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