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 <linux/page-flags-layout.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_PTE_PTLOCKS   (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
  27 #define USE_SPLIT_PMD_PTLOCKS   (USE_SPLIT_PTE_PTLOCKS && \
  28                 IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
  29
  30 /*
  31  * Each physical page in the system has a struct page associated with
  32  * it to keep track of whatever it is we are using the page for at the
  33  * moment. Note that we have no way to track which tasks are using
  34  * a page, though if it is a pagecache page, rmap structures can tell us
  35  * who is mapping it.
  36  *
  37  * The objects in struct page are organized in double word blocks in
  38  * order to allows us to use atomic double word operations on portions
  39  * of struct page. That is currently only used by slub but the arrangement
  40  * allows the use of atomic double word operations on the flags/mapping
  41  * and lru list pointers also.
  42  */
  43 struct page {
  44         /* First double word block */
  45         unsigned long flags;            /* Atomic flags, some possibly
  46                                          * updated asynchronously */
  47         struct address_space *mapping;  /* If low bit clear, points to
  48                                          * inode address_space, or NULL.
  49                                          * If page mapped as anonymous
  50                                          * memory, low bit is set, and
  51                                          * it points to anon_vma object:
  52                                          * see PAGE_MAPPING_ANON below.
  53                                          */
  54         /* Second double word */
  55         struct {
  56                 union {
  57                         pgoff_t index;          /* Our offset within mapping. */
  58                         void *freelist;         /* slub/slob first free object */
  59                         bool pfmemalloc;        /* If set by the page allocator,
  60                                                  * ALLOC_NO_WATERMARKS was set
  61                                                  * and the low watermark was not
  62                                                  * met implying that the system
  63                                                  * is under some pressure. The
  64                                                  * caller should try ensure
  65                                                  * this page is only used to
  66                                                  * free other pages.
  67                                                  */
  68 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
  69                 pgtable_t pmd_huge_pte; /* protected by page->ptl */
  70 #endif
  71                 };
  72
  73                 union {
  74 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  75         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  76                         /* Used for cmpxchg_double in slub */
  77                         unsigned long counters;
  78 #else
  79                         /*
  80                          * Keep _count separate from slub cmpxchg_double data.
  81                          * As the rest of the double word is protected by
  82                          * slab_lock but _count is not.
  83                          */
  84                         unsigned counters;
  85 #endif
  86
  87                         struct {
  88
  89                                 union {
  90                                         /*
  91                                          * Count of ptes mapped in
  92                                          * mms, to show when page is
  93                                          * mapped & limit reverse map
  94                                          * searches.
  95                                          *
  96                                          * Used also for tail pages
  97                                          * refcounting instead of
  98                                          * _count. Tail pages cannot
  99                                          * be mapped and keeping the
 100                                          * tail page _count zero at
 101                                          * all times guarantees
 102                                          * get_page_unless_zero() will
 103                                          * never succeed on tail
 104                                          * pages.
 105                                          */
 106                                         atomic_t _mapcount;
 107
 108                                         struct { /* SLUB */
 109                                                 unsigned inuse:16;
 110                                                 unsigned objects:15;
 111                                                 unsigned frozen:1;
 112                                         };
 113                                         int units;      /* SLOB */
 114                                 };
 115                                 atomic_t _count;                /* Usage count, see below. */
 116                         };
 117                 };
 118         };
 119
 120         /* Third double word block */
 121         union {
 122                 struct list_head lru;   /* Pageout list, eg. active_list
 123                                          * protected by zone->lru_lock !
 124                                          */
 125                 struct {                /* slub per cpu partial pages */
 126                         struct page *next;      /* Next partial slab */
 127 #ifdef CONFIG_64BIT
 128                         int pages;      /* Nr of partial slabs left */
 129                         int pobjects;   /* Approximate # of objects */
 130 #else
 131                         short int pages;
 132                         short int pobjects;
 133 #endif
 134                 };
 135
 136                 struct list_head list;  /* slobs list of pages */
 137                 struct slab *slab_page; /* slab fields */
 138         };
 139
 140         /* Remainder is not double word aligned */
 141         union {
 142                 unsigned long private;          /* Mapping-private opaque data:
 143                                                  * usually used for buffer_heads
 144                                                  * if PagePrivate set; used for
 145                                                  * swp_entry_t if PageSwapCache;
 146                                                  * indicates order in the buddy
 147                                                  * system if PG_buddy is set.
 148                                                  */
 149 #if USE_SPLIT_PTE_PTLOCKS
 150                 unsigned long ptl; /* It's spinlock_t if it fits to long,
 151                                     * otherwise it's pointer to dynamicaly
 152                                     * allocated spinlock_t.
 153                                     */
 154 #endif
 155                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 156                 struct page *first_page;        /* Compound tail pages */
 157         };
 158
 159         /*
 160          * On machines where all RAM is mapped into kernel address space,
 161          * we can simply calculate the virtual address. On machines with
 162          * highmem some memory is mapped into kernel virtual memory
 163          * dynamically, so we need a place to store that address.
 164          * Note that this field could be 16 bits on x86 ... ;)
 165          *
 166          * Architectures with slow multiplication can define
 167          * WANT_PAGE_VIRTUAL in asm/page.h
 168          */
 169 #if defined(WANT_PAGE_VIRTUAL)
 170         void *virtual;                  /* Kernel virtual address (NULL if
 171                                            not kmapped, ie. highmem) */
 172 #endif /* WANT_PAGE_VIRTUAL */
 173 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
 174         unsigned long debug_flags;      /* Use atomic bitops on this */
 175 #endif
 176
 177 #ifdef CONFIG_KMEMCHECK
 178         /*
 179          * kmemcheck wants to track the status of each byte in a page; this
 180          * is a pointer to such a status block. NULL if not tracked.
 181          */
 182         void *shadow;
 183 #endif
 184
 185 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 186         int _last_cpupid;
 187 #endif
 188 }
 189 /*
 190  * The struct page can be forced to be double word aligned so that atomic ops
 191  * on double words work. The SLUB allocator can make use of such a feature.
 192  */
 193 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 194         __aligned(2 * sizeof(unsigned long))
 195 #endif
 196 ;
 197
 198 struct page_frag {
 199         struct page *page;
 200 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 201         __u32 offset;
 202         __u32 size;
 203 #else
 204         __u16 offset;
 205         __u16 size;
 206 #endif
 207 };
 208
 209 typedef unsigned long __nocast vm_flags_t;
 210
 211 /*
 212  * A region containing a mapping of a non-memory backed file under NOMMU
 213  * conditions.  These are held in a global tree and are pinned by the VMAs that
 214  * map parts of them.
 215  */
 216 struct vm_region {
 217         struct rb_node  vm_rb;          /* link in global region tree */
 218         vm_flags_t      vm_flags;       /* VMA vm_flags */
 219         unsigned long   vm_start;       /* start address of region */
 220         unsigned long   vm_end;         /* region initialised to here */
 221         unsigned long   vm_top;         /* region allocated to here */
 222         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 223         struct file     *vm_file;       /* the backing file or NULL */
 224
 225         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 226         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 227                                                 * this region */
 228 };
 229
 230 /*
 231  * This struct defines a memory VMM memory area. There is one of these
 232  * per VM-area/task.  A VM area is any part of the process virtual memory
 233  * space that has a special rule for the page-fault handlers (ie a shared
 234  * library, the executable area etc).
 235  */
 236 struct vm_area_struct {
 237         /* The first cache line has the info for VMA tree walking. */
 238
 239         unsigned long vm_start;         /* Our start address within vm_mm. */
 240         unsigned long vm_end;           /* The first byte after our end address
 241                                            within vm_mm. */
 242
 243         /* linked list of VM areas per task, sorted by address */
 244         struct vm_area_struct *vm_next, *vm_prev;
 245
 246         struct rb_node vm_rb;
 247
 248         /*
 249          * Largest free memory gap in bytes to the left of this VMA.
 250          * Either between this VMA and vma->vm_prev, or between one of the
 251          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 252          * get_unmapped_area find a free area of the right size.
 253          */
 254         unsigned long rb_subtree_gap;
 255
 256         /* Second cache line starts here. */
 257
 258         struct mm_struct *vm_mm;        /* The address space we belong to. */
 259         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 260         unsigned long vm_flags;         /* Flags, see mm.h. */
 261
 262         /*
 263          * For areas with an address space and backing store,
 264          * linkage into the address_space->i_mmap interval tree, or
 265          * linkage of vma in the address_space->i_mmap_nonlinear list.
 266          */
 267         union {
 268                 struct {
 269                         struct rb_node rb;
 270                         unsigned long rb_subtree_last;
 271                 } linear;
 272                 struct list_head nonlinear;
 273         } shared;
 274
 275         /*
 276          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 277          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 278          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 279          * or brk vma (with NULL file) can only be in an anon_vma list.
 280          */
 281         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 282                                           * page_table_lock */
 283         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 284
 285         /* Function pointers to deal with this struct. */
 286         const struct vm_operations_struct *vm_ops;
 287
 288         /* Information about our backing store: */
 289         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 290                                            units, *not* PAGE_CACHE_SIZE */
 291         struct file * vm_file;          /* File we map to (can be NULL). */
 292         void * vm_private_data;         /* was vm_pte (shared mem) */
 293
 294 #ifndef CONFIG_MMU
 295         struct vm_region *vm_region;    /* NOMMU mapping region */
 296 #endif
 297 #ifdef CONFIG_NUMA
 298         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 299 #endif
 300 };
 301
 302 struct core_thread {
 303         struct task_struct *task;
 304         struct core_thread *next;
 305 };
 306
 307 struct core_state {
 308         atomic_t nr_threads;
 309         struct core_thread dumper;
 310         struct completion startup;
 311 };
 312
 313 enum {
 314         MM_FILEPAGES,
 315         MM_ANONPAGES,
 316         MM_SWAPENTS,
 317         NR_MM_COUNTERS
 318 };
 319
 320 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 321 #define SPLIT_RSS_COUNTING
 322 /* per-thread cached information, */
 323 struct task_rss_stat {
 324         int events;     /* for synchronization threshold */
 325         int count[NR_MM_COUNTERS];
 326 };
 327 #endif /* USE_SPLIT_PTE_PTLOCKS */
 328
 329 struct mm_rss_stat {
 330         atomic_long_t count[NR_MM_COUNTERS];
 331 };
 332
 333 struct kioctx_table;
 334 struct mm_struct {
 335         struct vm_area_struct * mmap;           /* list of VMAs */
 336         struct rb_root mm_rb;
 337         struct vm_area_struct * mmap_cache;     /* last find_vma result */
 338 #ifdef CONFIG_MMU
 339         unsigned long (*get_unmapped_area) (struct file *filp,
 340                                 unsigned long addr, unsigned long len,
 341                                 unsigned long pgoff, unsigned long flags);
 342 #endif
 343         unsigned long mmap_base;                /* base of mmap area */
 344         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 345         unsigned long task_size;                /* size of task vm space */
 346         unsigned long highest_vm_end;           /* highest vma end address */
 347         pgd_t * pgd;
 348         atomic_t mm_users;                      /* How many users with user space? */
 349         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 350         atomic_long_t nr_ptes;                  /* Page table pages */
 351         int map_count;                          /* number of VMAs */
 352
 353         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 354         struct rw_semaphore mmap_sem;
 355
 356         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 357                                                  * together off init_mm.mmlist, and are protected
 358                                                  * by mmlist_lock
 359                                                  */
 360
 361
 362         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 363         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 364
 365         unsigned long total_vm;         /* Total pages mapped */
 366         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 367         unsigned long pinned_vm;        /* Refcount permanently increased */
 368         unsigned long shared_vm;        /* Shared pages (files) */
 369         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 370         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 371         unsigned long def_flags;
 372         unsigned long start_code, end_code, start_data, end_data;
 373         unsigned long start_brk, brk, start_stack;
 374         unsigned long arg_start, arg_end, env_start, env_end;
 375
 376         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 377
 378         /*
 379          * Special counters, in some configurations protected by the
 380          * page_table_lock, in other configurations by being atomic.
 381          */
 382         struct mm_rss_stat rss_stat;
 383
 384         struct linux_binfmt *binfmt;
 385
 386         cpumask_var_t cpu_vm_mask_var;
 387
 388         /* Architecture-specific MM context */
 389         mm_context_t context;
 390
 391         unsigned long flags; /* Must use atomic bitops to access the bits */
 392
 393         struct core_state *core_state; /* coredumping support */
 394 #ifdef CONFIG_AIO
 395         spinlock_t                      ioctx_lock;
 396         struct kioctx_table __rcu       *ioctx_table;
 397 #endif
 398 #ifdef CONFIG_MM_OWNER
 399         /*
 400          * "owner" points to a task that is regarded as the canonical
 401          * user/owner of this mm. All of the following must be true in
 402          * order for it to be changed:
 403          *
 404          * current == mm->owner
 405          * current->mm != mm
 406          * new_owner->mm == mm
 407          * new_owner->alloc_lock is held
 408          */
 409         struct task_struct __rcu *owner;
 410 #endif
 411
 412         /* store ref to file /proc/<pid>/exe symlink points to */
 413         struct file *exe_file;
 414 #ifdef CONFIG_MMU_NOTIFIER
 415         struct mmu_notifier_mm *mmu_notifier_mm;
 416 #endif
 417 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 418         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 419 #endif
 420 #ifdef CONFIG_CPUMASK_OFFSTACK
 421         struct cpumask cpumask_allocation;
 422 #endif
 423 #ifdef CONFIG_NUMA_BALANCING
 424         /*
 425          * numa_next_scan is the next time that the PTEs will be marked
 426          * pte_numa. NUMA hinting faults will gather statistics and migrate
 427          * pages to new nodes if necessary.
 428          */
 429         unsigned long numa_next_scan;
 430
 431         /* Restart point for scanning and setting pte_numa */
 432         unsigned long numa_scan_offset;
 433
 434         /* numa_scan_seq prevents two threads setting pte_numa */
 435         int numa_scan_seq;
 436 #endif
 437         struct uprobes_state uprobes_state;
 438 };
 439
 440 static inline void mm_init_cpumask(struct mm_struct *mm)
 441 {
 442 #ifdef CONFIG_CPUMASK_OFFSTACK
 443         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 444 #endif
 445 }
 446
 447 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 448 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 449 {
 450         return mm->cpu_vm_mask_var;
 451 }
 452
 453 #endif /* _LINUX_MM_TYPES_H */