1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* internal.h: mm/ internal definitions 3 * 4 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved. 5 * Written by David Howells (dhowells@redhat.com) 6 */ 7 #ifndef __MM_INTERNAL_H 8 #define __MM_INTERNAL_H 9 10 #include <linux/fs.h> 11 #include <linux/khugepaged.h> 12 #include <linux/mm.h> 13 #include <linux/mm_inline.h> 14 #include <linux/pagemap.h> 15 #include <linux/rmap.h> 16 #include <linux/swap.h> 17 #include <linux/swapops.h> 18 #include <linux/swap_cgroup.h> 19 #include <linux/tracepoint-defs.h> 20 21 /* Internal core VMA manipulation functions. */ 22 #include "vma.h" 23 24 struct folio_batch; 25 26 /* 27 * The set of flags that only affect watermark checking and reclaim 28 * behaviour. This is used by the MM to obey the caller constraints 29 * about IO, FS and watermark checking while ignoring placement 30 * hints such as HIGHMEM usage. 31 */ 32 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\ 33 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\ 34 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\ 35 __GFP_NOLOCKDEP) 36 37 /* The GFP flags allowed during early boot */ 38 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS)) 39 40 /* Control allocation cpuset and node placement constraints */ 41 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE) 42 43 /* Do not use these with a slab allocator */ 44 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK) 45 46 /* 47 * Different from WARN_ON_ONCE(), no warning will be issued 48 * when we specify __GFP_NOWARN. 49 */ 50 #define WARN_ON_ONCE_GFP(cond, gfp) ({ \ 51 static bool __section(".data.once") __warned; \ 52 int __ret_warn_once = !!(cond); \ 53 \ 54 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \ 55 __warned = true; \ 56 WARN_ON(1); \ 57 } \ 58 unlikely(__ret_warn_once); \ 59 }) 60 61 void page_writeback_init(void); 62 63 /* 64 * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages, 65 * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit 66 * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE). Hugetlb currently 67 * leaves nr_pages_mapped at 0, but avoid surprise if it participates later. 68 */ 69 #define ENTIRELY_MAPPED 0x800000 70 #define FOLIO_PAGES_MAPPED (ENTIRELY_MAPPED - 1) 71 72 /* 73 * Flags passed to __show_mem() and show_free_areas() to suppress output in 74 * various contexts. 75 */ 76 #define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */ 77 78 /* 79 * How many individual pages have an elevated _mapcount. Excludes 80 * the folio's entire_mapcount. 81 * 82 * Don't use this function outside of debugging code. 83 */ 84 static inline int folio_nr_pages_mapped(const struct folio *folio) 85 { 86 return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED; 87 } 88 89 /* 90 * Retrieve the first entry of a folio based on a provided entry within the 91 * folio. We cannot rely on folio->swap as there is no guarantee that it has 92 * been initialized. Used for calling arch_swap_restore() 93 */ 94 static inline swp_entry_t folio_swap(swp_entry_t entry, 95 const struct folio *folio) 96 { 97 swp_entry_t swap = { 98 .val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)), 99 }; 100 101 return swap; 102 } 103 104 static inline void *folio_raw_mapping(const struct folio *folio) 105 { 106 unsigned long mapping = (unsigned long)folio->mapping; 107 108 return (void *)(mapping & ~PAGE_MAPPING_FLAGS); 109 } 110 111 #ifdef CONFIG_MMU 112 113 /* Flags for folio_pte_batch(). */ 114 typedef int __bitwise fpb_t; 115 116 /* Compare PTEs after pte_mkclean(), ignoring the dirty bit. */ 117 #define FPB_IGNORE_DIRTY ((__force fpb_t)BIT(0)) 118 119 /* Compare PTEs after pte_clear_soft_dirty(), ignoring the soft-dirty bit. */ 120 #define FPB_IGNORE_SOFT_DIRTY ((__force fpb_t)BIT(1)) 121 122 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags) 123 { 124 if (flags & FPB_IGNORE_DIRTY) 125 pte = pte_mkclean(pte); 126 if (likely(flags & FPB_IGNORE_SOFT_DIRTY)) 127 pte = pte_clear_soft_dirty(pte); 128 return pte_wrprotect(pte_mkold(pte)); 129 } 130 131 /** 132 * folio_pte_batch - detect a PTE batch for a large folio 133 * @folio: The large folio to detect a PTE batch for. 134 * @addr: The user virtual address the first page is mapped at. 135 * @start_ptep: Page table pointer for the first entry. 136 * @pte: Page table entry for the first page. 137 * @max_nr: The maximum number of table entries to consider. 138 * @flags: Flags to modify the PTE batch semantics. 139 * @any_writable: Optional pointer to indicate whether any entry except the 140 * first one is writable. 141 * @any_young: Optional pointer to indicate whether any entry except the 142 * first one is young. 143 * @any_dirty: Optional pointer to indicate whether any entry except the 144 * first one is dirty. 145 * 146 * Detect a PTE batch: consecutive (present) PTEs that map consecutive 147 * pages of the same large folio. 148 * 149 * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN, 150 * the accessed bit, writable bit, dirty bit (with FPB_IGNORE_DIRTY) and 151 * soft-dirty bit (with FPB_IGNORE_SOFT_DIRTY). 152 * 153 * start_ptep must map any page of the folio. max_nr must be at least one and 154 * must be limited by the caller so scanning cannot exceed a single page table. 155 * 156 * Return: the number of table entries in the batch. 157 */ 158 static inline int folio_pte_batch(struct folio *folio, unsigned long addr, 159 pte_t *start_ptep, pte_t pte, int max_nr, fpb_t flags, 160 bool *any_writable, bool *any_young, bool *any_dirty) 161 { 162 unsigned long folio_end_pfn = folio_pfn(folio) + folio_nr_pages(folio); 163 const pte_t *end_ptep = start_ptep + max_nr; 164 pte_t expected_pte, *ptep; 165 bool writable, young, dirty; 166 int nr; 167 168 if (any_writable) 169 *any_writable = false; 170 if (any_young) 171 *any_young = false; 172 if (any_dirty) 173 *any_dirty = false; 174 175 VM_WARN_ON_FOLIO(!pte_present(pte), folio); 176 VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio); 177 VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio); 178 179 nr = pte_batch_hint(start_ptep, pte); 180 expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags); 181 ptep = start_ptep + nr; 182 183 while (ptep < end_ptep) { 184 pte = ptep_get(ptep); 185 if (any_writable) 186 writable = !!pte_write(pte); 187 if (any_young) 188 young = !!pte_young(pte); 189 if (any_dirty) 190 dirty = !!pte_dirty(pte); 191 pte = __pte_batch_clear_ignored(pte, flags); 192 193 if (!pte_same(pte, expected_pte)) 194 break; 195 196 /* 197 * Stop immediately once we reached the end of the folio. In 198 * corner cases the next PFN might fall into a different 199 * folio. 200 */ 201 if (pte_pfn(pte) >= folio_end_pfn) 202 break; 203 204 if (any_writable) 205 *any_writable |= writable; 206 if (any_young) 207 *any_young |= young; 208 if (any_dirty) 209 *any_dirty |= dirty; 210 211 nr = pte_batch_hint(ptep, pte); 212 expected_pte = pte_advance_pfn(expected_pte, nr); 213 ptep += nr; 214 } 215 216 return min(ptep - start_ptep, max_nr); 217 } 218 219 /** 220 * pte_move_swp_offset - Move the swap entry offset field of a swap pte 221 * forward or backward by delta 222 * @pte: The initial pte state; is_swap_pte(pte) must be true and 223 * non_swap_entry() must be false. 224 * @delta: The direction and the offset we are moving; forward if delta 225 * is positive; backward if delta is negative 226 * 227 * Moves the swap offset, while maintaining all other fields, including 228 * swap type, and any swp pte bits. The resulting pte is returned. 229 */ 230 static inline pte_t pte_move_swp_offset(pte_t pte, long delta) 231 { 232 swp_entry_t entry = pte_to_swp_entry(pte); 233 pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry), 234 (swp_offset(entry) + delta))); 235 236 if (pte_swp_soft_dirty(pte)) 237 new = pte_swp_mksoft_dirty(new); 238 if (pte_swp_exclusive(pte)) 239 new = pte_swp_mkexclusive(new); 240 if (pte_swp_uffd_wp(pte)) 241 new = pte_swp_mkuffd_wp(new); 242 243 return new; 244 } 245 246 247 /** 248 * pte_next_swp_offset - Increment the swap entry offset field of a swap pte. 249 * @pte: The initial pte state; is_swap_pte(pte) must be true and 250 * non_swap_entry() must be false. 251 * 252 * Increments the swap offset, while maintaining all other fields, including 253 * swap type, and any swp pte bits. The resulting pte is returned. 254 */ 255 static inline pte_t pte_next_swp_offset(pte_t pte) 256 { 257 return pte_move_swp_offset(pte, 1); 258 } 259 260 /** 261 * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries 262 * @start_ptep: Page table pointer for the first entry. 263 * @max_nr: The maximum number of table entries to consider. 264 * @pte: Page table entry for the first entry. 265 * 266 * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs 267 * containing swap entries all with consecutive offsets and targeting the same 268 * swap type, all with matching swp pte bits. 269 * 270 * max_nr must be at least one and must be limited by the caller so scanning 271 * cannot exceed a single page table. 272 * 273 * Return: the number of table entries in the batch. 274 */ 275 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte) 276 { 277 pte_t expected_pte = pte_next_swp_offset(pte); 278 const pte_t *end_ptep = start_ptep + max_nr; 279 swp_entry_t entry = pte_to_swp_entry(pte); 280 pte_t *ptep = start_ptep + 1; 281 unsigned short cgroup_id; 282 283 VM_WARN_ON(max_nr < 1); 284 VM_WARN_ON(!is_swap_pte(pte)); 285 VM_WARN_ON(non_swap_entry(entry)); 286 287 cgroup_id = lookup_swap_cgroup_id(entry); 288 while (ptep < end_ptep) { 289 pte = ptep_get(ptep); 290 291 if (!pte_same(pte, expected_pte)) 292 break; 293 if (lookup_swap_cgroup_id(pte_to_swp_entry(pte)) != cgroup_id) 294 break; 295 expected_pte = pte_next_swp_offset(expected_pte); 296 ptep++; 297 } 298 299 return ptep - start_ptep; 300 } 301 #endif /* CONFIG_MMU */ 302 303 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio, 304 int nr_throttled); 305 static inline void acct_reclaim_writeback(struct folio *folio) 306 { 307 pg_data_t *pgdat = folio_pgdat(folio); 308 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled); 309 310 if (nr_throttled) 311 __acct_reclaim_writeback(pgdat, folio, nr_throttled); 312 } 313 314 static inline void wake_throttle_isolated(pg_data_t *pgdat) 315 { 316 wait_queue_head_t *wqh; 317 318 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED]; 319 if (waitqueue_active(wqh)) 320 wake_up(wqh); 321 } 322 323 vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf); 324 static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf) 325 { 326 vm_fault_t ret = __vmf_anon_prepare(vmf); 327 328 if (unlikely(ret & VM_FAULT_RETRY)) 329 vma_end_read(vmf->vma); 330 return ret; 331 } 332 333 vm_fault_t do_swap_page(struct vm_fault *vmf); 334 void folio_rotate_reclaimable(struct folio *folio); 335 bool __folio_end_writeback(struct folio *folio); 336 void deactivate_file_folio(struct folio *folio); 337 void folio_activate(struct folio *folio); 338 339 void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas, 340 struct vm_area_struct *start_vma, unsigned long floor, 341 unsigned long ceiling, bool mm_wr_locked); 342 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte); 343 344 struct zap_details; 345 void unmap_page_range(struct mmu_gather *tlb, 346 struct vm_area_struct *vma, 347 unsigned long addr, unsigned long end, 348 struct zap_details *details); 349 350 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *, 351 unsigned int order); 352 void force_page_cache_ra(struct readahead_control *, unsigned long nr); 353 static inline void force_page_cache_readahead(struct address_space *mapping, 354 struct file *file, pgoff_t index, unsigned long nr_to_read) 355 { 356 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index); 357 force_page_cache_ra(&ractl, nr_to_read); 358 } 359 360 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start, 361 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 362 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start, 363 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 364 void filemap_free_folio(struct address_space *mapping, struct folio *folio); 365 int truncate_inode_folio(struct address_space *mapping, struct folio *folio); 366 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, 367 loff_t end); 368 long mapping_evict_folio(struct address_space *mapping, struct folio *folio); 369 unsigned long mapping_try_invalidate(struct address_space *mapping, 370 pgoff_t start, pgoff_t end, unsigned long *nr_failed); 371 372 /** 373 * folio_evictable - Test whether a folio is evictable. 374 * @folio: The folio to test. 375 * 376 * Test whether @folio is evictable -- i.e., should be placed on 377 * active/inactive lists vs unevictable list. 378 * 379 * Reasons folio might not be evictable: 380 * 1. folio's mapping marked unevictable 381 * 2. One of the pages in the folio is part of an mlocked VMA 382 */ 383 static inline bool folio_evictable(struct folio *folio) 384 { 385 bool ret; 386 387 /* Prevent address_space of inode and swap cache from being freed */ 388 rcu_read_lock(); 389 ret = !mapping_unevictable(folio_mapping(folio)) && 390 !folio_test_mlocked(folio); 391 rcu_read_unlock(); 392 return ret; 393 } 394 395 /* 396 * Turn a non-refcounted page (->_refcount == 0) into refcounted with 397 * a count of one. 398 */ 399 static inline void set_page_refcounted(struct page *page) 400 { 401 VM_BUG_ON_PAGE(PageTail(page), page); 402 VM_BUG_ON_PAGE(page_ref_count(page), page); 403 set_page_count(page, 1); 404 } 405 406 /* 407 * Return true if a folio needs ->release_folio() calling upon it. 408 */ 409 static inline bool folio_needs_release(struct folio *folio) 410 { 411 struct address_space *mapping = folio_mapping(folio); 412 413 return folio_has_private(folio) || 414 (mapping && mapping_release_always(mapping)); 415 } 416 417 extern unsigned long highest_memmap_pfn; 418 419 /* 420 * Maximum number of reclaim retries without progress before the OOM 421 * killer is consider the only way forward. 422 */ 423 #define MAX_RECLAIM_RETRIES 16 424 425 /* 426 * in mm/vmscan.c: 427 */ 428 bool folio_isolate_lru(struct folio *folio); 429 void folio_putback_lru(struct folio *folio); 430 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason); 431 432 /* 433 * in mm/rmap.c: 434 */ 435 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address); 436 437 /* 438 * in mm/page_alloc.c 439 */ 440 #define K(x) ((x) << (PAGE_SHIFT-10)) 441 442 extern char * const zone_names[MAX_NR_ZONES]; 443 444 /* perform sanity checks on struct pages being allocated or freed */ 445 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled); 446 447 extern int min_free_kbytes; 448 449 void setup_per_zone_wmarks(void); 450 void calculate_min_free_kbytes(void); 451 int __meminit init_per_zone_wmark_min(void); 452 void page_alloc_sysctl_init(void); 453 454 /* 455 * Structure for holding the mostly immutable allocation parameters passed 456 * between functions involved in allocations, including the alloc_pages* 457 * family of functions. 458 * 459 * nodemask, migratetype and highest_zoneidx are initialized only once in 460 * __alloc_pages() and then never change. 461 * 462 * zonelist, preferred_zone and highest_zoneidx are set first in 463 * __alloc_pages() for the fast path, and might be later changed 464 * in __alloc_pages_slowpath(). All other functions pass the whole structure 465 * by a const pointer. 466 */ 467 struct alloc_context { 468 struct zonelist *zonelist; 469 nodemask_t *nodemask; 470 struct zoneref *preferred_zoneref; 471 int migratetype; 472 473 /* 474 * highest_zoneidx represents highest usable zone index of 475 * the allocation request. Due to the nature of the zone, 476 * memory on lower zone than the highest_zoneidx will be 477 * protected by lowmem_reserve[highest_zoneidx]. 478 * 479 * highest_zoneidx is also used by reclaim/compaction to limit 480 * the target zone since higher zone than this index cannot be 481 * usable for this allocation request. 482 */ 483 enum zone_type highest_zoneidx; 484 bool spread_dirty_pages; 485 }; 486 487 /* 488 * This function returns the order of a free page in the buddy system. In 489 * general, page_zone(page)->lock must be held by the caller to prevent the 490 * page from being allocated in parallel and returning garbage as the order. 491 * If a caller does not hold page_zone(page)->lock, it must guarantee that the 492 * page cannot be allocated or merged in parallel. Alternatively, it must 493 * handle invalid values gracefully, and use buddy_order_unsafe() below. 494 */ 495 static inline unsigned int buddy_order(struct page *page) 496 { 497 /* PageBuddy() must be checked by the caller */ 498 return page_private(page); 499 } 500 501 /* 502 * Like buddy_order(), but for callers who cannot afford to hold the zone lock. 503 * PageBuddy() should be checked first by the caller to minimize race window, 504 * and invalid values must be handled gracefully. 505 * 506 * READ_ONCE is used so that if the caller assigns the result into a local 507 * variable and e.g. tests it for valid range before using, the compiler cannot 508 * decide to remove the variable and inline the page_private(page) multiple 509 * times, potentially observing different values in the tests and the actual 510 * use of the result. 511 */ 512 #define buddy_order_unsafe(page) READ_ONCE(page_private(page)) 513 514 /* 515 * This function checks whether a page is free && is the buddy 516 * we can coalesce a page and its buddy if 517 * (a) the buddy is not in a hole (check before calling!) && 518 * (b) the buddy is in the buddy system && 519 * (c) a page and its buddy have the same order && 520 * (d) a page and its buddy are in the same zone. 521 * 522 * For recording whether a page is in the buddy system, we set PageBuddy. 523 * Setting, clearing, and testing PageBuddy is serialized by zone->lock. 524 * 525 * For recording page's order, we use page_private(page). 526 */ 527 static inline bool page_is_buddy(struct page *page, struct page *buddy, 528 unsigned int order) 529 { 530 if (!page_is_guard(buddy) && !PageBuddy(buddy)) 531 return false; 532 533 if (buddy_order(buddy) != order) 534 return false; 535 536 /* 537 * zone check is done late to avoid uselessly calculating 538 * zone/node ids for pages that could never merge. 539 */ 540 if (page_zone_id(page) != page_zone_id(buddy)) 541 return false; 542 543 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); 544 545 return true; 546 } 547 548 /* 549 * Locate the struct page for both the matching buddy in our 550 * pair (buddy1) and the combined O(n+1) page they form (page). 551 * 552 * 1) Any buddy B1 will have an order O twin B2 which satisfies 553 * the following equation: 554 * B2 = B1 ^ (1 << O) 555 * For example, if the starting buddy (buddy2) is #8 its order 556 * 1 buddy is #10: 557 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 558 * 559 * 2) Any buddy B will have an order O+1 parent P which 560 * satisfies the following equation: 561 * P = B & ~(1 << O) 562 * 563 * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER 564 */ 565 static inline unsigned long 566 __find_buddy_pfn(unsigned long page_pfn, unsigned int order) 567 { 568 return page_pfn ^ (1 << order); 569 } 570 571 /* 572 * Find the buddy of @page and validate it. 573 * @page: The input page 574 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the 575 * function is used in the performance-critical __free_one_page(). 576 * @order: The order of the page 577 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to 578 * page_to_pfn(). 579 * 580 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is 581 * not the same as @page. The validation is necessary before use it. 582 * 583 * Return: the found buddy page or NULL if not found. 584 */ 585 static inline struct page *find_buddy_page_pfn(struct page *page, 586 unsigned long pfn, unsigned int order, unsigned long *buddy_pfn) 587 { 588 unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order); 589 struct page *buddy; 590 591 buddy = page + (__buddy_pfn - pfn); 592 if (buddy_pfn) 593 *buddy_pfn = __buddy_pfn; 594 595 if (page_is_buddy(page, buddy, order)) 596 return buddy; 597 return NULL; 598 } 599 600 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn, 601 unsigned long end_pfn, struct zone *zone); 602 603 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn, 604 unsigned long end_pfn, struct zone *zone) 605 { 606 if (zone->contiguous) 607 return pfn_to_page(start_pfn); 608 609 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone); 610 } 611 612 void set_zone_contiguous(struct zone *zone); 613 614 static inline void clear_zone_contiguous(struct zone *zone) 615 { 616 zone->contiguous = false; 617 } 618 619 extern int __isolate_free_page(struct page *page, unsigned int order); 620 extern void __putback_isolated_page(struct page *page, unsigned int order, 621 int mt); 622 extern void memblock_free_pages(struct page *page, unsigned long pfn, 623 unsigned int order); 624 extern void __free_pages_core(struct page *page, unsigned int order, 625 enum meminit_context context); 626 627 /* 628 * This will have no effect, other than possibly generating a warning, if the 629 * caller passes in a non-large folio. 630 */ 631 static inline void folio_set_order(struct folio *folio, unsigned int order) 632 { 633 if (WARN_ON_ONCE(!order || !folio_test_large(folio))) 634 return; 635 636 folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order; 637 #ifdef CONFIG_64BIT 638 folio->_folio_nr_pages = 1U << order; 639 #endif 640 } 641 642 void __folio_undo_large_rmappable(struct folio *folio); 643 static inline void folio_undo_large_rmappable(struct folio *folio) 644 { 645 if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio)) 646 return; 647 648 /* 649 * At this point, there is no one trying to add the folio to 650 * deferred_list. If folio is not in deferred_list, it's safe 651 * to check without acquiring the split_queue_lock. 652 */ 653 if (data_race(list_empty(&folio->_deferred_list))) 654 return; 655 656 __folio_undo_large_rmappable(folio); 657 } 658 659 static inline struct folio *page_rmappable_folio(struct page *page) 660 { 661 struct folio *folio = (struct folio *)page; 662 663 if (folio && folio_test_large(folio)) 664 folio_set_large_rmappable(folio); 665 return folio; 666 } 667 668 static inline void prep_compound_head(struct page *page, unsigned int order) 669 { 670 struct folio *folio = (struct folio *)page; 671 672 folio_set_order(folio, order); 673 atomic_set(&folio->_large_mapcount, -1); 674 atomic_set(&folio->_entire_mapcount, -1); 675 atomic_set(&folio->_nr_pages_mapped, 0); 676 atomic_set(&folio->_pincount, 0); 677 if (order > 1) 678 INIT_LIST_HEAD(&folio->_deferred_list); 679 } 680 681 static inline void prep_compound_tail(struct page *head, int tail_idx) 682 { 683 struct page *p = head + tail_idx; 684 685 p->mapping = TAIL_MAPPING; 686 set_compound_head(p, head); 687 set_page_private(p, 0); 688 } 689 690 extern void prep_compound_page(struct page *page, unsigned int order); 691 692 extern void post_alloc_hook(struct page *page, unsigned int order, 693 gfp_t gfp_flags); 694 extern bool free_pages_prepare(struct page *page, unsigned int order); 695 696 extern int user_min_free_kbytes; 697 698 void free_unref_page(struct page *page, unsigned int order); 699 void free_unref_folios(struct folio_batch *fbatch); 700 701 extern void zone_pcp_reset(struct zone *zone); 702 extern void zone_pcp_disable(struct zone *zone); 703 extern void zone_pcp_enable(struct zone *zone); 704 extern void zone_pcp_init(struct zone *zone); 705 706 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align, 707 phys_addr_t min_addr, 708 int nid, bool exact_nid); 709 710 void memmap_init_range(unsigned long, int, unsigned long, unsigned long, 711 unsigned long, enum meminit_context, struct vmem_altmap *, int); 712 713 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 714 715 /* 716 * in mm/compaction.c 717 */ 718 /* 719 * compact_control is used to track pages being migrated and the free pages 720 * they are being migrated to during memory compaction. The free_pfn starts 721 * at the end of a zone and migrate_pfn begins at the start. Movable pages 722 * are moved to the end of a zone during a compaction run and the run 723 * completes when free_pfn <= migrate_pfn 724 */ 725 struct compact_control { 726 struct list_head freepages[NR_PAGE_ORDERS]; /* List of free pages to migrate to */ 727 struct list_head migratepages; /* List of pages being migrated */ 728 unsigned int nr_freepages; /* Number of isolated free pages */ 729 unsigned int nr_migratepages; /* Number of pages to migrate */ 730 unsigned long free_pfn; /* isolate_freepages search base */ 731 /* 732 * Acts as an in/out parameter to page isolation for migration. 733 * isolate_migratepages uses it as a search base. 734 * isolate_migratepages_block will update the value to the next pfn 735 * after the last isolated one. 736 */ 737 unsigned long migrate_pfn; 738 unsigned long fast_start_pfn; /* a pfn to start linear scan from */ 739 struct zone *zone; 740 unsigned long total_migrate_scanned; 741 unsigned long total_free_scanned; 742 unsigned short fast_search_fail;/* failures to use free list searches */ 743 short search_order; /* order to start a fast search at */ 744 const gfp_t gfp_mask; /* gfp mask of a direct compactor */ 745 int order; /* order a direct compactor needs */ 746 int migratetype; /* migratetype of direct compactor */ 747 const unsigned int alloc_flags; /* alloc flags of a direct compactor */ 748 const int highest_zoneidx; /* zone index of a direct compactor */ 749 enum migrate_mode mode; /* Async or sync migration mode */ 750 bool ignore_skip_hint; /* Scan blocks even if marked skip */ 751 bool no_set_skip_hint; /* Don't mark blocks for skipping */ 752 bool ignore_block_suitable; /* Scan blocks considered unsuitable */ 753 bool direct_compaction; /* False from kcompactd or /proc/... */ 754 bool proactive_compaction; /* kcompactd proactive compaction */ 755 bool whole_zone; /* Whole zone should/has been scanned */ 756 bool contended; /* Signal lock contention */ 757 bool finish_pageblock; /* Scan the remainder of a pageblock. Used 758 * when there are potentially transient 759 * isolation or migration failures to 760 * ensure forward progress. 761 */ 762 bool alloc_contig; /* alloc_contig_range allocation */ 763 }; 764 765 /* 766 * Used in direct compaction when a page should be taken from the freelists 767 * immediately when one is created during the free path. 768 */ 769 struct capture_control { 770 struct compact_control *cc; 771 struct page *page; 772 }; 773 774 unsigned long 775 isolate_freepages_range(struct compact_control *cc, 776 unsigned long start_pfn, unsigned long end_pfn); 777 int 778 isolate_migratepages_range(struct compact_control *cc, 779 unsigned long low_pfn, unsigned long end_pfn); 780 781 int __alloc_contig_migrate_range(struct compact_control *cc, 782 unsigned long start, unsigned long end, 783 int migratetype); 784 785 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ 786 void init_cma_reserved_pageblock(struct page *page); 787 788 #endif /* CONFIG_COMPACTION || CONFIG_CMA */ 789 790 int find_suitable_fallback(struct free_area *area, unsigned int order, 791 int migratetype, bool only_stealable, bool *can_steal); 792 793 static inline bool free_area_empty(struct free_area *area, int migratetype) 794 { 795 return list_empty(&area->free_list[migratetype]); 796 } 797 798 /* mm/util.c */ 799 struct anon_vma *folio_anon_vma(struct folio *folio); 800 801 #ifdef CONFIG_MMU 802 void unmap_mapping_folio(struct folio *folio); 803 extern long populate_vma_page_range(struct vm_area_struct *vma, 804 unsigned long start, unsigned long end, int *locked); 805 extern long faultin_page_range(struct mm_struct *mm, unsigned long start, 806 unsigned long end, bool write, int *locked); 807 extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 808 unsigned long bytes); 809 810 /* 811 * NOTE: This function can't tell whether the folio is "fully mapped" in the 812 * range. 813 * "fully mapped" means all the pages of folio is associated with the page 814 * table of range while this function just check whether the folio range is 815 * within the range [start, end). Function caller needs to do page table 816 * check if it cares about the page table association. 817 * 818 * Typical usage (like mlock or madvise) is: 819 * Caller knows at least 1 page of folio is associated with page table of VMA 820 * and the range [start, end) is intersect with the VMA range. Caller wants 821 * to know whether the folio is fully associated with the range. It calls 822 * this function to check whether the folio is in the range first. Then checks 823 * the page table to know whether the folio is fully mapped to the range. 824 */ 825 static inline bool 826 folio_within_range(struct folio *folio, struct vm_area_struct *vma, 827 unsigned long start, unsigned long end) 828 { 829 pgoff_t pgoff, addr; 830 unsigned long vma_pglen = vma_pages(vma); 831 832 VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio); 833 if (start > end) 834 return false; 835 836 if (start < vma->vm_start) 837 start = vma->vm_start; 838 839 if (end > vma->vm_end) 840 end = vma->vm_end; 841 842 pgoff = folio_pgoff(folio); 843 844 /* if folio start address is not in vma range */ 845 if (!in_range(pgoff, vma->vm_pgoff, vma_pglen)) 846 return false; 847 848 addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 849 850 return !(addr < start || end - addr < folio_size(folio)); 851 } 852 853 static inline bool 854 folio_within_vma(struct folio *folio, struct vm_area_struct *vma) 855 { 856 return folio_within_range(folio, vma, vma->vm_start, vma->vm_end); 857 } 858 859 /* 860 * mlock_vma_folio() and munlock_vma_folio(): 861 * should be called with vma's mmap_lock held for read or write, 862 * under page table lock for the pte/pmd being added or removed. 863 * 864 * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at 865 * the end of folio_remove_rmap_*(); but new anon folios are managed by 866 * folio_add_lru_vma() calling mlock_new_folio(). 867 */ 868 void mlock_folio(struct folio *folio); 869 static inline void mlock_vma_folio(struct folio *folio, 870 struct vm_area_struct *vma) 871 { 872 /* 873 * The VM_SPECIAL check here serves two purposes. 874 * 1) VM_IO check prevents migration from double-counting during mlock. 875 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED 876 * is never left set on a VM_SPECIAL vma, there is an interval while 877 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may 878 * still be set while VM_SPECIAL bits are added: so ignore it then. 879 */ 880 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED)) 881 mlock_folio(folio); 882 } 883 884 void munlock_folio(struct folio *folio); 885 static inline void munlock_vma_folio(struct folio *folio, 886 struct vm_area_struct *vma) 887 { 888 /* 889 * munlock if the function is called. Ideally, we should only 890 * do munlock if any page of folio is unmapped from VMA and 891 * cause folio not fully mapped to VMA. 892 * 893 * But it's not easy to confirm that's the situation. So we 894 * always munlock the folio and page reclaim will correct it 895 * if it's wrong. 896 */ 897 if (unlikely(vma->vm_flags & VM_LOCKED)) 898 munlock_folio(folio); 899 } 900 901 void mlock_new_folio(struct folio *folio); 902 bool need_mlock_drain(int cpu); 903 void mlock_drain_local(void); 904 void mlock_drain_remote(int cpu); 905 906 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma); 907 908 /** 909 * vma_address - Find the virtual address a page range is mapped at 910 * @vma: The vma which maps this object. 911 * @pgoff: The page offset within its object. 912 * @nr_pages: The number of pages to consider. 913 * 914 * If any page in this range is mapped by this VMA, return the first address 915 * where any of these pages appear. Otherwise, return -EFAULT. 916 */ 917 static inline unsigned long vma_address(struct vm_area_struct *vma, 918 pgoff_t pgoff, unsigned long nr_pages) 919 { 920 unsigned long address; 921 922 if (pgoff >= vma->vm_pgoff) { 923 address = vma->vm_start + 924 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 925 /* Check for address beyond vma (or wrapped through 0?) */ 926 if (address < vma->vm_start || address >= vma->vm_end) 927 address = -EFAULT; 928 } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) { 929 /* Test above avoids possibility of wrap to 0 on 32-bit */ 930 address = vma->vm_start; 931 } else { 932 address = -EFAULT; 933 } 934 return address; 935 } 936 937 /* 938 * Then at what user virtual address will none of the range be found in vma? 939 * Assumes that vma_address() already returned a good starting address. 940 */ 941 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw) 942 { 943 struct vm_area_struct *vma = pvmw->vma; 944 pgoff_t pgoff; 945 unsigned long address; 946 947 /* Common case, plus ->pgoff is invalid for KSM */ 948 if (pvmw->nr_pages == 1) 949 return pvmw->address + PAGE_SIZE; 950 951 pgoff = pvmw->pgoff + pvmw->nr_pages; 952 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 953 /* Check for address beyond vma (or wrapped through 0?) */ 954 if (address < vma->vm_start || address > vma->vm_end) 955 address = vma->vm_end; 956 return address; 957 } 958 959 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf, 960 struct file *fpin) 961 { 962 int flags = vmf->flags; 963 964 if (fpin) 965 return fpin; 966 967 /* 968 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or 969 * anything, so we only pin the file and drop the mmap_lock if only 970 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt. 971 */ 972 if (fault_flag_allow_retry_first(flags) && 973 !(flags & FAULT_FLAG_RETRY_NOWAIT)) { 974 fpin = get_file(vmf->vma->vm_file); 975 release_fault_lock(vmf); 976 } 977 return fpin; 978 } 979 #else /* !CONFIG_MMU */ 980 static inline void unmap_mapping_folio(struct folio *folio) { } 981 static inline void mlock_new_folio(struct folio *folio) { } 982 static inline bool need_mlock_drain(int cpu) { return false; } 983 static inline void mlock_drain_local(void) { } 984 static inline void mlock_drain_remote(int cpu) { } 985 static inline void vunmap_range_noflush(unsigned long start, unsigned long end) 986 { 987 } 988 #endif /* !CONFIG_MMU */ 989 990 /* Memory initialisation debug and verification */ 991 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 992 DECLARE_STATIC_KEY_TRUE(deferred_pages); 993 994 bool __init deferred_grow_zone(struct zone *zone, unsigned int order); 995 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 996 997 enum mminit_level { 998 MMINIT_WARNING, 999 MMINIT_VERIFY, 1000 MMINIT_TRACE 1001 }; 1002 1003 #ifdef CONFIG_DEBUG_MEMORY_INIT 1004 1005 extern int mminit_loglevel; 1006 1007 #define mminit_dprintk(level, prefix, fmt, arg...) \ 1008 do { \ 1009 if (level < mminit_loglevel) { \ 1010 if (level <= MMINIT_WARNING) \ 1011 pr_warn("mminit::" prefix " " fmt, ##arg); \ 1012 else \ 1013 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \ 1014 } \ 1015 } while (0) 1016 1017 extern void mminit_verify_pageflags_layout(void); 1018 extern void mminit_verify_zonelist(void); 1019 #else 1020 1021 static inline void mminit_dprintk(enum mminit_level level, 1022 const char *prefix, const char *fmt, ...) 1023 { 1024 } 1025 1026 static inline void mminit_verify_pageflags_layout(void) 1027 { 1028 } 1029 1030 static inline void mminit_verify_zonelist(void) 1031 { 1032 } 1033 #endif /* CONFIG_DEBUG_MEMORY_INIT */ 1034 1035 #define NODE_RECLAIM_NOSCAN -2 1036 #define NODE_RECLAIM_FULL -1 1037 #define NODE_RECLAIM_SOME 0 1038 #define NODE_RECLAIM_SUCCESS 1 1039 1040 #ifdef CONFIG_NUMA 1041 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int); 1042 extern int find_next_best_node(int node, nodemask_t *used_node_mask); 1043 #else 1044 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask, 1045 unsigned int order) 1046 { 1047 return NODE_RECLAIM_NOSCAN; 1048 } 1049 static inline int find_next_best_node(int node, nodemask_t *used_node_mask) 1050 { 1051 return NUMA_NO_NODE; 1052 } 1053 #endif 1054 1055 /* 1056 * mm/memory-failure.c 1057 */ 1058 #ifdef CONFIG_MEMORY_FAILURE 1059 void unmap_poisoned_folio(struct folio *folio, enum ttu_flags ttu); 1060 void shake_folio(struct folio *folio); 1061 extern int hwpoison_filter(struct page *p); 1062 1063 extern u32 hwpoison_filter_dev_major; 1064 extern u32 hwpoison_filter_dev_minor; 1065 extern u64 hwpoison_filter_flags_mask; 1066 extern u64 hwpoison_filter_flags_value; 1067 extern u64 hwpoison_filter_memcg; 1068 extern u32 hwpoison_filter_enable; 1069 #define MAGIC_HWPOISON 0x48575053U /* HWPS */ 1070 void SetPageHWPoisonTakenOff(struct page *page); 1071 void ClearPageHWPoisonTakenOff(struct page *page); 1072 bool take_page_off_buddy(struct page *page); 1073 bool put_page_back_buddy(struct page *page); 1074 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early); 1075 void add_to_kill_ksm(struct task_struct *tsk, struct page *p, 1076 struct vm_area_struct *vma, struct list_head *to_kill, 1077 unsigned long ksm_addr); 1078 unsigned long page_mapped_in_vma(struct page *page, struct vm_area_struct *vma); 1079 1080 #else 1081 static inline void unmap_poisoned_folio(struct folio *folio, enum ttu_flags ttu) 1082 { 1083 } 1084 #endif 1085 1086 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long, 1087 unsigned long, unsigned long, 1088 unsigned long, unsigned long); 1089 1090 extern void set_pageblock_order(void); 1091 struct folio *alloc_migrate_folio(struct folio *src, unsigned long private); 1092 unsigned long reclaim_pages(struct list_head *folio_list); 1093 unsigned int reclaim_clean_pages_from_list(struct zone *zone, 1094 struct list_head *folio_list); 1095 /* The ALLOC_WMARK bits are used as an index to zone->watermark */ 1096 #define ALLOC_WMARK_MIN WMARK_MIN 1097 #define ALLOC_WMARK_LOW WMARK_LOW 1098 #define ALLOC_WMARK_HIGH WMARK_HIGH 1099 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ 1100 1101 /* Mask to get the watermark bits */ 1102 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) 1103 1104 /* 1105 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we 1106 * cannot assume a reduced access to memory reserves is sufficient for 1107 * !MMU 1108 */ 1109 #ifdef CONFIG_MMU 1110 #define ALLOC_OOM 0x08 1111 #else 1112 #define ALLOC_OOM ALLOC_NO_WATERMARKS 1113 #endif 1114 1115 #define ALLOC_NON_BLOCK 0x10 /* Caller cannot block. Allow access 1116 * to 25% of the min watermark or 1117 * 62.5% if __GFP_HIGH is set. 1118 */ 1119 #define ALLOC_MIN_RESERVE 0x20 /* __GFP_HIGH set. Allow access to 50% 1120 * of the min watermark. 1121 */ 1122 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ 1123 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */ 1124 #ifdef CONFIG_ZONE_DMA32 1125 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */ 1126 #else 1127 #define ALLOC_NOFRAGMENT 0x0 1128 #endif 1129 #define ALLOC_HIGHATOMIC 0x200 /* Allows access to MIGRATE_HIGHATOMIC */ 1130 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */ 1131 1132 /* Flags that allow allocations below the min watermark. */ 1133 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM) 1134 1135 enum ttu_flags; 1136 struct tlbflush_unmap_batch; 1137 1138 1139 /* 1140 * only for MM internal work items which do not depend on 1141 * any allocations or locks which might depend on allocations 1142 */ 1143 extern struct workqueue_struct *mm_percpu_wq; 1144 1145 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 1146 void try_to_unmap_flush(void); 1147 void try_to_unmap_flush_dirty(void); 1148 void flush_tlb_batched_pending(struct mm_struct *mm); 1149 #else 1150 static inline void try_to_unmap_flush(void) 1151 { 1152 } 1153 static inline void try_to_unmap_flush_dirty(void) 1154 { 1155 } 1156 static inline void flush_tlb_batched_pending(struct mm_struct *mm) 1157 { 1158 } 1159 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ 1160 1161 extern const struct trace_print_flags pageflag_names[]; 1162 extern const struct trace_print_flags vmaflag_names[]; 1163 extern const struct trace_print_flags gfpflag_names[]; 1164 1165 static inline bool is_migrate_highatomic(enum migratetype migratetype) 1166 { 1167 return migratetype == MIGRATE_HIGHATOMIC; 1168 } 1169 1170 void setup_zone_pageset(struct zone *zone); 1171 1172 struct migration_target_control { 1173 int nid; /* preferred node id */ 1174 nodemask_t *nmask; 1175 gfp_t gfp_mask; 1176 enum migrate_reason reason; 1177 }; 1178 1179 /* 1180 * mm/filemap.c 1181 */ 1182 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe, 1183 struct folio *folio, loff_t fpos, size_t size); 1184 1185 /* 1186 * mm/vmalloc.c 1187 */ 1188 #ifdef CONFIG_MMU 1189 void __init vmalloc_init(void); 1190 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1191 pgprot_t prot, struct page **pages, unsigned int page_shift); 1192 #else 1193 static inline void vmalloc_init(void) 1194 { 1195 } 1196 1197 static inline 1198 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1199 pgprot_t prot, struct page **pages, unsigned int page_shift) 1200 { 1201 return -EINVAL; 1202 } 1203 #endif 1204 1205 int __must_check __vmap_pages_range_noflush(unsigned long addr, 1206 unsigned long end, pgprot_t prot, 1207 struct page **pages, unsigned int page_shift); 1208 1209 void vunmap_range_noflush(unsigned long start, unsigned long end); 1210 1211 void __vunmap_range_noflush(unsigned long start, unsigned long end); 1212 1213 int numa_migrate_check(struct folio *folio, struct vm_fault *vmf, 1214 unsigned long addr, int *flags, bool writable, 1215 int *last_cpupid); 1216 1217 void free_zone_device_folio(struct folio *folio); 1218 int migrate_device_coherent_folio(struct folio *folio); 1219 1220 /* 1221 * mm/gup.c 1222 */ 1223 int __must_check try_grab_folio(struct folio *folio, int refs, 1224 unsigned int flags); 1225 1226 /* 1227 * mm/huge_memory.c 1228 */ 1229 void touch_pud(struct vm_area_struct *vma, unsigned long addr, 1230 pud_t *pud, bool write); 1231 void touch_pmd(struct vm_area_struct *vma, unsigned long addr, 1232 pmd_t *pmd, bool write); 1233 1234 enum { 1235 /* mark page accessed */ 1236 FOLL_TOUCH = 1 << 16, 1237 /* a retry, previous pass started an IO */ 1238 FOLL_TRIED = 1 << 17, 1239 /* we are working on non-current tsk/mm */ 1240 FOLL_REMOTE = 1 << 18, 1241 /* pages must be released via unpin_user_page */ 1242 FOLL_PIN = 1 << 19, 1243 /* gup_fast: prevent fall-back to slow gup */ 1244 FOLL_FAST_ONLY = 1 << 20, 1245 /* allow unlocking the mmap lock */ 1246 FOLL_UNLOCKABLE = 1 << 21, 1247 /* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */ 1248 FOLL_MADV_POPULATE = 1 << 22, 1249 }; 1250 1251 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \ 1252 FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \ 1253 FOLL_MADV_POPULATE) 1254 1255 /* 1256 * Indicates for which pages that are write-protected in the page table, 1257 * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the 1258 * GUP pin will remain consistent with the pages mapped into the page tables 1259 * of the MM. 1260 * 1261 * Temporary unmapping of PageAnonExclusive() pages or clearing of 1262 * PageAnonExclusive() has to protect against concurrent GUP: 1263 * * Ordinary GUP: Using the PT lock 1264 * * GUP-fast and fork(): mm->write_protect_seq 1265 * * GUP-fast and KSM or temporary unmapping (swap, migration): see 1266 * folio_try_share_anon_rmap_*() 1267 * 1268 * Must be called with the (sub)page that's actually referenced via the 1269 * page table entry, which might not necessarily be the head page for a 1270 * PTE-mapped THP. 1271 * 1272 * If the vma is NULL, we're coming from the GUP-fast path and might have 1273 * to fallback to the slow path just to lookup the vma. 1274 */ 1275 static inline bool gup_must_unshare(struct vm_area_struct *vma, 1276 unsigned int flags, struct page *page) 1277 { 1278 /* 1279 * FOLL_WRITE is implicitly handled correctly as the page table entry 1280 * has to be writable -- and if it references (part of) an anonymous 1281 * folio, that part is required to be marked exclusive. 1282 */ 1283 if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN) 1284 return false; 1285 /* 1286 * Note: PageAnon(page) is stable until the page is actually getting 1287 * freed. 1288 */ 1289 if (!PageAnon(page)) { 1290 /* 1291 * We only care about R/O long-term pining: R/O short-term 1292 * pinning does not have the semantics to observe successive 1293 * changes through the process page tables. 1294 */ 1295 if (!(flags & FOLL_LONGTERM)) 1296 return false; 1297 1298 /* We really need the vma ... */ 1299 if (!vma) 1300 return true; 1301 1302 /* 1303 * ... because we only care about writable private ("COW") 1304 * mappings where we have to break COW early. 1305 */ 1306 return is_cow_mapping(vma->vm_flags); 1307 } 1308 1309 /* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */ 1310 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) 1311 smp_rmb(); 1312 1313 /* 1314 * Note that PageKsm() pages cannot be exclusive, and consequently, 1315 * cannot get pinned. 1316 */ 1317 return !PageAnonExclusive(page); 1318 } 1319 1320 extern bool mirrored_kernelcore; 1321 extern bool memblock_has_mirror(void); 1322 1323 static __always_inline void vma_set_range(struct vm_area_struct *vma, 1324 unsigned long start, unsigned long end, 1325 pgoff_t pgoff) 1326 { 1327 vma->vm_start = start; 1328 vma->vm_end = end; 1329 vma->vm_pgoff = pgoff; 1330 } 1331 1332 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma) 1333 { 1334 /* 1335 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty 1336 * enablements, because when without soft-dirty being compiled in, 1337 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY) 1338 * will be constantly true. 1339 */ 1340 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 1341 return false; 1342 1343 /* 1344 * Soft-dirty is kind of special: its tracking is enabled when the 1345 * vma flags not set. 1346 */ 1347 return !(vma->vm_flags & VM_SOFTDIRTY); 1348 } 1349 1350 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd) 1351 { 1352 return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd); 1353 } 1354 1355 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte) 1356 { 1357 return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte); 1358 } 1359 1360 void __meminit __init_single_page(struct page *page, unsigned long pfn, 1361 unsigned long zone, int nid); 1362 1363 /* shrinker related functions */ 1364 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg, 1365 int priority); 1366 1367 #ifdef CONFIG_64BIT 1368 static inline int can_do_mseal(unsigned long flags) 1369 { 1370 if (flags) 1371 return -EINVAL; 1372 1373 return 0; 1374 } 1375 1376 #else 1377 static inline int can_do_mseal(unsigned long flags) 1378 { 1379 return -EPERM; 1380 } 1381 #endif 1382 1383 #ifdef CONFIG_SHRINKER_DEBUG 1384 static inline __printf(2, 0) int shrinker_debugfs_name_alloc( 1385 struct shrinker *shrinker, const char *fmt, va_list ap) 1386 { 1387 shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap); 1388 1389 return shrinker->name ? 0 : -ENOMEM; 1390 } 1391 1392 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1393 { 1394 kfree_const(shrinker->name); 1395 shrinker->name = NULL; 1396 } 1397 1398 extern int shrinker_debugfs_add(struct shrinker *shrinker); 1399 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1400 int *debugfs_id); 1401 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1402 int debugfs_id); 1403 #else /* CONFIG_SHRINKER_DEBUG */ 1404 static inline int shrinker_debugfs_add(struct shrinker *shrinker) 1405 { 1406 return 0; 1407 } 1408 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker, 1409 const char *fmt, va_list ap) 1410 { 1411 return 0; 1412 } 1413 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1414 { 1415 } 1416 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1417 int *debugfs_id) 1418 { 1419 *debugfs_id = -1; 1420 return NULL; 1421 } 1422 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1423 int debugfs_id) 1424 { 1425 } 1426 #endif /* CONFIG_SHRINKER_DEBUG */ 1427 1428 /* Only track the nodes of mappings with shadow entries */ 1429 void workingset_update_node(struct xa_node *node); 1430 extern struct list_lru shadow_nodes; 1431 1432 /* mremap.c */ 1433 unsigned long move_page_tables(struct vm_area_struct *vma, 1434 unsigned long old_addr, struct vm_area_struct *new_vma, 1435 unsigned long new_addr, unsigned long len, 1436 bool need_rmap_locks, bool for_stack); 1437 1438 #ifdef CONFIG_UNACCEPTED_MEMORY 1439 void accept_page(struct page *page); 1440 #else /* CONFIG_UNACCEPTED_MEMORY */ 1441 static inline void accept_page(struct page *page) 1442 { 1443 } 1444 #endif /* CONFIG_UNACCEPTED_MEMORY */ 1445 1446 #endif /* __MM_INTERNAL_H */ 1447