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/rbtree.h>
  10 #include <linux/rwsem.h>
  11 #include <linux/completion.h>
  12 #include <linux/cpumask.h>
  13 #include <linux/page-debug-flags.h>
  14 #include <linux/uprobes.h>
  15 #include <asm/page.h>
  16 #include <asm/mmu.h>
  17
  18 #ifndef AT_VECTOR_SIZE_ARCH
  19 #define AT_VECTOR_SIZE_ARCH 0
  20 #endif
  21 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
  22
  23 struct address_space;
  24
  25 #define USE_SPLIT_PTLOCKS       (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
  26
  27 /*
  28  * Each physical page in the system has a struct page associated with
  29  * it to keep track of whatever it is we are using the page for at the
  30  * moment. Note that we have no way to track which tasks are using
  31  * a page, though if it is a pagecache page, rmap structures can tell us
  32  * who is mapping it.
  33  *
  34  * The objects in struct page are organized in double word blocks in
  35  * order to allows us to use atomic double word operations on portions
  36  * of struct page. That is currently only used by slub but the arrangement
  37  * allows the use of atomic double word operations on the flags/mapping
  38  * and lru list pointers also.
  39  */
  40 struct page {
  41         /* First double word block */
  42         unsigned long flags;            /* Atomic flags, some possibly
  43                                          * updated asynchronously */
  44         struct address_space *mapping;  /* If low bit clear, points to
  45                                          * inode address_space, or NULL.
  46                                          * If page mapped as anonymous
  47                                          * memory, low bit is set, and
  48                                          * it points to anon_vma object:
  49                                          * see PAGE_MAPPING_ANON below.
  50                                          */
  51         /* Second double word */
  52         struct {
  53                 union {
  54                         pgoff_t index;          /* Our offset within mapping. */
  55                         void *freelist;         /* slub/slob first free object */
  56                         bool pfmemalloc;        /* If set by the page allocator,
  57                                                  * ALLOC_NO_WATERMARKS was set
  58                                                  * and the low watermark was not
  59                                                  * met implying that the system
  60                                                  * is under some pressure. The
  61                                                  * caller should try ensure
  62                                                  * this page is only used to
  63                                                  * free other pages.
  64                                                  */
  65                 };
  66
  67                 union {
  68 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  69         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  70                         /* Used for cmpxchg_double in slub */
  71                         unsigned long counters;
  72 #else
  73                         /*
  74                          * Keep _count separate from slub cmpxchg_double data.
  75                          * As the rest of the double word is protected by
  76                          * slab_lock but _count is not.
  77                          */
  78                         unsigned counters;
  79 #endif
  80
  81                         struct {
  82
  83                                 union {
  84                                         /*
  85                                          * Count of ptes mapped in
  86                                          * mms, to show when page is
  87                                          * mapped & limit reverse map
  88                                          * searches.
  89                                          *
  90                                          * Used also for tail pages
  91                                          * refcounting instead of
  92                                          * _count. Tail pages cannot
  93                                          * be mapped and keeping the
  94                                          * tail page _count zero at
  95                                          * all times guarantees
  96                                          * get_page_unless_zero() will
  97                                          * never succeed on tail
  98                                          * pages.
  99                                          */
 100                                         atomic_t _mapcount;
 101
 102                                         struct { /* SLUB */
 103                                                 unsigned inuse:16;
 104                                                 unsigned objects:15;
 105                                                 unsigned frozen:1;
 106                                         };
 107                                         int units;      /* SLOB */
 108                                 };
 109                                 atomic_t _count;                /* Usage count, see below. */
 110                         };
 111                 };
 112         };
 113
 114         /* Third double word block */
 115         union {
 116                 struct list_head lru;   /* Pageout list, eg. active_list
 117                                          * protected by zone->lru_lock !
 118                                          */
 119                 struct {                /* slub per cpu partial pages */
 120                         struct page *next;      /* Next partial slab */
 121 #ifdef CONFIG_64BIT
 122                         int pages;      /* Nr of partial slabs left */
 123                         int pobjects;   /* Approximate # of objects */
 124 #else
 125                         short int pages;
 126                         short int pobjects;
 127 #endif
 128                 };
 129
 130                 struct list_head list;  /* slobs list of pages */
 131                 struct {                /* slab fields */
 132                         struct kmem_cache *slab_cache;
 133                         struct slab *slab_page;
 134                 };
 135         };
 136
 137         /* Remainder is not double word aligned */
 138         union {
 139                 unsigned long private;          /* Mapping-private opaque data:
 140                                                  * usually used for buffer_heads
 141                                                  * if PagePrivate set; used for
 142                                                  * swp_entry_t if PageSwapCache;
 143                                                  * indicates order in the buddy
 144                                                  * system if PG_buddy is set.
 145                                                  */
 146 #if USE_SPLIT_PTLOCKS
 147                 spinlock_t ptl;
 148 #endif
 149                 struct kmem_cache *slab;        /* SLUB: Pointer to slab */
 150                 struct page *first_page;        /* Compound tail pages */
 151         };
 152
 153         /*
 154          * On machines where all RAM is mapped into kernel address space,
 155          * we can simply calculate the virtual address. On machines with
 156          * highmem some memory is mapped into kernel virtual memory
 157          * dynamically, so we need a place to store that address.
 158          * Note that this field could be 16 bits on x86 ... ;)
 159          *
 160          * Architectures with slow multiplication can define
 161          * WANT_PAGE_VIRTUAL in asm/page.h
 162          */
 163 #if defined(WANT_PAGE_VIRTUAL)
 164         void *virtual;                  /* Kernel virtual address (NULL if
 165                                            not kmapped, ie. highmem) */
 166 #endif /* WANT_PAGE_VIRTUAL */
 167 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
 168         unsigned long debug_flags;      /* Use atomic bitops on this */
 169 #endif
 170
 171 #ifdef CONFIG_KMEMCHECK
 172         /*
 173          * kmemcheck wants to track the status of each byte in a page; this
 174          * is a pointer to such a status block. NULL if not tracked.
 175          */
 176         void *shadow;
 177 #endif
 178
 179 #ifdef CONFIG_NUMA_BALANCING
 180         int _last_nid;
 181 #endif
 182 }
 183 /*
 184  * The struct page can be forced to be double word aligned so that atomic ops
 185  * on double words work. The SLUB allocator can make use of such a feature.
 186  */
 187 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 188         __aligned(2 * sizeof(unsigned long))
 189 #endif
 190 ;
 191
 192 struct page_frag {
 193         struct page *page;
 194 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 195         __u32 offset;
 196         __u32 size;
 197 #else
 198         __u16 offset;
 199         __u16 size;
 200 #endif
 201 };
 202
 203 typedef unsigned long __nocast vm_flags_t;
 204
 205 /*
 206  * A region containing a mapping of a non-memory backed file under NOMMU
 207  * conditions.  These are held in a global tree and are pinned by the VMAs that
 208  * map parts of them.
 209  */
 210 struct vm_region {
 211         struct rb_node  vm_rb;          /* link in global region tree */
 212         vm_flags_t      vm_flags;       /* VMA vm_flags */
 213         unsigned long   vm_start;       /* start address of region */
 214         unsigned long   vm_end;         /* region initialised to here */
 215         unsigned long   vm_top;         /* region allocated to here */
 216         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 217         struct file     *vm_file;       /* the backing file or NULL */
 218
 219         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 220         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 221                                                 * this region */
 222 };
 223
 224 /*
 225  * This struct defines a memory VMM memory area. There is one of these
 226  * per VM-area/task.  A VM area is any part of the process virtual memory
 227  * space that has a special rule for the page-fault handlers (ie a shared
 228  * library, the executable area etc).
 229  */
 230 struct vm_area_struct {
 231         struct mm_struct * vm_mm;       /* The address space we belong to. */
 232         unsigned long vm_start;         /* Our start address within vm_mm. */
 233         unsigned long vm_end;           /* The first byte after our end address
 234                                            within vm_mm. */
 235
 236         /* linked list of VM areas per task, sorted by address */
 237         struct vm_area_struct *vm_next, *vm_prev;
 238
 239         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 240         unsigned long vm_flags;         /* Flags, see mm.h. */
 241
 242         struct rb_node vm_rb;
 243
 244         /*
 245          * For areas with an address space and backing store,
 246          * linkage into the address_space->i_mmap interval tree, or
 247          * linkage of vma in the address_space->i_mmap_nonlinear list.
 248          */
 249         union {
 250                 struct {
 251                         struct rb_node rb;
 252                         unsigned long rb_subtree_last;
 253                 } linear;
 254                 struct list_head nonlinear;
 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 def_flags;
 353         unsigned long nr_ptes;          /* Page table pages */
 354         unsigned long start_code, end_code, start_data, end_data;
 355         unsigned long start_brk, brk, start_stack;
 356         unsigned long arg_start, arg_end, env_start, env_end;
 357
 358         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 359
 360         /*
 361          * Special counters, in some configurations protected by the
 362          * page_table_lock, in other configurations by being atomic.
 363          */
 364         struct mm_rss_stat rss_stat;
 365
 366         struct linux_binfmt *binfmt;
 367
 368         cpumask_var_t cpu_vm_mask_var;
 369
 370         /* Architecture-specific MM context */
 371         mm_context_t context;
 372
 373         unsigned long flags; /* Must use atomic bitops to access the bits */
 374
 375         struct core_state *core_state; /* coredumping support */
 376 #ifdef CONFIG_AIO
 377         spinlock_t              ioctx_lock;
 378         struct hlist_head       ioctx_list;
 379 #endif
 380 #ifdef CONFIG_MM_OWNER
 381         /*
 382          * "owner" points to a task that is regarded as the canonical
 383          * user/owner of this mm. All of the following must be true in
 384          * order for it to be changed:
 385          *
 386          * current == mm->owner
 387          * current->mm != mm
 388          * new_owner->mm == mm
 389          * new_owner->alloc_lock is held
 390          */
 391         struct task_struct __rcu *owner;
 392 #endif
 393
 394         /* store ref to file /proc/<pid>/exe symlink points to */
 395         struct file *exe_file;
 396 #ifdef CONFIG_MMU_NOTIFIER
 397         struct mmu_notifier_mm *mmu_notifier_mm;
 398 #endif
 399 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 400         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 401 #endif
 402 #ifdef CONFIG_CPUMASK_OFFSTACK
 403         struct cpumask cpumask_allocation;
 404 #endif
 405 #ifdef CONFIG_NUMA_BALANCING
 406         /*
 407          * numa_next_scan is the next time when the PTEs will me marked
 408          * pte_numa to gather statistics and migrate pages to new nodes
 409          * if necessary
 410          */
 411         unsigned long numa_next_scan;
 412
 413         /* numa_next_reset is when the PTE scanner period will be reset */
 414         unsigned long numa_next_reset;
 415
 416         /* Restart point for scanning and setting pte_numa */
 417         unsigned long numa_scan_offset;
 418
 419         /* numa_scan_seq prevents two threads setting pte_numa */
 420         int numa_scan_seq;
 421
 422         /*
 423          * The first node a task was scheduled on. If a task runs on
 424          * a different node than Make PTE Scan Go Now.
 425          */
 426         int first_nid;
 427 #endif
 428         struct uprobes_state uprobes_state;
 429 };
 430
 431 /* first nid will either be a valid NID or one of these values */
 432 #define NUMA_PTE_SCAN_INIT      -1
 433 #define NUMA_PTE_SCAN_ACTIVE    -2
 434
 435 static inline void mm_init_cpumask(struct mm_struct *mm)
 436 {
 437 #ifdef CONFIG_CPUMASK_OFFSTACK
 438         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 439 #endif
 440 }
 441
 442 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 443 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 444 {
 445         return mm->cpu_vm_mask_var;
 446 }
 447
 448 #endif /* _LINUX_MM_TYPES_H */