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 * Maintains state across a page table move. The operation assumes both source 29 * and destination VMAs already exist and are specified by the user. 30 * 31 * Partial moves are permitted, but the old and new ranges must both reside 32 * within a VMA. 33 * 34 * mmap lock must be held in write and VMA write locks must be held on any VMA 35 * that is visible. 36 * 37 * Use the PAGETABLE_MOVE() macro to initialise this struct. 38 * 39 * The old_addr and new_addr fields are updated as the page table move is 40 * executed. 41 * 42 * NOTE: The page table move is affected by reading from [old_addr, old_end), 43 * and old_addr may be updated for better page table alignment, so len_in 44 * represents the length of the range being copied as specified by the user. 45 */ 46 struct pagetable_move_control { 47 struct vm_area_struct *old; /* Source VMA. */ 48 struct vm_area_struct *new; /* Destination VMA. */ 49 unsigned long old_addr; /* Address from which the move begins. */ 50 unsigned long old_end; /* Exclusive address at which old range ends. */ 51 unsigned long new_addr; /* Address to move page tables to. */ 52 unsigned long len_in; /* Bytes to remap specified by user. */ 53 54 bool need_rmap_locks; /* Do rmap locks need to be taken? */ 55 bool for_stack; /* Is this an early temp stack being moved? */ 56 }; 57 58 #define PAGETABLE_MOVE(name, old_, new_, old_addr_, new_addr_, len_) \ 59 struct pagetable_move_control name = { \ 60 .old = old_, \ 61 .new = new_, \ 62 .old_addr = old_addr_, \ 63 .old_end = (old_addr_) + (len_), \ 64 .new_addr = new_addr_, \ 65 .len_in = len_, \ 66 } 67 68 /* 69 * The set of flags that only affect watermark checking and reclaim 70 * behaviour. This is used by the MM to obey the caller constraints 71 * about IO, FS and watermark checking while ignoring placement 72 * hints such as HIGHMEM usage. 73 */ 74 #define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\ 75 __GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\ 76 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\ 77 __GFP_NOLOCKDEP) 78 79 /* The GFP flags allowed during early boot */ 80 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS)) 81 82 /* Control allocation cpuset and node placement constraints */ 83 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE) 84 85 /* Do not use these with a slab allocator */ 86 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK) 87 88 /* 89 * Different from WARN_ON_ONCE(), no warning will be issued 90 * when we specify __GFP_NOWARN. 91 */ 92 #define WARN_ON_ONCE_GFP(cond, gfp) ({ \ 93 static bool __section(".data..once") __warned; \ 94 int __ret_warn_once = !!(cond); \ 95 \ 96 if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \ 97 __warned = true; \ 98 WARN_ON(1); \ 99 } \ 100 unlikely(__ret_warn_once); \ 101 }) 102 103 void page_writeback_init(void); 104 105 /* 106 * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages, 107 * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit 108 * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE). Hugetlb currently 109 * leaves nr_pages_mapped at 0, but avoid surprise if it participates later. 110 */ 111 #define ENTIRELY_MAPPED 0x800000 112 #define FOLIO_PAGES_MAPPED (ENTIRELY_MAPPED - 1) 113 114 /* 115 * Flags passed to __show_mem() and show_free_areas() to suppress output in 116 * various contexts. 117 */ 118 #define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */ 119 120 /* 121 * How many individual pages have an elevated _mapcount. Excludes 122 * the folio's entire_mapcount. 123 * 124 * Don't use this function outside of debugging code. 125 */ 126 static inline int folio_nr_pages_mapped(const struct folio *folio) 127 { 128 if (IS_ENABLED(CONFIG_NO_PAGE_MAPCOUNT)) 129 return -1; 130 return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED; 131 } 132 133 /* 134 * Retrieve the first entry of a folio based on a provided entry within the 135 * folio. We cannot rely on folio->swap as there is no guarantee that it has 136 * been initialized. Used for calling arch_swap_restore() 137 */ 138 static inline swp_entry_t folio_swap(swp_entry_t entry, 139 const struct folio *folio) 140 { 141 swp_entry_t swap = { 142 .val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)), 143 }; 144 145 return swap; 146 } 147 148 static inline void *folio_raw_mapping(const struct folio *folio) 149 { 150 unsigned long mapping = (unsigned long)folio->mapping; 151 152 return (void *)(mapping & ~PAGE_MAPPING_FLAGS); 153 } 154 155 /* 156 * This is a file-backed mapping, and is about to be memory mapped - invoke its 157 * mmap hook and safely handle error conditions. On error, VMA hooks will be 158 * mutated. 159 * 160 * @file: File which backs the mapping. 161 * @vma: VMA which we are mapping. 162 * 163 * Returns: 0 if success, error otherwise. 164 */ 165 static inline int mmap_file(struct file *file, struct vm_area_struct *vma) 166 { 167 int err = call_mmap(file, vma); 168 169 if (likely(!err)) 170 return 0; 171 172 /* 173 * OK, we tried to call the file hook for mmap(), but an error 174 * arose. The mapping is in an inconsistent state and we most not invoke 175 * any further hooks on it. 176 */ 177 vma->vm_ops = &vma_dummy_vm_ops; 178 179 return err; 180 } 181 182 /* 183 * If the VMA has a close hook then close it, and since closing it might leave 184 * it in an inconsistent state which makes the use of any hooks suspect, clear 185 * them down by installing dummy empty hooks. 186 */ 187 static inline void vma_close(struct vm_area_struct *vma) 188 { 189 if (vma->vm_ops && vma->vm_ops->close) { 190 vma->vm_ops->close(vma); 191 192 /* 193 * The mapping is in an inconsistent state, and no further hooks 194 * may be invoked upon it. 195 */ 196 vma->vm_ops = &vma_dummy_vm_ops; 197 } 198 } 199 200 #ifdef CONFIG_MMU 201 202 /* Flags for folio_pte_batch(). */ 203 typedef int __bitwise fpb_t; 204 205 /* Compare PTEs after pte_mkclean(), ignoring the dirty bit. */ 206 #define FPB_IGNORE_DIRTY ((__force fpb_t)BIT(0)) 207 208 /* Compare PTEs after pte_clear_soft_dirty(), ignoring the soft-dirty bit. */ 209 #define FPB_IGNORE_SOFT_DIRTY ((__force fpb_t)BIT(1)) 210 211 static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags) 212 { 213 if (flags & FPB_IGNORE_DIRTY) 214 pte = pte_mkclean(pte); 215 if (likely(flags & FPB_IGNORE_SOFT_DIRTY)) 216 pte = pte_clear_soft_dirty(pte); 217 return pte_wrprotect(pte_mkold(pte)); 218 } 219 220 /** 221 * folio_pte_batch - detect a PTE batch for a large folio 222 * @folio: The large folio to detect a PTE batch for. 223 * @addr: The user virtual address the first page is mapped at. 224 * @start_ptep: Page table pointer for the first entry. 225 * @pte: Page table entry for the first page. 226 * @max_nr: The maximum number of table entries to consider. 227 * @flags: Flags to modify the PTE batch semantics. 228 * @any_writable: Optional pointer to indicate whether any entry except the 229 * first one is writable. 230 * @any_young: Optional pointer to indicate whether any entry except the 231 * first one is young. 232 * @any_dirty: Optional pointer to indicate whether any entry except the 233 * first one is dirty. 234 * 235 * Detect a PTE batch: consecutive (present) PTEs that map consecutive 236 * pages of the same large folio. 237 * 238 * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN, 239 * the accessed bit, writable bit, dirty bit (with FPB_IGNORE_DIRTY) and 240 * soft-dirty bit (with FPB_IGNORE_SOFT_DIRTY). 241 * 242 * start_ptep must map any page of the folio. max_nr must be at least one and 243 * must be limited by the caller so scanning cannot exceed a single page table. 244 * 245 * Return: the number of table entries in the batch. 246 */ 247 static inline int folio_pte_batch(struct folio *folio, unsigned long addr, 248 pte_t *start_ptep, pte_t pte, int max_nr, fpb_t flags, 249 bool *any_writable, bool *any_young, bool *any_dirty) 250 { 251 pte_t expected_pte, *ptep; 252 bool writable, young, dirty; 253 int nr, cur_nr; 254 255 if (any_writable) 256 *any_writable = false; 257 if (any_young) 258 *any_young = false; 259 if (any_dirty) 260 *any_dirty = false; 261 262 VM_WARN_ON_FOLIO(!pte_present(pte), folio); 263 VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio); 264 VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio); 265 266 /* Limit max_nr to the actual remaining PFNs in the folio we could batch. */ 267 max_nr = min_t(unsigned long, max_nr, 268 folio_pfn(folio) + folio_nr_pages(folio) - pte_pfn(pte)); 269 270 nr = pte_batch_hint(start_ptep, pte); 271 expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags); 272 ptep = start_ptep + nr; 273 274 while (nr < max_nr) { 275 pte = ptep_get(ptep); 276 if (any_writable) 277 writable = !!pte_write(pte); 278 if (any_young) 279 young = !!pte_young(pte); 280 if (any_dirty) 281 dirty = !!pte_dirty(pte); 282 pte = __pte_batch_clear_ignored(pte, flags); 283 284 if (!pte_same(pte, expected_pte)) 285 break; 286 287 if (any_writable) 288 *any_writable |= writable; 289 if (any_young) 290 *any_young |= young; 291 if (any_dirty) 292 *any_dirty |= dirty; 293 294 cur_nr = pte_batch_hint(ptep, pte); 295 expected_pte = pte_advance_pfn(expected_pte, cur_nr); 296 ptep += cur_nr; 297 nr += cur_nr; 298 } 299 300 return min(nr, max_nr); 301 } 302 303 /** 304 * pte_move_swp_offset - Move the swap entry offset field of a swap pte 305 * forward or backward by delta 306 * @pte: The initial pte state; is_swap_pte(pte) must be true and 307 * non_swap_entry() must be false. 308 * @delta: The direction and the offset we are moving; forward if delta 309 * is positive; backward if delta is negative 310 * 311 * Moves the swap offset, while maintaining all other fields, including 312 * swap type, and any swp pte bits. The resulting pte is returned. 313 */ 314 static inline pte_t pte_move_swp_offset(pte_t pte, long delta) 315 { 316 swp_entry_t entry = pte_to_swp_entry(pte); 317 pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry), 318 (swp_offset(entry) + delta))); 319 320 if (pte_swp_soft_dirty(pte)) 321 new = pte_swp_mksoft_dirty(new); 322 if (pte_swp_exclusive(pte)) 323 new = pte_swp_mkexclusive(new); 324 if (pte_swp_uffd_wp(pte)) 325 new = pte_swp_mkuffd_wp(new); 326 327 return new; 328 } 329 330 331 /** 332 * pte_next_swp_offset - Increment the swap entry offset field of a swap pte. 333 * @pte: The initial pte state; is_swap_pte(pte) must be true and 334 * non_swap_entry() must be false. 335 * 336 * Increments the swap offset, while maintaining all other fields, including 337 * swap type, and any swp pte bits. The resulting pte is returned. 338 */ 339 static inline pte_t pte_next_swp_offset(pte_t pte) 340 { 341 return pte_move_swp_offset(pte, 1); 342 } 343 344 /** 345 * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries 346 * @start_ptep: Page table pointer for the first entry. 347 * @max_nr: The maximum number of table entries to consider. 348 * @pte: Page table entry for the first entry. 349 * 350 * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs 351 * containing swap entries all with consecutive offsets and targeting the same 352 * swap type, all with matching swp pte bits. 353 * 354 * max_nr must be at least one and must be limited by the caller so scanning 355 * cannot exceed a single page table. 356 * 357 * Return: the number of table entries in the batch. 358 */ 359 static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte) 360 { 361 pte_t expected_pte = pte_next_swp_offset(pte); 362 const pte_t *end_ptep = start_ptep + max_nr; 363 swp_entry_t entry = pte_to_swp_entry(pte); 364 pte_t *ptep = start_ptep + 1; 365 unsigned short cgroup_id; 366 367 VM_WARN_ON(max_nr < 1); 368 VM_WARN_ON(!is_swap_pte(pte)); 369 VM_WARN_ON(non_swap_entry(entry)); 370 371 cgroup_id = lookup_swap_cgroup_id(entry); 372 while (ptep < end_ptep) { 373 pte = ptep_get(ptep); 374 375 if (!pte_same(pte, expected_pte)) 376 break; 377 if (lookup_swap_cgroup_id(pte_to_swp_entry(pte)) != cgroup_id) 378 break; 379 expected_pte = pte_next_swp_offset(expected_pte); 380 ptep++; 381 } 382 383 return ptep - start_ptep; 384 } 385 #endif /* CONFIG_MMU */ 386 387 void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio, 388 int nr_throttled); 389 static inline void acct_reclaim_writeback(struct folio *folio) 390 { 391 pg_data_t *pgdat = folio_pgdat(folio); 392 int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled); 393 394 if (nr_throttled) 395 __acct_reclaim_writeback(pgdat, folio, nr_throttled); 396 } 397 398 static inline void wake_throttle_isolated(pg_data_t *pgdat) 399 { 400 wait_queue_head_t *wqh; 401 402 wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED]; 403 if (waitqueue_active(wqh)) 404 wake_up(wqh); 405 } 406 407 vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf); 408 static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf) 409 { 410 vm_fault_t ret = __vmf_anon_prepare(vmf); 411 412 if (unlikely(ret & VM_FAULT_RETRY)) 413 vma_end_read(vmf->vma); 414 return ret; 415 } 416 417 vm_fault_t do_swap_page(struct vm_fault *vmf); 418 void folio_rotate_reclaimable(struct folio *folio); 419 bool __folio_end_writeback(struct folio *folio); 420 void deactivate_file_folio(struct folio *folio); 421 void folio_activate(struct folio *folio); 422 423 void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas, 424 struct vm_area_struct *start_vma, unsigned long floor, 425 unsigned long ceiling, bool mm_wr_locked); 426 void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte); 427 428 struct zap_details; 429 void unmap_page_range(struct mmu_gather *tlb, 430 struct vm_area_struct *vma, 431 unsigned long addr, unsigned long end, 432 struct zap_details *details); 433 void zap_page_range_single_batched(struct mmu_gather *tlb, 434 struct vm_area_struct *vma, unsigned long addr, 435 unsigned long size, struct zap_details *details); 436 int folio_unmap_invalidate(struct address_space *mapping, struct folio *folio, 437 gfp_t gfp); 438 439 void page_cache_ra_order(struct readahead_control *, struct file_ra_state *, 440 unsigned int order); 441 void force_page_cache_ra(struct readahead_control *, unsigned long nr); 442 static inline void force_page_cache_readahead(struct address_space *mapping, 443 struct file *file, pgoff_t index, unsigned long nr_to_read) 444 { 445 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index); 446 force_page_cache_ra(&ractl, nr_to_read); 447 } 448 449 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start, 450 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 451 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start, 452 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices); 453 void filemap_free_folio(struct address_space *mapping, struct folio *folio); 454 int truncate_inode_folio(struct address_space *mapping, struct folio *folio); 455 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, 456 loff_t end); 457 long mapping_evict_folio(struct address_space *mapping, struct folio *folio); 458 unsigned long mapping_try_invalidate(struct address_space *mapping, 459 pgoff_t start, pgoff_t end, unsigned long *nr_failed); 460 461 /** 462 * folio_evictable - Test whether a folio is evictable. 463 * @folio: The folio to test. 464 * 465 * Test whether @folio is evictable -- i.e., should be placed on 466 * active/inactive lists vs unevictable list. 467 * 468 * Reasons folio might not be evictable: 469 * 1. folio's mapping marked unevictable 470 * 2. One of the pages in the folio is part of an mlocked VMA 471 */ 472 static inline bool folio_evictable(struct folio *folio) 473 { 474 bool ret; 475 476 /* Prevent address_space of inode and swap cache from being freed */ 477 rcu_read_lock(); 478 ret = !mapping_unevictable(folio_mapping(folio)) && 479 !folio_test_mlocked(folio); 480 rcu_read_unlock(); 481 return ret; 482 } 483 484 /* 485 * Turn a non-refcounted page (->_refcount == 0) into refcounted with 486 * a count of one. 487 */ 488 static inline void set_page_refcounted(struct page *page) 489 { 490 VM_BUG_ON_PAGE(PageTail(page), page); 491 VM_BUG_ON_PAGE(page_ref_count(page), page); 492 set_page_count(page, 1); 493 } 494 495 /* 496 * Return true if a folio needs ->release_folio() calling upon it. 497 */ 498 static inline bool folio_needs_release(struct folio *folio) 499 { 500 struct address_space *mapping = folio_mapping(folio); 501 502 return folio_has_private(folio) || 503 (mapping && mapping_release_always(mapping)); 504 } 505 506 extern unsigned long highest_memmap_pfn; 507 508 /* 509 * Maximum number of reclaim retries without progress before the OOM 510 * killer is consider the only way forward. 511 */ 512 #define MAX_RECLAIM_RETRIES 16 513 514 /* 515 * in mm/vmscan.c: 516 */ 517 bool folio_isolate_lru(struct folio *folio); 518 void folio_putback_lru(struct folio *folio); 519 extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason); 520 521 /* 522 * in mm/rmap.c: 523 */ 524 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address); 525 526 /* 527 * in mm/page_alloc.c 528 */ 529 #define K(x) ((x) << (PAGE_SHIFT-10)) 530 531 extern char * const zone_names[MAX_NR_ZONES]; 532 533 /* perform sanity checks on struct pages being allocated or freed */ 534 DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled); 535 536 extern int min_free_kbytes; 537 extern int defrag_mode; 538 539 void setup_per_zone_wmarks(void); 540 void calculate_min_free_kbytes(void); 541 int __meminit init_per_zone_wmark_min(void); 542 void page_alloc_sysctl_init(void); 543 544 /* 545 * Structure for holding the mostly immutable allocation parameters passed 546 * between functions involved in allocations, including the alloc_pages* 547 * family of functions. 548 * 549 * nodemask, migratetype and highest_zoneidx are initialized only once in 550 * __alloc_pages() and then never change. 551 * 552 * zonelist, preferred_zone and highest_zoneidx are set first in 553 * __alloc_pages() for the fast path, and might be later changed 554 * in __alloc_pages_slowpath(). All other functions pass the whole structure 555 * by a const pointer. 556 */ 557 struct alloc_context { 558 struct zonelist *zonelist; 559 nodemask_t *nodemask; 560 struct zoneref *preferred_zoneref; 561 int migratetype; 562 563 /* 564 * highest_zoneidx represents highest usable zone index of 565 * the allocation request. Due to the nature of the zone, 566 * memory on lower zone than the highest_zoneidx will be 567 * protected by lowmem_reserve[highest_zoneidx]. 568 * 569 * highest_zoneidx is also used by reclaim/compaction to limit 570 * the target zone since higher zone than this index cannot be 571 * usable for this allocation request. 572 */ 573 enum zone_type highest_zoneidx; 574 bool spread_dirty_pages; 575 }; 576 577 /* 578 * This function returns the order of a free page in the buddy system. In 579 * general, page_zone(page)->lock must be held by the caller to prevent the 580 * page from being allocated in parallel and returning garbage as the order. 581 * If a caller does not hold page_zone(page)->lock, it must guarantee that the 582 * page cannot be allocated or merged in parallel. Alternatively, it must 583 * handle invalid values gracefully, and use buddy_order_unsafe() below. 584 */ 585 static inline unsigned int buddy_order(struct page *page) 586 { 587 /* PageBuddy() must be checked by the caller */ 588 return page_private(page); 589 } 590 591 /* 592 * Like buddy_order(), but for callers who cannot afford to hold the zone lock. 593 * PageBuddy() should be checked first by the caller to minimize race window, 594 * and invalid values must be handled gracefully. 595 * 596 * READ_ONCE is used so that if the caller assigns the result into a local 597 * variable and e.g. tests it for valid range before using, the compiler cannot 598 * decide to remove the variable and inline the page_private(page) multiple 599 * times, potentially observing different values in the tests and the actual 600 * use of the result. 601 */ 602 #define buddy_order_unsafe(page) READ_ONCE(page_private(page)) 603 604 /* 605 * This function checks whether a page is free && is the buddy 606 * we can coalesce a page and its buddy if 607 * (a) the buddy is not in a hole (check before calling!) && 608 * (b) the buddy is in the buddy system && 609 * (c) a page and its buddy have the same order && 610 * (d) a page and its buddy are in the same zone. 611 * 612 * For recording whether a page is in the buddy system, we set PageBuddy. 613 * Setting, clearing, and testing PageBuddy is serialized by zone->lock. 614 * 615 * For recording page's order, we use page_private(page). 616 */ 617 static inline bool page_is_buddy(struct page *page, struct page *buddy, 618 unsigned int order) 619 { 620 if (!page_is_guard(buddy) && !PageBuddy(buddy)) 621 return false; 622 623 if (buddy_order(buddy) != order) 624 return false; 625 626 /* 627 * zone check is done late to avoid uselessly calculating 628 * zone/node ids for pages that could never merge. 629 */ 630 if (page_zone_id(page) != page_zone_id(buddy)) 631 return false; 632 633 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); 634 635 return true; 636 } 637 638 /* 639 * Locate the struct page for both the matching buddy in our 640 * pair (buddy1) and the combined O(n+1) page they form (page). 641 * 642 * 1) Any buddy B1 will have an order O twin B2 which satisfies 643 * the following equation: 644 * B2 = B1 ^ (1 << O) 645 * For example, if the starting buddy (buddy2) is #8 its order 646 * 1 buddy is #10: 647 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 648 * 649 * 2) Any buddy B will have an order O+1 parent P which 650 * satisfies the following equation: 651 * P = B & ~(1 << O) 652 * 653 * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER 654 */ 655 static inline unsigned long 656 __find_buddy_pfn(unsigned long page_pfn, unsigned int order) 657 { 658 return page_pfn ^ (1 << order); 659 } 660 661 /* 662 * Find the buddy of @page and validate it. 663 * @page: The input page 664 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the 665 * function is used in the performance-critical __free_one_page(). 666 * @order: The order of the page 667 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to 668 * page_to_pfn(). 669 * 670 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is 671 * not the same as @page. The validation is necessary before use it. 672 * 673 * Return: the found buddy page or NULL if not found. 674 */ 675 static inline struct page *find_buddy_page_pfn(struct page *page, 676 unsigned long pfn, unsigned int order, unsigned long *buddy_pfn) 677 { 678 unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order); 679 struct page *buddy; 680 681 buddy = page + (__buddy_pfn - pfn); 682 if (buddy_pfn) 683 *buddy_pfn = __buddy_pfn; 684 685 if (page_is_buddy(page, buddy, order)) 686 return buddy; 687 return NULL; 688 } 689 690 extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn, 691 unsigned long end_pfn, struct zone *zone); 692 693 static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn, 694 unsigned long end_pfn, struct zone *zone) 695 { 696 if (zone->contiguous) 697 return pfn_to_page(start_pfn); 698 699 return __pageblock_pfn_to_page(start_pfn, end_pfn, zone); 700 } 701 702 void set_zone_contiguous(struct zone *zone); 703 bool pfn_range_intersects_zones(int nid, unsigned long start_pfn, 704 unsigned long nr_pages); 705 706 static inline void clear_zone_contiguous(struct zone *zone) 707 { 708 zone->contiguous = false; 709 } 710 711 extern int __isolate_free_page(struct page *page, unsigned int order); 712 extern void __putback_isolated_page(struct page *page, unsigned int order, 713 int mt); 714 extern void memblock_free_pages(struct page *page, unsigned long pfn, 715 unsigned int order); 716 extern void __free_pages_core(struct page *page, unsigned int order, 717 enum meminit_context context); 718 719 /* 720 * This will have no effect, other than possibly generating a warning, if the 721 * caller passes in a non-large folio. 722 */ 723 static inline void folio_set_order(struct folio *folio, unsigned int order) 724 { 725 if (WARN_ON_ONCE(!order || !folio_test_large(folio))) 726 return; 727 728 folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order; 729 #ifdef NR_PAGES_IN_LARGE_FOLIO 730 folio->_nr_pages = 1U << order; 731 #endif 732 } 733 734 bool __folio_unqueue_deferred_split(struct folio *folio); 735 static inline bool folio_unqueue_deferred_split(struct folio *folio) 736 { 737 if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio)) 738 return false; 739 740 /* 741 * At this point, there is no one trying to add the folio to 742 * deferred_list. If folio is not in deferred_list, it's safe 743 * to check without acquiring the split_queue_lock. 744 */ 745 if (data_race(list_empty(&folio->_deferred_list))) 746 return false; 747 748 return __folio_unqueue_deferred_split(folio); 749 } 750 751 static inline struct folio *page_rmappable_folio(struct page *page) 752 { 753 struct folio *folio = (struct folio *)page; 754 755 if (folio && folio_test_large(folio)) 756 folio_set_large_rmappable(folio); 757 return folio; 758 } 759 760 static inline void prep_compound_head(struct page *page, unsigned int order) 761 { 762 struct folio *folio = (struct folio *)page; 763 764 folio_set_order(folio, order); 765 atomic_set(&folio->_large_mapcount, -1); 766 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT)) 767 atomic_set(&folio->_nr_pages_mapped, 0); 768 if (IS_ENABLED(CONFIG_MM_ID)) { 769 folio->_mm_ids = 0; 770 folio->_mm_id_mapcount[0] = -1; 771 folio->_mm_id_mapcount[1] = -1; 772 } 773 if (IS_ENABLED(CONFIG_64BIT) || order > 1) { 774 atomic_set(&folio->_pincount, 0); 775 atomic_set(&folio->_entire_mapcount, -1); 776 } 777 if (order > 1) 778 INIT_LIST_HEAD(&folio->_deferred_list); 779 } 780 781 static inline void prep_compound_tail(struct page *head, int tail_idx) 782 { 783 struct page *p = head + tail_idx; 784 785 p->mapping = TAIL_MAPPING; 786 set_compound_head(p, head); 787 set_page_private(p, 0); 788 } 789 790 void post_alloc_hook(struct page *page, unsigned int order, gfp_t gfp_flags); 791 extern bool free_pages_prepare(struct page *page, unsigned int order); 792 793 extern int user_min_free_kbytes; 794 795 struct page *__alloc_frozen_pages_noprof(gfp_t, unsigned int order, int nid, 796 nodemask_t *); 797 #define __alloc_frozen_pages(...) \ 798 alloc_hooks(__alloc_frozen_pages_noprof(__VA_ARGS__)) 799 void free_frozen_pages(struct page *page, unsigned int order); 800 void free_unref_folios(struct folio_batch *fbatch); 801 802 #ifdef CONFIG_NUMA 803 struct page *alloc_frozen_pages_noprof(gfp_t, unsigned int order); 804 #else 805 static inline struct page *alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order) 806 { 807 return __alloc_frozen_pages_noprof(gfp, order, numa_node_id(), NULL); 808 } 809 #endif 810 811 #define alloc_frozen_pages(...) \ 812 alloc_hooks(alloc_frozen_pages_noprof(__VA_ARGS__)) 813 814 extern void zone_pcp_reset(struct zone *zone); 815 extern void zone_pcp_disable(struct zone *zone); 816 extern void zone_pcp_enable(struct zone *zone); 817 extern void zone_pcp_init(struct zone *zone); 818 819 extern void *memmap_alloc(phys_addr_t size, phys_addr_t align, 820 phys_addr_t min_addr, 821 int nid, bool exact_nid); 822 823 void memmap_init_range(unsigned long, int, unsigned long, unsigned long, 824 unsigned long, enum meminit_context, struct vmem_altmap *, int); 825 826 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 827 828 /* 829 * in mm/compaction.c 830 */ 831 /* 832 * compact_control is used to track pages being migrated and the free pages 833 * they are being migrated to during memory compaction. The free_pfn starts 834 * at the end of a zone and migrate_pfn begins at the start. Movable pages 835 * are moved to the end of a zone during a compaction run and the run 836 * completes when free_pfn <= migrate_pfn 837 */ 838 struct compact_control { 839 struct list_head freepages[NR_PAGE_ORDERS]; /* List of free pages to migrate to */ 840 struct list_head migratepages; /* List of pages being migrated */ 841 unsigned int nr_freepages; /* Number of isolated free pages */ 842 unsigned int nr_migratepages; /* Number of pages to migrate */ 843 unsigned long free_pfn; /* isolate_freepages search base */ 844 /* 845 * Acts as an in/out parameter to page isolation for migration. 846 * isolate_migratepages uses it as a search base. 847 * isolate_migratepages_block will update the value to the next pfn 848 * after the last isolated one. 849 */ 850 unsigned long migrate_pfn; 851 unsigned long fast_start_pfn; /* a pfn to start linear scan from */ 852 struct zone *zone; 853 unsigned long total_migrate_scanned; 854 unsigned long total_free_scanned; 855 unsigned short fast_search_fail;/* failures to use free list searches */ 856 short search_order; /* order to start a fast search at */ 857 const gfp_t gfp_mask; /* gfp mask of a direct compactor */ 858 int order; /* order a direct compactor needs */ 859 int migratetype; /* migratetype of direct compactor */ 860 const unsigned int alloc_flags; /* alloc flags of a direct compactor */ 861 const int highest_zoneidx; /* zone index of a direct compactor */ 862 enum migrate_mode mode; /* Async or sync migration mode */ 863 bool ignore_skip_hint; /* Scan blocks even if marked skip */ 864 bool no_set_skip_hint; /* Don't mark blocks for skipping */ 865 bool ignore_block_suitable; /* Scan blocks considered unsuitable */ 866 bool direct_compaction; /* False from kcompactd or /proc/... */ 867 bool proactive_compaction; /* kcompactd proactive compaction */ 868 bool whole_zone; /* Whole zone should/has been scanned */ 869 bool contended; /* Signal lock contention */ 870 bool finish_pageblock; /* Scan the remainder of a pageblock. Used 871 * when there are potentially transient 872 * isolation or migration failures to 873 * ensure forward progress. 874 */ 875 bool alloc_contig; /* alloc_contig_range allocation */ 876 }; 877 878 /* 879 * Used in direct compaction when a page should be taken from the freelists 880 * immediately when one is created during the free path. 881 */ 882 struct capture_control { 883 struct compact_control *cc; 884 struct page *page; 885 }; 886 887 unsigned long 888 isolate_freepages_range(struct compact_control *cc, 889 unsigned long start_pfn, unsigned long end_pfn); 890 int 891 isolate_migratepages_range(struct compact_control *cc, 892 unsigned long low_pfn, unsigned long end_pfn); 893 894 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ 895 void init_cma_reserved_pageblock(struct page *page); 896 897 #endif /* CONFIG_COMPACTION || CONFIG_CMA */ 898 899 struct cma; 900 901 #ifdef CONFIG_CMA 902 void *cma_reserve_early(struct cma *cma, unsigned long size); 903 void init_cma_pageblock(struct page *page); 904 #else 905 static inline void *cma_reserve_early(struct cma *cma, unsigned long size) 906 { 907 return NULL; 908 } 909 static inline void init_cma_pageblock(struct page *page) 910 { 911 } 912 #endif 913 914 915 int find_suitable_fallback(struct free_area *area, unsigned int order, 916 int migratetype, bool claimable); 917 918 static inline bool free_area_empty(struct free_area *area, int migratetype) 919 { 920 return list_empty(&area->free_list[migratetype]); 921 } 922 923 /* mm/util.c */ 924 struct anon_vma *folio_anon_vma(const struct folio *folio); 925 926 #ifdef CONFIG_MMU 927 void unmap_mapping_folio(struct folio *folio); 928 extern long populate_vma_page_range(struct vm_area_struct *vma, 929 unsigned long start, unsigned long end, int *locked); 930 extern long faultin_page_range(struct mm_struct *mm, unsigned long start, 931 unsigned long end, bool write, int *locked); 932 extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 933 unsigned long bytes); 934 935 /* 936 * NOTE: This function can't tell whether the folio is "fully mapped" in the 937 * range. 938 * "fully mapped" means all the pages of folio is associated with the page 939 * table of range while this function just check whether the folio range is 940 * within the range [start, end). Function caller needs to do page table 941 * check if it cares about the page table association. 942 * 943 * Typical usage (like mlock or madvise) is: 944 * Caller knows at least 1 page of folio is associated with page table of VMA 945 * and the range [start, end) is intersect with the VMA range. Caller wants 946 * to know whether the folio is fully associated with the range. It calls 947 * this function to check whether the folio is in the range first. Then checks 948 * the page table to know whether the folio is fully mapped to the range. 949 */ 950 static inline bool 951 folio_within_range(struct folio *folio, struct vm_area_struct *vma, 952 unsigned long start, unsigned long end) 953 { 954 pgoff_t pgoff, addr; 955 unsigned long vma_pglen = vma_pages(vma); 956 957 VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio); 958 if (start > end) 959 return false; 960 961 if (start < vma->vm_start) 962 start = vma->vm_start; 963 964 if (end > vma->vm_end) 965 end = vma->vm_end; 966 967 pgoff = folio_pgoff(folio); 968 969 /* if folio start address is not in vma range */ 970 if (!in_range(pgoff, vma->vm_pgoff, vma_pglen)) 971 return false; 972 973 addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 974 975 return !(addr < start || end - addr < folio_size(folio)); 976 } 977 978 static inline bool 979 folio_within_vma(struct folio *folio, struct vm_area_struct *vma) 980 { 981 return folio_within_range(folio, vma, vma->vm_start, vma->vm_end); 982 } 983 984 /* 985 * mlock_vma_folio() and munlock_vma_folio(): 986 * should be called with vma's mmap_lock held for read or write, 987 * under page table lock for the pte/pmd being added or removed. 988 * 989 * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at 990 * the end of folio_remove_rmap_*(); but new anon folios are managed by 991 * folio_add_lru_vma() calling mlock_new_folio(). 992 */ 993 void mlock_folio(struct folio *folio); 994 static inline void mlock_vma_folio(struct folio *folio, 995 struct vm_area_struct *vma) 996 { 997 /* 998 * The VM_SPECIAL check here serves two purposes. 999 * 1) VM_IO check prevents migration from double-counting during mlock. 1000 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED 1001 * is never left set on a VM_SPECIAL vma, there is an interval while 1002 * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may 1003 * still be set while VM_SPECIAL bits are added: so ignore it then. 1004 */ 1005 if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED)) 1006 mlock_folio(folio); 1007 } 1008 1009 void munlock_folio(struct folio *folio); 1010 static inline void munlock_vma_folio(struct folio *folio, 1011 struct vm_area_struct *vma) 1012 { 1013 /* 1014 * munlock if the function is called. Ideally, we should only 1015 * do munlock if any page of folio is unmapped from VMA and 1016 * cause folio not fully mapped to VMA. 1017 * 1018 * But it's not easy to confirm that's the situation. So we 1019 * always munlock the folio and page reclaim will correct it 1020 * if it's wrong. 1021 */ 1022 if (unlikely(vma->vm_flags & VM_LOCKED)) 1023 munlock_folio(folio); 1024 } 1025 1026 void mlock_new_folio(struct folio *folio); 1027 bool need_mlock_drain(int cpu); 1028 void mlock_drain_local(void); 1029 void mlock_drain_remote(int cpu); 1030 1031 extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma); 1032 1033 /** 1034 * vma_address - Find the virtual address a page range is mapped at 1035 * @vma: The vma which maps this object. 1036 * @pgoff: The page offset within its object. 1037 * @nr_pages: The number of pages to consider. 1038 * 1039 * If any page in this range is mapped by this VMA, return the first address 1040 * where any of these pages appear. Otherwise, return -EFAULT. 1041 */ 1042 static inline unsigned long vma_address(const struct vm_area_struct *vma, 1043 pgoff_t pgoff, unsigned long nr_pages) 1044 { 1045 unsigned long address; 1046 1047 if (pgoff >= vma->vm_pgoff) { 1048 address = vma->vm_start + 1049 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 1050 /* Check for address beyond vma (or wrapped through 0?) */ 1051 if (address < vma->vm_start || address >= vma->vm_end) 1052 address = -EFAULT; 1053 } else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) { 1054 /* Test above avoids possibility of wrap to 0 on 32-bit */ 1055 address = vma->vm_start; 1056 } else { 1057 address = -EFAULT; 1058 } 1059 return address; 1060 } 1061 1062 /* 1063 * Then at what user virtual address will none of the range be found in vma? 1064 * Assumes that vma_address() already returned a good starting address. 1065 */ 1066 static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw) 1067 { 1068 struct vm_area_struct *vma = pvmw->vma; 1069 pgoff_t pgoff; 1070 unsigned long address; 1071 1072 /* Common case, plus ->pgoff is invalid for KSM */ 1073 if (pvmw->nr_pages == 1) 1074 return pvmw->address + PAGE_SIZE; 1075 1076 pgoff = pvmw->pgoff + pvmw->nr_pages; 1077 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 1078 /* Check for address beyond vma (or wrapped through 0?) */ 1079 if (address < vma->vm_start || address > vma->vm_end) 1080 address = vma->vm_end; 1081 return address; 1082 } 1083 1084 static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf, 1085 struct file *fpin) 1086 { 1087 int flags = vmf->flags; 1088 1089 if (fpin) 1090 return fpin; 1091 1092 /* 1093 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or 1094 * anything, so we only pin the file and drop the mmap_lock if only 1095 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt. 1096 */ 1097 if (fault_flag_allow_retry_first(flags) && 1098 !(flags & FAULT_FLAG_RETRY_NOWAIT)) { 1099 fpin = get_file(vmf->vma->vm_file); 1100 release_fault_lock(vmf); 1101 } 1102 return fpin; 1103 } 1104 #else /* !CONFIG_MMU */ 1105 static inline void unmap_mapping_folio(struct folio *folio) { } 1106 static inline void mlock_new_folio(struct folio *folio) { } 1107 static inline bool need_mlock_drain(int cpu) { return false; } 1108 static inline void mlock_drain_local(void) { } 1109 static inline void mlock_drain_remote(int cpu) { } 1110 static inline void vunmap_range_noflush(unsigned long start, unsigned long end) 1111 { 1112 } 1113 #endif /* !CONFIG_MMU */ 1114 1115 /* Memory initialisation debug and verification */ 1116 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1117 DECLARE_STATIC_KEY_TRUE(deferred_pages); 1118 1119 bool __init deferred_grow_zone(struct zone *zone, unsigned int order); 1120 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1121 1122 void init_deferred_page(unsigned long pfn, int nid); 1123 1124 enum mminit_level { 1125 MMINIT_WARNING, 1126 MMINIT_VERIFY, 1127 MMINIT_TRACE 1128 }; 1129 1130 #ifdef CONFIG_DEBUG_MEMORY_INIT 1131 1132 extern int mminit_loglevel; 1133 1134 #define mminit_dprintk(level, prefix, fmt, arg...) \ 1135 do { \ 1136 if (level < mminit_loglevel) { \ 1137 if (level <= MMINIT_WARNING) \ 1138 pr_warn("mminit::" prefix " " fmt, ##arg); \ 1139 else \ 1140 printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \ 1141 } \ 1142 } while (0) 1143 1144 extern void mminit_verify_pageflags_layout(void); 1145 extern void mminit_verify_zonelist(void); 1146 #else 1147 1148 static inline void mminit_dprintk(enum mminit_level level, 1149 const char *prefix, const char *fmt, ...) 1150 { 1151 } 1152 1153 static inline void mminit_verify_pageflags_layout(void) 1154 { 1155 } 1156 1157 static inline void mminit_verify_zonelist(void) 1158 { 1159 } 1160 #endif /* CONFIG_DEBUG_MEMORY_INIT */ 1161 1162 #define NODE_RECLAIM_NOSCAN -2 1163 #define NODE_RECLAIM_FULL -1 1164 #define NODE_RECLAIM_SOME 0 1165 #define NODE_RECLAIM_SUCCESS 1 1166 1167 #ifdef CONFIG_NUMA 1168 extern int node_reclaim_mode; 1169 1170 extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int); 1171 extern int find_next_best_node(int node, nodemask_t *used_node_mask); 1172 #else 1173 #define node_reclaim_mode 0 1174 1175 static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask, 1176 unsigned int order) 1177 { 1178 return NODE_RECLAIM_NOSCAN; 1179 } 1180 static inline int find_next_best_node(int node, nodemask_t *used_node_mask) 1181 { 1182 return NUMA_NO_NODE; 1183 } 1184 #endif 1185 1186 static inline bool node_reclaim_enabled(void) 1187 { 1188 /* Is any node_reclaim_mode bit set? */ 1189 return node_reclaim_mode & (RECLAIM_ZONE|RECLAIM_WRITE|RECLAIM_UNMAP); 1190 } 1191 1192 /* 1193 * mm/memory-failure.c 1194 */ 1195 #ifdef CONFIG_MEMORY_FAILURE 1196 int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill); 1197 void shake_folio(struct folio *folio); 1198 extern int hwpoison_filter(struct page *p); 1199 1200 extern u32 hwpoison_filter_dev_major; 1201 extern u32 hwpoison_filter_dev_minor; 1202 extern u64 hwpoison_filter_flags_mask; 1203 extern u64 hwpoison_filter_flags_value; 1204 extern u64 hwpoison_filter_memcg; 1205 extern u32 hwpoison_filter_enable; 1206 #define MAGIC_HWPOISON 0x48575053U /* HWPS */ 1207 void SetPageHWPoisonTakenOff(struct page *page); 1208 void ClearPageHWPoisonTakenOff(struct page *page); 1209 bool take_page_off_buddy(struct page *page); 1210 bool put_page_back_buddy(struct page *page); 1211 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early); 1212 void add_to_kill_ksm(struct task_struct *tsk, const struct page *p, 1213 struct vm_area_struct *vma, struct list_head *to_kill, 1214 unsigned long ksm_addr); 1215 unsigned long page_mapped_in_vma(const struct page *page, 1216 struct vm_area_struct *vma); 1217 1218 #else 1219 static inline int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill) 1220 { 1221 return -EBUSY; 1222 } 1223 #endif 1224 1225 extern unsigned long __must_check vm_mmap_pgoff(struct file *, unsigned long, 1226 unsigned long, unsigned long, 1227 unsigned long, unsigned long); 1228 1229 extern void set_pageblock_order(void); 1230 struct folio *alloc_migrate_folio(struct folio *src, unsigned long private); 1231 unsigned long reclaim_pages(struct list_head *folio_list); 1232 unsigned int reclaim_clean_pages_from_list(struct zone *zone, 1233 struct list_head *folio_list); 1234 /* The ALLOC_WMARK bits are used as an index to zone->watermark */ 1235 #define ALLOC_WMARK_MIN WMARK_MIN 1236 #define ALLOC_WMARK_LOW WMARK_LOW 1237 #define ALLOC_WMARK_HIGH WMARK_HIGH 1238 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ 1239 1240 /* Mask to get the watermark bits */ 1241 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) 1242 1243 /* 1244 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we 1245 * cannot assume a reduced access to memory reserves is sufficient for 1246 * !MMU 1247 */ 1248 #ifdef CONFIG_MMU 1249 #define ALLOC_OOM 0x08 1250 #else 1251 #define ALLOC_OOM ALLOC_NO_WATERMARKS 1252 #endif 1253 1254 #define ALLOC_NON_BLOCK 0x10 /* Caller cannot block. Allow access 1255 * to 25% of the min watermark or 1256 * 62.5% if __GFP_HIGH is set. 1257 */ 1258 #define ALLOC_MIN_RESERVE 0x20 /* __GFP_HIGH set. Allow access to 50% 1259 * of the min watermark. 1260 */ 1261 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ 1262 #define ALLOC_CMA 0x80 /* allow allocations from CMA areas */ 1263 #ifdef CONFIG_ZONE_DMA32 1264 #define ALLOC_NOFRAGMENT 0x100 /* avoid mixing pageblock types */ 1265 #else 1266 #define ALLOC_NOFRAGMENT 0x0 1267 #endif 1268 #define ALLOC_HIGHATOMIC 0x200 /* Allows access to MIGRATE_HIGHATOMIC */ 1269 #define ALLOC_TRYLOCK 0x400 /* Only use spin_trylock in allocation path */ 1270 #define ALLOC_KSWAPD 0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */ 1271 1272 /* Flags that allow allocations below the min watermark. */ 1273 #define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM) 1274 1275 enum ttu_flags; 1276 struct tlbflush_unmap_batch; 1277 1278 1279 /* 1280 * only for MM internal work items which do not depend on 1281 * any allocations or locks which might depend on allocations 1282 */ 1283 extern struct workqueue_struct *mm_percpu_wq; 1284 1285 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 1286 void try_to_unmap_flush(void); 1287 void try_to_unmap_flush_dirty(void); 1288 void flush_tlb_batched_pending(struct mm_struct *mm); 1289 #else 1290 static inline void try_to_unmap_flush(void) 1291 { 1292 } 1293 static inline void try_to_unmap_flush_dirty(void) 1294 { 1295 } 1296 static inline void flush_tlb_batched_pending(struct mm_struct *mm) 1297 { 1298 } 1299 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ 1300 1301 extern const struct trace_print_flags pageflag_names[]; 1302 extern const struct trace_print_flags vmaflag_names[]; 1303 extern const struct trace_print_flags gfpflag_names[]; 1304 1305 static inline bool is_migrate_highatomic(enum migratetype migratetype) 1306 { 1307 return migratetype == MIGRATE_HIGHATOMIC; 1308 } 1309 1310 void setup_zone_pageset(struct zone *zone); 1311 1312 struct migration_target_control { 1313 int nid; /* preferred node id */ 1314 nodemask_t *nmask; 1315 gfp_t gfp_mask; 1316 enum migrate_reason reason; 1317 }; 1318 1319 /* 1320 * mm/filemap.c 1321 */ 1322 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe, 1323 struct folio *folio, loff_t fpos, size_t size); 1324 1325 /* 1326 * mm/vmalloc.c 1327 */ 1328 #ifdef CONFIG_MMU 1329 void __init vmalloc_init(void); 1330 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1331 pgprot_t prot, struct page **pages, unsigned int page_shift); 1332 unsigned int get_vm_area_page_order(struct vm_struct *vm); 1333 #else 1334 static inline void vmalloc_init(void) 1335 { 1336 } 1337 1338 static inline 1339 int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end, 1340 pgprot_t prot, struct page **pages, unsigned int page_shift) 1341 { 1342 return -EINVAL; 1343 } 1344 #endif 1345 1346 int __must_check __vmap_pages_range_noflush(unsigned long addr, 1347 unsigned long end, pgprot_t prot, 1348 struct page **pages, unsigned int page_shift); 1349 1350 void vunmap_range_noflush(unsigned long start, unsigned long end); 1351 1352 void __vunmap_range_noflush(unsigned long start, unsigned long end); 1353 1354 int numa_migrate_check(struct folio *folio, struct vm_fault *vmf, 1355 unsigned long addr, int *flags, bool writable, 1356 int *last_cpupid); 1357 1358 void free_zone_device_folio(struct folio *folio); 1359 int migrate_device_coherent_folio(struct folio *folio); 1360 1361 struct vm_struct *__get_vm_area_node(unsigned long size, 1362 unsigned long align, unsigned long shift, 1363 unsigned long flags, unsigned long start, 1364 unsigned long end, int node, gfp_t gfp_mask, 1365 const void *caller); 1366 1367 /* 1368 * mm/gup.c 1369 */ 1370 int __must_check try_grab_folio(struct folio *folio, int refs, 1371 unsigned int flags); 1372 1373 /* 1374 * mm/huge_memory.c 1375 */ 1376 void touch_pud(struct vm_area_struct *vma, unsigned long addr, 1377 pud_t *pud, bool write); 1378 void touch_pmd(struct vm_area_struct *vma, unsigned long addr, 1379 pmd_t *pmd, bool write); 1380 1381 /* 1382 * Parses a string with mem suffixes into its order. Useful to parse kernel 1383 * parameters. 1384 */ 1385 static inline int get_order_from_str(const char *size_str, 1386 unsigned long valid_orders) 1387 { 1388 unsigned long size; 1389 char *endptr; 1390 int order; 1391 1392 size = memparse(size_str, &endptr); 1393 1394 if (!is_power_of_2(size)) 1395 return -EINVAL; 1396 order = get_order(size); 1397 if (BIT(order) & ~valid_orders) 1398 return -EINVAL; 1399 1400 return order; 1401 } 1402 1403 enum { 1404 /* mark page accessed */ 1405 FOLL_TOUCH = 1 << 16, 1406 /* a retry, previous pass started an IO */ 1407 FOLL_TRIED = 1 << 17, 1408 /* we are working on non-current tsk/mm */ 1409 FOLL_REMOTE = 1 << 18, 1410 /* pages must be released via unpin_user_page */ 1411 FOLL_PIN = 1 << 19, 1412 /* gup_fast: prevent fall-back to slow gup */ 1413 FOLL_FAST_ONLY = 1 << 20, 1414 /* allow unlocking the mmap lock */ 1415 FOLL_UNLOCKABLE = 1 << 21, 1416 /* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */ 1417 FOLL_MADV_POPULATE = 1 << 22, 1418 }; 1419 1420 #define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \ 1421 FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \ 1422 FOLL_MADV_POPULATE) 1423 1424 /* 1425 * Indicates for which pages that are write-protected in the page table, 1426 * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the 1427 * GUP pin will remain consistent with the pages mapped into the page tables 1428 * of the MM. 1429 * 1430 * Temporary unmapping of PageAnonExclusive() pages or clearing of 1431 * PageAnonExclusive() has to protect against concurrent GUP: 1432 * * Ordinary GUP: Using the PT lock 1433 * * GUP-fast and fork(): mm->write_protect_seq 1434 * * GUP-fast and KSM or temporary unmapping (swap, migration): see 1435 * folio_try_share_anon_rmap_*() 1436 * 1437 * Must be called with the (sub)page that's actually referenced via the 1438 * page table entry, which might not necessarily be the head page for a 1439 * PTE-mapped THP. 1440 * 1441 * If the vma is NULL, we're coming from the GUP-fast path and might have 1442 * to fallback to the slow path just to lookup the vma. 1443 */ 1444 static inline bool gup_must_unshare(struct vm_area_struct *vma, 1445 unsigned int flags, struct page *page) 1446 { 1447 /* 1448 * FOLL_WRITE is implicitly handled correctly as the page table entry 1449 * has to be writable -- and if it references (part of) an anonymous 1450 * folio, that part is required to be marked exclusive. 1451 */ 1452 if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN) 1453 return false; 1454 /* 1455 * Note: PageAnon(page) is stable until the page is actually getting 1456 * freed. 1457 */ 1458 if (!PageAnon(page)) { 1459 /* 1460 * We only care about R/O long-term pining: R/O short-term 1461 * pinning does not have the semantics to observe successive 1462 * changes through the process page tables. 1463 */ 1464 if (!(flags & FOLL_LONGTERM)) 1465 return false; 1466 1467 /* We really need the vma ... */ 1468 if (!vma) 1469 return true; 1470 1471 /* 1472 * ... because we only care about writable private ("COW") 1473 * mappings where we have to break COW early. 1474 */ 1475 return is_cow_mapping(vma->vm_flags); 1476 } 1477 1478 /* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */ 1479 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) 1480 smp_rmb(); 1481 1482 /* 1483 * Note that KSM pages cannot be exclusive, and consequently, 1484 * cannot get pinned. 1485 */ 1486 return !PageAnonExclusive(page); 1487 } 1488 1489 extern bool mirrored_kernelcore; 1490 bool memblock_has_mirror(void); 1491 void memblock_free_all(void); 1492 1493 static __always_inline void vma_set_range(struct vm_area_struct *vma, 1494 unsigned long start, unsigned long end, 1495 pgoff_t pgoff) 1496 { 1497 vma->vm_start = start; 1498 vma->vm_end = end; 1499 vma->vm_pgoff = pgoff; 1500 } 1501 1502 static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma) 1503 { 1504 /* 1505 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty 1506 * enablements, because when without soft-dirty being compiled in, 1507 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY) 1508 * will be constantly true. 1509 */ 1510 if (!IS_ENABLED(CONFIG_MEM_SOFT_DIRTY)) 1511 return false; 1512 1513 /* 1514 * Soft-dirty is kind of special: its tracking is enabled when the 1515 * vma flags not set. 1516 */ 1517 return !(vma->vm_flags & VM_SOFTDIRTY); 1518 } 1519 1520 static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd) 1521 { 1522 return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd); 1523 } 1524 1525 static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte) 1526 { 1527 return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte); 1528 } 1529 1530 void __meminit __init_single_page(struct page *page, unsigned long pfn, 1531 unsigned long zone, int nid); 1532 void __meminit __init_page_from_nid(unsigned long pfn, int nid); 1533 1534 /* shrinker related functions */ 1535 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg, 1536 int priority); 1537 1538 #ifdef CONFIG_SHRINKER_DEBUG 1539 static inline __printf(2, 0) int shrinker_debugfs_name_alloc( 1540 struct shrinker *shrinker, const char *fmt, va_list ap) 1541 { 1542 shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap); 1543 1544 return shrinker->name ? 0 : -ENOMEM; 1545 } 1546 1547 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1548 { 1549 kfree_const(shrinker->name); 1550 shrinker->name = NULL; 1551 } 1552 1553 extern int shrinker_debugfs_add(struct shrinker *shrinker); 1554 extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1555 int *debugfs_id); 1556 extern void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1557 int debugfs_id); 1558 #else /* CONFIG_SHRINKER_DEBUG */ 1559 static inline int shrinker_debugfs_add(struct shrinker *shrinker) 1560 { 1561 return 0; 1562 } 1563 static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker, 1564 const char *fmt, va_list ap) 1565 { 1566 return 0; 1567 } 1568 static inline void shrinker_debugfs_name_free(struct shrinker *shrinker) 1569 { 1570 } 1571 static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker, 1572 int *debugfs_id) 1573 { 1574 *debugfs_id = -1; 1575 return NULL; 1576 } 1577 static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry, 1578 int debugfs_id) 1579 { 1580 } 1581 #endif /* CONFIG_SHRINKER_DEBUG */ 1582 1583 /* Only track the nodes of mappings with shadow entries */ 1584 void workingset_update_node(struct xa_node *node); 1585 extern struct list_lru shadow_nodes; 1586 #define mapping_set_update(xas, mapping) do { \ 1587 if (!dax_mapping(mapping) && !shmem_mapping(mapping)) { \ 1588 xas_set_update(xas, workingset_update_node); \ 1589 xas_set_lru(xas, &shadow_nodes); \ 1590 } \ 1591 } while (0) 1592 1593 /* mremap.c */ 1594 unsigned long move_page_tables(struct pagetable_move_control *pmc); 1595 1596 #ifdef CONFIG_UNACCEPTED_MEMORY 1597 void accept_page(struct page *page); 1598 #else /* CONFIG_UNACCEPTED_MEMORY */ 1599 static inline void accept_page(struct page *page) 1600 { 1601 } 1602 #endif /* CONFIG_UNACCEPTED_MEMORY */ 1603 1604 /* pagewalk.c */ 1605 int walk_page_range_mm(struct mm_struct *mm, unsigned long start, 1606 unsigned long end, const struct mm_walk_ops *ops, 1607 void *private); 1608 1609 /* pt_reclaim.c */ 1610 bool try_get_and_clear_pmd(struct mm_struct *mm, pmd_t *pmd, pmd_t *pmdval); 1611 void free_pte(struct mm_struct *mm, unsigned long addr, struct mmu_gather *tlb, 1612 pmd_t pmdval); 1613 void try_to_free_pte(struct mm_struct *mm, pmd_t *pmd, unsigned long addr, 1614 struct mmu_gather *tlb); 1615 1616 #ifdef CONFIG_PT_RECLAIM 1617 bool reclaim_pt_is_enabled(unsigned long start, unsigned long end, 1618 struct zap_details *details); 1619 #else 1620 static inline bool reclaim_pt_is_enabled(unsigned long start, unsigned long end, 1621 struct zap_details *details) 1622 { 1623 return false; 1624 } 1625 #endif /* CONFIG_PT_RECLAIM */ 1626 1627 void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm); 1628 int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm); 1629 1630 #endif /* __MM_INTERNAL_H */ 1631