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/uprobes.h>
  14 #include <linux/page-flags-layout.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 struct mem_cgroup;
  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 #define ALLOC_SPLIT_PTLOCKS     (SPINLOCK_SIZE > BITS_PER_LONG/8)
  30
  31 /*
  32  * Each physical page in the system has a struct page associated with
  33  * it to keep track of whatever it is we are using the page for at the
  34  * moment. Note that we have no way to track which tasks are using
  35  * a page, though if it is a pagecache page, rmap structures can tell us
  36  * who is mapping it.
  37  *
  38  * The objects in struct page are organized in double word blocks in
  39  * order to allows us to use atomic double word operations on portions
  40  * of struct page. That is currently only used by slub but the arrangement
  41  * allows the use of atomic double word operations on the flags/mapping
  42  * and lru list pointers also.
  43  */
  44 struct page {
  45         /* First double word block */
  46         unsigned long flags;            /* Atomic flags, some possibly
  47                                          * updated asynchronously */
  48         union {
  49                 struct address_space *mapping;  /* If low bit clear, points to
  50                                                  * inode address_space, or NULL.
  51                                                  * If page mapped as anonymous
  52                                                  * memory, low bit is set, and
  53                                                  * it points to anon_vma object:
  54                                                  * see PAGE_MAPPING_ANON below.
  55                                                  */
  56                 void *s_mem;                    /* slab first object */
  57         };
  58
  59         /* Second double word */
  60         struct {
  61                 union {
  62                         pgoff_t index;          /* Our offset within mapping. */
  63                         void *freelist;         /* sl[aou]b first free object */
  64                         bool pfmemalloc;        /* If set by the page allocator,
  65                                                  * ALLOC_NO_WATERMARKS was set
  66                                                  * and the low watermark was not
  67                                                  * met implying that the system
  68                                                  * is under some pressure. The
  69                                                  * caller should try ensure
  70                                                  * this page is only used to
  71                                                  * free other pages.
  72                                                  */
  73                 };
  74
  75                 union {
  76 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  77         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  78                         /* Used for cmpxchg_double in slub */
  79                         unsigned long counters;
  80 #else
  81                         /*
  82                          * Keep _count separate from slub cmpxchg_double data.
  83                          * As the rest of the double word is protected by
  84                          * slab_lock but _count is not.
  85                          */
  86                         unsigned counters;
  87 #endif
  88
  89                         struct {
  90
  91                                 union {
  92                                         /*
  93                                          * Count of ptes mapped in
  94                                          * mms, to show when page is
  95                                          * mapped & limit reverse map
  96                                          * searches.
  97                                          *
  98                                          * Used also for tail pages
  99                                          * refcounting instead of
 100                                          * _count. Tail pages cannot
 101                                          * be mapped and keeping the
 102                                          * tail page _count zero at
 103                                          * all times guarantees
 104                                          * get_page_unless_zero() will
 105                                          * never succeed on tail
 106                                          * pages.
 107                                          */
 108                                         atomic_t _mapcount;
 109
 110                                         struct { /* SLUB */
 111                                                 unsigned inuse:16;
 112                                                 unsigned objects:15;
 113                                                 unsigned frozen:1;
 114                                         };
 115                                         int units;      /* SLOB */
 116                                 };
 117                                 atomic_t _count;                /* Usage count, see below. */
 118                         };
 119                         unsigned int active;    /* SLAB */
 120                 };
 121         };
 122
 123         /* Third double word block */
 124         union {
 125                 struct list_head lru;   /* Pageout list, eg. active_list
 126                                          * protected by zone->lru_lock !
 127                                          * Can be used as a generic list
 128                                          * by the page owner.
 129                                          */
 130                 struct {                /* slub per cpu partial pages */
 131                         struct page *next;      /* Next partial slab */
 132 #ifdef CONFIG_64BIT
 133                         int pages;      /* Nr of partial slabs left */
 134                         int pobjects;   /* Approximate # of objects */
 135 #else
 136                         short int pages;
 137                         short int pobjects;
 138 #endif
 139                 };
 140
 141                 struct slab *slab_page; /* slab fields */
 142                 struct rcu_head rcu_head;       /* Used by SLAB
 143                                                  * when destroying via RCU
 144                                                  */
 145 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
 146                 pgtable_t pmd_huge_pte; /* protected by page->ptl */
 147 #endif
 148         };
 149
 150         /* Remainder is not double word aligned */
 151         union {
 152                 unsigned long private;          /* Mapping-private opaque data:
 153                                                  * usually used for buffer_heads
 154                                                  * if PagePrivate set; used for
 155                                                  * swp_entry_t if PageSwapCache;
 156                                                  * indicates order in the buddy
 157                                                  * system if PG_buddy is set.
 158                                                  */
 159 #if USE_SPLIT_PTE_PTLOCKS
 160 #if ALLOC_SPLIT_PTLOCKS
 161                 spinlock_t *ptl;
 162 #else
 163                 spinlock_t ptl;
 164 #endif
 165 #endif
 166                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 167                 struct page *first_page;        /* Compound tail pages */
 168         };
 169
 170 #ifdef CONFIG_MEMCG
 171         struct mem_cgroup *mem_cgroup;
 172 #endif
 173
 174         /*
 175          * On machines where all RAM is mapped into kernel address space,
 176          * we can simply calculate the virtual address. On machines with
 177          * highmem some memory is mapped into kernel virtual memory
 178          * dynamically, so we need a place to store that address.
 179          * Note that this field could be 16 bits on x86 ... ;)
 180          *
 181          * Architectures with slow multiplication can define
 182          * WANT_PAGE_VIRTUAL in asm/page.h
 183          */
 184 #if defined(WANT_PAGE_VIRTUAL)
 185         void *virtual;                  /* Kernel virtual address (NULL if
 186                                            not kmapped, ie. highmem) */
 187 #endif /* WANT_PAGE_VIRTUAL */
 188
 189 #ifdef CONFIG_KMEMCHECK
 190         /*
 191          * kmemcheck wants to track the status of each byte in a page; this
 192          * is a pointer to such a status block. NULL if not tracked.
 193          */
 194         void *shadow;
 195 #endif
 196
 197 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 198         int _last_cpupid;
 199 #endif
 200 }
 201 /*
 202  * The struct page can be forced to be double word aligned so that atomic ops
 203  * on double words work. The SLUB allocator can make use of such a feature.
 204  */
 205 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 206         __aligned(2 * sizeof(unsigned long))
 207 #endif
 208 ;
 209
 210 struct page_frag {
 211         struct page *page;
 212 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 213         __u32 offset;
 214         __u32 size;
 215 #else
 216         __u16 offset;
 217         __u16 size;
 218 #endif
 219 };
 220
 221 typedef unsigned long __nocast vm_flags_t;
 222
 223 /*
 224  * A region containing a mapping of a non-memory backed file under NOMMU
 225  * conditions.  These are held in a global tree and are pinned by the VMAs that
 226  * map parts of them.
 227  */
 228 struct vm_region {
 229         struct rb_node  vm_rb;          /* link in global region tree */
 230         vm_flags_t      vm_flags;       /* VMA vm_flags */
 231         unsigned long   vm_start;       /* start address of region */
 232         unsigned long   vm_end;         /* region initialised to here */
 233         unsigned long   vm_top;         /* region allocated to here */
 234         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 235         struct file     *vm_file;       /* the backing file or NULL */
 236
 237         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 238         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 239                                                 * this region */
 240 };
 241
 242 /*
 243  * This struct defines a memory VMM memory area. There is one of these
 244  * per VM-area/task.  A VM area is any part of the process virtual memory
 245  * space that has a special rule for the page-fault handlers (ie a shared
 246  * library, the executable area etc).
 247  */
 248 struct vm_area_struct {
 249         /* The first cache line has the info for VMA tree walking. */
 250
 251         unsigned long vm_start;         /* Our start address within vm_mm. */
 252         unsigned long vm_end;           /* The first byte after our end address
 253                                            within vm_mm. */
 254
 255         /* linked list of VM areas per task, sorted by address */
 256         struct vm_area_struct *vm_next, *vm_prev;
 257
 258         struct rb_node vm_rb;
 259
 260         /*
 261          * Largest free memory gap in bytes to the left of this VMA.
 262          * Either between this VMA and vma->vm_prev, or between one of the
 263          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 264          * get_unmapped_area find a free area of the right size.
 265          */
 266         unsigned long rb_subtree_gap;
 267
 268         /* Second cache line starts here. */
 269
 270         struct mm_struct *vm_mm;        /* The address space we belong to. */
 271         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 272         unsigned long vm_flags;         /* Flags, see mm.h. */
 273
 274         /*
 275          * For areas with an address space and backing store,
 276          * linkage into the address_space->i_mmap interval tree, or
 277          * linkage of vma in the address_space->i_mmap_nonlinear list.
 278          */
 279         union {
 280                 struct {
 281                         struct rb_node rb;
 282                         unsigned long rb_subtree_last;
 283                 } linear;
 284                 struct list_head nonlinear;
 285         } shared;
 286
 287         /*
 288          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 289          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 290          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 291          * or brk vma (with NULL file) can only be in an anon_vma list.
 292          */
 293         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 294                                           * page_table_lock */
 295         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 296
 297         /* Function pointers to deal with this struct. */
 298         const struct vm_operations_struct *vm_ops;
 299
 300         /* Information about our backing store: */
 301         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 302                                            units, *not* PAGE_CACHE_SIZE */
 303         struct file * vm_file;          /* File we map to (can be NULL). */
 304         void * vm_private_data;         /* was vm_pte (shared mem) */
 305
 306 #ifndef CONFIG_MMU
 307         struct vm_region *vm_region;    /* NOMMU mapping region */
 308 #endif
 309 #ifdef CONFIG_NUMA
 310         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 311 #endif
 312 };
 313
 314 struct core_thread {
 315         struct task_struct *task;
 316         struct core_thread *next;
 317 };
 318
 319 struct core_state {
 320         atomic_t nr_threads;
 321         struct core_thread dumper;
 322         struct completion startup;
 323 };
 324
 325 enum {
 326         MM_FILEPAGES,
 327         MM_ANONPAGES,
 328         MM_SWAPENTS,
 329         NR_MM_COUNTERS
 330 };
 331
 332 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 333 #define SPLIT_RSS_COUNTING
 334 /* per-thread cached information, */
 335 struct task_rss_stat {
 336         int events;     /* for synchronization threshold */
 337         int count[NR_MM_COUNTERS];
 338 };
 339 #endif /* USE_SPLIT_PTE_PTLOCKS */
 340
 341 struct mm_rss_stat {
 342         atomic_long_t count[NR_MM_COUNTERS];
 343 };
 344
 345 struct kioctx_table;
 346 struct mm_struct {
 347         struct vm_area_struct *mmap;            /* list of VMAs */
 348         struct rb_root mm_rb;
 349         u32 vmacache_seqnum;                   /* per-thread vmacache */
 350 #ifdef CONFIG_MMU
 351         unsigned long (*get_unmapped_area) (struct file *filp,
 352                                 unsigned long addr, unsigned long len,
 353                                 unsigned long pgoff, unsigned long flags);
 354 #endif
 355         unsigned long mmap_base;                /* base of mmap area */
 356         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 357         unsigned long task_size;                /* size of task vm space */
 358         unsigned long highest_vm_end;           /* highest vma end address */
 359         pgd_t * pgd;
 360         atomic_t mm_users;                      /* How many users with user space? */
 361         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 362         atomic_long_t nr_ptes;                  /* Page table pages */
 363         int map_count;                          /* number of VMAs */
 364
 365         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 366         struct rw_semaphore mmap_sem;
 367
 368         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 369                                                  * together off init_mm.mmlist, and are protected
 370                                                  * by mmlist_lock
 371                                                  */
 372
 373
 374         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 375         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 376
 377         unsigned long total_vm;         /* Total pages mapped */
 378         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 379         unsigned long pinned_vm;        /* Refcount permanently increased */
 380         unsigned long shared_vm;        /* Shared pages (files) */
 381         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 382         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 383         unsigned long def_flags;
 384         unsigned long start_code, end_code, start_data, end_data;
 385         unsigned long start_brk, brk, start_stack;
 386         unsigned long arg_start, arg_end, env_start, env_end;
 387
 388         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 389
 390         /*
 391          * Special counters, in some configurations protected by the
 392          * page_table_lock, in other configurations by being atomic.
 393          */
 394         struct mm_rss_stat rss_stat;
 395
 396         struct linux_binfmt *binfmt;
 397
 398         cpumask_var_t cpu_vm_mask_var;
 399
 400         /* Architecture-specific MM context */
 401         mm_context_t context;
 402
 403         unsigned long flags; /* Must use atomic bitops to access the bits */
 404
 405         struct core_state *core_state; /* coredumping support */
 406 #ifdef CONFIG_AIO
 407         spinlock_t                      ioctx_lock;
 408         struct kioctx_table __rcu       *ioctx_table;
 409 #endif
 410 #ifdef CONFIG_MEMCG
 411         /*
 412          * "owner" points to a task that is regarded as the canonical
 413          * user/owner of this mm. All of the following must be true in
 414          * order for it to be changed:
 415          *
 416          * current == mm->owner
 417          * current->mm != mm
 418          * new_owner->mm == mm
 419          * new_owner->alloc_lock is held
 420          */
 421         struct task_struct __rcu *owner;
 422 #endif
 423
 424         /* store ref to file /proc/<pid>/exe symlink points to */
 425         struct file *exe_file;
 426 #ifdef CONFIG_MMU_NOTIFIER
 427         struct mmu_notifier_mm *mmu_notifier_mm;
 428 #endif
 429 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 430         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 431 #endif
 432 #ifdef CONFIG_CPUMASK_OFFSTACK
 433         struct cpumask cpumask_allocation;
 434 #endif
 435 #ifdef CONFIG_NUMA_BALANCING
 436         /*
 437          * numa_next_scan is the next time that the PTEs will be marked
 438          * pte_numa. NUMA hinting faults will gather statistics and migrate
 439          * pages to new nodes if necessary.
 440          */
 441         unsigned long numa_next_scan;
 442
 443         /* Restart point for scanning and setting pte_numa */
 444         unsigned long numa_scan_offset;
 445
 446         /* numa_scan_seq prevents two threads setting pte_numa */
 447         int numa_scan_seq;
 448 #endif
 449 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 450         /*
 451          * An operation with batched TLB flushing is going on. Anything that
 452          * can move process memory needs to flush the TLB when moving a
 453          * PROT_NONE or PROT_NUMA mapped page.
 454          */
 455         bool tlb_flush_pending;
 456 #endif
 457         struct uprobes_state uprobes_state;
 458 #ifdef CONFIG_X86_INTEL_MPX
 459         /* address of the bounds directory */
 460         void __user *bd_addr;
 461 #endif
 462 };
 463
 464 static inline void mm_init_cpumask(struct mm_struct *mm)
 465 {
 466 #ifdef CONFIG_CPUMASK_OFFSTACK
 467         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 468 #endif
 469         cpumask_clear(mm->cpu_vm_mask_var);
 470 }
 471
 472 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 473 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 474 {
 475         return mm->cpu_vm_mask_var;
 476 }
 477
 478 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 479 /*
 480  * Memory barriers to keep this state in sync are graciously provided by
 481  * the page table locks, outside of which no page table modifications happen.
 482  * The barriers below prevent the compiler from re-ordering the instructions
 483  * around the memory barriers that are already present in the code.
 484  */
 485 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 486 {
 487         barrier();
 488         return mm->tlb_flush_pending;
 489 }
 490 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 491 {
 492         mm->tlb_flush_pending = true;
 493
 494         /*
 495          * Guarantee that the tlb_flush_pending store does not leak into the
 496          * critical section updating the page tables
 497          */
 498         smp_mb__before_spinlock();
 499 }
 500 /* Clearing is done after a TLB flush, which also provides a barrier. */
 501 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 502 {
 503         barrier();
 504         mm->tlb_flush_pending = false;
 505 }
 506 #else
 507 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 508 {
 509         return false;
 510 }
 511 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 512 {
 513 }
 514 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 515 {
 516 }
 517 #endif
 518
 519 struct vm_special_mapping
 520 {
 521         const char *name;
 522         struct page **pages;
 523 };
 524
 525 enum tlb_flush_reason {
 526         TLB_FLUSH_ON_TASK_SWITCH,
 527         TLB_REMOTE_SHOOTDOWN,
 528         TLB_LOCAL_SHOOTDOWN,
 529         TLB_LOCAL_MM_SHOOTDOWN,
 530         NR_TLB_FLUSH_REASONS,
 531 };
 532
 533  /*
 534   * A swap entry has to fit into a "unsigned long", as the entry is hidden
 535   * in the "index" field of the swapper address space.
 536   */
 537 typedef struct {
 538         unsigned long val;
 539 } swp_entry_t;
 540
 541 #endif /* _LINUX_MM_TYPES_H */