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_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 *slab_page; /* slab fields */
 133         };
 134
 135         /* Remainder is not double word aligned */
 136         union {
 137                 unsigned long private;          /* Mapping-private opaque data:
 138                                                  * usually used for buffer_heads
 139                                                  * if PagePrivate set; used for
 140                                                  * swp_entry_t if PageSwapCache;
 141                                                  * indicates order in the buddy
 142                                                  * system if PG_buddy is set.
 143                                                  */
 144 #if USE_SPLIT_PTLOCKS
 145                 spinlock_t ptl;
 146 #endif
 147                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 148                 struct page *first_page;        /* Compound tail pages */
 149         };
 150
 151         /*
 152          * On machines where all RAM is mapped into kernel address space,
 153          * we can simply calculate the virtual address. On machines with
 154          * highmem some memory is mapped into kernel virtual memory
 155          * dynamically, so we need a place to store that address.
 156          * Note that this field could be 16 bits on x86 ... ;)
 157          *
 158          * Architectures with slow multiplication can define
 159          * WANT_PAGE_VIRTUAL in asm/page.h
 160          */
 161 #if defined(WANT_PAGE_VIRTUAL)
 162         void *virtual;                  /* Kernel virtual address (NULL if
 163                                            not kmapped, ie. highmem) */
 164 #endif /* WANT_PAGE_VIRTUAL */
 165 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
 166         unsigned long debug_flags;      /* Use atomic bitops on this */
 167 #endif
 168
 169 #ifdef CONFIG_KMEMCHECK
 170         /*
 171          * kmemcheck wants to track the status of each byte in a page; this
 172          * is a pointer to such a status block. NULL if not tracked.
 173          */
 174         void *shadow;
 175 #endif
 176
 177 #ifdef LAST_NID_NOT_IN_PAGE_FLAGS
 178         int _last_nid;
 179 #endif
 180 }
 181 /*
 182  * The struct page can be forced to be double word aligned so that atomic ops
 183  * on double words work. The SLUB allocator can make use of such a feature.
 184  */
 185 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 186         __aligned(2 * sizeof(unsigned long))
 187 #endif
 188 ;
 189
 190 struct page_frag {
 191         struct page *page;
 192 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 193         __u32 offset;
 194         __u32 size;
 195 #else
 196         __u16 offset;
 197         __u16 size;
 198 #endif
 199 };
 200
 201 typedef unsigned long __nocast vm_flags_t;
 202
 203 /*
 204  * A region containing a mapping of a non-memory backed file under NOMMU
 205  * conditions.  These are held in a global tree and are pinned by the VMAs that
 206  * map parts of them.
 207  */
 208 struct vm_region {
 209         struct rb_node  vm_rb;          /* link in global region tree */
 210         vm_flags_t      vm_flags;       /* VMA vm_flags */
 211         unsigned long   vm_start;       /* start address of region */
 212         unsigned long   vm_end;         /* region initialised to here */
 213         unsigned long   vm_top;         /* region allocated to here */
 214         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 215         struct file     *vm_file;       /* the backing file or NULL */
 216
 217         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 218         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 219                                                 * this region */
 220 };
 221
 222 /*
 223  * This struct defines a memory VMM memory area. There is one of these
 224  * per VM-area/task.  A VM area is any part of the process virtual memory
 225  * space that has a special rule for the page-fault handlers (ie a shared
 226  * library, the executable area etc).
 227  */
 228 struct vm_area_struct {
 229         /* The first cache line has the info for VMA tree walking. */
 230
 231         unsigned long vm_start;         /* Our start address within vm_mm. */
 232         unsigned long vm_end;           /* The first byte after our end address
 233                                            within vm_mm. */
 234
 235         /* linked list of VM areas per task, sorted by address */
 236         struct vm_area_struct *vm_next, *vm_prev;
 237
 238         struct rb_node vm_rb;
 239
 240         /*
 241          * Largest free memory gap in bytes to the left of this VMA.
 242          * Either between this VMA and vma->vm_prev, or between one of the
 243          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 244          * get_unmapped_area find a free area of the right size.
 245          */
 246         unsigned long rb_subtree_gap;
 247
 248         /* Second cache line starts here. */
 249
 250         struct mm_struct *vm_mm;        /* The address space we belong to. */
 251         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 252         unsigned long vm_flags;         /* Flags, see mm.h. */
 253
 254         /*
 255          * For areas with an address space and backing store,
 256          * linkage into the address_space->i_mmap interval tree, or
 257          * linkage of vma in the address_space->i_mmap_nonlinear list.
 258          */
 259         union {
 260                 struct {
 261                         struct rb_node rb;
 262                         unsigned long rb_subtree_last;
 263                 } linear;
 264                 struct list_head nonlinear;
 265         } shared;
 266
 267         /*
 268          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 269          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 270          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 271          * or brk vma (with NULL file) can only be in an anon_vma list.
 272          */
 273         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 274                                           * page_table_lock */
 275         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 276
 277         /* Function pointers to deal with this struct. */
 278         const struct vm_operations_struct *vm_ops;
 279
 280         /* Information about our backing store: */
 281         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 282                                            units, *not* PAGE_CACHE_SIZE */
 283         struct file * vm_file;          /* File we map to (can be NULL). */
 284         void * vm_private_data;         /* was vm_pte (shared mem) */
 285
 286 #ifndef CONFIG_MMU
 287         struct vm_region *vm_region;    /* NOMMU mapping region */
 288 #endif
 289 #ifdef CONFIG_NUMA
 290         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 291 #endif
 292 };
 293
 294 struct core_thread {
 295         struct task_struct *task;
 296         struct core_thread *next;
 297 };
 298
 299 struct core_state {
 300         atomic_t nr_threads;
 301         struct core_thread dumper;
 302         struct completion startup;
 303 };
 304
 305 enum {
 306         MM_FILEPAGES,
 307         MM_ANONPAGES,
 308         MM_SWAPENTS,
 309         NR_MM_COUNTERS
 310 };
 311
 312 #if USE_SPLIT_PTLOCKS && defined(CONFIG_MMU)
 313 #define SPLIT_RSS_COUNTING
 314 /* per-thread cached information, */
 315 struct task_rss_stat {
 316         int events;     /* for synchronization threshold */
 317         int count[NR_MM_COUNTERS];
 318 };
 319 #endif /* USE_SPLIT_PTLOCKS */
 320
 321 struct mm_rss_stat {
 322         atomic_long_t count[NR_MM_COUNTERS];
 323 };
 324
 325 struct mm_struct {
 326         struct vm_area_struct * mmap;           /* list of VMAs */
 327         struct rb_root mm_rb;
 328         struct vm_area_struct * mmap_cache;     /* last find_vma result */
 329 #ifdef CONFIG_MMU
 330         unsigned long (*get_unmapped_area) (struct file *filp,
 331                                 unsigned long addr, unsigned long len,
 332                                 unsigned long pgoff, unsigned long flags);
 333         void (*unmap_area) (struct mm_struct *mm, unsigned long addr);
 334 #endif
 335         unsigned long mmap_base;                /* base of mmap area */
 336         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 337         unsigned long task_size;                /* size of task vm space */
 338         unsigned long cached_hole_size;         /* if non-zero, the largest hole below free_area_cache */
 339         unsigned long free_area_cache;          /* first hole of size cached_hole_size or larger */
 340         unsigned long highest_vm_end;           /* highest vma end address */
 341         pgd_t * pgd;
 342         atomic_t mm_users;                      /* How many users with user space? */
 343         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 344         int map_count;                          /* number of VMAs */
 345
 346         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 347         struct rw_semaphore mmap_sem;
 348
 349         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 350                                                  * together off init_mm.mmlist, and are protected
 351                                                  * by mmlist_lock
 352                                                  */
 353
 354
 355         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 356         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 357
 358         unsigned long total_vm;         /* Total pages mapped */
 359         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 360         unsigned long pinned_vm;        /* Refcount permanently increased */
 361         unsigned long shared_vm;        /* Shared pages (files) */
 362         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 363         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 364         unsigned long def_flags;
 365         unsigned long nr_ptes;          /* Page table pages */
 366         unsigned long start_code, end_code, start_data, end_data;
 367         unsigned long start_brk, brk, start_stack;
 368         unsigned long arg_start, arg_end, env_start, env_end;
 369
 370         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 371
 372         /*
 373          * Special counters, in some configurations protected by the
 374          * page_table_lock, in other configurations by being atomic.
 375          */
 376         struct mm_rss_stat rss_stat;
 377
 378         struct linux_binfmt *binfmt;
 379
 380         cpumask_var_t cpu_vm_mask_var;
 381
 382         /* Architecture-specific MM context */
 383         mm_context_t context;
 384
 385         unsigned long flags; /* Must use atomic bitops to access the bits */
 386
 387         struct core_state *core_state; /* coredumping support */
 388 #ifdef CONFIG_AIO
 389         spinlock_t              ioctx_lock;
 390         struct hlist_head       ioctx_list;
 391 #endif
 392 #ifdef CONFIG_MM_OWNER
 393         /*
 394          * "owner" points to a task that is regarded as the canonical
 395          * user/owner of this mm. All of the following must be true in
 396          * order for it to be changed:
 397          *
 398          * current == mm->owner
 399          * current->mm != mm
 400          * new_owner->mm == mm
 401          * new_owner->alloc_lock is held
 402          */
 403         struct task_struct __rcu *owner;
 404 #endif
 405
 406         /* store ref to file /proc/<pid>/exe symlink points to */
 407         struct file *exe_file;
 408 #ifdef CONFIG_MMU_NOTIFIER
 409         struct mmu_notifier_mm *mmu_notifier_mm;
 410 #endif
 411 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 412         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 413 #endif
 414 #ifdef CONFIG_CPUMASK_OFFSTACK
 415         struct cpumask cpumask_allocation;
 416 #endif
 417 #ifdef CONFIG_NUMA_BALANCING
 418         /*
 419          * numa_next_scan is the next time that the PTEs will be marked
 420          * pte_numa. NUMA hinting faults will gather statistics and migrate
 421          * pages to new nodes if necessary.
 422          */
 423         unsigned long numa_next_scan;
 424
 425         /* numa_next_reset is when the PTE scanner period will be reset */
 426         unsigned long numa_next_reset;
 427
 428         /* Restart point for scanning and setting pte_numa */
 429         unsigned long numa_scan_offset;
 430
 431         /* numa_scan_seq prevents two threads setting pte_numa */
 432         int numa_scan_seq;
 433
 434         /*
 435          * The first node a task was scheduled on. If a task runs on
 436          * a different node than Make PTE Scan Go Now.
 437          */
 438         int first_nid;
 439 #endif
 440 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 441         /*
 442          * An operation with batched TLB flushing is going on. Anything that
 443          * can move process memory needs to flush the TLB when moving a
 444          * PROT_NONE or PROT_NUMA mapped page.
 445          */
 446         bool tlb_flush_pending;
 447 #endif
 448         struct uprobes_state uprobes_state;
 449 };
 450
 451 /* first nid will either be a valid NID or one of these values */
 452 #define NUMA_PTE_SCAN_INIT      -1
 453 #define NUMA_PTE_SCAN_ACTIVE    -2
 454
 455 static inline void mm_init_cpumask(struct mm_struct *mm)
 456 {
 457 #ifdef CONFIG_CPUMASK_OFFSTACK
 458         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 459 #endif
 460 }
 461
 462 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 463 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 464 {
 465         return mm->cpu_vm_mask_var;
 466 }
 467
 468 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 469 /*
 470  * Memory barriers to keep this state in sync are graciously provided by
 471  * the page table locks, outside of which no page table modifications happen.
 472  * The barriers below prevent the compiler from re-ordering the instructions
 473  * around the memory barriers that are already present in the code.
 474  */
 475 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 476 {
 477         barrier();
 478         return mm->tlb_flush_pending;
 479 }
 480 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 481 {
 482         mm->tlb_flush_pending = true;
 483
 484         /*
 485          * Guarantee that the tlb_flush_pending store does not leak into the
 486          * critical section updating the page tables
 487          */
 488         smp_mb__before_spinlock();
 489 }
 490 /* Clearing is done after a TLB flush, which also provides a barrier. */
 491 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 492 {
 493         barrier();
 494         mm->tlb_flush_pending = false;
 495 }
 496 #else
 497 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 498 {
 499         return false;
 500 }
 501 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 502 {
 503 }
 504 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 505 {
 506 }
 507 #endif
 508
 509 #endif /* _LINUX_MM_TYPES_H */