mmu_gather *tlb, 1533 pmd_t pmdval); 1534 void try_to_free_pte(struct mm_struct *mm, pmd_t *pmd, unsigned long addr, 1535 struct mmu_gather *tlb); 1536 1537 #ifdef CONFIG_PT_RECLAIM 1538 bool reclaim_pt_is_enabled(unsigned long start, unsigned long end, 1539 struct zap_details *details); 1540 #else 1541 static inline bool reclaim_pt_is_enabled(unsigned long start, unsigned long end, 1542 struct zap_details *details) 1543 { 1544 return false; 1545 } 1546 #endif /* CONFIG_PT_RECLAIM */ 1547 1548 1549 #endif /* __MM_INTERNAL_H */ 1550