1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_MM_H 3 #define _LINUX_MM_H 4 5 #include <linux/errno.h> 6 #include <linux/mmdebug.h> 7 #include <linux/gfp.h> 8 #include <linux/bug.h> 9 #include <linux/list.h> 10 #include <linux/mmzone.h> 11 #include <linux/rbtree.h> 12 #include <linux/atomic.h> 13 #include <linux/debug_locks.h> 14 #include <linux/mm_types.h> 15 #include <linux/mmap_lock.h> 16 #include <linux/range.h> 17 #include <linux/pfn.h> 18 #include <linux/percpu-refcount.h> 19 #include <linux/bit_spinlock.h> 20 #include <linux/shrinker.h> 21 #include <linux/resource.h> 22 #include <linux/page_ext.h> 23 #include <linux/err.h> 24 #include <linux/page-flags.h> 25 #include <linux/page_ref.h> 26 #include <linux/overflow.h> 27 #include <linux/sizes.h> 28 #include <linux/sched.h> 29 #include <linux/pgtable.h> 30 #include <linux/kasan.h> 31 #include <linux/memremap.h> 32 #include <linux/slab.h> 33 34 struct mempolicy; 35 struct anon_vma; 36 struct anon_vma_chain; 37 struct user_struct; 38 struct pt_regs; 39 40 extern int sysctl_page_lock_unfairness; 41 42 void mm_core_init(void); 43 void init_mm_internals(void); 44 45 #ifndef CONFIG_NUMA /* Don't use mapnrs, do it properly */ 46 extern unsigned long max_mapnr; 47 48 static inline void set_max_mapnr(unsigned long limit) 49 { 50 max_mapnr = limit; 51 } 52 #else 53 static inline void set_max_mapnr(unsigned long limit) { } 54 #endif 55 56 extern atomic_long_t _totalram_pages; 57 static inline unsigned long totalram_pages(void) 58 { 59 return (unsigned long)atomic_long_read(&_totalram_pages); 60 } 61 62 static inline void totalram_pages_inc(void) 63 { 64 atomic_long_inc(&_totalram_pages); 65 } 66 67 static inline void totalram_pages_dec(void) 68 { 69 atomic_long_dec(&_totalram_pages); 70 } 71 72 static inline void totalram_pages_add(long count) 73 { 74 atomic_long_add(count, &_totalram_pages); 75 } 76 77 extern void * high_memory; 78 extern int page_cluster; 79 extern const int page_cluster_max; 80 81 #ifdef CONFIG_SYSCTL 82 extern int sysctl_legacy_va_layout; 83 #else 84 #define sysctl_legacy_va_layout 0 85 #endif 86 87 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 88 extern const int mmap_rnd_bits_min; 89 extern const int mmap_rnd_bits_max; 90 extern int mmap_rnd_bits __read_mostly; 91 #endif 92 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 93 extern const int mmap_rnd_compat_bits_min; 94 extern const int mmap_rnd_compat_bits_max; 95 extern int mmap_rnd_compat_bits __read_mostly; 96 #endif 97 98 #include <asm/page.h> 99 #include <asm/processor.h> 100 101 #ifndef __pa_symbol 102 #define __pa_symbol(x) __pa(RELOC_HIDE((unsigned long)(x), 0)) 103 #endif 104 105 #ifndef page_to_virt 106 #define page_to_virt(x) __va(PFN_PHYS(page_to_pfn(x))) 107 #endif 108 109 #ifndef lm_alias 110 #define lm_alias(x) __va(__pa_symbol(x)) 111 #endif 112 113 /* 114 * To prevent common memory management code establishing 115 * a zero page mapping on a read fault. 116 * This macro should be defined within <asm/pgtable.h>. 117 * s390 does this to prevent multiplexing of hardware bits 118 * related to the physical page in case of virtualization. 119 */ 120 #ifndef mm_forbids_zeropage 121 #define mm_forbids_zeropage(X) (0) 122 #endif 123 124 /* 125 * On some architectures it is expensive to call memset() for small sizes. 126 * If an architecture decides to implement their own version of 127 * mm_zero_struct_page they should wrap the defines below in a #ifndef and 128 * define their own version of this macro in <asm/pgtable.h> 129 */ 130 #if BITS_PER_LONG == 64 131 /* This function must be updated when the size of struct page grows above 96 132 * or reduces below 56. The idea that compiler optimizes out switch() 133 * statement, and only leaves move/store instructions. Also the compiler can 134 * combine write statements if they are both assignments and can be reordered, 135 * this can result in several of the writes here being dropped. 136 */ 137 #define mm_zero_struct_page(pp) __mm_zero_struct_page(pp) 138 static inline void __mm_zero_struct_page(struct page *page) 139 { 140 unsigned long *_pp = (void *)page; 141 142 /* Check that struct page is either 56, 64, 72, 80, 88 or 96 bytes */ 143 BUILD_BUG_ON(sizeof(struct page) & 7); 144 BUILD_BUG_ON(sizeof(struct page) < 56); 145 BUILD_BUG_ON(sizeof(struct page) > 96); 146 147 switch (sizeof(struct page)) { 148 case 96: 149 _pp[11] = 0; 150 fallthrough; 151 case 88: 152 _pp[10] = 0; 153 fallthrough; 154 case 80: 155 _pp[9] = 0; 156 fallthrough; 157 case 72: 158 _pp[8] = 0; 159 fallthrough; 160 case 64: 161 _pp[7] = 0; 162 fallthrough; 163 case 56: 164 _pp[6] = 0; 165 _pp[5] = 0; 166 _pp[4] = 0; 167 _pp[3] = 0; 168 _pp[2] = 0; 169 _pp[1] = 0; 170 _pp[0] = 0; 171 } 172 } 173 #else 174 #define mm_zero_struct_page(pp) ((void)memset((pp), 0, sizeof(struct page))) 175 #endif 176 177 /* 178 * Default maximum number of active map areas, this limits the number of vmas 179 * per mm struct. Users can overwrite this number by sysctl but there is a 180 * problem. 181 * 182 * When a program's coredump is generated as ELF format, a section is created 183 * per a vma. In ELF, the number of sections is represented in unsigned short. 184 * This means the number of sections should be smaller than 65535 at coredump. 185 * Because the kernel adds some informative sections to a image of program at 186 * generating coredump, we need some margin. The number of extra sections is 187 * 1-3 now and depends on arch. We use "5" as safe margin, here. 188 * 189 * ELF extended numbering allows more than 65535 sections, so 16-bit bound is 190 * not a hard limit any more. Although some userspace tools can be surprised by 191 * that. 192 */ 193 #define MAPCOUNT_ELF_CORE_MARGIN (5) 194 #define DEFAULT_MAX_MAP_COUNT (USHRT_MAX - MAPCOUNT_ELF_CORE_MARGIN) 195 196 extern int sysctl_max_map_count; 197 198 extern unsigned long sysctl_user_reserve_kbytes; 199 extern unsigned long sysctl_admin_reserve_kbytes; 200 201 extern int sysctl_overcommit_memory; 202 extern int sysctl_overcommit_ratio; 203 extern unsigned long sysctl_overcommit_kbytes; 204 205 int overcommit_ratio_handler(struct ctl_table *, int, void *, size_t *, 206 loff_t *); 207 int overcommit_kbytes_handler(struct ctl_table *, int, void *, size_t *, 208 loff_t *); 209 int overcommit_policy_handler(struct ctl_table *, int, void *, size_t *, 210 loff_t *); 211 212 #if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP) 213 #define nth_page(page,n) pfn_to_page(page_to_pfn((page)) + (n)) 214 #define folio_page_idx(folio, p) (page_to_pfn(p) - folio_pfn(folio)) 215 #else 216 #define nth_page(page,n) ((page) + (n)) 217 #define folio_page_idx(folio, p) ((p) - &(folio)->page) 218 #endif 219 220 /* to align the pointer to the (next) page boundary */ 221 #define PAGE_ALIGN(addr) ALIGN(addr, PAGE_SIZE) 222 223 /* to align the pointer to the (prev) page boundary */ 224 #define PAGE_ALIGN_DOWN(addr) ALIGN_DOWN(addr, PAGE_SIZE) 225 226 /* test whether an address (unsigned long or pointer) is aligned to PAGE_SIZE */ 227 #define PAGE_ALIGNED(addr) IS_ALIGNED((unsigned long)(addr), PAGE_SIZE) 228 229 #define lru_to_page(head) (list_entry((head)->prev, struct page, lru)) 230 static inline struct folio *lru_to_folio(struct list_head *head) 231 { 232 return list_entry((head)->prev, struct folio, lru); 233 } 234 235 void setup_initial_init_mm(void *start_code, void *end_code, 236 void *end_data, void *brk); 237 238 /* 239 * Linux kernel virtual memory manager primitives. 240 * The idea being to have a "virtual" mm in the same way 241 * we have a virtual fs - giving a cleaner interface to the 242 * mm details, and allowing different kinds of memory mappings 243 * (from shared memory to executable loading to arbitrary 244 * mmap() functions). 245 */ 246 247 struct vm_area_struct *vm_area_alloc(struct mm_struct *); 248 struct vm_area_struct *vm_area_dup(struct vm_area_struct *); 249 void vm_area_free(struct vm_area_struct *); 250 /* Use only if VMA has no other users */ 251 void __vm_area_free(struct vm_area_struct *vma); 252 253 #ifndef CONFIG_MMU 254 extern struct rb_root nommu_region_tree; 255 extern struct rw_semaphore nommu_region_sem; 256 257 extern unsigned int kobjsize(const void *objp); 258 #endif 259 260 /* 261 * vm_flags in vm_area_struct, see mm_types.h. 262 * When changing, update also include/trace/events/mmflags.h 263 */ 264 #define VM_NONE 0x00000000 265 266 #define VM_READ 0x00000001 /* currently active flags */ 267 #define VM_WRITE 0x00000002 268 #define VM_EXEC 0x00000004 269 #define VM_SHARED 0x00000008 270 271 /* mprotect() hardcodes VM_MAYREAD >> 4 == VM_READ, and so for r/w/x bits. */ 272 #define VM_MAYREAD 0x00000010 /* limits for mprotect() etc */ 273 #define VM_MAYWRITE 0x00000020 274 #define VM_MAYEXEC 0x00000040 275 #define VM_MAYSHARE 0x00000080 276 277 #define VM_GROWSDOWN 0x00000100 /* general info on the segment */ 278 #ifdef CONFIG_MMU 279 #define VM_UFFD_MISSING 0x00000200 /* missing pages tracking */ 280 #else /* CONFIG_MMU */ 281 #define VM_MAYOVERLAY 0x00000200 /* nommu: R/O MAP_PRIVATE mapping that might overlay a file mapping */ 282 #define VM_UFFD_MISSING 0 283 #endif /* CONFIG_MMU */ 284 #define VM_PFNMAP 0x00000400 /* Page-ranges managed without "struct page", just pure PFN */ 285 #define VM_UFFD_WP 0x00001000 /* wrprotect pages tracking */ 286 287 #define VM_LOCKED 0x00002000 288 #define VM_IO 0x00004000 /* Memory mapped I/O or similar */ 289 290 /* Used by sys_madvise() */ 291 #define VM_SEQ_READ 0x00008000 /* App will access data sequentially */ 292 #define VM_RAND_READ 0x00010000 /* App will not benefit from clustered reads */ 293 294 #define VM_DONTCOPY 0x00020000 /* Do not copy this vma on fork */ 295 #define VM_DONTEXPAND 0x00040000 /* Cannot expand with mremap() */ 296 #define VM_LOCKONFAULT 0x00080000 /* Lock the pages covered when they are faulted in */ 297 #define VM_ACCOUNT 0x00100000 /* Is a VM accounted object */ 298 #define VM_NORESERVE 0x00200000 /* should the VM suppress accounting */ 299 #define VM_HUGETLB 0x00400000 /* Huge TLB Page VM */ 300 #define VM_SYNC 0x00800000 /* Synchronous page faults */ 301 #define VM_ARCH_1 0x01000000 /* Architecture-specific flag */ 302 #define VM_WIPEONFORK 0x02000000 /* Wipe VMA contents in child. */ 303 #define VM_DONTDUMP 0x04000000 /* Do not include in the core dump */ 304 305 #ifdef CONFIG_MEM_SOFT_DIRTY 306 # define VM_SOFTDIRTY 0x08000000 /* Not soft dirty clean area */ 307 #else 308 # define VM_SOFTDIRTY 0 309 #endif 310 311 #define VM_MIXEDMAP 0x10000000 /* Can contain "struct page" and pure PFN pages */ 312 #define VM_HUGEPAGE 0x20000000 /* MADV_HUGEPAGE marked this vma */ 313 #define VM_NOHUGEPAGE 0x40000000 /* MADV_NOHUGEPAGE marked this vma */ 314 #define VM_MERGEABLE 0x80000000 /* KSM may merge identical pages */ 315 316 #ifdef CONFIG_ARCH_USES_HIGH_VMA_FLAGS 317 #define VM_HIGH_ARCH_BIT_0 32 /* bit only usable on 64-bit architectures */ 318 #define VM_HIGH_ARCH_BIT_1 33 /* bit only usable on 64-bit architectures */ 319 #define VM_HIGH_ARCH_BIT_2 34 /* bit only usable on 64-bit architectures */ 320 #define VM_HIGH_ARCH_BIT_3 35 /* bit only usable on 64-bit architectures */ 321 #define VM_HIGH_ARCH_BIT_4 36 /* bit only usable on 64-bit architectures */ 322 #define VM_HIGH_ARCH_0 BIT(VM_HIGH_ARCH_BIT_0) 323 #define VM_HIGH_ARCH_1 BIT(VM_HIGH_ARCH_BIT_1) 324 #define VM_HIGH_ARCH_2 BIT(VM_HIGH_ARCH_BIT_2) 325 #define VM_HIGH_ARCH_3 BIT(VM_HIGH_ARCH_BIT_3) 326 #define VM_HIGH_ARCH_4 BIT(VM_HIGH_ARCH_BIT_4) 327 #endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */ 328 329 #ifdef CONFIG_ARCH_HAS_PKEYS 330 # define VM_PKEY_SHIFT VM_HIGH_ARCH_BIT_0 331 # define VM_PKEY_BIT0 VM_HIGH_ARCH_0 /* A protection key is a 4-bit value */ 332 # define VM_PKEY_BIT1 VM_HIGH_ARCH_1 /* on x86 and 5-bit value on ppc64 */ 333 # define VM_PKEY_BIT2 VM_HIGH_ARCH_2 334 # define VM_PKEY_BIT3 VM_HIGH_ARCH_3 335 #ifdef CONFIG_PPC 336 # define VM_PKEY_BIT4 VM_HIGH_ARCH_4 337 #else 338 # define VM_PKEY_BIT4 0 339 #endif 340 #endif /* CONFIG_ARCH_HAS_PKEYS */ 341 342 #if defined(CONFIG_X86) 343 # define VM_PAT VM_ARCH_1 /* PAT reserves whole VMA at once (x86) */ 344 #elif defined(CONFIG_PPC) 345 # define VM_SAO VM_ARCH_1 /* Strong Access Ordering (powerpc) */ 346 #elif defined(CONFIG_PARISC) 347 # define VM_GROWSUP VM_ARCH_1 348 #elif defined(CONFIG_IA64) 349 # define VM_GROWSUP VM_ARCH_1 350 #elif defined(CONFIG_SPARC64) 351 # define VM_SPARC_ADI VM_ARCH_1 /* Uses ADI tag for access control */ 352 # define VM_ARCH_CLEAR VM_SPARC_ADI 353 #elif defined(CONFIG_ARM64) 354 # define VM_ARM64_BTI VM_ARCH_1 /* BTI guarded page, a.k.a. GP bit */ 355 # define VM_ARCH_CLEAR VM_ARM64_BTI 356 #elif !defined(CONFIG_MMU) 357 # define VM_MAPPED_COPY VM_ARCH_1 /* T if mapped copy of data (nommu mmap) */ 358 #endif 359 360 #if defined(CONFIG_ARM64_MTE) 361 # define VM_MTE VM_HIGH_ARCH_0 /* Use Tagged memory for access control */ 362 # define VM_MTE_ALLOWED VM_HIGH_ARCH_1 /* Tagged memory permitted */ 363 #else 364 # define VM_MTE VM_NONE 365 # define VM_MTE_ALLOWED VM_NONE 366 #endif 367 368 #ifndef VM_GROWSUP 369 # define VM_GROWSUP VM_NONE 370 #endif 371 372 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 373 # define VM_UFFD_MINOR_BIT 37 374 # define VM_UFFD_MINOR BIT(VM_UFFD_MINOR_BIT) /* UFFD minor faults */ 375 #else /* !CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 376 # define VM_UFFD_MINOR VM_NONE 377 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 378 379 /* Bits set in the VMA until the stack is in its final location */ 380 #define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ) 381 382 #define TASK_EXEC ((current->personality & READ_IMPLIES_EXEC) ? VM_EXEC : 0) 383 384 /* Common data flag combinations */ 385 #define VM_DATA_FLAGS_TSK_EXEC (VM_READ | VM_WRITE | TASK_EXEC | \ 386 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC) 387 #define VM_DATA_FLAGS_NON_EXEC (VM_READ | VM_WRITE | VM_MAYREAD | \ 388 VM_MAYWRITE | VM_MAYEXEC) 389 #define VM_DATA_FLAGS_EXEC (VM_READ | VM_WRITE | VM_EXEC | \ 390 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC) 391 392 #ifndef VM_DATA_DEFAULT_FLAGS /* arch can override this */ 393 #define VM_DATA_DEFAULT_FLAGS VM_DATA_FLAGS_EXEC 394 #endif 395 396 #ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */ 397 #define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS 398 #endif 399 400 #ifdef CONFIG_STACK_GROWSUP 401 #define VM_STACK VM_GROWSUP 402 #else 403 #define VM_STACK VM_GROWSDOWN 404 #endif 405 406 #define VM_STACK_FLAGS (VM_STACK | VM_STACK_DEFAULT_FLAGS | VM_ACCOUNT) 407 408 /* VMA basic access permission flags */ 409 #define VM_ACCESS_FLAGS (VM_READ | VM_WRITE | VM_EXEC) 410 411 412 /* 413 * Special vmas that are non-mergable, non-mlock()able. 414 */ 415 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_PFNMAP | VM_MIXEDMAP) 416 417 /* This mask prevents VMA from being scanned with khugepaged */ 418 #define VM_NO_KHUGEPAGED (VM_SPECIAL | VM_HUGETLB) 419 420 /* This mask defines which mm->def_flags a process can inherit its parent */ 421 #define VM_INIT_DEF_MASK VM_NOHUGEPAGE 422 423 /* This mask represents all the VMA flag bits used by mlock */ 424 #define VM_LOCKED_MASK (VM_LOCKED | VM_LOCKONFAULT) 425 426 /* Arch-specific flags to clear when updating VM flags on protection change */ 427 #ifndef VM_ARCH_CLEAR 428 # define VM_ARCH_CLEAR VM_NONE 429 #endif 430 #define VM_FLAGS_CLEAR (ARCH_VM_PKEY_FLAGS | VM_ARCH_CLEAR) 431 432 /* 433 * mapping from the currently active vm_flags protection bits (the 434 * low four bits) to a page protection mask.. 435 */ 436 437 /* 438 * The default fault flags that should be used by most of the 439 * arch-specific page fault handlers. 440 */ 441 #define FAULT_FLAG_DEFAULT (FAULT_FLAG_ALLOW_RETRY | \ 442 FAULT_FLAG_KILLABLE | \ 443 FAULT_FLAG_INTERRUPTIBLE) 444 445 /** 446 * fault_flag_allow_retry_first - check ALLOW_RETRY the first time 447 * @flags: Fault flags. 448 * 449 * This is mostly used for places where we want to try to avoid taking 450 * the mmap_lock for too long a time when waiting for another condition 451 * to change, in which case we can try to be polite to release the 452 * mmap_lock in the first round to avoid potential starvation of other 453 * processes that would also want the mmap_lock. 454 * 455 * Return: true if the page fault allows retry and this is the first 456 * attempt of the fault handling; false otherwise. 457 */ 458 static inline bool fault_flag_allow_retry_first(enum fault_flag flags) 459 { 460 return (flags & FAULT_FLAG_ALLOW_RETRY) && 461 (!(flags & FAULT_FLAG_TRIED)); 462 } 463 464 #define FAULT_FLAG_TRACE \ 465 { FAULT_FLAG_WRITE, "WRITE" }, \ 466 { FAULT_FLAG_MKWRITE, "MKWRITE" }, \ 467 { FAULT_FLAG_ALLOW_RETRY, "ALLOW_RETRY" }, \ 468 { FAULT_FLAG_RETRY_NOWAIT, "RETRY_NOWAIT" }, \ 469 { FAULT_FLAG_KILLABLE, "KILLABLE" }, \ 470 { FAULT_FLAG_TRIED, "TRIED" }, \ 471 { FAULT_FLAG_USER, "USER" }, \ 472 { FAULT_FLAG_REMOTE, "REMOTE" }, \ 473 { FAULT_FLAG_INSTRUCTION, "INSTRUCTION" }, \ 474 { FAULT_FLAG_INTERRUPTIBLE, "INTERRUPTIBLE" }, \ 475 { FAULT_FLAG_VMA_LOCK, "VMA_LOCK" } 476 477 /* 478 * vm_fault is filled by the pagefault handler and passed to the vma's 479 * ->fault function. The vma's ->fault is responsible for returning a bitmask 480 * of VM_FAULT_xxx flags that give details about how the fault was handled. 481 * 482 * MM layer fills up gfp_mask for page allocations but fault handler might 483 * alter it if its implementation requires a different allocation context. 484 * 485 * pgoff should be used in favour of virtual_address, if possible. 486 */ 487 struct vm_fault { 488 const struct { 489 struct vm_area_struct *vma; /* Target VMA */ 490 gfp_t gfp_mask; /* gfp mask to be used for allocations */ 491 pgoff_t pgoff; /* Logical page offset based on vma */ 492 unsigned long address; /* Faulting virtual address - masked */ 493 unsigned long real_address; /* Faulting virtual address - unmasked */ 494 }; 495 enum fault_flag flags; /* FAULT_FLAG_xxx flags 496 * XXX: should really be 'const' */ 497 pmd_t *pmd; /* Pointer to pmd entry matching 498 * the 'address' */ 499 pud_t *pud; /* Pointer to pud entry matching 500 * the 'address' 501 */ 502 union { 503 pte_t orig_pte; /* Value of PTE at the time of fault */ 504 pmd_t orig_pmd; /* Value of PMD at the time of fault, 505 * used by PMD fault only. 506 */ 507 }; 508 509 struct page *cow_page; /* Page handler may use for COW fault */ 510 struct page *page; /* ->fault handlers should return a 511 * page here, unless VM_FAULT_NOPAGE 512 * is set (which is also implied by 513 * VM_FAULT_ERROR). 514 */ 515 /* These three entries are valid only while holding ptl lock */ 516 pte_t *pte; /* Pointer to pte entry matching 517 * the 'address'. NULL if the page 518 * table hasn't been allocated. 519 */ 520 spinlock_t *ptl; /* Page table lock. 521 * Protects pte page table if 'pte' 522 * is not NULL, otherwise pmd. 523 */ 524 pgtable_t prealloc_pte; /* Pre-allocated pte page table. 525 * vm_ops->map_pages() sets up a page 526 * table from atomic context. 527 * do_fault_around() pre-allocates 528 * page table to avoid allocation from 529 * atomic context. 530 */ 531 }; 532 533 /* page entry size for vm->huge_fault() */ 534 enum page_entry_size { 535 PE_SIZE_PTE = 0, 536 PE_SIZE_PMD, 537 PE_SIZE_PUD, 538 }; 539 540 /* 541 * These are the virtual MM functions - opening of an area, closing and 542 * unmapping it (needed to keep files on disk up-to-date etc), pointer 543 * to the functions called when a no-page or a wp-page exception occurs. 544 */ 545 struct vm_operations_struct { 546 void (*open)(struct vm_area_struct * area); 547 /** 548 * @close: Called when the VMA is being removed from the MM. 549 * Context: User context. May sleep. Caller holds mmap_lock. 550 */ 551 void (*close)(struct vm_area_struct * area); 552 /* Called any time before splitting to check if it's allowed */ 553 int (*may_split)(struct vm_area_struct *area, unsigned long addr); 554 int (*mremap)(struct vm_area_struct *area); 555 /* 556 * Called by mprotect() to make driver-specific permission 557 * checks before mprotect() is finalised. The VMA must not 558 * be modified. Returns 0 if mprotect() can proceed. 559 */ 560 int (*mprotect)(struct vm_area_struct *vma, unsigned long start, 561 unsigned long end, unsigned long newflags); 562 vm_fault_t (*fault)(struct vm_fault *vmf); 563 vm_fault_t (*huge_fault)(struct vm_fault *vmf, 564 enum page_entry_size pe_size); 565 vm_fault_t (*map_pages)(struct vm_fault *vmf, 566 pgoff_t start_pgoff, pgoff_t end_pgoff); 567 unsigned long (*pagesize)(struct vm_area_struct * area); 568 569 /* notification that a previously read-only page is about to become 570 * writable, if an error is returned it will cause a SIGBUS */ 571 vm_fault_t (*page_mkwrite)(struct vm_fault *vmf); 572 573 /* same as page_mkwrite when using VM_PFNMAP|VM_MIXEDMAP */ 574 vm_fault_t (*pfn_mkwrite)(struct vm_fault *vmf); 575 576 /* called by access_process_vm when get_user_pages() fails, typically 577 * for use by special VMAs. See also generic_access_phys() for a generic 578 * implementation useful for any iomem mapping. 579 */ 580 int (*access)(struct vm_area_struct *vma, unsigned long addr, 581 void *buf, int len, int write); 582 583 /* Called by the /proc/PID/maps code to ask the vma whether it 584 * has a special name. Returning non-NULL will also cause this 585 * vma to be dumped unconditionally. */ 586 const char *(*name)(struct vm_area_struct *vma); 587 588 #ifdef CONFIG_NUMA 589 /* 590 * set_policy() op must add a reference to any non-NULL @new mempolicy 591 * to hold the policy upon return. Caller should pass NULL @new to 592 * remove a policy and fall back to surrounding context--i.e. do not 593 * install a MPOL_DEFAULT policy, nor the task or system default 594 * mempolicy. 595 */ 596 int (*set_policy)(struct vm_area_struct *vma, struct mempolicy *new); 597 598 /* 599 * get_policy() op must add reference [mpol_get()] to any policy at 600 * (vma,addr) marked as MPOL_SHARED. The shared policy infrastructure 601 * in mm/mempolicy.c will do this automatically. 602 * get_policy() must NOT add a ref if the policy at (vma,addr) is not 603 * marked as MPOL_SHARED. vma policies are protected by the mmap_lock. 604 * If no [shared/vma] mempolicy exists at the addr, get_policy() op 605 * must return NULL--i.e., do not "fallback" to task or system default 606 * policy. 607 */ 608 struct mempolicy *(*get_policy)(struct vm_area_struct *vma, 609 unsigned long addr); 610 #endif 611 /* 612 * Called by vm_normal_page() for special PTEs to find the 613 * page for @addr. This is useful if the default behavior 614 * (using pte_page()) would not find the correct page. 615 */ 616 struct page *(*find_special_page)(struct vm_area_struct *vma, 617 unsigned long addr); 618 }; 619 620 #ifdef CONFIG_NUMA_BALANCING 621 static inline void vma_numab_state_init(struct vm_area_struct *vma) 622 { 623 vma->numab_state = NULL; 624 } 625 static inline void vma_numab_state_free(struct vm_area_struct *vma) 626 { 627 kfree(vma->numab_state); 628 } 629 #else 630 static inline void vma_numab_state_init(struct vm_area_struct *vma) {} 631 static inline void vma_numab_state_free(struct vm_area_struct *vma) {} 632 #endif /* CONFIG_NUMA_BALANCING */ 633 634 #ifdef CONFIG_PER_VMA_LOCK 635 /* 636 * Try to read-lock a vma. The function is allowed to occasionally yield false 637 * locked result to avoid performance overhead, in which case we fall back to 638 * using mmap_lock. The function should never yield false unlocked result. 639 */ 640 static inline bool vma_start_read(struct vm_area_struct *vma) 641 { 642 /* Check before locking. A race might cause false locked result. */ 643 if (vma->vm_lock_seq == READ_ONCE(vma->vm_mm->mm_lock_seq)) 644 return false; 645 646 if (unlikely(down_read_trylock(&vma->vm_lock->lock) == 0)) 647 return false; 648 649 /* 650 * Overflow might produce false locked result. 651 * False unlocked result is impossible because we modify and check 652 * vma->vm_lock_seq under vma->vm_lock protection and mm->mm_lock_seq 653 * modification invalidates all existing locks. 654 */ 655 if (unlikely(vma->vm_lock_seq == READ_ONCE(vma->vm_mm->mm_lock_seq))) { 656 up_read(&vma->vm_lock->lock); 657 return false; 658 } 659 return true; 660 } 661 662 static inline void vma_end_read(struct vm_area_struct *vma) 663 { 664 rcu_read_lock(); /* keeps vma alive till the end of up_read */ 665 up_read(&vma->vm_lock->lock); 666 rcu_read_unlock(); 667 } 668 669 static bool __is_vma_write_locked(struct vm_area_struct *vma, int *mm_lock_seq) 670 { 671 mmap_assert_write_locked(vma->vm_mm); 672 673 /* 674 * current task is holding mmap_write_lock, both vma->vm_lock_seq and 675 * mm->mm_lock_seq can't be concurrently modified. 676 */ 677 *mm_lock_seq = READ_ONCE(vma->vm_mm->mm_lock_seq); 678 return (vma->vm_lock_seq == *mm_lock_seq); 679 } 680 681 static inline void vma_start_write(struct vm_area_struct *vma) 682 { 683 int mm_lock_seq; 684 685 if (__is_vma_write_locked(vma, &mm_lock_seq)) 686 return; 687 688 down_write(&vma->vm_lock->lock); 689 vma->vm_lock_seq = mm_lock_seq; 690 up_write(&vma->vm_lock->lock); 691 } 692 693 static inline bool vma_try_start_write(struct vm_area_struct *vma) 694 { 695 int mm_lock_seq; 696 697 if (__is_vma_write_locked(vma, &mm_lock_seq)) 698 return true; 699 700 if (!down_write_trylock(&vma->vm_lock->lock)) 701 return false; 702 703 vma->vm_lock_seq = mm_lock_seq; 704 up_write(&vma->vm_lock->lock); 705 return true; 706 } 707 708 static inline void vma_assert_write_locked(struct vm_area_struct *vma) 709 { 710 int mm_lock_seq; 711 712 VM_BUG_ON_VMA(!__is_vma_write_locked(vma, &mm_lock_seq), vma); 713 } 714 715 static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached) 716 { 717 /* When detaching vma should be write-locked */ 718 if (detached) 719 vma_assert_write_locked(vma); 720 vma->detached = detached; 721 } 722 723 struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm, 724 unsigned long address); 725 726 #else /* CONFIG_PER_VMA_LOCK */ 727 728 static inline void vma_init_lock(struct vm_area_struct *vma) {} 729 static inline bool vma_start_read(struct vm_area_struct *vma) 730 { return false; } 731 static inline void vma_end_read(struct vm_area_struct *vma) {} 732 static inline void vma_start_write(struct vm_area_struct *vma) {} 733 static inline bool vma_try_start_write(struct vm_area_struct *vma) 734 { return true; } 735 static inline void vma_assert_write_locked(struct vm_area_struct *vma) {} 736 static inline void vma_mark_detached(struct vm_area_struct *vma, 737 bool detached) {} 738 739 #endif /* CONFIG_PER_VMA_LOCK */ 740 741 /* 742 * WARNING: vma_init does not initialize vma->vm_lock. 743 * Use vm_area_alloc()/vm_area_free() if vma needs locking. 744 */ 745 static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm) 746 { 747 static const struct vm_operations_struct dummy_vm_ops = {}; 748 749 memset(vma, 0, sizeof(*vma)); 750 vma->vm_mm = mm; 751 vma->vm_ops = &dummy_vm_ops; 752 INIT_LIST_HEAD(&vma->anon_vma_chain); 753 vma_mark_detached(vma, false); 754 vma_numab_state_init(vma); 755 } 756 757 /* Use when VMA is not part of the VMA tree and needs no locking */ 758 static inline void vm_flags_init(struct vm_area_struct *vma, 759 vm_flags_t flags) 760 { 761 ACCESS_PRIVATE(vma, __vm_flags) = flags; 762 } 763 764 /* Use when VMA is part of the VMA tree and modifications need coordination */ 765 static inline void vm_flags_reset(struct vm_area_struct *vma, 766 vm_flags_t flags) 767 { 768 vma_start_write(vma); 769 vm_flags_init(vma, flags); 770 } 771 772 static inline void vm_flags_reset_once(struct vm_area_struct *vma, 773 vm_flags_t flags) 774 { 775 vma_start_write(vma); 776 WRITE_ONCE(ACCESS_PRIVATE(vma, __vm_flags), flags); 777 } 778 779 static inline void vm_flags_set(struct vm_area_struct *vma, 780 vm_flags_t flags) 781 { 782 vma_start_write(vma); 783 ACCESS_PRIVATE(vma, __vm_flags) |= flags; 784 } 785 786 static inline void vm_flags_clear(struct vm_area_struct *vma, 787 vm_flags_t flags) 788 { 789 vma_start_write(vma); 790 ACCESS_PRIVATE(vma, __vm_flags) &= ~flags; 791 } 792 793 /* 794 * Use only if VMA is not part of the VMA tree or has no other users and 795 * therefore needs no locking. 796 */ 797 static inline void __vm_flags_mod(struct vm_area_struct *vma, 798 vm_flags_t set, vm_flags_t clear) 799 { 800 vm_flags_init(vma, (vma->vm_flags | set) & ~clear); 801 } 802 803 /* 804 * Use only when the order of set/clear operations is unimportant, otherwise 805 * use vm_flags_{set|clear} explicitly. 806 */ 807 static inline void vm_flags_mod(struct vm_area_struct *vma, 808 vm_flags_t set, vm_flags_t clear) 809 { 810 vma_start_write(vma); 811 __vm_flags_mod(vma, set, clear); 812 } 813 814 static inline void vma_set_anonymous(struct vm_area_struct *vma) 815 { 816 vma->vm_ops = NULL; 817 } 818 819 static inline bool vma_is_anonymous(struct vm_area_struct *vma) 820 { 821 return !vma->vm_ops; 822 } 823 824 static inline bool vma_is_temporary_stack(struct vm_area_struct *vma) 825 { 826 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); 827 828 if (!maybe_stack) 829 return false; 830 831 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == 832 VM_STACK_INCOMPLETE_SETUP) 833 return true; 834 835 return false; 836 } 837 838 static inline bool vma_is_foreign(struct vm_area_struct *vma) 839 { 840 if (!current->mm) 841 return true; 842 843 if (current->mm != vma->vm_mm) 844 return true; 845 846 return false; 847 } 848 849 static inline bool vma_is_accessible(struct vm_area_struct *vma) 850 { 851 return vma->vm_flags & VM_ACCESS_FLAGS; 852 } 853 854 static inline 855 struct vm_area_struct *vma_find(struct vma_iterator *vmi, unsigned long max) 856 { 857 return mas_find(&vmi->mas, max - 1); 858 } 859 860 static inline struct vm_area_struct *vma_next(struct vma_iterator *vmi) 861 { 862 /* 863 * Uses mas_find() to get the first VMA when the iterator starts. 864 * Calling mas_next() could skip the first entry. 865 */ 866 return mas_find(&vmi->mas, ULONG_MAX); 867 } 868 869 static inline struct vm_area_struct *vma_prev(struct vma_iterator *vmi) 870 { 871 return mas_prev(&vmi->mas, 0); 872 } 873 874 static inline unsigned long vma_iter_addr(struct vma_iterator *vmi) 875 { 876 return vmi->mas.index; 877 } 878 879 static inline unsigned long vma_iter_end(struct vma_iterator *vmi) 880 { 881 return vmi->mas.last + 1; 882 } 883 static inline int vma_iter_bulk_alloc(struct vma_iterator *vmi, 884 unsigned long count) 885 { 886 return mas_expected_entries(&vmi->mas, count); 887 } 888 889 /* Free any unused preallocations */ 890 static inline void vma_iter_free(struct vma_iterator *vmi) 891 { 892 mas_destroy(&vmi->mas); 893 } 894 895 static inline int vma_iter_bulk_store(struct vma_iterator *vmi, 896 struct vm_area_struct *vma) 897 { 898 vmi->mas.index = vma->vm_start; 899 vmi->mas.last = vma->vm_end - 1; 900 mas_store(&vmi->mas, vma); 901 if (unlikely(mas_is_err(&vmi->mas))) 902 return -ENOMEM; 903 904 return 0; 905 } 906 907 static inline void vma_iter_invalidate(struct vma_iterator *vmi) 908 { 909 mas_pause(&vmi->mas); 910 } 911 912 static inline void vma_iter_set(struct vma_iterator *vmi, unsigned long addr) 913 { 914 mas_set(&vmi->mas, addr); 915 } 916 917 #define for_each_vma(__vmi, __vma) \ 918 while (((__vma) = vma_next(&(__vmi))) != NULL) 919 920 /* The MM code likes to work with exclusive end addresses */ 921 #define for_each_vma_range(__vmi, __vma, __end) \ 922 while (((__vma) = vma_find(&(__vmi), (__end))) != NULL) 923 924 #ifdef CONFIG_SHMEM 925 /* 926 * The vma_is_shmem is not inline because it is used only by slow 927 * paths in userfault. 928 */ 929 bool vma_is_shmem(struct vm_area_struct *vma); 930 bool vma_is_anon_shmem(struct vm_area_struct *vma); 931 #else 932 static inline bool vma_is_shmem(struct vm_area_struct *vma) { return false; } 933 static inline bool vma_is_anon_shmem(struct vm_area_struct *vma) { return false; } 934 #endif 935 936 int vma_is_stack_for_current(struct vm_area_struct *vma); 937 938 /* flush_tlb_range() takes a vma, not a mm, and can care about flags */ 939 #define TLB_FLUSH_VMA(mm,flags) { .vm_mm = (mm), .vm_flags = (flags) } 940 941 struct mmu_gather; 942 struct inode; 943 944 /* 945 * compound_order() can be called without holding a reference, which means 946 * that niceties like page_folio() don't work. These callers should be 947 * prepared to handle wild return values. For example, PG_head may be 948 * set before _folio_order is initialised, or this may be a tail page. 949 * See compaction.c for some good examples. 950 */ 951 static inline unsigned int compound_order(struct page *page) 952 { 953 struct folio *folio = (struct folio *)page; 954 955 if (!test_bit(PG_head, &folio->flags)) 956 return 0; 957 return folio->_folio_order; 958 } 959 960 /** 961 * folio_order - The allocation order of a folio. 962 * @folio: The folio. 963 * 964 * A folio is composed of 2^order pages. See get_order() for the definition 965 * of order. 966 * 967 * Return: The order of the folio. 968 */ 969 static inline unsigned int folio_order(struct folio *folio) 970 { 971 if (!folio_test_large(folio)) 972 return 0; 973 return folio->_folio_order; 974 } 975 976 #include <linux/huge_mm.h> 977 978 /* 979 * Methods to modify the page usage count. 980 * 981 * What counts for a page usage: 982 * - cache mapping (page->mapping) 983 * - private data (page->private) 984 * - page mapped in a task's page tables, each mapping 985 * is counted separately 986 * 987 * Also, many kernel routines increase the page count before a critical 988 * routine so they can be sure the page doesn't go away from under them. 989 */ 990 991 /* 992 * Drop a ref, return true if the refcount fell to zero (the page has no users) 993 */ 994 static inline int put_page_testzero(struct page *page) 995 { 996 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); 997 return page_ref_dec_and_test(page); 998 } 999 1000 static inline int folio_put_testzero(struct folio *folio) 1001 { 1002 return put_page_testzero(&folio->page); 1003 } 1004 1005 /* 1006 * Try to grab a ref unless the page has a refcount of zero, return false if 1007 * that is the case. 1008 * This can be called when MMU is off so it must not access 1009 * any of the virtual mappings. 1010 */ 1011 static inline bool get_page_unless_zero(struct page *page) 1012 { 1013 return page_ref_add_unless(page, 1, 0); 1014 } 1015 1016 static inline struct folio *folio_get_nontail_page(struct page *page) 1017 { 1018 if (unlikely(!get_page_unless_zero(page))) 1019 return NULL; 1020 return (struct folio *)page; 1021 } 1022 1023 extern int page_is_ram(unsigned long pfn); 1024 1025 enum { 1026 REGION_INTERSECTS, 1027 REGION_DISJOINT, 1028 REGION_MIXED, 1029 }; 1030 1031 int region_intersects(resource_size_t offset, size_t size, unsigned long flags, 1032 unsigned long desc); 1033 1034 /* Support for virtually mapped pages */ 1035 struct page *vmalloc_to_page(const void *addr); 1036 unsigned long vmalloc_to_pfn(const void *addr); 1037 1038 /* 1039 * Determine if an address is within the vmalloc range 1040 * 1041 * On nommu, vmalloc/vfree wrap through kmalloc/kfree directly, so there 1042 * is no special casing required. 1043 */ 1044 1045 #ifndef is_ioremap_addr 1046 #define is_ioremap_addr(x) is_vmalloc_addr(x) 1047 #endif 1048 1049 #ifdef CONFIG_MMU 1050 extern bool is_vmalloc_addr(const void *x); 1051 extern int is_vmalloc_or_module_addr(const void *x); 1052 #else 1053 static inline bool is_vmalloc_addr(const void *x) 1054 { 1055 return false; 1056 } 1057 static inline int is_vmalloc_or_module_addr(const void *x) 1058 { 1059 return 0; 1060 } 1061 #endif 1062 1063 /* 1064 * How many times the entire folio is mapped as a single unit (eg by a 1065 * PMD or PUD entry). This is probably not what you want, except for 1066 * debugging purposes - it does not include PTE-mapped sub-pages; look 1067 * at folio_mapcount() or page_mapcount() or total_mapcount() instead. 1068 */ 1069 static inline int folio_entire_mapcount(struct folio *folio) 1070 { 1071 VM_BUG_ON_FOLIO(!folio_test_large(folio), folio); 1072 return atomic_read(&folio->_entire_mapcount) + 1; 1073 } 1074 1075 /* 1076 * The atomic page->_mapcount, starts from -1: so that transitions 1077 * both from it and to it can be tracked, using atomic_inc_and_test 1078 * and atomic_add_negative(-1). 1079 */ 1080 static inline void page_mapcount_reset(struct page *page) 1081 { 1082 atomic_set(&(page)->_mapcount, -1); 1083 } 1084 1085 /** 1086 * page_mapcount() - Number of times this precise page is mapped. 1087 * @page: The page. 1088 * 1089 * The number of times this page is mapped. If this page is part of 1090 * a large folio, it includes the number of times this page is mapped 1091 * as part of that folio. 1092 * 1093 * The result is undefined for pages which cannot be mapped into userspace. 1094 * For example SLAB or special types of pages. See function page_has_type(). 1095 * They use this field in struct page differently. 1096 */ 1097 static inline int page_mapcount(struct page *page) 1098 { 1099 int mapcount = atomic_read(&page->_mapcount) + 1; 1100 1101 if (unlikely(PageCompound(page))) 1102 mapcount += folio_entire_mapcount(page_folio(page)); 1103 1104 return mapcount; 1105 } 1106 1107 int folio_total_mapcount(struct folio *folio); 1108 1109 /** 1110 * folio_mapcount() - Calculate the number of mappings of this folio. 1111 * @folio: The folio. 1112 * 1113 * A large folio tracks both how many times the entire folio is mapped, 1114 * and how many times each individual page in the folio is mapped. 1115 * This function calculates the total number of times the folio is 1116 * mapped. 1117 * 1118 * Return: The number of times this folio is mapped. 1119 */ 1120 static inline int folio_mapcount(struct folio *folio) 1121 { 1122 if (likely(!folio_test_large(folio))) 1123 return atomic_read(&folio->_mapcount) + 1; 1124 return folio_total_mapcount(folio); 1125 } 1126 1127 static inline int total_mapcount(struct page *page) 1128 { 1129 if (likely(!PageCompound(page))) 1130 return atomic_read(&page->_mapcount) + 1; 1131 return folio_total_mapcount(page_folio(page)); 1132 } 1133 1134 static inline bool folio_large_is_mapped(struct folio *folio) 1135 { 1136 /* 1137 * Reading _entire_mapcount below could be omitted if hugetlb 1138 * participated in incrementing nr_pages_mapped when compound mapped. 1139 */ 1140 return atomic_read(&folio->_nr_pages_mapped) > 0 || 1141 atomic_read(&folio->_entire_mapcount) >= 0; 1142 } 1143 1144 /** 1145 * folio_mapped - Is this folio mapped into userspace? 1146 * @folio: The folio. 1147 * 1148 * Return: True if any page in this folio is referenced by user page tables. 1149 */ 1150 static inline bool folio_mapped(struct folio *folio) 1151 { 1152 if (likely(!folio_test_large(folio))) 1153 return atomic_read(&folio->_mapcount) >= 0; 1154 return folio_large_is_mapped(folio); 1155 } 1156 1157 /* 1158 * Return true if this page is mapped into pagetables. 1159 * For compound page it returns true if any sub-page of compound page is mapped, 1160 * even if this particular sub-page is not itself mapped by any PTE or PMD. 1161 */ 1162 static inline bool page_mapped(struct page *page) 1163 { 1164 if (likely(!PageCompound(page))) 1165 return atomic_read(&page->_mapcount) >= 0; 1166 return folio_large_is_mapped(page_folio(page)); 1167 } 1168 1169 static inline struct page *virt_to_head_page(const void *x) 1170 { 1171 struct page *page = virt_to_page(x); 1172 1173 return compound_head(page); 1174 } 1175 1176 static inline struct folio *virt_to_folio(const void *x) 1177 { 1178 struct page *page = virt_to_page(x); 1179 1180 return page_folio(page); 1181 } 1182 1183 void __folio_put(struct folio *folio); 1184 1185 void put_pages_list(struct list_head *pages); 1186 1187 void split_page(struct page *page, unsigned int order); 1188 void folio_copy(struct folio *dst, struct folio *src); 1189 1190 unsigned long nr_free_buffer_pages(void); 1191 1192 /* 1193 * Compound pages have a destructor function. Provide a 1194 * prototype for that function and accessor functions. 1195 * These are _only_ valid on the head of a compound page. 1196 */ 1197 typedef void compound_page_dtor(struct page *); 1198 1199 /* Keep the enum in sync with compound_page_dtors array in mm/page_alloc.c */ 1200 enum compound_dtor_id { 1201 NULL_COMPOUND_DTOR, 1202 COMPOUND_PAGE_DTOR, 1203 #ifdef CONFIG_HUGETLB_PAGE 1204 HUGETLB_PAGE_DTOR, 1205 #endif 1206 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1207 TRANSHUGE_PAGE_DTOR, 1208 #endif 1209 NR_COMPOUND_DTORS, 1210 }; 1211 extern compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS]; 1212 1213 static inline void set_compound_page_dtor(struct page *page, 1214 enum compound_dtor_id compound_dtor) 1215 { 1216 struct folio *folio = (struct folio *)page; 1217 1218 VM_BUG_ON_PAGE(compound_dtor >= NR_COMPOUND_DTORS, page); 1219 VM_BUG_ON_PAGE(!PageHead(page), page); 1220 folio->_folio_dtor = compound_dtor; 1221 } 1222 1223 static inline void folio_set_compound_dtor(struct folio *folio, 1224 enum compound_dtor_id compound_dtor) 1225 { 1226 VM_BUG_ON_FOLIO(compound_dtor >= NR_COMPOUND_DTORS, folio); 1227 folio->_folio_dtor = compound_dtor; 1228 } 1229 1230 void destroy_large_folio(struct folio *folio); 1231 1232 static inline void set_compound_order(struct page *page, unsigned int order) 1233 { 1234 struct folio *folio = (struct folio *)page; 1235 1236 folio->_folio_order = order; 1237 #ifdef CONFIG_64BIT 1238 folio->_folio_nr_pages = 1U << order; 1239 #endif 1240 } 1241 1242 /* Returns the number of bytes in this potentially compound page. */ 1243 static inline unsigned long page_size(struct page *page) 1244 { 1245 return PAGE_SIZE << compound_order(page); 1246 } 1247 1248 /* Returns the number of bits needed for the number of bytes in a page */ 1249 static inline unsigned int page_shift(struct page *page) 1250 { 1251 return PAGE_SHIFT + compound_order(page); 1252 } 1253 1254 /** 1255 * thp_order - Order of a transparent huge page. 1256 * @page: Head page of a transparent huge page. 1257 */ 1258 static inline unsigned int thp_order(struct page *page) 1259 { 1260 VM_BUG_ON_PGFLAGS(PageTail(page), page); 1261 return compound_order(page); 1262 } 1263 1264 /** 1265 * thp_size - Size of a transparent huge page. 1266 * @page: Head page of a transparent huge page. 1267 * 1268 * Return: Number of bytes in this page. 1269 */ 1270 static inline unsigned long thp_size(struct page *page) 1271 { 1272 return PAGE_SIZE << thp_order(page); 1273 } 1274 1275 void free_compound_page(struct page *page); 1276 1277 #ifdef CONFIG_MMU 1278 /* 1279 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when 1280 * servicing faults for write access. In the normal case, do always want 1281 * pte_mkwrite. But get_user_pages can cause write faults for mappings 1282 * that do not have writing enabled, when used by access_process_vm. 1283 */ 1284 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) 1285 { 1286 if (likely(vma->vm_flags & VM_WRITE)) 1287 pte = pte_mkwrite(pte); 1288 return pte; 1289 } 1290 1291 vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page); 1292 void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr); 1293 1294 vm_fault_t finish_fault(struct vm_fault *vmf); 1295 vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf); 1296 #endif 1297 1298 /* 1299 * Multiple processes may "see" the same page. E.g. for untouched 1300 * mappings of /dev/null, all processes see the same page full of 1301 * zeroes, and text pages of executables and shared libraries have 1302 * only one copy in memory, at most, normally. 1303 * 1304 * For the non-reserved pages, page_count(page) denotes a reference count. 1305 * page_count() == 0 means the page is free. page->lru is then used for 1306 * freelist management in the buddy allocator. 1307 * page_count() > 0 means the page has been allocated. 1308 * 1309 * Pages are allocated by the slab allocator in order to provide memory 1310 * to kmalloc and kmem_cache_alloc. In this case, the management of the 1311 * page, and the fields in 'struct page' are the responsibility of mm/slab.c 1312 * unless a particular usage is carefully commented. (the responsibility of 1313 * freeing the kmalloc memory is the caller's, of course). 1314 * 1315 * A page may be used by anyone else who does a __get_free_page(). 1316 * In this case, page_count still tracks the references, and should only 1317 * be used through the normal accessor functions. The top bits of page->flags 1318 * and page->virtual store page management information, but all other fields 1319 * are unused and could be used privately, carefully. The management of this 1320 * page is the responsibility of the one who allocated it, and those who have 1321 * subsequently been given references to it. 1322 * 1323 * The other pages (we may call them "pagecache pages") are completely 1324 * managed by the Linux memory manager: I/O, buffers, swapping etc. 1325 * The following discussion applies only to them. 1326 * 1327 * A pagecache page contains an opaque `private' member, which belongs to the 1328 * page's address_space. Usually, this is the address of a circular list of 1329 * the page's disk buffers. PG_private must be set to tell the VM to call 1330 * into the filesystem to release these pages. 1331 * 1332 * A page may belong to an inode's memory mapping. In this case, page->mapping 1333 * is the pointer to the inode, and page->index is the file offset of the page, 1334 * in units of PAGE_SIZE. 1335 * 1336 * If pagecache pages are not associated with an inode, they are said to be 1337 * anonymous pages. These may become associated with the swapcache, and in that 1338 * case PG_swapcache is set, and page->private is an offset into the swapcache. 1339 * 1340 * In either case (swapcache or inode backed), the pagecache itself holds one 1341 * reference to the page. Setting PG_private should also increment the 1342 * refcount. The each user mapping also has a reference to the page. 1343 * 1344 * The pagecache pages are stored in a per-mapping radix tree, which is 1345 * rooted at mapping->i_pages, and indexed by offset. 1346 * Where 2.4 and early 2.6 kernels kept dirty/clean pages in per-address_space 1347 * lists, we instead now tag pages as dirty/writeback in the radix tree. 1348 * 1349 * All pagecache pages may be subject to I/O: 1350 * - inode pages may need to be read from disk, 1351 * - inode pages which have been modified and are MAP_SHARED may need 1352 * to be written back to the inode on disk, 1353 * - anonymous pages (including MAP_PRIVATE file mappings) which have been 1354 * modified may need to be swapped out to swap space and (later) to be read 1355 * back into memory. 1356 */ 1357 1358 #if defined(CONFIG_ZONE_DEVICE) && defined(CONFIG_FS_DAX) 1359 DECLARE_STATIC_KEY_FALSE(devmap_managed_key); 1360 1361 bool __put_devmap_managed_page_refs(struct page *page, int refs); 1362 static inline bool put_devmap_managed_page_refs(struct page *page, int refs) 1363 { 1364 if (!static_branch_unlikely(&devmap_managed_key)) 1365 return false; 1366 if (!is_zone_device_page(page)) 1367 return false; 1368 return __put_devmap_managed_page_refs(page, refs); 1369 } 1370 #else /* CONFIG_ZONE_DEVICE && CONFIG_FS_DAX */ 1371 static inline bool put_devmap_managed_page_refs(struct page *page, int refs) 1372 { 1373 return false; 1374 } 1375 #endif /* CONFIG_ZONE_DEVICE && CONFIG_FS_DAX */ 1376 1377 static inline bool put_devmap_managed_page(struct page *page) 1378 { 1379 return put_devmap_managed_page_refs(page, 1); 1380 } 1381 1382 /* 127: arbitrary random number, small enough to assemble well */ 1383 #define folio_ref_zero_or_close_to_overflow(folio) \ 1384 ((unsigned int) folio_ref_count(folio) + 127u <= 127u) 1385 1386 /** 1387 * folio_get - Increment the reference count on a folio. 1388 * @folio: The folio. 1389 * 1390 * Context: May be called in any context, as long as you know that 1391 * you have a refcount on the folio. If you do not already have one, 1392 * folio_try_get() may be the right interface for you to use. 1393 */ 1394 static inline void folio_get(struct folio *folio) 1395 { 1396 VM_BUG_ON_FOLIO(folio_ref_zero_or_close_to_overflow(folio), folio); 1397 folio_ref_inc(folio); 1398 } 1399 1400 static inline void get_page(struct page *page) 1401 { 1402 folio_get(page_folio(page)); 1403 } 1404 1405 static inline __must_check bool try_get_page(struct page *page) 1406 { 1407 page = compound_head(page); 1408 if (WARN_ON_ONCE(page_ref_count(page) <= 0)) 1409 return false; 1410 page_ref_inc(page); 1411 return true; 1412 } 1413 1414 /** 1415 * folio_put - Decrement the reference count on a folio. 1416 * @folio: The folio. 1417 * 1418 * If the folio's reference count reaches zero, the memory will be 1419 * released back to the page allocator and may be used by another 1420 * allocation immediately. Do not access the memory or the struct folio 1421 * after calling folio_put() unless you can be sure that it wasn't the 1422 * last reference. 1423 * 1424 * Context: May be called in process or interrupt context, but not in NMI 1425 * context. May be called while holding a spinlock. 1426 */ 1427 static inline void folio_put(struct folio *folio) 1428 { 1429 if (folio_put_testzero(folio)) 1430 __folio_put(folio); 1431 } 1432 1433 /** 1434 * folio_put_refs - Reduce the reference count on a folio. 1435 * @folio: The folio. 1436 * @refs: The amount to subtract from the folio's reference count. 1437 * 1438 * If the folio's reference count reaches zero, the memory will be 1439 * released back to the page allocator and may be used by another 1440 * allocation immediately. Do not access the memory or the struct folio 1441 * after calling folio_put_refs() unless you can be sure that these weren't 1442 * the last references. 1443 * 1444 * Context: May be called in process or interrupt context, but not in NMI 1445 * context. May be called while holding a spinlock. 1446 */ 1447 static inline void folio_put_refs(struct folio *folio, int refs) 1448 { 1449 if (folio_ref_sub_and_test(folio, refs)) 1450 __folio_put(folio); 1451 } 1452 1453 /* 1454 * union release_pages_arg - an array of pages or folios 1455 * 1456 * release_pages() releases a simple array of multiple pages, and 1457 * accepts various different forms of said page array: either 1458 * a regular old boring array of pages, an array of folios, or 1459 * an array of encoded page pointers. 1460 * 1461 * The transparent union syntax for this kind of "any of these 1462 * argument types" is all kinds of ugly, so look away. 1463 */ 1464 typedef union { 1465 struct page **pages; 1466 struct folio **folios; 1467 struct encoded_page **encoded_pages; 1468 } release_pages_arg __attribute__ ((__transparent_union__)); 1469 1470 void release_pages(release_pages_arg, int nr); 1471 1472 /** 1473 * folios_put - Decrement the reference count on an array of folios. 1474 * @folios: The folios. 1475 * @nr: How many folios there are. 1476 * 1477 * Like folio_put(), but for an array of folios. This is more efficient 1478 * than writing the loop yourself as it will optimise the locks which 1479 * need to be taken if the folios are freed. 1480 * 1481 * Context: May be called in process or interrupt context, but not in NMI 1482 * context. May be called while holding a spinlock. 1483 */ 1484 static inline void folios_put(struct folio **folios, unsigned int nr) 1485 { 1486 release_pages(folios, nr); 1487 } 1488 1489 static inline void put_page(struct page *page) 1490 { 1491 struct folio *folio = page_folio(page); 1492 1493 /* 1494 * For some devmap managed pages we need to catch refcount transition 1495 * from 2 to 1: 1496 */ 1497 if (put_devmap_managed_page(&folio->page)) 1498 return; 1499 folio_put(folio); 1500 } 1501 1502 /* 1503 * GUP_PIN_COUNTING_BIAS, and the associated functions that use it, overload 1504 * the page's refcount so that two separate items are tracked: the original page 1505 * reference count, and also a new count of how many pin_user_pages() calls were 1506 * made against the page. ("gup-pinned" is another term for the latter). 1507 * 1508 * With this scheme, pin_user_pages() becomes special: such pages are marked as 1509 * distinct from normal pages. As such, the unpin_user_page() call (and its 1510 * variants) must be used in order to release gup-pinned pages. 1511 * 1512 * Choice of value: 1513 * 1514 * By making GUP_PIN_COUNTING_BIAS a power of two, debugging of page reference 1515 * counts with respect to pin_user_pages() and unpin_user_page() becomes 1516 * simpler, due to the fact that adding an even power of two to the page 1517 * refcount has the effect of using only the upper N bits, for the code that 1518 * counts up using the bias value. This means that the lower bits are left for 1519 * the exclusive use of the original code that increments and decrements by one 1520 * (or at least, by much smaller values than the bias value). 1521 * 1522 * Of course, once the lower bits overflow into the upper bits (and this is 1523 * OK, because subtraction recovers the original values), then visual inspection 1524 * no longer suffices to directly view the separate counts. However, for normal 1525 * applications that don't have huge page reference counts, this won't be an 1526 * issue. 1527 * 1528 * Locking: the lockless algorithm described in folio_try_get_rcu() 1529 * provides safe operation for get_user_pages(), page_mkclean() and 1530 * other calls that race to set up page table entries. 1531 */ 1532 #define GUP_PIN_COUNTING_BIAS (1U << 10) 1533 1534 void unpin_user_page(struct page *page); 1535 void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, 1536 bool make_dirty); 1537 void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages, 1538 bool make_dirty); 1539 void unpin_user_pages(struct page **pages, unsigned long npages); 1540 1541 static inline bool is_cow_mapping(vm_flags_t flags) 1542 { 1543 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 1544 } 1545 1546 #ifndef CONFIG_MMU 1547 static inline bool is_nommu_shared_mapping(vm_flags_t flags) 1548 { 1549 /* 1550 * NOMMU shared mappings are ordinary MAP_SHARED mappings and selected 1551 * R/O MAP_PRIVATE file mappings that are an effective R/O overlay of 1552 * a file mapping. R/O MAP_PRIVATE mappings might still modify 1553 * underlying memory if ptrace is active, so this is only possible if 1554 * ptrace does not apply. Note that there is no mprotect() to upgrade 1555 * write permissions later. 1556 */ 1557 return flags & (VM_MAYSHARE | VM_MAYOVERLAY); 1558 } 1559 #endif 1560 1561 #if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP) 1562 #define SECTION_IN_PAGE_FLAGS 1563 #endif 1564 1565 /* 1566 * The identification function is mainly used by the buddy allocator for 1567 * determining if two pages could be buddies. We are not really identifying 1568 * the zone since we could be using the section number id if we do not have 1569 * node id available in page flags. 1570 * We only guarantee that it will return the same value for two combinable 1571 * pages in a zone. 1572 */ 1573 static inline int page_zone_id(struct page *page) 1574 { 1575 return (page->flags >> ZONEID_PGSHIFT) & ZONEID_MASK; 1576 } 1577 1578 #ifdef NODE_NOT_IN_PAGE_FLAGS 1579 extern int page_to_nid(const struct page *page); 1580 #else 1581 static inline int page_to_nid(const struct page *page) 1582 { 1583 struct page *p = (struct page *)page; 1584 1585 return (PF_POISONED_CHECK(p)->flags >> NODES_PGSHIFT) & NODES_MASK; 1586 } 1587 #endif 1588 1589 static inline int folio_nid(const struct folio *folio) 1590 { 1591 return page_to_nid(&folio->page); 1592 } 1593 1594 #ifdef CONFIG_NUMA_BALANCING 1595 /* page access time bits needs to hold at least 4 seconds */ 1596 #define PAGE_ACCESS_TIME_MIN_BITS 12 1597 #if LAST_CPUPID_SHIFT < PAGE_ACCESS_TIME_MIN_BITS 1598 #define PAGE_ACCESS_TIME_BUCKETS \ 1599 (PAGE_ACCESS_TIME_MIN_BITS - LAST_CPUPID_SHIFT) 1600 #else 1601 #define PAGE_ACCESS_TIME_BUCKETS 0 1602 #endif 1603 1604 #define PAGE_ACCESS_TIME_MASK \ 1605 (LAST_CPUPID_MASK << PAGE_ACCESS_TIME_BUCKETS) 1606 1607 static inline int cpu_pid_to_cpupid(int cpu, int pid) 1608 { 1609 return ((cpu & LAST__CPU_MASK) << LAST__PID_SHIFT) | (pid & LAST__PID_MASK); 1610 } 1611 1612 static inline int cpupid_to_pid(int cpupid) 1613 { 1614 return cpupid & LAST__PID_MASK; 1615 } 1616 1617 static inline int cpupid_to_cpu(int cpupid) 1618 { 1619 return (cpupid >> LAST__PID_SHIFT) & LAST__CPU_MASK; 1620 } 1621 1622 static inline int cpupid_to_nid(int cpupid) 1623 { 1624 return cpu_to_node(cpupid_to_cpu(cpupid)); 1625 } 1626 1627 static inline bool cpupid_pid_unset(int cpupid) 1628 { 1629 return cpupid_to_pid(cpupid) == (-1 & LAST__PID_MASK); 1630 } 1631 1632 static inline bool cpupid_cpu_unset(int cpupid) 1633 { 1634 return cpupid_to_cpu(cpupid) == (-1 & LAST__CPU_MASK); 1635 } 1636 1637 static inline bool __cpupid_match_pid(pid_t task_pid, int cpupid) 1638 { 1639 return (task_pid & LAST__PID_MASK) == cpupid_to_pid(cpupid); 1640 } 1641 1642 #define cpupid_match_pid(task, cpupid) __cpupid_match_pid(task->pid, cpupid) 1643 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS 1644 static inline int page_cpupid_xchg_last(struct page *page, int cpupid) 1645 { 1646 return xchg(&page->_last_cpupid, cpupid & LAST_CPUPID_MASK); 1647 } 1648 1649 static inline int page_cpupid_last(struct page *page) 1650 { 1651 return page->_last_cpupid; 1652 } 1653 static inline void page_cpupid_reset_last(struct page *page) 1654 { 1655 page->_last_cpupid = -1 & LAST_CPUPID_MASK; 1656 } 1657 #else 1658 static inline int page_cpupid_last(struct page *page) 1659 { 1660 return (page->flags >> LAST_CPUPID_PGSHIFT) & LAST_CPUPID_MASK; 1661 } 1662 1663 extern int page_cpupid_xchg_last(struct page *page, int cpupid); 1664 1665 static inline void page_cpupid_reset_last(struct page *page) 1666 { 1667 page->flags |= LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT; 1668 } 1669 #endif /* LAST_CPUPID_NOT_IN_PAGE_FLAGS */ 1670 1671 static inline int xchg_page_access_time(struct page *page, int time) 1672 { 1673 int last_time; 1674 1675 last_time = page_cpupid_xchg_last(page, time >> PAGE_ACCESS_TIME_BUCKETS); 1676 return last_time << PAGE_ACCESS_TIME_BUCKETS; 1677 } 1678 1679 static inline void vma_set_access_pid_bit(struct vm_area_struct *vma) 1680 { 1681 unsigned int pid_bit; 1682 1683 pid_bit = hash_32(current->pid, ilog2(BITS_PER_LONG)); 1684 if (vma->numab_state && !test_bit(pid_bit, &vma->numab_state->access_pids[1])) { 1685 __set_bit(pid_bit, &vma->numab_state->access_pids[1]); 1686 } 1687 } 1688 #else /* !CONFIG_NUMA_BALANCING */ 1689 static inline int page_cpupid_xchg_last(struct page *page, int cpupid) 1690 { 1691 return page_to_nid(page); /* XXX */ 1692 } 1693 1694 static inline int xchg_page_access_time(struct page *page, int time) 1695 { 1696 return 0; 1697 } 1698 1699 static inline int page_cpupid_last(struct page *page) 1700 { 1701 return page_to_nid(page); /* XXX */ 1702 } 1703 1704 static inline int cpupid_to_nid(int cpupid) 1705 { 1706 return -1; 1707 } 1708 1709 static inline int cpupid_to_pid(int cpupid) 1710 { 1711 return -1; 1712 } 1713 1714 static inline int cpupid_to_cpu(int cpupid) 1715 { 1716 return -1; 1717 } 1718 1719 static inline int cpu_pid_to_cpupid(int nid, int pid) 1720 { 1721 return -1; 1722 } 1723 1724 static inline bool cpupid_pid_unset(int cpupid) 1725 { 1726 return true; 1727 } 1728 1729 static inline void page_cpupid_reset_last(struct page *page) 1730 { 1731 } 1732 1733 static inline bool cpupid_match_pid(struct task_struct *task, int cpupid) 1734 { 1735 return false; 1736 } 1737 1738 static inline void vma_set_access_pid_bit(struct vm_area_struct *vma) 1739 { 1740 } 1741 #endif /* CONFIG_NUMA_BALANCING */ 1742 1743 #if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS) 1744 1745 /* 1746 * KASAN per-page tags are stored xor'ed with 0xff. This allows to avoid 1747 * setting tags for all pages to native kernel tag value 0xff, as the default 1748 * value 0x00 maps to 0xff. 1749 */ 1750 1751 static inline u8 page_kasan_tag(const struct page *page) 1752 { 1753 u8 tag = 0xff; 1754 1755 if (kasan_enabled()) { 1756 tag = (page->flags >> KASAN_TAG_PGSHIFT) & KASAN_TAG_MASK; 1757 tag ^= 0xff; 1758 } 1759 1760 return tag; 1761 } 1762 1763 static inline void page_kasan_tag_set(struct page *page, u8 tag) 1764 { 1765 unsigned long old_flags, flags; 1766 1767 if (!kasan_enabled()) 1768 return; 1769 1770 tag ^= 0xff; 1771 old_flags = READ_ONCE(page->flags); 1772 do { 1773 flags = old_flags; 1774 flags &= ~(KASAN_TAG_MASK << KASAN_TAG_PGSHIFT); 1775 flags |= (tag & KASAN_TAG_MASK) << KASAN_TAG_PGSHIFT; 1776 } while (unlikely(!try_cmpxchg(&page->flags, &old_flags, flags))); 1777 } 1778 1779 static inline void page_kasan_tag_reset(struct page *page) 1780 { 1781 if (kasan_enabled()) 1782 page_kasan_tag_set(page, 0xff); 1783 } 1784 1785 #else /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */ 1786 1787 static inline u8 page_kasan_tag(const struct page *page) 1788 { 1789 return 0xff; 1790 } 1791 1792 static inline void page_kasan_tag_set(struct page *page, u8 tag) { } 1793 static inline void page_kasan_tag_reset(struct page *page) { } 1794 1795 #endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */ 1796 1797 static inline struct zone *page_zone(const struct page *page) 1798 { 1799 return &NODE_DATA(page_to_nid(page))->node_zones[page_zonenum(page)]; 1800 } 1801 1802 static inline pg_data_t *page_pgdat(const struct page *page) 1803 { 1804 return NODE_DATA(page_to_nid(page)); 1805 } 1806 1807 static inline struct zone *folio_zone(const struct folio *folio) 1808 { 1809 return page_zone(&folio->page); 1810 } 1811 1812 static inline pg_data_t *folio_pgdat(const struct folio *folio) 1813 { 1814 return page_pgdat(&folio->page); 1815 } 1816 1817 #ifdef SECTION_IN_PAGE_FLAGS 1818 static inline void set_page_section(struct page *page, unsigned long section) 1819 { 1820 page->flags &= ~(SECTIONS_MASK << SECTIONS_PGSHIFT); 1821 page->flags |= (section & SECTIONS_MASK) << SECTIONS_PGSHIFT; 1822 } 1823 1824 static inline unsigned long page_to_section(const struct page *page) 1825 { 1826 return (page->flags >> SECTIONS_PGSHIFT) & SECTIONS_MASK; 1827 } 1828 #endif 1829 1830 /** 1831 * folio_pfn - Return the Page Frame Number of a folio. 1832 * @folio: The folio. 1833 * 1834 * A folio may contain multiple pages. The pages have consecutive 1835 * Page Frame Numbers. 1836 * 1837 * Return: The Page Frame Number of the first page in the folio. 1838 */ 1839 static inline unsigned long folio_pfn(struct folio *folio) 1840 { 1841 return page_to_pfn(&folio->page); 1842 } 1843 1844 static inline struct folio *pfn_folio(unsigned long pfn) 1845 { 1846 return page_folio(pfn_to_page(pfn)); 1847 } 1848 1849 /** 1850 * folio_maybe_dma_pinned - Report if a folio may be pinned for DMA. 1851 * @folio: The folio. 1852 * 1853 * This function checks if a folio has been pinned via a call to 1854 * a function in the pin_user_pages() family. 1855 * 1856 * For small folios, the return value is partially fuzzy: false is not fuzzy, 1857 * because it means "definitely not pinned for DMA", but true means "probably 1858 * pinned for DMA, but possibly a false positive due to having at least 1859 * GUP_PIN_COUNTING_BIAS worth of normal folio references". 1860 * 1861 * False positives are OK, because: a) it's unlikely for a folio to 1862 * get that many refcounts, and b) all the callers of this routine are 1863 * expected to be able to deal gracefully with a false positive. 1864 * 1865 * For large folios, the result will be exactly correct. That's because 1866 * we have more tracking data available: the _pincount field is used 1867 * instead of the GUP_PIN_COUNTING_BIAS scheme. 1868 * 1869 * For more information, please see Documentation/core-api/pin_user_pages.rst. 1870 * 1871 * Return: True, if it is likely that the page has been "dma-pinned". 1872 * False, if the page is definitely not dma-pinned. 1873 */ 1874 static inline bool folio_maybe_dma_pinned(struct folio *folio) 1875 { 1876 if (folio_test_large(folio)) 1877 return atomic_read(&folio->_pincount) > 0; 1878 1879 /* 1880 * folio_ref_count() is signed. If that refcount overflows, then 1881 * folio_ref_count() returns a negative value, and callers will avoid 1882 * further incrementing the refcount. 1883 * 1884 * Here, for that overflow case, use the sign bit to count a little 1885 * bit higher via unsigned math, and thus still get an accurate result. 1886 */ 1887 return ((unsigned int)folio_ref_count(folio)) >= 1888 GUP_PIN_COUNTING_BIAS; 1889 } 1890 1891 static inline bool page_maybe_dma_pinned(struct page *page) 1892 { 1893 return folio_maybe_dma_pinned(page_folio(page)); 1894 } 1895 1896 /* 1897 * This should most likely only be called during fork() to see whether we 1898 * should break the cow immediately for an anon page on the src mm. 1899 * 1900 * The caller has to hold the PT lock and the vma->vm_mm->->write_protect_seq. 1901 */ 1902 static inline bool page_needs_cow_for_dma(struct vm_area_struct *vma, 1903 struct page *page) 1904 { 1905 VM_BUG_ON(!(raw_read_seqcount(&vma->vm_mm->write_protect_seq) & 1)); 1906 1907 if (!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)) 1908 return false; 1909 1910 return page_maybe_dma_pinned(page); 1911 } 1912 1913 /* MIGRATE_CMA and ZONE_MOVABLE do not allow pin pages */ 1914 #ifdef CONFIG_MIGRATION 1915 static inline bool is_longterm_pinnable_page(struct page *page) 1916 { 1917 #ifdef CONFIG_CMA 1918 int mt = get_pageblock_migratetype(page); 1919 1920 if (mt == MIGRATE_CMA || mt == MIGRATE_ISOLATE) 1921 return false; 1922 #endif 1923 /* The zero page may always be pinned */ 1924 if (is_zero_pfn(page_to_pfn(page))) 1925 return true; 1926 1927 /* Coherent device memory must always allow eviction. */ 1928 if (is_device_coherent_page(page)) 1929 return false; 1930 1931 /* Otherwise, non-movable zone pages can be pinned. */ 1932 return !is_zone_movable_page(page); 1933 } 1934 #else 1935 static inline bool is_longterm_pinnable_page(struct page *page) 1936 { 1937 return true; 1938 } 1939 #endif 1940 1941 static inline bool folio_is_longterm_pinnable(struct folio *folio) 1942 { 1943 return is_longterm_pinnable_page(&folio->page); 1944 } 1945 1946 static inline void set_page_zone(struct page *page, enum zone_type zone) 1947 { 1948 page->flags &= ~(ZONES_MASK << ZONES_PGSHIFT); 1949 page->flags |= (zone & ZONES_MASK) << ZONES_PGSHIFT; 1950 } 1951 1952 static inline void set_page_node(struct page *page, unsigned long node) 1953 { 1954 page->flags &= ~(NODES_MASK << NODES_PGSHIFT); 1955 page->flags |= (node & NODES_MASK) << NODES_PGSHIFT; 1956 } 1957 1958 static inline void set_page_links(struct page *page, enum zone_type zone, 1959 unsigned long node, unsigned long pfn) 1960 { 1961 set_page_zone(page, zone); 1962 set_page_node(page, node); 1963 #ifdef SECTION_IN_PAGE_FLAGS 1964 set_page_section(page, pfn_to_section_nr(pfn)); 1965 #endif 1966 } 1967 1968 /** 1969 * folio_nr_pages - The number of pages in the folio. 1970 * @folio: The folio. 1971 * 1972 * Return: A positive power of two. 1973 */ 1974 static inline long folio_nr_pages(struct folio *folio) 1975 { 1976 if (!folio_test_large(folio)) 1977 return 1; 1978 #ifdef CONFIG_64BIT 1979 return folio->_folio_nr_pages; 1980 #else 1981 return 1L << folio->_folio_order; 1982 #endif 1983 } 1984 1985 /* 1986 * compound_nr() returns the number of pages in this potentially compound 1987 * page. compound_nr() can be called on a tail page, and is defined to 1988 * return 1 in that case. 1989 */ 1990 static inline unsigned long compound_nr(struct page *page) 1991 { 1992 struct folio *folio = (struct folio *)page; 1993 1994 if (!test_bit(PG_head, &folio->flags)) 1995 return 1; 1996 #ifdef CONFIG_64BIT 1997 return folio->_folio_nr_pages; 1998 #else 1999 return 1L << folio->_folio_order; 2000 #endif 2001 } 2002 2003 /** 2004 * thp_nr_pages - The number of regular pages in this huge page. 2005 * @page: The head page of a huge page. 2006 */ 2007 static inline int thp_nr_pages(struct page *page) 2008 { 2009 return folio_nr_pages((struct folio *)page); 2010 } 2011 2012 /** 2013 * folio_next - Move to the next physical folio. 2014 * @folio: The folio we're currently operating on. 2015 * 2016 * If you have physically contiguous memory which may span more than 2017 * one folio (eg a &struct bio_vec), use this function to move from one 2018 * folio to the next. Do not use it if the memory is only virtually 2019 * contiguous as the folios are almost certainly not adjacent to each 2020 * other. This is the folio equivalent to writing ``page++``. 2021 * 2022 * Context: We assume that the folios are refcounted and/or locked at a 2023 * higher level and do not adjust the reference counts. 2024 * Return: The next struct folio. 2025 */ 2026 static inline struct folio *folio_next(struct folio *folio) 2027 { 2028 return (struct folio *)folio_page(folio, folio_nr_pages(folio)); 2029 } 2030 2031 /** 2032 * folio_shift - The size of the memory described by this folio. 2033 * @folio: The folio. 2034 * 2035 * A folio represents a number of bytes which is a power-of-two in size. 2036 * This function tells you which power-of-two the folio is. See also 2037 * folio_size() and folio_order(). 2038 * 2039 * Context: The caller should have a reference on the folio to prevent 2040 * it from being split. It is not necessary for the folio to be locked. 2041 * Return: The base-2 logarithm of the size of this folio. 2042 */ 2043 static inline unsigned int folio_shift(struct folio *folio) 2044 { 2045 return PAGE_SHIFT + folio_order(folio); 2046 } 2047 2048 /** 2049 * folio_size - The number of bytes in a folio. 2050 * @folio: The folio. 2051 * 2052 * Context: The caller should have a reference on the folio to prevent 2053 * it from being split. It is not necessary for the folio to be locked. 2054 * Return: The number of bytes in this folio. 2055 */ 2056 static inline size_t folio_size(struct folio *folio) 2057 { 2058 return PAGE_SIZE << folio_order(folio); 2059 } 2060 2061 /** 2062 * folio_estimated_sharers - Estimate the number of sharers of a folio. 2063 * @folio: The folio. 2064 * 2065 * folio_estimated_sharers() aims to serve as a function to efficiently 2066 * estimate the number of processes sharing a folio. This is done by 2067 * looking at the precise mapcount of the first subpage in the folio, and 2068 * assuming the other subpages are the same. This may not be true for large 2069 * folios. If you want exact mapcounts for exact calculations, look at 2070 * page_mapcount() or folio_total_mapcount(). 2071 * 2072 * Return: The estimated number of processes sharing a folio. 2073 */ 2074 static inline int folio_estimated_sharers(struct folio *folio) 2075 { 2076 return page_mapcount(folio_page(folio, 0)); 2077 } 2078 2079 #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE 2080 static inline int arch_make_page_accessible(struct page *page) 2081 { 2082 return 0; 2083 } 2084 #endif 2085 2086 #ifndef HAVE_ARCH_MAKE_FOLIO_ACCESSIBLE 2087 static inline int arch_make_folio_accessible(struct folio *folio) 2088 { 2089 int ret; 2090 long i, nr = folio_nr_pages(folio); 2091 2092 for (i = 0; i < nr; i++) { 2093 ret = arch_make_page_accessible(folio_page(folio, i)); 2094 if (ret) 2095 break; 2096 } 2097 2098 return ret; 2099 } 2100 #endif 2101 2102 /* 2103 * Some inline functions in vmstat.h depend on page_zone() 2104 */ 2105 #include <linux/vmstat.h> 2106 2107 static __always_inline void *lowmem_page_address(const struct page *page) 2108 { 2109 return page_to_virt(page); 2110 } 2111 2112 #if defined(CONFIG_HIGHMEM) && !defined(WANT_PAGE_VIRTUAL) 2113 #define HASHED_PAGE_VIRTUAL 2114 #endif 2115 2116 #if defined(WANT_PAGE_VIRTUAL) 2117 static inline void *page_address(const struct page *page) 2118 { 2119 return page->virtual; 2120 } 2121 static inline void set_page_address(struct page *page, void *address) 2122 { 2123 page->virtual = address; 2124 } 2125 #define page_address_init() do { } while(0) 2126 #endif 2127 2128 #if defined(HASHED_PAGE_VIRTUAL) 2129 void *page_address(const struct page *page); 2130 void set_page_address(struct page *page, void *virtual); 2131 void page_address_init(void); 2132 #endif 2133 2134 #if !defined(HASHED_PAGE_VIRTUAL) && !defined(WANT_PAGE_VIRTUAL) 2135 #define page_address(page) lowmem_page_address(page) 2136 #define set_page_address(page, address) do { } while(0) 2137 #define page_address_init() do { } while(0) 2138 #endif 2139 2140 static inline void *folio_address(const struct folio *folio) 2141 { 2142 return page_address(&folio->page); 2143 } 2144 2145 extern void *page_rmapping(struct page *page); 2146 extern pgoff_t __page_file_index(struct page *page); 2147 2148 /* 2149 * Return the pagecache index of the passed page. Regular pagecache pages 2150 * use ->index whereas swapcache pages use swp_offset(->private) 2151 */ 2152 static inline pgoff_t page_index(struct page *page) 2153 { 2154 if (unlikely(PageSwapCache(page))) 2155 return __page_file_index(page); 2156 return page->index; 2157 } 2158 2159 /* 2160 * Return true only if the page has been allocated with 2161 * ALLOC_NO_WATERMARKS and the low watermark was not 2162 * met implying that the system is under some pressure. 2163 */ 2164 static inline bool page_is_pfmemalloc(const struct page *page) 2165 { 2166 /* 2167 * lru.next has bit 1 set if the page is allocated from the 2168 * pfmemalloc reserves. Callers may simply overwrite it if 2169 * they do not need to preserve that information. 2170 */ 2171 return (uintptr_t)page->lru.next & BIT(1); 2172 } 2173 2174 /* 2175 * Return true only if the folio has been allocated with 2176 * ALLOC_NO_WATERMARKS and the low watermark was not 2177 * met implying that the system is under some pressure. 2178 */ 2179 static inline bool folio_is_pfmemalloc(const struct folio *folio) 2180 { 2181 /* 2182 * lru.next has bit 1 set if the page is allocated from the 2183 * pfmemalloc reserves. Callers may simply overwrite it if 2184 * they do not need to preserve that information. 2185 */ 2186 return (uintptr_t)folio->lru.next & BIT(1); 2187 } 2188 2189 /* 2190 * Only to be called by the page allocator on a freshly allocated 2191 * page. 2192 */ 2193 static inline void set_page_pfmemalloc(struct page *page) 2194 { 2195 page->lru.next = (void *)BIT(1); 2196 } 2197 2198 static inline void clear_page_pfmemalloc(struct page *page) 2199 { 2200 page->lru.next = NULL; 2201 } 2202 2203 /* 2204 * Can be called by the pagefault handler when it gets a VM_FAULT_OOM. 2205 */ 2206 extern void pagefault_out_of_memory(void); 2207 2208 #define offset_in_page(p) ((unsigned long)(p) & ~PAGE_MASK) 2209 #define offset_in_thp(page, p) ((unsigned long)(p) & (thp_size(page) - 1)) 2210 #define offset_in_folio(folio, p) ((unsigned long)(p) & (folio_size(folio) - 1)) 2211 2212 /* 2213 * Flags passed to show_mem() and show_free_areas() to suppress output in 2214 * various contexts. 2215 */ 2216 #define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */ 2217 2218 extern void __show_free_areas(unsigned int flags, nodemask_t *nodemask, int max_zone_idx); 2219 static void __maybe_unused show_free_areas(unsigned int flags, nodemask_t *nodemask) 2220 { 2221 __show_free_areas(flags, nodemask, MAX_NR_ZONES - 1); 2222 } 2223 2224 /* 2225 * Parameter block passed down to zap_pte_range in exceptional cases. 2226 */ 2227 struct zap_details { 2228 struct folio *single_folio; /* Locked folio to be unmapped */ 2229 bool even_cows; /* Zap COWed private pages too? */ 2230 zap_flags_t zap_flags; /* Extra flags for zapping */ 2231 }; 2232 2233 /* 2234 * Whether to drop the pte markers, for example, the uffd-wp information for 2235 * file-backed memory. This should only be specified when we will completely 2236 * drop the page in the mm, either by truncation or unmapping of the vma. By 2237 * default, the flag is not set. 2238 */ 2239 #define ZAP_FLAG_DROP_MARKER ((__force zap_flags_t) BIT(0)) 2240 /* Set in unmap_vmas() to indicate a final unmap call. Only used by hugetlb */ 2241 #define ZAP_FLAG_UNMAP ((__force zap_flags_t) BIT(1)) 2242 2243 #ifdef CONFIG_SCHED_MM_CID 2244 void sched_mm_cid_before_execve(struct task_struct *t); 2245 void sched_mm_cid_after_execve(struct task_struct *t); 2246 void sched_mm_cid_fork(struct task_struct *t); 2247 void sched_mm_cid_exit_signals(struct task_struct *t); 2248 static inline int task_mm_cid(struct task_struct *t) 2249 { 2250 return t->mm_cid; 2251 } 2252 #else 2253 static inline void sched_mm_cid_before_execve(struct task_struct *t) { } 2254 static inline void sched_mm_cid_after_execve(struct task_struct *t) { } 2255 static inline void sched_mm_cid_fork(struct task_struct *t) { } 2256 static inline void sched_mm_cid_exit_signals(struct task_struct *t) { } 2257 static inline int task_mm_cid(struct task_struct *t) 2258 { 2259 /* 2260 * Use the processor id as a fall-back when the mm cid feature is 2261 * disabled. This provides functional per-cpu data structure accesses 2262 * in user-space, althrough it won't provide the memory usage benefits. 2263 */ 2264 return raw_smp_processor_id(); 2265 } 2266 #endif 2267 2268 #ifdef CONFIG_MMU 2269 extern bool can_do_mlock(void); 2270 #else 2271 static inline bool can_do_mlock(void) { return false; } 2272 #endif 2273 extern int user_shm_lock(size_t, struct ucounts *); 2274 extern void user_shm_unlock(size_t, struct ucounts *); 2275 2276 struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr, 2277 pte_t pte); 2278 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 2279 pte_t pte); 2280 struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, 2281 pmd_t pmd); 2282 2283 void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 2284 unsigned long size); 2285 void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 2286 unsigned long size, struct zap_details *details); 2287 static inline void zap_vma_pages(struct vm_area_struct *vma) 2288 { 2289 zap_page_range_single(vma, vma->vm_start, 2290 vma->vm_end - vma->vm_start, NULL); 2291 } 2292 void unmap_vmas(struct mmu_gather *tlb, struct maple_tree *mt, 2293 struct vm_area_struct *start_vma, unsigned long start, 2294 unsigned long end, bool mm_wr_locked); 2295 2296 struct mmu_notifier_range; 2297 2298 void free_pgd_range(struct mmu_gather *tlb, unsigned long addr, 2299 unsigned long end, unsigned long floor, unsigned long ceiling); 2300 int 2301 copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma); 2302 int follow_pte(struct mm_struct *mm, unsigned long address, 2303 pte_t **ptepp, spinlock_t **ptlp); 2304 int follow_pfn(struct vm_area_struct *vma, unsigned long address, 2305 unsigned long *pfn); 2306 int follow_phys(struct vm_area_struct *vma, unsigned long address, 2307 unsigned int flags, unsigned long *prot, resource_size_t *phys); 2308 int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 2309 void *buf, int len, int write); 2310 2311 extern void truncate_pagecache(struct inode *inode, loff_t new); 2312 extern void truncate_setsize(struct inode *inode, loff_t newsize); 2313 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to); 2314 void truncate_pagecache_range(struct inode *inode, loff_t offset, loff_t end); 2315 int generic_error_remove_page(struct address_space *mapping, struct page *page); 2316 2317 #ifdef CONFIG_MMU 2318 extern vm_fault_t handle_mm_fault(struct vm_area_struct *vma, 2319 unsigned long address, unsigned int flags, 2320 struct pt_regs *regs); 2321 extern int fixup_user_fault(struct mm_struct *mm, 2322 unsigned long address, unsigned int fault_flags, 2323 bool *unlocked); 2324 void unmap_mapping_pages(struct address_space *mapping, 2325 pgoff_t start, pgoff_t nr, bool even_cows); 2326 void unmap_mapping_range(struct address_space *mapping, 2327 loff_t const holebegin, loff_t const holelen, int even_cows); 2328 #else 2329 static inline vm_fault_t handle_mm_fault(struct vm_area_struct *vma, 2330 unsigned long address, unsigned int flags, 2331 struct pt_regs *regs) 2332 { 2333 /* should never happen if there's no MMU */ 2334 BUG(); 2335 return VM_FAULT_SIGBUS; 2336 } 2337 static inline int fixup_user_fault(struct mm_struct *mm, unsigned long address, 2338 unsigned int fault_flags, bool *unlocked) 2339 { 2340 /* should never happen if there's no MMU */ 2341 BUG(); 2342 return -EFAULT; 2343 } 2344 static inline void unmap_mapping_pages(struct address_space *mapping, 2345 pgoff_t start, pgoff_t nr, bool even_cows) { } 2346 static inline void unmap_mapping_range(struct address_space *mapping, 2347 loff_t const holebegin, loff_t const holelen, int even_cows) { } 2348 #endif 2349 2350 static inline void unmap_shared_mapping_range(struct address_space *mapping, 2351 loff_t const holebegin, loff_t const holelen) 2352 { 2353 unmap_mapping_range(mapping, holebegin, holelen, 0); 2354 } 2355 2356 extern int access_process_vm(struct task_struct *tsk, unsigned long addr, 2357 void *buf, int len, unsigned int gup_flags); 2358 extern int access_remote_vm(struct mm_struct *mm, unsigned long addr, 2359 void *buf, int len, unsigned int gup_flags); 2360 extern int __access_remote_vm(struct mm_struct *mm, unsigned long addr, 2361 void *buf, int len, unsigned int gup_flags); 2362 2363 long get_user_pages_remote(struct mm_struct *mm, 2364 unsigned long start, unsigned long nr_pages, 2365 unsigned int gup_flags, struct page **pages, 2366 struct vm_area_struct **vmas, int *locked); 2367 long pin_user_pages_remote(struct mm_struct *mm, 2368 unsigned long start, unsigned long nr_pages, 2369 unsigned int gup_flags, struct page **pages, 2370 struct vm_area_struct **vmas, int *locked); 2371 long get_user_pages(unsigned long start, unsigned long nr_pages, 2372 unsigned int gup_flags, struct page **pages, 2373 struct vm_area_struct **vmas); 2374 long pin_user_pages(unsigned long start, unsigned long nr_pages, 2375 unsigned int gup_flags, struct page **pages, 2376 struct vm_area_struct **vmas); 2377 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, 2378 struct page **pages, unsigned int gup_flags); 2379 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, 2380 struct page **pages, unsigned int gup_flags); 2381 2382 int get_user_pages_fast(unsigned long start, int nr_pages, 2383 unsigned int gup_flags, struct page **pages); 2384 int pin_user_pages_fast(unsigned long start, int nr_pages, 2385 unsigned int gup_flags, struct page **pages); 2386 2387 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc); 2388 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc, 2389 struct task_struct *task, bool bypass_rlim); 2390 2391 struct kvec; 2392 struct page *get_dump_page(unsigned long addr); 2393 2394 bool folio_mark_dirty(struct folio *folio); 2395 bool set_page_dirty(struct page *page); 2396 int set_page_dirty_lock(struct page *page); 2397 2398 int get_cmdline(struct task_struct *task, char *buffer, int buflen); 2399 2400 extern unsigned long move_page_tables(struct vm_area_struct *vma, 2401 unsigned long old_addr, struct vm_area_struct *new_vma, 2402 unsigned long new_addr, unsigned long len, 2403 bool need_rmap_locks); 2404 2405 /* 2406 * Flags used by change_protection(). For now we make it a bitmap so 2407 * that we can pass in multiple flags just like parameters. However 2408 * for now all the callers are only use one of the flags at the same 2409 * time. 2410 */ 2411 /* 2412 * Whether we should manually check if we can map individual PTEs writable, 2413 * because something (e.g., COW, uffd-wp) blocks that from happening for all 2414 * PTEs automatically in a writable mapping. 2415 */ 2416 #define MM_CP_TRY_CHANGE_WRITABLE (1UL << 0) 2417 /* Whether this protection change is for NUMA hints */ 2418 #define MM_CP_PROT_NUMA (1UL << 1) 2419 /* Whether this change is for write protecting */ 2420 #define MM_CP_UFFD_WP (1UL << 2) /* do wp */ 2421 #define MM_CP_UFFD_WP_RESOLVE (1UL << 3) /* Resolve wp */ 2422 #define MM_CP_UFFD_WP_ALL (MM_CP_UFFD_WP | \ 2423 MM_CP_UFFD_WP_RESOLVE) 2424 2425 int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot); 2426 static inline bool vma_wants_manual_pte_write_upgrade(struct vm_area_struct *vma) 2427 { 2428 /* 2429 * We want to check manually if we can change individual PTEs writable 2430 * if we can't do that automatically for all PTEs in a mapping. For 2431 * private mappings, that's always the case when we have write 2432 * permissions as we properly have to handle COW. 2433 */ 2434 if (vma->vm_flags & VM_SHARED) 2435 return vma_wants_writenotify(vma, vma->vm_page_prot); 2436 return !!(vma->vm_flags & VM_WRITE); 2437 2438 } 2439 bool can_change_pte_writable(struct vm_area_struct *vma, unsigned long addr, 2440 pte_t pte); 2441 extern long change_protection(struct mmu_gather *tlb, 2442 struct vm_area_struct *vma, unsigned long start, 2443 unsigned long end, unsigned long cp_flags); 2444 extern int mprotect_fixup(struct vma_iterator *vmi, struct mmu_gather *tlb, 2445 struct vm_area_struct *vma, struct vm_area_struct **pprev, 2446 unsigned long start, unsigned long end, unsigned long newflags); 2447 2448 /* 2449 * doesn't attempt to fault and will return short. 2450 */ 2451 int get_user_pages_fast_only(unsigned long start, int nr_pages, 2452 unsigned int gup_flags, struct page **pages); 2453 2454 static inline bool get_user_page_fast_only(unsigned long addr, 2455 unsigned int gup_flags, struct page **pagep) 2456 { 2457 return get_user_pages_fast_only(addr, 1, gup_flags, pagep) == 1; 2458 } 2459 /* 2460 * per-process(per-mm_struct) statistics. 2461 */ 2462 static inline unsigned long get_mm_counter(struct mm_struct *mm, int member) 2463 { 2464 return percpu_counter_read_positive(&mm->rss_stat[member]); 2465 } 2466 2467 void mm_trace_rss_stat(struct mm_struct *mm, int member); 2468 2469 static inline void add_mm_counter(struct mm_struct *mm, int member, long value) 2470 { 2471 percpu_counter_add(&mm->rss_stat[member], value); 2472 2473 mm_trace_rss_stat(mm, member); 2474 } 2475 2476 static inline void inc_mm_counter(struct mm_struct *mm, int member) 2477 { 2478 percpu_counter_inc(&mm->rss_stat[member]); 2479 2480 mm_trace_rss_stat(mm, member); 2481 } 2482 2483 static inline void dec_mm_counter(struct mm_struct *mm, int member) 2484 { 2485 percpu_counter_dec(&mm->rss_stat[member]); 2486 2487 mm_trace_rss_stat(mm, member); 2488 } 2489 2490 /* Optimized variant when page is already known not to be PageAnon */ 2491 static inline int mm_counter_file(struct page *page) 2492 { 2493 if (PageSwapBacked(page)) 2494 return MM_SHMEMPAGES; 2495 return MM_FILEPAGES; 2496 } 2497 2498 static inline int mm_counter(struct page *page) 2499 { 2500 if (PageAnon(page)) 2501 return MM_ANONPAGES; 2502 return mm_counter_file(page); 2503 } 2504 2505 static inline unsigned long get_mm_rss(struct mm_struct *mm) 2506 { 2507 return get_mm_counter(mm, MM_FILEPAGES) + 2508 get_mm_counter(mm, MM_ANONPAGES) + 2509 get_mm_counter(mm, MM_SHMEMPAGES); 2510 } 2511 2512 static inline unsigned long get_mm_hiwater_rss(struct mm_struct *mm) 2513 { 2514 return max(mm->hiwater_rss, get_mm_rss(mm)); 2515 } 2516 2517 static inline unsigned long get_mm_hiwater_vm(struct mm_struct *mm) 2518 { 2519 return max(mm->hiwater_vm, mm->total_vm); 2520 } 2521 2522 static inline void update_hiwater_rss(struct mm_struct *mm) 2523 { 2524 unsigned long _rss = get_mm_rss(mm); 2525 2526 if ((mm)->hiwater_rss < _rss) 2527 (mm)->hiwater_rss = _rss; 2528 } 2529 2530 static inline void update_hiwater_vm(struct mm_struct *mm) 2531 { 2532 if (mm->hiwater_vm < mm->total_vm) 2533 mm->hiwater_vm = mm->total_vm; 2534 } 2535 2536 static inline void reset_mm_hiwater_rss(struct mm_struct *mm) 2537 { 2538 mm->hiwater_rss = get_mm_rss(mm); 2539 } 2540 2541 static inline void setmax_mm_hiwater_rss(unsigned long *maxrss, 2542 struct mm_struct *mm) 2543 { 2544 unsigned long hiwater_rss = get_mm_hiwater_rss(mm); 2545 2546 if (*maxrss < hiwater_rss) 2547 *maxrss = hiwater_rss; 2548 } 2549 2550 #if defined(SPLIT_RSS_COUNTING) 2551 void sync_mm_rss(struct mm_struct *mm); 2552 #else 2553 static inline void sync_mm_rss(struct mm_struct *mm) 2554 { 2555 } 2556 #endif 2557 2558 #ifndef CONFIG_ARCH_HAS_PTE_SPECIAL 2559 static inline int pte_special(pte_t pte) 2560 { 2561 return 0; 2562 } 2563 2564 static inline pte_t pte_mkspecial(pte_t pte) 2565 { 2566 return pte; 2567 } 2568 #endif 2569 2570 #ifndef CONFIG_ARCH_HAS_PTE_DEVMAP 2571 static inline int pte_devmap(pte_t pte) 2572 { 2573 return 0; 2574 } 2575 #endif 2576 2577 extern pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 2578 spinlock_t **ptl); 2579 static inline pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr, 2580 spinlock_t **ptl) 2581 { 2582 pte_t *ptep; 2583 __cond_lock(*ptl, ptep = __get_locked_pte(mm, addr, ptl)); 2584 return ptep; 2585 } 2586 2587 #ifdef __PAGETABLE_P4D_FOLDED 2588 static inline int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, 2589 unsigned long address) 2590 { 2591 return 0; 2592 } 2593 #else 2594 int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address); 2595 #endif 2596 2597 #if defined(__PAGETABLE_PUD_FOLDED) || !defined(CONFIG_MMU) 2598 static inline int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, 2599 unsigned long address) 2600 { 2601 return 0; 2602 } 2603 static inline void mm_inc_nr_puds(struct mm_struct *mm) {} 2604 static inline void mm_dec_nr_puds(struct mm_struct *mm) {} 2605 2606 #else 2607 int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address); 2608 2609 static inline void mm_inc_nr_puds(struct mm_struct *mm) 2610 { 2611 if (mm_pud_folded(mm)) 2612 return; 2613 atomic_long_add(PTRS_PER_PUD * sizeof(pud_t), &mm->pgtables_bytes); 2614 } 2615 2616 static inline void mm_dec_nr_puds(struct mm_struct *mm) 2617 { 2618 if (mm_pud_folded(mm)) 2619 return; 2620 atomic_long_sub(PTRS_PER_PUD * sizeof(pud_t), &mm->pgtables_bytes); 2621 } 2622 #endif 2623 2624 #if defined(__PAGETABLE_PMD_FOLDED) || !defined(CONFIG_MMU) 2625 static inline int __pmd_alloc(struct mm_struct *mm, pud_t *pud, 2626 unsigned long address) 2627 { 2628 return 0; 2629 } 2630 2631 static inline void mm_inc_nr_pmds(struct mm_struct *mm) {} 2632 static inline void mm_dec_nr_pmds(struct mm_struct *mm) {} 2633 2634 #else 2635 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address); 2636 2637 static inline void mm_inc_nr_pmds(struct mm_struct *mm) 2638 { 2639 if (mm_pmd_folded(mm)) 2640 return; 2641 atomic_long_add(PTRS_PER_PMD * sizeof(pmd_t), &mm->pgtables_bytes); 2642 } 2643 2644 static inline void mm_dec_nr_pmds(struct mm_struct *mm) 2645 { 2646 if (mm_pmd_folded(mm)) 2647 return; 2648 atomic_long_sub(PTRS_PER_PMD * sizeof(pmd_t), &mm->pgtables_bytes); 2649 } 2650 #endif 2651 2652 #ifdef CONFIG_MMU 2653 static inline void mm_pgtables_bytes_init(struct mm_struct *mm) 2654 { 2655 atomic_long_set(&mm->pgtables_bytes, 0); 2656 } 2657 2658 static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm) 2659 { 2660 return atomic_long_read(&mm->pgtables_bytes); 2661 } 2662 2663 static inline void mm_inc_nr_ptes(struct mm_struct *mm) 2664 { 2665 atomic_long_add(PTRS_PER_PTE * sizeof(pte_t), &mm->pgtables_bytes); 2666 } 2667 2668 static inline void mm_dec_nr_ptes(struct mm_struct *mm) 2669 { 2670 atomic_long_sub(PTRS_PER_PTE * sizeof(pte_t), &mm->pgtables_bytes); 2671 } 2672 #else 2673 2674 static inline void mm_pgtables_bytes_init(struct mm_struct *mm) {} 2675 static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm) 2676 { 2677 return 0; 2678 } 2679 2680 static inline void mm_inc_nr_ptes(struct mm_struct *mm) {} 2681 static inline void mm_dec_nr_ptes(struct mm_struct *mm) {} 2682 #endif 2683 2684 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd); 2685 int __pte_alloc_kernel(pmd_t *pmd); 2686 2687 #if defined(CONFIG_MMU) 2688 2689 static inline p4d_t *p4d_alloc(struct mm_struct *mm, pgd_t *pgd, 2690 unsigned long address) 2691 { 2692 return (unlikely(pgd_none(*pgd)) && __p4d_alloc(mm, pgd, address)) ? 2693 NULL : p4d_offset(pgd, address); 2694 } 2695 2696 static inline pud_t *pud_alloc(struct mm_struct *mm, p4d_t *p4d, 2697 unsigned long address) 2698 { 2699 return (unlikely(p4d_none(*p4d)) && __pud_alloc(mm, p4d, address)) ? 2700 NULL : pud_offset(p4d, address); 2701 } 2702 2703 static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 2704 { 2705 return (unlikely(pud_none(*pud)) && __pmd_alloc(mm, pud, address))? 2706 NULL: pmd_offset(pud, address); 2707 } 2708 #endif /* CONFIG_MMU */ 2709 2710 #if USE_SPLIT_PTE_PTLOCKS 2711 #if ALLOC_SPLIT_PTLOCKS 2712 void __init ptlock_cache_init(void); 2713 extern bool ptlock_alloc(struct page *page); 2714 extern void ptlock_free(struct page *page); 2715 2716 static inline spinlock_t *ptlock_ptr(struct page *page) 2717 { 2718 return page->ptl; 2719 } 2720 #else /* ALLOC_SPLIT_PTLOCKS */ 2721 static inline void ptlock_cache_init(void) 2722 { 2723 } 2724 2725 static inline bool ptlock_alloc(struct page *page) 2726 { 2727 return true; 2728 } 2729 2730 static inline void ptlock_free(struct page *page) 2731 { 2732 } 2733 2734 static inline spinlock_t *ptlock_ptr(struct page *page) 2735 { 2736 return &page->ptl; 2737 } 2738 #endif /* ALLOC_SPLIT_PTLOCKS */ 2739 2740 static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd) 2741 { 2742 return ptlock_ptr(pmd_page(*pmd)); 2743 } 2744 2745 static inline bool ptlock_init(struct page *page) 2746 { 2747 /* 2748 * prep_new_page() initialize page->private (and therefore page->ptl) 2749 * with 0. Make sure nobody took it in use in between. 2750 * 2751 * It can happen if arch try to use slab for page table allocation: 2752 * slab code uses page->slab_cache, which share storage with page->ptl. 2753 */ 2754 VM_BUG_ON_PAGE(*(unsigned long *)&page->ptl, page); 2755 if (!ptlock_alloc(page)) 2756 return false; 2757 spin_lock_init(ptlock_ptr(page)); 2758 return true; 2759 } 2760 2761 #else /* !USE_SPLIT_PTE_PTLOCKS */ 2762 /* 2763 * We use mm->page_table_lock to guard all pagetable pages of the mm. 2764 */ 2765 static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd) 2766 { 2767 return &mm->page_table_lock; 2768 } 2769 static inline void ptlock_cache_init(void) {} 2770 static inline bool ptlock_init(struct page *page) { return true; } 2771 static inline void ptlock_free(struct page *page) {} 2772 #endif /* USE_SPLIT_PTE_PTLOCKS */ 2773 2774 static inline bool pgtable_pte_page_ctor(struct page *page) 2775 { 2776 if (!ptlock_init(page)) 2777 return false; 2778 __SetPageTable(page); 2779 inc_lruvec_page_state(page, NR_PAGETABLE); 2780 return true; 2781 } 2782 2783 static inline void pgtable_pte_page_dtor(struct page *page) 2784 { 2785 ptlock_free(page); 2786 __ClearPageTable(page); 2787 dec_lruvec_page_state(page, NR_PAGETABLE); 2788 } 2789 2790 #define pte_offset_map_lock(mm, pmd, address, ptlp) \ 2791 ({ \ 2792 spinlock_t *__ptl = pte_lockptr(mm, pmd); \ 2793 pte_t *__pte = pte_offset_map(pmd, address); \ 2794 *(ptlp) = __ptl; \ 2795 spin_lock(__ptl); \ 2796 __pte; \ 2797 }) 2798 2799 #define pte_unmap_unlock(pte, ptl) do { \ 2800 spin_unlock(ptl); \ 2801 pte_unmap(pte); \ 2802 } while (0) 2803 2804 #define pte_alloc(mm, pmd) (unlikely(pmd_none(*(pmd))) && __pte_alloc(mm, pmd)) 2805 2806 #define pte_alloc_map(mm, pmd, address) \ 2807 (pte_alloc(mm, pmd) ? NULL : pte_offset_map(pmd, address)) 2808 2809 #define pte_alloc_map_lock(mm, pmd, address, ptlp) \ 2810 (pte_alloc(mm, pmd) ? \ 2811 NULL : pte_offset_map_lock(mm, pmd, address, ptlp)) 2812 2813 #define pte_alloc_kernel(pmd, address) \ 2814 ((unlikely(pmd_none(*(pmd))) && __pte_alloc_kernel(pmd))? \ 2815 NULL: pte_offset_kernel(pmd, address)) 2816 2817 #if USE_SPLIT_PMD_PTLOCKS 2818 2819 static inline struct page *pmd_pgtable_page(pmd_t *pmd) 2820 { 2821 unsigned long mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1); 2822 return virt_to_page((void *)((unsigned long) pmd & mask)); 2823 } 2824 2825 static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd) 2826 { 2827 return ptlock_ptr(pmd_pgtable_page(pmd)); 2828 } 2829 2830 static inline bool pmd_ptlock_init(struct page *page) 2831 { 2832 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2833 page->pmd_huge_pte = NULL; 2834 #endif 2835 return ptlock_init(page); 2836 } 2837 2838 static inline void pmd_ptlock_free(struct page *page) 2839 { 2840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2841 VM_BUG_ON_PAGE(page->pmd_huge_pte, page); 2842 #endif 2843 ptlock_free(page); 2844 } 2845 2846 #define pmd_huge_pte(mm, pmd) (pmd_pgtable_page(pmd)->pmd_huge_pte) 2847 2848 #else 2849 2850 static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd) 2851 { 2852 return &mm->page_table_lock; 2853 } 2854 2855 static inline bool pmd_ptlock_init(struct page *page) { return true; } 2856 static inline void pmd_ptlock_free(struct page *page) {} 2857 2858 #define pmd_huge_pte(mm, pmd) ((mm)->pmd_huge_pte) 2859 2860 #endif 2861 2862 static inline spinlock_t *pmd_lock(struct mm_struct *mm, pmd_t *pmd) 2863 { 2864 spinlock_t *ptl = pmd_lockptr(mm, pmd); 2865 spin_lock(ptl); 2866 return ptl; 2867 } 2868 2869 static inline bool pgtable_pmd_page_ctor(struct page *page) 2870 { 2871 if (!pmd_ptlock_init(page)) 2872 return false; 2873 __SetPageTable(page); 2874 inc_lruvec_page_state(page, NR_PAGETABLE); 2875 return true; 2876 } 2877 2878 static inline void pgtable_pmd_page_dtor(struct page *page) 2879 { 2880 pmd_ptlock_free(page); 2881 __ClearPageTable(page); 2882 dec_lruvec_page_state(page, NR_PAGETABLE); 2883 } 2884 2885 /* 2886 * No scalability reason to split PUD locks yet, but follow the same pattern 2887 * as the PMD locks to make it easier if we decide to. The VM should not be 2888 * considered ready to switch to split PUD locks yet; there may be places 2889 * which need to be converted from page_table_lock. 2890 */ 2891 static inline spinlock_t *pud_lockptr(struct mm_struct *mm, pud_t *pud) 2892 { 2893 return &mm->page_table_lock; 2894 } 2895 2896 static inline spinlock_t *pud_lock(struct mm_struct *mm, pud_t *pud) 2897 { 2898 spinlock_t *ptl = pud_lockptr(mm, pud); 2899 2900 spin_lock(ptl); 2901 return ptl; 2902 } 2903 2904 extern void __init pagecache_init(void); 2905 extern void free_initmem(void); 2906 2907 /* 2908 * Free reserved pages within range [PAGE_ALIGN(start), end & PAGE_MASK) 2909 * into the buddy system. The freed pages will be poisoned with pattern 2910 * "poison" if it's within range [0, UCHAR_MAX]. 2911 * Return pages freed into the buddy system. 2912 */ 2913 extern unsigned long free_reserved_area(void *start, void *end, 2914 int poison, const char *s); 2915 2916 extern void adjust_managed_page_count(struct page *page, long count); 2917 2918 extern void reserve_bootmem_region(phys_addr_t start, phys_addr_t end); 2919 2920 /* Free the reserved page into the buddy system, so it gets managed. */ 2921 static inline void free_reserved_page(struct page *page) 2922 { 2923 ClearPageReserved(page); 2924 init_page_count(page); 2925 __free_page(page); 2926 adjust_managed_page_count(page, 1); 2927 } 2928 #define free_highmem_page(page) free_reserved_page(page) 2929 2930 static inline void mark_page_reserved(struct page *page) 2931 { 2932 SetPageReserved(page); 2933 adjust_managed_page_count(page, -1); 2934 } 2935 2936 /* 2937 * Default method to free all the __init memory into the buddy system. 2938 * The freed pages will be poisoned with pattern "poison" if it's within 2939 * range [0, UCHAR_MAX]. 2940 * Return pages freed into the buddy system. 2941 */ 2942 static inline unsigned long free_initmem_default(int poison) 2943 { 2944 extern char __init_begin[], __init_end[]; 2945 2946 return free_reserved_area(&__init_begin, &__init_end, 2947 poison, "unused kernel image (initmem)"); 2948 } 2949 2950 static inline unsigned long get_num_physpages(void) 2951 { 2952 int nid; 2953 unsigned long phys_pages = 0; 2954 2955 for_each_online_node(nid) 2956 phys_pages += node_present_pages(nid); 2957 2958 return phys_pages; 2959 } 2960 2961 /* 2962 * Using memblock node mappings, an architecture may initialise its 2963 * zones, allocate the backing mem_map and account for memory holes in an 2964 * architecture independent manner. 2965 * 2966 * An architecture is expected to register range of page frames backed by 2967 * physical memory with memblock_add[_node]() before calling 2968 * free_area_init() passing in the PFN each zone ends at. At a basic 2969 * usage, an architecture is expected to do something like 2970 * 2971 * unsigned long max_zone_pfns[MAX_NR_ZONES] = {max_dma, max_normal_pfn, 2972 * max_highmem_pfn}; 2973 * for_each_valid_physical_page_range() 2974 * memblock_add_node(base, size, nid, MEMBLOCK_NONE) 2975 * free_area_init(max_zone_pfns); 2976 */ 2977 void free_area_init(unsigned long *max_zone_pfn); 2978 unsigned long node_map_pfn_alignment(void); 2979 unsigned long __absent_pages_in_range(int nid, unsigned long start_pfn, 2980 unsigned long end_pfn); 2981 extern unsigned long absent_pages_in_range(unsigned long start_pfn, 2982 unsigned long end_pfn); 2983 extern void get_pfn_range_for_nid(unsigned int nid, 2984 unsigned long *start_pfn, unsigned long *end_pfn); 2985 2986 #ifndef CONFIG_NUMA 2987 static inline int early_pfn_to_nid(unsigned long pfn) 2988 { 2989 return 0; 2990 } 2991 #else 2992 /* please see mm/page_alloc.c */ 2993 extern int __meminit early_pfn_to_nid(unsigned long pfn); 2994 #endif 2995 2996 extern void set_dma_reserve(unsigned long new_dma_reserve); 2997 extern void memmap_init_range(unsigned long, int, unsigned long, 2998 unsigned long, unsigned long, enum meminit_context, 2999 struct vmem_altmap *, int migratetype); 3000 extern void setup_per_zone_wmarks(void); 3001 extern void calculate_min_free_kbytes(void); 3002 extern int __meminit init_per_zone_wmark_min(void); 3003 extern void mem_init(void); 3004 extern void __init mmap_init(void); 3005 3006 extern void __show_mem(unsigned int flags, nodemask_t *nodemask, int max_zone_idx); 3007 static inline void show_mem(unsigned int flags, nodemask_t *nodemask) 3008 { 3009 __show_mem(flags, nodemask, MAX_NR_ZONES - 1); 3010 } 3011 extern long si_mem_available(void); 3012 extern void si_meminfo(struct sysinfo * val); 3013 extern void si_meminfo_node(struct sysinfo *val, int nid); 3014 #ifdef __HAVE_ARCH_RESERVED_KERNEL_PAGES 3015 extern unsigned long arch_reserved_kernel_pages(void); 3016 #endif 3017 3018 extern __printf(3, 4) 3019 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...); 3020 3021 extern void setup_per_cpu_pageset(void); 3022 3023 /* page_alloc.c */ 3024 extern int min_free_kbytes; 3025 extern int watermark_boost_factor; 3026 extern int watermark_scale_factor; 3027 3028 /* nommu.c */ 3029 extern atomic_long_t mmap_pages_allocated; 3030 extern int nommu_shrink_inode_mappings(struct inode *, size_t, size_t); 3031 3032 /* interval_tree.c */ 3033 void vma_interval_tree_insert(struct vm_area_struct *node, 3034 struct rb_root_cached *root); 3035 void vma_interval_tree_insert_after(struct vm_area_struct *node, 3036 struct vm_area_struct *prev, 3037 struct rb_root_cached *root); 3038 void vma_interval_tree_remove(struct vm_area_struct *node, 3039 struct rb_root_cached *root); 3040 struct vm_area_struct *vma_interval_tree_iter_first(struct rb_root_cached *root, 3041 unsigned long start, unsigned long last); 3042 struct vm_area_struct *vma_interval_tree_iter_next(struct vm_area_struct *node, 3043 unsigned long start, unsigned long last); 3044 3045 #define vma_interval_tree_foreach(vma, root, start, last) \ 3046 for (vma = vma_interval_tree_iter_first(root, start, last); \ 3047 vma; vma = vma_interval_tree_iter_next(vma, start, last)) 3048 3049 void anon_vma_interval_tree_insert(struct anon_vma_chain *node, 3050 struct rb_root_cached *root); 3051 void anon_vma_interval_tree_remove(struct anon_vma_chain *node, 3052 struct rb_root_cached *root); 3053 struct anon_vma_chain * 3054 anon_vma_interval_tree_iter_first(struct rb_root_cached *root, 3055 unsigned long start, unsigned long last); 3056 struct anon_vma_chain *anon_vma_interval_tree_iter_next( 3057 struct anon_vma_chain *node, unsigned long start, unsigned long last); 3058 #ifdef CONFIG_DEBUG_VM_RB 3059 void anon_vma_interval_tree_verify(struct anon_vma_chain *node); 3060 #endif 3061 3062 #define anon_vma_interval_tree_foreach(avc, root, start, last) \ 3063 for (avc = anon_vma_interval_tree_iter_first(root, start, last); \ 3064 avc; avc = anon_vma_interval_tree_iter_next(avc, start, last)) 3065 3066 /* mmap.c */ 3067 extern int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin); 3068 extern int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, 3069 unsigned long start, unsigned long end, pgoff_t pgoff, 3070 struct vm_area_struct *next); 3071 extern int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, 3072 unsigned long start, unsigned long end, pgoff_t pgoff); 3073 extern struct vm_area_struct *vma_merge(struct vma_iterator *vmi, 3074 struct mm_struct *, struct vm_area_struct *prev, unsigned long addr, 3075 unsigned long end, unsigned long vm_flags, struct anon_vma *, 3076 struct file *, pgoff_t, struct mempolicy *, struct vm_userfaultfd_ctx, 3077 struct anon_vma_name *); 3078 extern struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *); 3079 extern int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *, 3080 unsigned long addr, int new_below); 3081 extern int split_vma(struct vma_iterator *vmi, struct vm_area_struct *, 3082 unsigned long addr, int new_below); 3083 extern int insert_vm_struct(struct mm_struct *, struct vm_area_struct *); 3084 extern void unlink_file_vma(struct vm_area_struct *); 3085 extern struct vm_area_struct *copy_vma(struct vm_area_struct **, 3086 unsigned long addr, unsigned long len, pgoff_t pgoff, 3087 bool *need_rmap_locks); 3088 extern void exit_mmap(struct mm_struct *); 3089 3090 static inline int check_data_rlimit(unsigned long rlim, 3091 unsigned long new, 3092 unsigned long start, 3093 unsigned long end_data, 3094 unsigned long start_data) 3095 { 3096 if (rlim < RLIM_INFINITY) { 3097 if (((new - start) + (end_data - start_data)) > rlim) 3098 return -ENOSPC; 3099 } 3100 3101 return 0; 3102 } 3103 3104 extern int mm_take_all_locks(struct mm_struct *mm); 3105 extern void mm_drop_all_locks(struct mm_struct *mm); 3106 3107 extern int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file); 3108 extern int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file); 3109 extern struct file *get_mm_exe_file(struct mm_struct *mm); 3110 extern struct file *get_task_exe_file(struct task_struct *task); 3111 3112 extern bool may_expand_vm(struct mm_struct *, vm_flags_t, unsigned long npages); 3113 extern void vm_stat_account(struct mm_struct *, vm_flags_t, long npages); 3114 3115 extern bool vma_is_special_mapping(const struct vm_area_struct *vma, 3116 const struct vm_special_mapping *sm); 3117 extern struct vm_area_struct *_install_special_mapping(struct mm_struct *mm, 3118 unsigned long addr, unsigned long len, 3119 unsigned long flags, 3120 const struct vm_special_mapping *spec); 3121 /* This is an obsolete alternative to _install_special_mapping. */ 3122 extern int install_special_mapping(struct mm_struct *mm, 3123 unsigned long addr, unsigned long len, 3124 unsigned long flags, struct page **pages); 3125 3126 unsigned long randomize_stack_top(unsigned long stack_top); 3127 unsigned long randomize_page(unsigned long start, unsigned long range); 3128 3129 extern unsigned long get_unmapped_area(struct file *, unsigned long, unsigned long, unsigned long, unsigned long); 3130 3131 extern unsigned long mmap_region(struct file *file, unsigned long addr, 3132 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 3133 struct list_head *uf); 3134 extern unsigned long do_mmap(struct file *file, unsigned long addr, 3135 unsigned long len, unsigned long prot, unsigned long flags, 3136 unsigned long pgoff, unsigned long *populate, struct list_head *uf); 3137 extern int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 3138 unsigned long start, size_t len, struct list_head *uf, 3139 bool downgrade); 3140 extern int do_munmap(struct mm_struct *, unsigned long, size_t, 3141 struct list_head *uf); 3142 extern int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behavior); 3143 3144 #ifdef CONFIG_MMU 3145 extern int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 3146 unsigned long start, unsigned long end, 3147 struct list_head *uf, bool downgrade); 3148 extern int __mm_populate(unsigned long addr, unsigned long len, 3149 int ignore_errors); 3150 static inline void mm_populate(unsigned long addr, unsigned long len) 3151 { 3152 /* Ignore errors */ 3153 (void) __mm_populate(addr, len, 1); 3154 } 3155 #else 3156 static inline void mm_populate(unsigned long addr, unsigned long len) {} 3157 #endif 3158 3159 /* These take the mm semaphore themselves */ 3160 extern int __must_check vm_brk(unsigned long, unsigned long); 3161 extern int __must_check vm_brk_flags(unsigned long, unsigned long, unsigned long); 3162 extern int vm_munmap(unsigned long, size_t); 3163 extern unsigned long __must_check vm_mmap(struct file *, unsigned long, 3164 unsigned long, unsigned long, 3165 unsigned long, unsigned long); 3166 3167 struct vm_unmapped_area_info { 3168 #define VM_UNMAPPED_AREA_TOPDOWN 1 3169 unsigned long flags; 3170 unsigned long length; 3171 unsigned long low_limit; 3172 unsigned long high_limit; 3173 unsigned long align_mask; 3174 unsigned long align_offset; 3175 }; 3176 3177 extern unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info); 3178 3179 /* truncate.c */ 3180 extern void truncate_inode_pages(struct address_space *, loff_t); 3181 extern void truncate_inode_pages_range(struct address_space *, 3182 loff_t lstart, loff_t lend); 3183 extern void truncate_inode_pages_final(struct address_space *); 3184 3185 /* generic vm_area_ops exported for stackable file systems */ 3186 extern vm_fault_t filemap_fault(struct vm_fault *vmf); 3187 extern vm_fault_t filemap_map_pages(struct vm_fault *vmf, 3188 pgoff_t start_pgoff, pgoff_t end_pgoff); 3189 extern vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf); 3190 3191 extern unsigned long stack_guard_gap; 3192 /* Generic expand stack which grows the stack according to GROWS{UP,DOWN} */ 3193 extern int expand_stack(struct vm_area_struct *vma, unsigned long address); 3194 3195 /* CONFIG_STACK_GROWSUP still needs to grow downwards at some places */ 3196 extern int expand_downwards(struct vm_area_struct *vma, 3197 unsigned long address); 3198 #if VM_GROWSUP 3199 extern int expand_upwards(struct vm_area_struct *vma, unsigned long address); 3200 #else 3201 #define expand_upwards(vma, address) (0) 3202 #endif 3203 3204 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 3205 extern struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr); 3206 extern struct vm_area_struct * find_vma_prev(struct mm_struct * mm, unsigned long addr, 3207 struct vm_area_struct **pprev); 3208 3209 /* 3210 * Look up the first VMA which intersects the interval [start_addr, end_addr) 3211 * NULL if none. Assume start_addr < end_addr. 3212 */ 3213 struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 3214 unsigned long start_addr, unsigned long end_addr); 3215 3216 /** 3217 * vma_lookup() - Find a VMA at a specific address 3218 * @mm: The process address space. 3219 * @addr: The user address. 3220 * 3221 * Return: The vm_area_struct at the given address, %NULL otherwise. 3222 */ 3223 static inline 3224 struct vm_area_struct *vma_lookup(struct mm_struct *mm, unsigned long addr) 3225 { 3226 return mtree_load(&mm->mm_mt, addr); 3227 } 3228 3229 static inline unsigned long vm_start_gap(struct vm_area_struct *vma) 3230 { 3231 unsigned long vm_start = vma->vm_start; 3232 3233 if (vma->vm_flags & VM_GROWSDOWN) { 3234 vm_start -= stack_guard_gap; 3235 if (vm_start > vma->vm_start) 3236 vm_start = 0; 3237 } 3238 return vm_start; 3239 } 3240 3241 static inline unsigned long vm_end_gap(struct vm_area_struct *vma) 3242 { 3243 unsigned long vm_end = vma->vm_end; 3244 3245 if (vma->vm_flags & VM_GROWSUP) { 3246 vm_end += stack_guard_gap; 3247 if (vm_end < vma->vm_end) 3248 vm_end = -PAGE_SIZE; 3249 } 3250 return vm_end; 3251 } 3252 3253 static inline unsigned long vma_pages(struct vm_area_struct *vma) 3254 { 3255 return (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 3256 } 3257 3258 /* Look up the first VMA which exactly match the interval vm_start ... vm_end */ 3259 static inline struct vm_area_struct *find_exact_vma(struct mm_struct *mm, 3260 unsigned long vm_start, unsigned long vm_end) 3261 { 3262 struct vm_area_struct *vma = vma_lookup(mm, vm_start); 3263 3264 if (vma && (vma->vm_start != vm_start || vma->vm_end != vm_end)) 3265 vma = NULL; 3266 3267 return vma; 3268 } 3269 3270 static inline bool range_in_vma(struct vm_area_struct *vma, 3271 unsigned long start, unsigned long end) 3272 { 3273 return (vma && vma->vm_start <= start && end <= vma->vm_end); 3274 } 3275 3276 #ifdef CONFIG_MMU 3277 pgprot_t vm_get_page_prot(unsigned long vm_flags); 3278 void vma_set_page_prot(struct vm_area_struct *vma); 3279 #else 3280 static inline pgprot_t vm_get_page_prot(unsigned long vm_flags) 3281 { 3282 return __pgprot(0); 3283 } 3284 static inline void vma_set_page_prot(struct vm_area_struct *vma) 3285 { 3286 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3287 } 3288 #endif 3289 3290 void vma_set_file(struct vm_area_struct *vma, struct file *file); 3291 3292 #ifdef CONFIG_NUMA_BALANCING 3293 unsigned long change_prot_numa(struct vm_area_struct *vma, 3294 unsigned long start, unsigned long end); 3295 #endif 3296 3297 struct vm_area_struct *find_extend_vma(struct mm_struct *, unsigned long addr); 3298 int remap_pfn_range(struct vm_area_struct *, unsigned long addr, 3299 unsigned long pfn, unsigned long size, pgprot_t); 3300 int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, 3301 unsigned long pfn, unsigned long size, pgprot_t prot); 3302 int vm_insert_page(struct vm_area_struct *, unsigned long addr, struct page *); 3303 int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, 3304 struct page **pages, unsigned long *num); 3305 int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 3306 unsigned long num); 3307 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 3308 unsigned long num); 3309 vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 3310 unsigned long pfn); 3311 vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, 3312 unsigned long pfn, pgprot_t pgprot); 3313 vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 3314 pfn_t pfn); 3315 vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, 3316 unsigned long addr, pfn_t pfn); 3317 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len); 3318 3319 static inline vm_fault_t vmf_insert_page(struct vm_area_struct *vma, 3320 unsigned long addr, struct page *page) 3321 { 3322 int err = vm_insert_page(vma, addr, page); 3323 3324 if (err == -ENOMEM) 3325 return VM_FAULT_OOM; 3326 if (err < 0 && err != -EBUSY) 3327 return VM_FAULT_SIGBUS; 3328 3329 return VM_FAULT_NOPAGE; 3330 } 3331 3332 #ifndef io_remap_pfn_range 3333 static inline int io_remap_pfn_range(struct vm_area_struct *vma, 3334 unsigned long addr, unsigned long pfn, 3335 unsigned long size, pgprot_t prot) 3336 { 3337 return remap_pfn_range(vma, addr, pfn, size, pgprot_decrypted(prot)); 3338 } 3339 #endif 3340 3341 static inline vm_fault_t vmf_error(int err) 3342 { 3343 if (err == -ENOMEM) 3344 return VM_FAULT_OOM; 3345 return VM_FAULT_SIGBUS; 3346 } 3347 3348 struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 3349 unsigned int foll_flags); 3350 3351 static inline int vm_fault_to_errno(vm_fault_t vm_fault, int foll_flags) 3352 { 3353 if (vm_fault & VM_FAULT_OOM) 3354 return -ENOMEM; 3355 if (vm_fault & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) 3356 return (foll_flags & FOLL_HWPOISON) ? -EHWPOISON : -EFAULT; 3357 if (vm_fault & (VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV)) 3358 return -EFAULT; 3359 return 0; 3360 } 3361 3362 /* 3363 * Indicates whether GUP can follow a PROT_NONE mapped page, or whether 3364 * a (NUMA hinting) fault is required. 3365 */ 3366 static inline bool gup_can_follow_protnone(unsigned int flags) 3367 { 3368 /* 3369 * FOLL_FORCE has to be able to make progress even if the VMA is 3370 * inaccessible. Further, FOLL_FORCE access usually does not represent 3371 * application behaviour and we should avoid triggering NUMA hinting 3372 * faults. 3373 */ 3374 return flags & FOLL_FORCE; 3375 } 3376 3377 typedef int (*pte_fn_t)(pte_t *pte, unsigned long addr, void *data); 3378 extern int apply_to_page_range(struct mm_struct *mm, unsigned long address, 3379 unsigned long size, pte_fn_t fn, void *data); 3380 extern int apply_to_existing_page_range(struct mm_struct *mm, 3381 unsigned long address, unsigned long size, 3382 pte_fn_t fn, void *data); 3383 3384 #ifdef CONFIG_PAGE_POISONING 3385 extern void __kernel_poison_pages(struct page *page, int numpages); 3386 extern void __kernel_unpoison_pages(struct page *page, int numpages); 3387 extern bool _page_poisoning_enabled_early; 3388 DECLARE_STATIC_KEY_FALSE(_page_poisoning_enabled); 3389 static inline bool page_poisoning_enabled(void) 3390 { 3391 return _page_poisoning_enabled_early; 3392 } 3393 /* 3394 * For use in fast paths after init_mem_debugging() has run, or when a 3395 * false negative result is not harmful when called too early. 3396 */ 3397 static inline bool page_poisoning_enabled_static(void) 3398 { 3399 return static_branch_unlikely(&_page_poisoning_enabled); 3400 } 3401 static inline void kernel_poison_pages(struct page *page, int numpages) 3402 { 3403 if (page_poisoning_enabled_static()) 3404 __kernel_poison_pages(page, numpages); 3405 } 3406 static inline void kernel_unpoison_pages(struct page *page, int numpages) 3407 { 3408 if (page_poisoning_enabled_static()) 3409 __kernel_unpoison_pages(page, numpages); 3410 } 3411 #else 3412 static inline bool page_poisoning_enabled(void) { return false; } 3413 static inline bool page_poisoning_enabled_static(void) { return false; } 3414 static inline void __kernel_poison_pages(struct page *page, int nunmpages) { } 3415 static inline void kernel_poison_pages(struct page *page, int numpages) { } 3416 static inline void kernel_unpoison_pages(struct page *page, int numpages) { } 3417 #endif 3418 3419 DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc); 3420 static inline bool want_init_on_alloc(gfp_t flags) 3421 { 3422 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, 3423 &init_on_alloc)) 3424 return true; 3425 return flags & __GFP_ZERO; 3426 } 3427 3428 DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free); 3429 static inline bool want_init_on_free(void) 3430 { 3431 return static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON, 3432 &init_on_free); 3433 } 3434 3435 extern bool _debug_pagealloc_enabled_early; 3436 DECLARE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); 3437 3438 static inline bool debug_pagealloc_enabled(void) 3439 { 3440 return IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && 3441 _debug_pagealloc_enabled_early; 3442 } 3443 3444 /* 3445 * For use in fast paths after init_debug_pagealloc() has run, or when a 3446 * false negative result is not harmful when called too early. 3447 */ 3448 static inline bool debug_pagealloc_enabled_static(void) 3449 { 3450 if (!IS_ENABLED(CONFIG_DEBUG_PAGEALLOC)) 3451 return false; 3452 3453 return static_branch_unlikely(&_debug_pagealloc_enabled); 3454 } 3455 3456 #ifdef CONFIG_DEBUG_PAGEALLOC 3457 /* 3458 * To support DEBUG_PAGEALLOC architecture must ensure that 3459 * __kernel_map_pages() never fails 3460 */ 3461 extern void __kernel_map_pages(struct page *page, int numpages, int enable); 3462 3463 static inline void debug_pagealloc_map_pages(struct page *page, int numpages) 3464 { 3465 if (debug_pagealloc_enabled_static()) 3466 __kernel_map_pages(page, numpages, 1); 3467 } 3468 3469 static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages) 3470 { 3471 if (debug_pagealloc_enabled_static()) 3472 __kernel_map_pages(page, numpages, 0); 3473 } 3474 #else /* CONFIG_DEBUG_PAGEALLOC */ 3475 static inline void debug_pagealloc_map_pages(struct page *page, int numpages) {} 3476 static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages) {} 3477 #endif /* CONFIG_DEBUG_PAGEALLOC */ 3478 3479 #ifdef __HAVE_ARCH_GATE_AREA 3480 extern struct vm_area_struct *get_gate_vma(struct mm_struct *mm); 3481 extern int in_gate_area_no_mm(unsigned long addr); 3482 extern int in_gate_area(struct mm_struct *mm, unsigned long addr); 3483 #else 3484 static inline struct vm_area_struct *get_gate_vma(struct mm_struct *mm) 3485 { 3486 return NULL; 3487 } 3488 static inline int in_gate_area_no_mm(unsigned long addr) { return 0; } 3489 static inline int in_gate_area(struct mm_struct *mm, unsigned long addr) 3490 { 3491 return 0; 3492 } 3493 #endif /* __HAVE_ARCH_GATE_AREA */ 3494 3495 extern bool process_shares_mm(struct task_struct *p, struct mm_struct *mm); 3496 3497 #ifdef CONFIG_SYSCTL 3498 extern int sysctl_drop_caches; 3499 int drop_caches_sysctl_handler(struct ctl_table *, int, void *, size_t *, 3500 loff_t *); 3501 #endif 3502 3503 void drop_slab(void); 3504 3505 #ifndef CONFIG_MMU 3506 #define randomize_va_space 0 3507 #else 3508 extern int randomize_va_space; 3509 #endif 3510 3511 const char * arch_vma_name(struct vm_area_struct *vma); 3512 #ifdef CONFIG_MMU 3513 void print_vma_addr(char *prefix, unsigned long rip); 3514 #else 3515 static inline void print_vma_addr(char *prefix, unsigned long rip) 3516 { 3517 } 3518 #endif 3519 3520 void *sparse_buffer_alloc(unsigned long size); 3521 struct page * __populate_section_memmap(unsigned long pfn, 3522 unsigned long nr_pages, int nid, struct vmem_altmap *altmap, 3523 struct dev_pagemap *pgmap); 3524 void pmd_init(void *addr); 3525 void pud_init(void *addr); 3526 pgd_t *vmemmap_pgd_populate(unsigned long addr, int node); 3527 p4d_t *vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node); 3528 pud_t *vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node); 3529 pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node); 3530 pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node, 3531 struct vmem_altmap *altmap, struct page *reuse); 3532 void *vmemmap_alloc_block(unsigned long size, int node); 3533 struct vmem_altmap; 3534 void *vmemmap_alloc_block_buf(unsigned long size, int node, 3535 struct vmem_altmap *altmap); 3536 void vmemmap_verify(pte_t *, int, unsigned long, unsigned long); 3537 void vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 3538 unsigned long addr, unsigned long next); 3539 int vmemmap_check_pmd(pmd_t *pmd, int node, 3540 unsigned long addr, unsigned long next); 3541 int vmemmap_populate_basepages(unsigned long start, unsigned long end, 3542 int node, struct vmem_altmap *altmap); 3543 int vmemmap_populate_hugepages(unsigned long start, unsigned long end, 3544 int node, struct vmem_altmap *altmap); 3545 int vmemmap_populate(unsigned long start, unsigned long end, int node, 3546 struct vmem_altmap *altmap); 3547 void vmemmap_populate_print_last(void); 3548 #ifdef CONFIG_MEMORY_HOTPLUG 3549 void vmemmap_free(unsigned long start, unsigned long end, 3550 struct vmem_altmap *altmap); 3551 #endif 3552 3553 #ifdef CONFIG_ARCH_WANT_OPTIMIZE_VMEMMAP 3554 static inline bool vmemmap_can_optimize(struct vmem_altmap *altmap, 3555 struct dev_pagemap *pgmap) 3556 { 3557 return is_power_of_2(sizeof(struct page)) && 3558 pgmap && (pgmap_vmemmap_nr(pgmap) > 1) && !altmap; 3559 } 3560 #else 3561 static inline bool vmemmap_can_optimize(struct vmem_altmap *altmap, 3562 struct dev_pagemap *pgmap) 3563 { 3564 return false; 3565 } 3566 #endif 3567 3568 void register_page_bootmem_memmap(unsigned long section_nr, struct page *map, 3569 unsigned long nr_pages); 3570 3571 enum mf_flags { 3572 MF_COUNT_INCREASED = 1 << 0, 3573 MF_ACTION_REQUIRED = 1 << 1, 3574 MF_MUST_KILL = 1 << 2, 3575 MF_SOFT_OFFLINE = 1 << 3, 3576 MF_UNPOISON = 1 << 4, 3577 MF_SW_SIMULATED = 1 << 5, 3578 MF_NO_RETRY = 1 << 6, 3579 }; 3580 int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index, 3581 unsigned long count, int mf_flags); 3582 extern int memory_failure(unsigned long pfn, int flags); 3583 extern void memory_failure_queue_kick(int cpu); 3584 extern int unpoison_memory(unsigned long pfn); 3585 extern void shake_page(struct page *p); 3586 extern atomic_long_t num_poisoned_pages __read_mostly; 3587 extern int soft_offline_page(unsigned long pfn, int flags); 3588 #ifdef CONFIG_MEMORY_FAILURE 3589 extern void memory_failure_queue(unsigned long pfn, int flags); 3590 extern int __get_huge_page_for_hwpoison(unsigned long pfn, int flags, 3591 bool *migratable_cleared); 3592 void num_poisoned_pages_inc(unsigned long pfn); 3593 void num_poisoned_pages_sub(unsigned long pfn, long i); 3594 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early); 3595 #else 3596 static inline void memory_failure_queue(unsigned long pfn, int flags) 3597 { 3598 } 3599 3600 static inline int __get_huge_page_for_hwpoison(unsigned long pfn, int flags, 3601 bool *migratable_cleared) 3602 { 3603 return 0; 3604 } 3605 3606 static inline void num_poisoned_pages_inc(unsigned long pfn) 3607 { 3608 } 3609 3610 static inline void num_poisoned_pages_sub(unsigned long pfn, long i) 3611 { 3612 } 3613 #endif 3614 3615 #if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_KSM) 3616 void add_to_kill_ksm(struct task_struct *tsk, struct page *p, 3617 struct vm_area_struct *vma, struct list_head *to_kill, 3618 unsigned long ksm_addr); 3619 #endif 3620 3621 #if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_MEMORY_HOTPLUG) 3622 extern void memblk_nr_poison_inc(unsigned long pfn); 3623 extern void memblk_nr_poison_sub(unsigned long pfn, long i); 3624 #else 3625 static inline void memblk_nr_poison_inc(unsigned long pfn) 3626 { 3627 } 3628 3629 static inline void memblk_nr_poison_sub(unsigned long pfn, long i) 3630 { 3631 } 3632 #endif 3633 3634 #ifndef arch_memory_failure 3635 static inline int arch_memory_failure(unsigned long pfn, int flags) 3636 { 3637 return -ENXIO; 3638 } 3639 #endif 3640 3641 #ifndef arch_is_platform_page 3642 static inline bool arch_is_platform_page(u64 paddr) 3643 { 3644 return false; 3645 } 3646 #endif 3647 3648 /* 3649 * Error handlers for various types of pages. 3650 */ 3651 enum mf_result { 3652 MF_IGNORED, /* Error: cannot be handled */ 3653 MF_FAILED, /* Error: handling failed */ 3654 MF_DELAYED, /* Will be handled later */ 3655 MF_RECOVERED, /* Successfully recovered */ 3656 }; 3657 3658 enum mf_action_page_type { 3659 MF_MSG_KERNEL, 3660 MF_MSG_KERNEL_HIGH_ORDER, 3661 MF_MSG_SLAB, 3662 MF_MSG_DIFFERENT_COMPOUND, 3663 MF_MSG_HUGE, 3664 MF_MSG_FREE_HUGE, 3665 MF_MSG_UNMAP_FAILED, 3666 MF_MSG_DIRTY_SWAPCACHE, 3667 MF_MSG_CLEAN_SWAPCACHE, 3668 MF_MSG_DIRTY_MLOCKED_LRU, 3669 MF_MSG_CLEAN_MLOCKED_LRU, 3670 MF_MSG_DIRTY_UNEVICTABLE_LRU, 3671 MF_MSG_CLEAN_UNEVICTABLE_LRU, 3672 MF_MSG_DIRTY_LRU, 3673 MF_MSG_CLEAN_LRU, 3674 MF_MSG_TRUNCATED_LRU, 3675 MF_MSG_BUDDY, 3676 MF_MSG_DAX, 3677 MF_MSG_UNSPLIT_THP, 3678 MF_MSG_UNKNOWN, 3679 }; 3680 3681 /* 3682 * Sysfs entries for memory failure handling statistics. 3683 */ 3684 extern const struct attribute_group memory_failure_attr_group; 3685 3686 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 3687 extern void clear_huge_page(struct page *page, 3688 unsigned long addr_hint, 3689 unsigned int pages_per_huge_page); 3690 int copy_user_large_folio(struct folio *dst, struct folio *src, 3691 unsigned long addr_hint, 3692 struct vm_area_struct *vma); 3693 long copy_folio_from_user(struct folio *dst_folio, 3694 const void __user *usr_src, 3695 bool allow_pagefault); 3696 3697 /** 3698 * vma_is_special_huge - Are transhuge page-table entries considered special? 3699 * @vma: Pointer to the struct vm_area_struct to consider 3700 * 3701 * Whether transhuge page-table entries are considered "special" following 3702 * the definition in vm_normal_page(). 3703 * 3704 * Return: true if transhuge page-table entries should be considered special, 3705 * false otherwise. 3706 */ 3707 static inline bool vma_is_special_huge(const struct vm_area_struct *vma) 3708 { 3709 return vma_is_dax(vma) || (vma->vm_file && 3710 (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))); 3711 } 3712 3713 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 3714 3715 #ifdef CONFIG_DEBUG_PAGEALLOC 3716 extern unsigned int _debug_guardpage_minorder; 3717 DECLARE_STATIC_KEY_FALSE(_debug_guardpage_enabled); 3718 3719 static inline unsigned int debug_guardpage_minorder(void) 3720 { 3721 return _debug_guardpage_minorder; 3722 } 3723 3724 static inline bool debug_guardpage_enabled(void) 3725 { 3726 return static_branch_unlikely(&_debug_guardpage_enabled); 3727 } 3728 3729 static inline bool page_is_guard(struct page *page) 3730 { 3731 if (!debug_guardpage_enabled()) 3732 return false; 3733 3734 return PageGuard(page); 3735 } 3736 #else 3737 static inline unsigned int debug_guardpage_minorder(void) { return 0; } 3738 static inline bool debug_guardpage_enabled(void) { return false; } 3739 static inline bool page_is_guard(struct page *page) { return false; } 3740 #endif /* CONFIG_DEBUG_PAGEALLOC */ 3741 3742 #if MAX_NUMNODES > 1 3743 void __init setup_nr_node_ids(void); 3744 #else 3745 static inline void setup_nr_node_ids(void) {} 3746 #endif 3747 3748 extern int memcmp_pages(struct page *page1, struct page *page2); 3749 3750 static inline int pages_identical(struct page *page1, struct page *page2) 3751 { 3752 return !memcmp_pages(page1, page2); 3753 } 3754 3755 #ifdef CONFIG_MAPPING_DIRTY_HELPERS 3756 unsigned long clean_record_shared_mapping_range(struct address_space *mapping, 3757 pgoff_t first_index, pgoff_t nr, 3758 pgoff_t bitmap_pgoff, 3759 unsigned long *bitmap, 3760 pgoff_t *start, 3761 pgoff_t *end); 3762 3763 unsigned long wp_shared_mapping_range(struct address_space *mapping, 3764 pgoff_t first_index, pgoff_t nr); 3765 #endif 3766 3767 extern int sysctl_nr_trim_pages; 3768 3769 #ifdef CONFIG_PRINTK 3770 void mem_dump_obj(void *object); 3771 #else 3772 static inline void mem_dump_obj(void *object) {} 3773 #endif 3774 3775 /** 3776 * seal_check_future_write - Check for F_SEAL_FUTURE_WRITE flag and handle it 3777 * @seals: the seals to check 3778 * @vma: the vma to operate on 3779 * 3780 * Check whether F_SEAL_FUTURE_WRITE is set; if so, do proper check/handling on 3781 * the vma flags. Return 0 if check pass, or <0 for errors. 3782 */ 3783 static inline int seal_check_future_write(int seals, struct vm_area_struct *vma) 3784 { 3785 if (seals & F_SEAL_FUTURE_WRITE) { 3786 /* 3787 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when 3788 * "future write" seal active. 3789 */ 3790 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE)) 3791 return -EPERM; 3792 3793 /* 3794 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as 3795 * MAP_SHARED and read-only, take care to not allow mprotect to 3796 * revert protections on such mappings. Do this only for shared 3797 * mappings. For private mappings, don't need to mask 3798 * VM_MAYWRITE as we still want them to be COW-writable. 3799 */ 3800 if (vma->vm_flags & VM_SHARED) 3801 vm_flags_clear(vma, VM_MAYWRITE); 3802 } 3803 3804 return 0; 3805 } 3806 3807 #ifdef CONFIG_ANON_VMA_NAME 3808 int madvise_set_anon_name(struct mm_struct *mm, unsigned long start, 3809 unsigned long len_in, 3810 struct anon_vma_name *anon_name); 3811 #else 3812 static inline int 3813 madvise_set_anon_name(struct mm_struct *mm, unsigned long start, 3814 unsigned long len_in, struct anon_vma_name *anon_name) { 3815 return 0; 3816 } 3817 #endif 3818 3819 #ifdef CONFIG_UNACCEPTED_MEMORY 3820 3821 bool range_contains_unaccepted_memory(phys_addr_t start, phys_addr_t end); 3822 void accept_memory(phys_addr_t start, phys_addr_t end); 3823 3824 #else 3825 3826 static inline bool range_contains_unaccepted_memory(phys_addr_t start, 3827 phys_addr_t end) 3828 { 3829 return false; 3830 } 3831 3832 static inline void accept_memory(phys_addr_t start, phys_addr_t end) 3833 { 3834 } 3835 3836 #endif 3837 3838 #endif /* _LINUX_MM_H */ 3839