include/linux/mm_types.h

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