include/linux/mm_types.h

   1 #ifndef _LINUX_MM_TYPES_H
   2 #define _LINUX_MM_TYPES_H
   3
   4 #include <linux/auxvec.h>
   5 #include <linux/types.h>
   6 #include <linux/threads.h>
   7 #include <linux/list.h>
   8 #include <linux/spinlock.h>
   9 #include <linux/prio_tree.h>
  10 #include <linux/rbtree.h>
  11 #include <linux/rwsem.h>
  12 #include <linux/completion.h>
  13 #include <linux/cpumask.h>
  14 #include <linux/page-debug-flags.h>
  15 #include <linux/uprobes.h>
  16 #include <asm/page.h>
  17 #include <asm/mmu.h>
  18
  19 #ifndef AT_VECTOR_SIZE_ARCH
  20 #define AT_VECTOR_SIZE_ARCH 0
  21 #endif
  22 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
  23
  24 struct address_space;
  25
  26 #define USE_SPLIT_PTLOCKS       (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
  27
  28 /*
  29  * Each physical page in the system has a struct page associated with
  30  * it to keep track of whatever it is we are using the page for at the
  31  * moment. Note that we have no way to track which tasks are using
  32  * a page, though if it is a pagecache page, rmap structures can tell us
  33  * who is mapping it.
  34  *
  35  * The objects in struct page are organized in double word blocks in
  36  * order to allows us to use atomic double word operations on portions
  37  * of struct page. That is currently only used by slub but the arrangement
  38  * allows the use of atomic double word operations on the flags/mapping
  39  * and lru list pointers also.
  40  */
  41 struct page {
  42         /* First double word block */
  43         unsigned long flags;            /* Atomic flags, some possibly
  44                                          * updated asynchronously */
  45         struct address_space *mapping;  /* If low bit clear, points to
  46                                          * inode address_space, or NULL.
  47                                          * If page mapped as anonymous
  48                                          * memory, low bit is set, and
  49                                          * it points to anon_vma object:
  50                                          * see PAGE_MAPPING_ANON below.
  51                                          */
  52         /* Second double word */
  53         struct {
  54                 union {
  55                         pgoff_t index;          /* Our offset within mapping. */
  56                         void *freelist;         /* slub/slob first free object */
  57                         bool pfmemalloc;        /* If set by the page allocator,
  58                                                  * ALLOC_NO_WATERMARKS was set
  59                                                  * and the low watermark was not
  60                                                  * met implying that the system
  61                                                  * is under some pressure. The
  62                                                  * caller should try ensure
  63                                                  * this page is only used to
  64                                                  * free other pages.
  65                                                  */
  66                 };
  67
  68                 union {
  69 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  70         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  71                         /* Used for cmpxchg_double in slub */
  72                         unsigned long counters;
  73 #else
  74                         /*
  75                          * Keep _count separate from slub cmpxchg_double data.
  76                          * As the rest of the double word is protected by
  77                          * slab_lock but _count is not.
  78                          */
  79                         unsigned counters;
  80 #endif
  81
  82                         struct {
  83
  84                                 union {
  85                                         /*
  86                                          * Count of ptes mapped in
  87                                          * mms, to show when page is
  88                                          * mapped & limit reverse map
  89                                          * searches.
  90                                          *
  91                                          * Used also for tail pages
  92                                          * refcounting instead of
  93                                          * _count. Tail pages cannot
  94                                          * be mapped and keeping the
  95                                          * tail page _count zero at
  96                                          * all times guarantees
  97                                          * get_page_unless_zero() will
  98                                          * never succeed on tail
  99                                          * pages.
 100                                          */
 101                                         atomic_t _mapcount;
 102
 103                                         struct { /* SLUB */
 104                                                 unsigned inuse:16;
 105                                                 unsigned objects:15;
 106                                                 unsigned frozen:1;
 107                                         };
 108                                         int units;      /* SLOB */
 109                                 };
 110                                 atomic_t _count;                /* Usage count, see below. */
 111                         };
 112                 };
 113         };
 114
 115         /* Third double word block */
 116         union {
 117                 struct list_head lru;   /* Pageout list, eg. active_list
 118                                          * protected by zone->lru_lock !
 119                                          */
 120                 struct {                /* slub per cpu partial pages */
 121                         struct page *next;      /* Next partial slab */
 122 #ifdef CONFIG_64BIT
 123                         int pages;      /* Nr of partial slabs left */
 124                         int pobjects;   /* Approximate # of objects */
 125 #else
 126                         short int pages;
 127                         short int pobjects;
 128 #endif
 129                 };
 130
 131                 struct list_head list;  /* slobs list of pages */
 132                 struct {                /* slab fields */
 133                         struct kmem_cache *slab_cache;
 134                         struct slab *slab_page;
 135                 };
 136         };
 137
 138         /* Remainder is not double word aligned */
 139         union {
 140                 unsigned long private;          /* Mapping-private opaque data:
 141                                                  * usually used for buffer_heads
 142                                                  * if PagePrivate set; used for
 143                                                  * swp_entry_t if PageSwapCache;
 144                                                  * indicates order in the buddy
 145                                                  * system if PG_buddy is set.
 146                                                  */
 147 #if USE_SPLIT_PTLOCKS
 148                 spinlock_t ptl;
 149 #endif
 150                 struct kmem_cache *slab;        /* SLUB: Pointer to slab */
 151                 struct page *first_page;        /* Compound tail pages */
 152         };
 153
 154         /*
 155          * On machines where all RAM is mapped into kernel address space,
 156          * we can simply calculate the virtual address. On machines with
 157          * highmem some memory is mapped into kernel virtual memory
 158          * dynamically, so we need a place to store that address.
 159          * Note that this field could be 16 bits on x86 ... ;)
 160          *
 161          * Architectures with slow multiplication can define
 162          * WANT_PAGE_VIRTUAL in asm/page.h
 163          */
 164 #if defined(WANT_PAGE_VIRTUAL)
 165         void *virtual;                  /* Kernel virtual address (NULL if
 166                                            not kmapped, ie. highmem) */
 167 #endif /* WANT_PAGE_VIRTUAL */
 168 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
 169         unsigned long debug_flags;      /* Use atomic bitops on this */
 170 #endif
 171
 172 #ifdef CONFIG_KMEMCHECK
 173         /*
 174          * kmemcheck wants to track the status of each byte in a page; this
 175          * is a pointer to such a status block. NULL if not tracked.
 176          */
 177         void *shadow;
 178 #endif
 179 }
 180 /*
 181  * The struct page can be forced to be double word aligned so that atomic ops
 182  * on double words work. The SLUB allocator can make use of such a feature.
 183  */
 184 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 185         __aligned(2 * sizeof(unsigned long))
 186 #endif
 187 ;
 188
 189 struct page_frag {
 190         struct page *page;
 191 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 192         __u32 offset;
 193         __u32 size;
 194 #else
 195         __u16 offset;
 196         __u16 size;
 197 #endif
 198 };
 199
 200 typedef unsigned long __nocast vm_flags_t;
 201
 202 /*
 203  * A region containing a mapping of a non-memory backed file under NOMMU
 204  * conditions.  These are held in a global tree and are pinned by the VMAs that
 205  * map parts of them.
 206  */
 207 struct vm_region {
 208         struct rb_node  vm_rb;          /* link in global region tree */
 209         vm_flags_t      vm_flags;       /* VMA vm_flags */
 210         unsigned long   vm_start;       /* start address of region */
 211         unsigned long   vm_end;         /* region initialised to here */
 212         unsigned long   vm_top;         /* region allocated to here */
 213         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 214         struct file     *vm_file;       /* the backing file or NULL */
 215
 216         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 217         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 218                                                 * this region */
 219 };
 220
 221 /*
 222  * This struct defines a memory VMM memory area. There is one of these
 223  * per VM-area/task.  A VM area is any part of the process virtual memory
 224  * space that has a special rule for the page-fault handlers (ie a shared
 225  * library, the executable area etc).
 226  */
 227 struct vm_area_struct {
 228         struct mm_struct * vm_mm;       /* The address space we belong to. */
 229         unsigned long vm_start;         /* Our start address within vm_mm. */
 230         unsigned long vm_end;           /* The first byte after our end address
 231                                            within vm_mm. */
 232
 233         /* linked list of VM areas per task, sorted by address */
 234         struct vm_area_struct *vm_next, *vm_prev;
 235
 236         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 237         unsigned long vm_flags;         /* Flags, see mm.h. */
 238
 239         struct rb_node vm_rb;
 240
 241         /*
 242          * For areas with an address space and backing store,
 243          * linkage into the address_space->i_mmap prio tree, or
 244          * linkage to the list of like vmas hanging off its node, or
 245          * linkage of vma in the address_space->i_mmap_nonlinear list.
 246          */
 247         union {
 248                 struct {
 249                         struct list_head list;
 250                         void *parent;   /* aligns with prio_tree_node parent */
 251                         struct vm_area_struct *head;
 252                 } vm_set;
 253
 254                 struct raw_prio_tree_node prio_tree_node;
 255         } shared;
 256
 257         /*
 258          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 259          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 260          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 261          * or brk vma (with NULL file) can only be in an anon_vma list.
 262          */
 263         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 264                                           * page_table_lock */
 265         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 266
 267         /* Function pointers to deal with this struct. */
 268         const struct vm_operations_struct *vm_ops;
 269
 270         /* Information about our backing store: */
 271         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 272                                            units, *not* PAGE_CACHE_SIZE */
 273         struct file * vm_file;          /* File we map to (can be NULL). */
 274         void * vm_private_data;         /* was vm_pte (shared mem) */
 275
 276 #ifndef CONFIG_MMU
 277         struct vm_region *vm_region;    /* NOMMU mapping region */
 278 #endif
 279 #ifdef CONFIG_NUMA
 280         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 281 #endif
 282 };
 283
 284 struct core_thread {
 285         struct task_struct *task;
 286         struct core_thread *next;
 287 };
 288
 289 struct core_state {
 290         atomic_t nr_threads;
 291         struct core_thread dumper;
 292         struct completion startup;
 293 };
 294
 295 enum {
 296         MM_FILEPAGES,
 297         MM_ANONPAGES,
 298         MM_SWAPENTS,
 299         NR_MM_COUNTERS
 300 };
 301
 302 #if USE_SPLIT_PTLOCKS && defined(CONFIG_MMU)
 303 #define SPLIT_RSS_COUNTING
 304 /* per-thread cached information, */
 305 struct task_rss_stat {
 306         int events;     /* for synchronization threshold */
 307         int count[NR_MM_COUNTERS];
 308 };
 309 #endif /* USE_SPLIT_PTLOCKS */
 310
 311 struct mm_rss_stat {
 312         atomic_long_t count[NR_MM_COUNTERS];
 313 };
 314
 315 struct mm_struct {
 316         struct vm_area_struct * mmap;           /* list of VMAs */
 317         struct rb_root mm_rb;
 318         struct vm_area_struct * mmap_cache;     /* last find_vma result */
 319 #ifdef CONFIG_MMU
 320         unsigned long (*get_unmapped_area) (struct file *filp,
 321                                 unsigned long addr, unsigned long len,
 322                                 unsigned long pgoff, unsigned long flags);
 323         void (*unmap_area) (struct mm_struct *mm, unsigned long addr);
 324 #endif
 325         unsigned long mmap_base;                /* base of mmap area */
 326         unsigned long task_size;                /* size of task vm space */
 327         unsigned long cached_hole_size;         /* if non-zero, the largest hole below free_area_cache */
 328         unsigned long free_area_cache;          /* first hole of size cached_hole_size or larger */
 329         pgd_t * pgd;
 330         atomic_t mm_users;                      /* How many users with user space? */
 331         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 332         int map_count;                          /* number of VMAs */
 333
 334         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 335         struct rw_semaphore mmap_sem;
 336
 337         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 338                                                  * together off init_mm.mmlist, and are protected
 339                                                  * by mmlist_lock
 340                                                  */
 341
 342
 343         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 344         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 345
 346         unsigned long total_vm;         /* Total pages mapped */
 347         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 348         unsigned long pinned_vm;        /* Refcount permanently increased */
 349         unsigned long shared_vm;        /* Shared pages (files) */
 350         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 351         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 352         unsigned long reserved_vm;      /* VM_RESERVED|VM_IO pages */
 353         unsigned long def_flags;
 354         unsigned long nr_ptes;          /* Page table pages */
 355         unsigned long start_code, end_code, start_data, end_data;
 356         unsigned long start_brk, brk, start_stack;
 357         unsigned long arg_start, arg_end, env_start, env_end;
 358
 359         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 360
 361         /*
 362          * Special counters, in some configurations protected by the
 363          * page_table_lock, in other configurations by being atomic.
 364          */
 365         struct mm_rss_stat rss_stat;
 366
 367         struct linux_binfmt *binfmt;
 368
 369         cpumask_var_t cpu_vm_mask_var;
 370
 371         /* Architecture-specific MM context */
 372         mm_context_t context;
 373
 374         unsigned long flags; /* Must use atomic bitops to access the bits */
 375
 376         struct core_state *core_state; /* coredumping support */
 377 #ifdef CONFIG_AIO
 378         spinlock_t              ioctx_lock;
 379         struct hlist_head       ioctx_list;
 380 #endif
 381 #ifdef CONFIG_MM_OWNER
 382         /*
 383          * "owner" points to a task that is regarded as the canonical
 384          * user/owner of this mm. All of the following must be true in
 385          * order for it to be changed:
 386          *
 387          * current == mm->owner
 388          * current->mm != mm
 389          * new_owner->mm == mm
 390          * new_owner->alloc_lock is held
 391          */
 392         struct task_struct __rcu *owner;
 393 #endif
 394
 395         /* store ref to file /proc/<pid>/exe symlink points to */
 396         struct file *exe_file;
 397         unsigned long num_exe_file_vmas;
 398 #ifdef CONFIG_MMU_NOTIFIER
 399         struct mmu_notifier_mm *mmu_notifier_mm;
 400 #endif
 401 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 402         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 403 #endif
 404 #ifdef CONFIG_CPUMASK_OFFSTACK
 405         struct cpumask cpumask_allocation;
 406 #endif
 407         struct uprobes_state uprobes_state;
 408 };
 409
 410 static inline void mm_init_cpumask(struct mm_struct *mm)
 411 {
 412 #ifdef CONFIG_CPUMASK_OFFSTACK
 413         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 414 #endif
 415 }
 416
 417 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 418 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 419 {
 420         return mm->cpu_vm_mask_var;
 421 }
 422
 423 #endif /* _LINUX_MM_TYPES_H */