1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_MM_TYPES_H 3 #define _LINUX_MM_TYPES_H 4 5 #include <linux/mm_types_task.h> 6 7 #include <linux/auxvec.h> 8 #include <linux/kref.h> 9 #include <linux/list.h> 10 #include <linux/spinlock.h> 11 #include <linux/rbtree.h> 12 #include <linux/rwsem.h> 13 #include <linux/completion.h> 14 #include <linux/cpumask.h> 15 #include <linux/uprobes.h> 16 #include <linux/rcupdate.h> 17 #include <linux/page-flags-layout.h> 18 #include <linux/workqueue.h> 19 #include <linux/seqlock.h> 20 21 #include <asm/mmu.h> 22 23 #ifndef AT_VECTOR_SIZE_ARCH 24 #define AT_VECTOR_SIZE_ARCH 0 25 #endif 26 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1)) 27 28 #define INIT_PASID 0 29 30 struct address_space; 31 struct mem_cgroup; 32 33 /* 34 * Each physical page in the system has a struct page associated with 35 * it to keep track of whatever it is we are using the page for at the 36 * moment. Note that we have no way to track which tasks are using 37 * a page, though if it is a pagecache page, rmap structures can tell us 38 * who is mapping it. 39 * 40 * If you allocate the page using alloc_pages(), you can use some of the 41 * space in struct page for your own purposes. The five words in the main 42 * union are available, except for bit 0 of the first word which must be 43 * kept clear. Many users use this word to store a pointer to an object 44 * which is guaranteed to be aligned. If you use the same storage as 45 * page->mapping, you must restore it to NULL before freeing the page. 46 * 47 * If your page will not be mapped to userspace, you can also use the four 48 * bytes in the mapcount union, but you must call page_mapcount_reset() 49 * before freeing it. 50 * 51 * If you want to use the refcount field, it must be used in such a way 52 * that other CPUs temporarily incrementing and then decrementing the 53 * refcount does not cause problems. On receiving the page from 54 * alloc_pages(), the refcount will be positive. 55 * 56 * If you allocate pages of order > 0, you can use some of the fields 57 * in each subpage, but you may need to restore some of their values 58 * afterwards. 59 * 60 * SLUB uses cmpxchg_double() to atomically update its freelist and counters. 61 * That requires that freelist & counters in struct slab be adjacent and 62 * double-word aligned. Because struct slab currently just reinterprets the 63 * bits of struct page, we align all struct pages to double-word boundaries, 64 * and ensure that 'freelist' is aligned within struct slab. 65 */ 66 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE 67 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long)) 68 #else 69 #define _struct_page_alignment 70 #endif 71 72 struct page { 73 unsigned long flags; /* Atomic flags, some possibly 74 * updated asynchronously */ 75 /* 76 * Five words (20/40 bytes) are available in this union. 77 * WARNING: bit 0 of the first word is used for PageTail(). That 78 * means the other users of this union MUST NOT use the bit to 79 * avoid collision and false-positive PageTail(). 80 */ 81 union { 82 struct { /* Page cache and anonymous pages */ 83 /** 84 * @lru: Pageout list, eg. active_list protected by 85 * lruvec->lru_lock. Sometimes used as a generic list 86 * by the page owner. 87 */ 88 struct list_head lru; 89 /* See page-flags.h for PAGE_MAPPING_FLAGS */ 90 struct address_space *mapping; 91 pgoff_t index; /* Our offset within mapping. */ 92 /** 93 * @private: Mapping-private opaque data. 94 * Usually used for buffer_heads if PagePrivate. 95 * Used for swp_entry_t if PageSwapCache. 96 * Indicates order in the buddy system if PageBuddy. 97 */ 98 unsigned long private; 99 }; 100 struct { /* page_pool used by netstack */ 101 /** 102 * @pp_magic: magic value to avoid recycling non 103 * page_pool allocated pages. 104 */ 105 unsigned long pp_magic; 106 struct page_pool *pp; 107 unsigned long _pp_mapping_pad; 108 unsigned long dma_addr; 109 union { 110 /** 111 * dma_addr_upper: might require a 64-bit 112 * value on 32-bit architectures. 113 */ 114 unsigned long dma_addr_upper; 115 /** 116 * For frag page support, not supported in 117 * 32-bit architectures with 64-bit DMA. 118 */ 119 atomic_long_t pp_frag_count; 120 }; 121 }; 122 struct { /* slab, slob and slub */ 123 union { 124 struct list_head slab_list; 125 struct { /* Partial pages */ 126 struct page *next; 127 #ifdef CONFIG_64BIT 128 int pages; /* Nr of pages left */ 129 #else 130 short int pages; 131 #endif 132 }; 133 }; 134 struct kmem_cache *slab_cache; /* not slob */ 135 /* Double-word boundary */ 136 void *freelist; /* first free object */ 137 union { 138 void *s_mem; /* slab: first object */ 139 unsigned long counters; /* SLUB */ 140 struct { /* SLUB */ 141 unsigned inuse:16; 142 unsigned objects:15; 143 unsigned frozen:1; 144 }; 145 }; 146 }; 147 struct { /* Tail pages of compound page */ 148 unsigned long compound_head; /* Bit zero is set */ 149 150 /* First tail page only */ 151 unsigned char compound_dtor; 152 unsigned char compound_order; 153 atomic_t compound_mapcount; 154 unsigned int compound_nr; /* 1 << compound_order */ 155 }; 156 struct { /* Second tail page of compound page */ 157 unsigned long _compound_pad_1; /* compound_head */ 158 atomic_t hpage_pinned_refcount; 159 /* For both global and memcg */ 160 struct list_head deferred_list; 161 }; 162 struct { /* Page table pages */ 163 unsigned long _pt_pad_1; /* compound_head */ 164 pgtable_t pmd_huge_pte; /* protected by page->ptl */ 165 unsigned long _pt_pad_2; /* mapping */ 166 union { 167 struct mm_struct *pt_mm; /* x86 pgds only */ 168 atomic_t pt_frag_refcount; /* powerpc */ 169 }; 170 #if ALLOC_SPLIT_PTLOCKS 171 spinlock_t *ptl; 172 #else 173 spinlock_t ptl; 174 #endif 175 }; 176 struct { /* ZONE_DEVICE pages */ 177 /** @pgmap: Points to the hosting device page map. */ 178 struct dev_pagemap *pgmap; 179 void *zone_device_data; 180 /* 181 * ZONE_DEVICE private pages are counted as being 182 * mapped so the next 3 words hold the mapping, index, 183 * and private fields from the source anonymous or 184 * page cache page while the page is migrated to device 185 * private memory. 186 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also 187 * use the mapping, index, and private fields when 188 * pmem backed DAX files are mapped. 189 */ 190 }; 191 192 /** @rcu_head: You can use this to free a page by RCU. */ 193 struct rcu_head rcu_head; 194 }; 195 196 union { /* This union is 4 bytes in size. */ 197 /* 198 * If the page can be mapped to userspace, encodes the number 199 * of times this page is referenced by a page table. 200 */ 201 atomic_t _mapcount; 202 203 /* 204 * If the page is neither PageSlab nor mappable to userspace, 205 * the value stored here may help determine what this page 206 * is used for. See page-flags.h for a list of page types 207 * which are currently stored here. 208 */ 209 unsigned int page_type; 210 211 unsigned int active; /* SLAB */ 212 int units; /* SLOB */ 213 }; 214 215 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */ 216 atomic_t _refcount; 217 218 #ifdef CONFIG_MEMCG 219 unsigned long memcg_data; 220 #endif 221 222 /* 223 * On machines where all RAM is mapped into kernel address space, 224 * we can simply calculate the virtual address. On machines with 225 * highmem some memory is mapped into kernel virtual memory 226 * dynamically, so we need a place to store that address. 227 * Note that this field could be 16 bits on x86 ... ;) 228 * 229 * Architectures with slow multiplication can define 230 * WANT_PAGE_VIRTUAL in asm/page.h 231 */ 232 #if defined(WANT_PAGE_VIRTUAL) 233 void *virtual; /* Kernel virtual address (NULL if 234 not kmapped, ie. highmem) */ 235 #endif /* WANT_PAGE_VIRTUAL */ 236 237 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS 238 int _last_cpupid; 239 #endif 240 } _struct_page_alignment; 241 242 /** 243 * struct folio - Represents a contiguous set of bytes. 244 * @flags: Identical to the page flags. 245 * @lru: Least Recently Used list; tracks how recently this folio was used. 246 * @mapping: The file this page belongs to, or refers to the anon_vma for 247 * anonymous memory. 248 * @index: Offset within the file, in units of pages. For anonymous memory, 249 * this is the index from the beginning of the mmap. 250 * @private: Filesystem per-folio data (see folio_attach_private()). 251 * Used for swp_entry_t if folio_test_swapcache(). 252 * @_mapcount: Do not access this member directly. Use folio_mapcount() to 253 * find out how many times this folio is mapped by userspace. 254 * @_refcount: Do not access this member directly. Use folio_ref_count() 255 * to find how many references there are to this folio. 256 * @memcg_data: Memory Control Group data. 257 * 258 * A folio is a physically, virtually and logically contiguous set 259 * of bytes. It is a power-of-two in size, and it is aligned to that 260 * same power-of-two. It is at least as large as %PAGE_SIZE. If it is 261 * in the page cache, it is at a file offset which is a multiple of that 262 * power-of-two. It may be mapped into userspace at an address which is 263 * at an arbitrary page offset, but its kernel virtual address is aligned 264 * to its size. 265 */ 266 struct folio { 267 /* private: don't document the anon union */ 268 union { 269 struct { 270 /* public: */ 271 unsigned long flags; 272 struct list_head lru; 273 struct address_space *mapping; 274 pgoff_t index; 275 void *private; 276 atomic_t _mapcount; 277 atomic_t _refcount; 278 #ifdef CONFIG_MEMCG 279 unsigned long memcg_data; 280 #endif 281 /* private: the union with struct page is transitional */ 282 }; 283 struct page page; 284 }; 285 }; 286 287 static_assert(sizeof(struct page) == sizeof(struct folio)); 288 #define FOLIO_MATCH(pg, fl) \ 289 static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl)) 290 FOLIO_MATCH(flags, flags); 291 FOLIO_MATCH(lru, lru); 292 FOLIO_MATCH(compound_head, lru); 293 FOLIO_MATCH(index, index); 294 FOLIO_MATCH(private, private); 295 FOLIO_MATCH(_mapcount, _mapcount); 296 FOLIO_MATCH(_refcount, _refcount); 297 #ifdef CONFIG_MEMCG 298 FOLIO_MATCH(memcg_data, memcg_data); 299 #endif 300 #undef FOLIO_MATCH 301 302 static inline atomic_t *folio_mapcount_ptr(struct folio *folio) 303 { 304 struct page *tail = &folio->page + 1; 305 return &tail->compound_mapcount; 306 } 307 308 static inline atomic_t *compound_mapcount_ptr(struct page *page) 309 { 310 return &page[1].compound_mapcount; 311 } 312 313 static inline atomic_t *compound_pincount_ptr(struct page *page) 314 { 315 return &page[2].hpage_pinned_refcount; 316 } 317 318 /* 319 * Used for sizing the vmemmap region on some architectures 320 */ 321 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page))) 322 323 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK) 324 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE) 325 326 /* 327 * page_private can be used on tail pages. However, PagePrivate is only 328 * checked by the VM on the head page. So page_private on the tail pages 329 * should be used for data that's ancillary to the head page (eg attaching 330 * buffer heads to tail pages after attaching buffer heads to the head page) 331 */ 332 #define page_private(page) ((page)->private) 333 334 static inline void set_page_private(struct page *page, unsigned long private) 335 { 336 page->private = private; 337 } 338 339 static inline void *folio_get_private(struct folio *folio) 340 { 341 return folio->private; 342 } 343 344 struct page_frag_cache { 345 void * va; 346 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) 347 __u16 offset; 348 __u16 size; 349 #else 350 __u32 offset; 351 #endif 352 /* we maintain a pagecount bias, so that we dont dirty cache line 353 * containing page->_refcount every time we allocate a fragment. 354 */ 355 unsigned int pagecnt_bias; 356 bool pfmemalloc; 357 }; 358 359 typedef unsigned long vm_flags_t; 360 361 /* 362 * A region containing a mapping of a non-memory backed file under NOMMU 363 * conditions. These are held in a global tree and are pinned by the VMAs that 364 * map parts of them. 365 */ 366 struct vm_region { 367 struct rb_node vm_rb; /* link in global region tree */ 368 vm_flags_t vm_flags; /* VMA vm_flags */ 369 unsigned long vm_start; /* start address of region */ 370 unsigned long vm_end; /* region initialised to here */ 371 unsigned long vm_top; /* region allocated to here */ 372 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */ 373 struct file *vm_file; /* the backing file or NULL */ 374 375 int vm_usage; /* region usage count (access under nommu_region_sem) */ 376 bool vm_icache_flushed : 1; /* true if the icache has been flushed for 377 * this region */ 378 }; 379 380 #ifdef CONFIG_USERFAULTFD 381 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, }) 382 struct vm_userfaultfd_ctx { 383 struct userfaultfd_ctx *ctx; 384 }; 385 #else /* CONFIG_USERFAULTFD */ 386 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {}) 387 struct vm_userfaultfd_ctx {}; 388 #endif /* CONFIG_USERFAULTFD */ 389 390 struct anon_vma_name { 391 struct kref kref; 392 /* The name needs to be at the end because it is dynamically sized. */ 393 char name[]; 394 }; 395 396 /* 397 * This struct describes a virtual memory area. There is one of these 398 * per VM-area/task. A VM area is any part of the process virtual memory 399 * space that has a special rule for the page-fault handlers (ie a shared 400 * library, the executable area etc). 401 */ 402 struct vm_area_struct { 403 /* The first cache line has the info for VMA tree walking. */ 404 405 unsigned long vm_start; /* Our start address within vm_mm. */ 406 unsigned long vm_end; /* The first byte after our end address 407 within vm_mm. */ 408 409 /* linked list of VM areas per task, sorted by address */ 410 struct vm_area_struct *vm_next, *vm_prev; 411 412 struct rb_node vm_rb; 413 414 /* 415 * Largest free memory gap in bytes to the left of this VMA. 416 * Either between this VMA and vma->vm_prev, or between one of the 417 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps 418 * get_unmapped_area find a free area of the right size. 419 */ 420 unsigned long rb_subtree_gap; 421 422 /* Second cache line starts here. */ 423 424 struct mm_struct *vm_mm; /* The address space we belong to. */ 425 426 /* 427 * Access permissions of this VMA. 428 * See vmf_insert_mixed_prot() for discussion. 429 */ 430 pgprot_t vm_page_prot; 431 unsigned long vm_flags; /* Flags, see mm.h. */ 432 433 /* 434 * For areas with an address space and backing store, 435 * linkage into the address_space->i_mmap interval tree. 436 * 437 * For private anonymous mappings, a pointer to a null terminated string 438 * containing the name given to the vma, or NULL if unnamed. 439 */ 440 441 union { 442 struct { 443 struct rb_node rb; 444 unsigned long rb_subtree_last; 445 } shared; 446 /* Serialized by mmap_sem. */ 447 struct anon_vma_name *anon_name; 448 }; 449 450 /* 451 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma 452 * list, after a COW of one of the file pages. A MAP_SHARED vma 453 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack 454 * or brk vma (with NULL file) can only be in an anon_vma list. 455 */ 456 struct list_head anon_vma_chain; /* Serialized by mmap_lock & 457 * page_table_lock */ 458 struct anon_vma *anon_vma; /* Serialized by page_table_lock */ 459 460 /* Function pointers to deal with this struct. */ 461 const struct vm_operations_struct *vm_ops; 462 463 /* Information about our backing store: */ 464 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE 465 units */ 466 struct file * vm_file; /* File we map to (can be NULL). */ 467 void * vm_private_data; /* was vm_pte (shared mem) */ 468 469 #ifdef CONFIG_SWAP 470 atomic_long_t swap_readahead_info; 471 #endif 472 #ifndef CONFIG_MMU 473 struct vm_region *vm_region; /* NOMMU mapping region */ 474 #endif 475 #ifdef CONFIG_NUMA 476 struct mempolicy *vm_policy; /* NUMA policy for the VMA */ 477 #endif 478 struct vm_userfaultfd_ctx vm_userfaultfd_ctx; 479 } __randomize_layout; 480 481 struct kioctx_table; 482 struct mm_struct { 483 struct { 484 struct vm_area_struct *mmap; /* list of VMAs */ 485 struct rb_root mm_rb; 486 u64 vmacache_seqnum; /* per-thread vmacache */ 487 #ifdef CONFIG_MMU 488 unsigned long (*get_unmapped_area) (struct file *filp, 489 unsigned long addr, unsigned long len, 490 unsigned long pgoff, unsigned long flags); 491 #endif 492 unsigned long mmap_base; /* base of mmap area */ 493 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */ 494 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES 495 /* Base addresses for compatible mmap() */ 496 unsigned long mmap_compat_base; 497 unsigned long mmap_compat_legacy_base; 498 #endif 499 unsigned long task_size; /* size of task vm space */ 500 unsigned long highest_vm_end; /* highest vma end address */ 501 pgd_t * pgd; 502 503 #ifdef CONFIG_MEMBARRIER 504 /** 505 * @membarrier_state: Flags controlling membarrier behavior. 506 * 507 * This field is close to @pgd to hopefully fit in the same 508 * cache-line, which needs to be touched by switch_mm(). 509 */ 510 atomic_t membarrier_state; 511 #endif 512 513 /** 514 * @mm_users: The number of users including userspace. 515 * 516 * Use mmget()/mmget_not_zero()/mmput() to modify. When this 517 * drops to 0 (i.e. when the task exits and there are no other 518 * temporary reference holders), we also release a reference on 519 * @mm_count (which may then free the &struct mm_struct if 520 * @mm_count also drops to 0). 521 */ 522 atomic_t mm_users; 523 524 /** 525 * @mm_count: The number of references to &struct mm_struct 526 * (@mm_users count as 1). 527 * 528 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the 529 * &struct mm_struct is freed. 530 */ 531 atomic_t mm_count; 532 533 #ifdef CONFIG_MMU 534 atomic_long_t pgtables_bytes; /* PTE page table pages */ 535 #endif 536 int map_count; /* number of VMAs */ 537 538 spinlock_t page_table_lock; /* Protects page tables and some 539 * counters 540 */ 541 /* 542 * With some kernel config, the current mmap_lock's offset 543 * inside 'mm_struct' is at 0x120, which is very optimal, as 544 * its two hot fields 'count' and 'owner' sit in 2 different 545 * cachelines, and when mmap_lock is highly contended, both 546 * of the 2 fields will be accessed frequently, current layout 547 * will help to reduce cache bouncing. 548 * 549 * So please be careful with adding new fields before 550 * mmap_lock, which can easily push the 2 fields into one 551 * cacheline. 552 */ 553 struct rw_semaphore mmap_lock; 554 555 struct list_head mmlist; /* List of maybe swapped mm's. These 556 * are globally strung together off 557 * init_mm.mmlist, and are protected 558 * by mmlist_lock 559 */ 560 561 562 unsigned long hiwater_rss; /* High-watermark of RSS usage */ 563 unsigned long hiwater_vm; /* High-water virtual memory usage */ 564 565 unsigned long total_vm; /* Total pages mapped */ 566 unsigned long locked_vm; /* Pages that have PG_mlocked set */ 567 atomic64_t pinned_vm; /* Refcount permanently increased */ 568 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */ 569 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */ 570 unsigned long stack_vm; /* VM_STACK */ 571 unsigned long def_flags; 572 573 /** 574 * @write_protect_seq: Locked when any thread is write 575 * protecting pages mapped by this mm to enforce a later COW, 576 * for instance during page table copying for fork(). 577 */ 578 seqcount_t write_protect_seq; 579 580 spinlock_t arg_lock; /* protect the below fields */ 581 582 unsigned long start_code, end_code, start_data, end_data; 583 unsigned long start_brk, brk, start_stack; 584 unsigned long arg_start, arg_end, env_start, env_end; 585 586 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */ 587 588 /* 589 * Special counters, in some configurations protected by the 590 * page_table_lock, in other configurations by being atomic. 591 */ 592 struct mm_rss_stat rss_stat; 593 594 struct linux_binfmt *binfmt; 595 596 /* Architecture-specific MM context */ 597 mm_context_t context; 598 599 unsigned long flags; /* Must use atomic bitops to access */ 600 601 #ifdef CONFIG_AIO 602 spinlock_t ioctx_lock; 603 struct kioctx_table __rcu *ioctx_table; 604 #endif 605 #ifdef CONFIG_MEMCG 606 /* 607 * "owner" points to a task that is regarded as the canonical 608 * user/owner of this mm. All of the following must be true in 609 * order for it to be changed: 610 * 611 * current == mm->owner 612 * current->mm != mm 613 * new_owner->mm == mm 614 * new_owner->alloc_lock is held 615 */ 616 struct task_struct __rcu *owner; 617 #endif 618 struct user_namespace *user_ns; 619 620 /* store ref to file /proc/<pid>/exe symlink points to */ 621 struct file __rcu *exe_file; 622 #ifdef CONFIG_MMU_NOTIFIER 623 struct mmu_notifier_subscriptions *notifier_subscriptions; 624 #endif 625 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 626 pgtable_t pmd_huge_pte; /* protected by page_table_lock */ 627 #endif 628 #ifdef CONFIG_NUMA_BALANCING 629 /* 630 * numa_next_scan is the next time that the PTEs will be marked 631 * pte_numa. NUMA hinting faults will gather statistics and 632 * migrate pages to new nodes if necessary. 633 */ 634 unsigned long numa_next_scan; 635 636 /* Restart point for scanning and setting pte_numa */ 637 unsigned long numa_scan_offset; 638 639 /* numa_scan_seq prevents two threads setting pte_numa */ 640 int numa_scan_seq; 641 #endif 642 /* 643 * An operation with batched TLB flushing is going on. Anything 644 * that can move process memory needs to flush the TLB when 645 * moving a PROT_NONE or PROT_NUMA mapped page. 646 */ 647 atomic_t tlb_flush_pending; 648 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 649 /* See flush_tlb_batched_pending() */ 650 atomic_t tlb_flush_batched; 651 #endif 652 struct uprobes_state uprobes_state; 653 #ifdef CONFIG_PREEMPT_RT 654 struct rcu_head delayed_drop; 655 #endif 656 #ifdef CONFIG_HUGETLB_PAGE 657 atomic_long_t hugetlb_usage; 658 #endif 659 struct work_struct async_put_work; 660 661 #ifdef CONFIG_IOMMU_SUPPORT 662 u32 pasid; 663 #endif 664 } __randomize_layout; 665 666 /* 667 * The mm_cpumask needs to be at the end of mm_struct, because it 668 * is dynamically sized based on nr_cpu_ids. 669 */ 670 unsigned long cpu_bitmap[]; 671 }; 672 673 extern struct mm_struct init_mm; 674 675 /* Pointer magic because the dynamic array size confuses some compilers. */ 676 static inline void mm_init_cpumask(struct mm_struct *mm) 677 { 678 unsigned long cpu_bitmap = (unsigned long)mm; 679 680 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap); 681 cpumask_clear((struct cpumask *)cpu_bitmap); 682 } 683 684 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */ 685 static inline cpumask_t *mm_cpumask(struct mm_struct *mm) 686 { 687 return (struct cpumask *)&mm->cpu_bitmap; 688 } 689 690 struct mmu_gather; 691 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm); 692 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm); 693 extern void tlb_finish_mmu(struct mmu_gather *tlb); 694 695 struct vm_fault; 696 697 /** 698 * typedef vm_fault_t - Return type for page fault handlers. 699 * 700 * Page fault handlers return a bitmask of %VM_FAULT values. 701 */ 702 typedef __bitwise unsigned int vm_fault_t; 703 704 /** 705 * enum vm_fault_reason - Page fault handlers return a bitmask of 706 * these values to tell the core VM what happened when handling the 707 * fault. Used to decide whether a process gets delivered SIGBUS or 708 * just gets major/minor fault counters bumped up. 709 * 710 * @VM_FAULT_OOM: Out Of Memory 711 * @VM_FAULT_SIGBUS: Bad access 712 * @VM_FAULT_MAJOR: Page read from storage 713 * @VM_FAULT_WRITE: Special case for get_user_pages 714 * @VM_FAULT_HWPOISON: Hit poisoned small page 715 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded 716 * in upper bits 717 * @VM_FAULT_SIGSEGV: segmentation fault 718 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page 719 * @VM_FAULT_LOCKED: ->fault locked the returned page 720 * @VM_FAULT_RETRY: ->fault blocked, must retry 721 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small 722 * @VM_FAULT_DONE_COW: ->fault has fully handled COW 723 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs 724 * fsync() to complete (for synchronous page faults 725 * in DAX) 726 * @VM_FAULT_HINDEX_MASK: mask HINDEX value 727 * 728 */ 729 enum vm_fault_reason { 730 VM_FAULT_OOM = (__force vm_fault_t)0x000001, 731 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002, 732 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004, 733 VM_FAULT_WRITE = (__force vm_fault_t)0x000008, 734 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010, 735 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020, 736 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040, 737 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100, 738 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200, 739 VM_FAULT_RETRY = (__force vm_fault_t)0x000400, 740 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800, 741 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000, 742 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000, 743 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000, 744 }; 745 746 /* Encode hstate index for a hwpoisoned large page */ 747 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16)) 748 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf) 749 750 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \ 751 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \ 752 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK) 753 754 #define VM_FAULT_RESULT_TRACE \ 755 { VM_FAULT_OOM, "OOM" }, \ 756 { VM_FAULT_SIGBUS, "SIGBUS" }, \ 757 { VM_FAULT_MAJOR, "MAJOR" }, \ 758 { VM_FAULT_WRITE, "WRITE" }, \ 759 { VM_FAULT_HWPOISON, "HWPOISON" }, \ 760 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \ 761 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \ 762 { VM_FAULT_NOPAGE, "NOPAGE" }, \ 763 { VM_FAULT_LOCKED, "LOCKED" }, \ 764 { VM_FAULT_RETRY, "RETRY" }, \ 765 { VM_FAULT_FALLBACK, "FALLBACK" }, \ 766 { VM_FAULT_DONE_COW, "DONE_COW" }, \ 767 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" } 768 769 struct vm_special_mapping { 770 const char *name; /* The name, e.g. "[vdso]". */ 771 772 /* 773 * If .fault is not provided, this points to a 774 * NULL-terminated array of pages that back the special mapping. 775 * 776 * This must not be NULL unless .fault is provided. 777 */ 778 struct page **pages; 779 780 /* 781 * If non-NULL, then this is called to resolve page faults 782 * on the special mapping. If used, .pages is not checked. 783 */ 784 vm_fault_t (*fault)(const struct vm_special_mapping *sm, 785 struct vm_area_struct *vma, 786 struct vm_fault *vmf); 787 788 int (*mremap)(const struct vm_special_mapping *sm, 789 struct vm_area_struct *new_vma); 790 }; 791 792 enum tlb_flush_reason { 793 TLB_FLUSH_ON_TASK_SWITCH, 794 TLB_REMOTE_SHOOTDOWN, 795 TLB_LOCAL_SHOOTDOWN, 796 TLB_LOCAL_MM_SHOOTDOWN, 797 TLB_REMOTE_SEND_IPI, 798 NR_TLB_FLUSH_REASONS, 799 }; 800 801 /* 802 * A swap entry has to fit into a "unsigned long", as the entry is hidden 803 * in the "index" field of the swapper address space. 804 */ 805 typedef struct { 806 unsigned long val; 807 } swp_entry_t; 808 809 /** 810 * enum fault_flag - Fault flag definitions. 811 * @FAULT_FLAG_WRITE: Fault was a write fault. 812 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE. 813 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked. 814 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying. 815 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region. 816 * @FAULT_FLAG_TRIED: The fault has been tried once. 817 * @FAULT_FLAG_USER: The fault originated in userspace. 818 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm. 819 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch. 820 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals. 821 * 822 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify 823 * whether we would allow page faults to retry by specifying these two 824 * fault flags correctly. Currently there can be three legal combinations: 825 * 826 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and 827 * this is the first try 828 * 829 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and 830 * we've already tried at least once 831 * 832 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry 833 * 834 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never 835 * be used. Note that page faults can be allowed to retry for multiple times, 836 * in which case we'll have an initial fault with flags (a) then later on 837 * continuous faults with flags (b). We should always try to detect pending 838 * signals before a retry to make sure the continuous page faults can still be 839 * interrupted if necessary. 840 */ 841 enum fault_flag { 842 FAULT_FLAG_WRITE = 1 << 0, 843 FAULT_FLAG_MKWRITE = 1 << 1, 844 FAULT_FLAG_ALLOW_RETRY = 1 << 2, 845 FAULT_FLAG_RETRY_NOWAIT = 1 << 3, 846 FAULT_FLAG_KILLABLE = 1 << 4, 847 FAULT_FLAG_TRIED = 1 << 5, 848 FAULT_FLAG_USER = 1 << 6, 849 FAULT_FLAG_REMOTE = 1 << 7, 850 FAULT_FLAG_INSTRUCTION = 1 << 8, 851 FAULT_FLAG_INTERRUPTIBLE = 1 << 9, 852 }; 853 854 #endif /* _LINUX_MM_TYPES_H */ 855