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 typedef void compound_page_dtor(struct page *);
  32
  33 /*
  34  * Each physical page in the system has a struct page associated with
  35  * it to keep track of whatever it is we are using the page for at the
  36  * moment. Note that we have no way to track which tasks are using
  37  * a page, though if it is a pagecache page, rmap structures can tell us
  38  * who is mapping it.
  39  *
  40  * The objects in struct page are organized in double word blocks in
  41  * order to allows us to use atomic double word operations on portions
  42  * of struct page. That is currently only used by slub but the arrangement
  43  * allows the use of atomic double word operations on the flags/mapping
  44  * and lru list pointers also.
  45  */
  46 struct page {
  47         /* First double word block */
  48         unsigned long flags;            /* Atomic flags, some possibly
  49                                          * updated asynchronously */
  50         union {
  51                 struct address_space *mapping;  /* If low bit clear, points to
  52                                                  * inode address_space, or NULL.
  53                                                  * If page mapped as anonymous
  54                                                  * memory, low bit is set, and
  55                                                  * it points to anon_vma object:
  56                                                  * see PAGE_MAPPING_ANON below.
  57                                                  */
  58                 void *s_mem;                    /* slab first object */
  59         };
  60
  61         /* Second double word */
  62         struct {
  63                 union {
  64                         pgoff_t index;          /* Our offset within mapping. */
  65                         void *freelist;         /* sl[aou]b first free object */
  66                         bool pfmemalloc;        /* If set by the page allocator,
  67                                                  * ALLOC_NO_WATERMARKS was set
  68                                                  * and the low watermark was not
  69                                                  * met implying that the system
  70                                                  * is under some pressure. The
  71                                                  * caller should try ensure
  72                                                  * this page is only used to
  73                                                  * free other pages.
  74                                                  */
  75                 };
  76
  77                 union {
  78 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  79         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  80                         /* Used for cmpxchg_double in slub */
  81                         unsigned long counters;
  82 #else
  83                         /*
  84                          * Keep _count separate from slub cmpxchg_double data.
  85                          * As the rest of the double word is protected by
  86                          * slab_lock but _count is not.
  87                          */
  88                         unsigned counters;
  89 #endif
  90
  91                         struct {
  92
  93                                 union {
  94                                         /*
  95                                          * Count of ptes mapped in
  96                                          * mms, to show when page is
  97                                          * mapped & limit reverse map
  98                                          * searches.
  99                                          *
 100                                          * Used also for tail pages
 101                                          * refcounting instead of
 102                                          * _count. Tail pages cannot
 103                                          * be mapped and keeping the
 104                                          * tail page _count zero at
 105                                          * all times guarantees
 106                                          * get_page_unless_zero() will
 107                                          * never succeed on tail
 108                                          * pages.
 109                                          */
 110                                         atomic_t _mapcount;
 111
 112                                         struct { /* SLUB */
 113                                                 unsigned inuse:16;
 114                                                 unsigned objects:15;
 115                                                 unsigned frozen:1;
 116                                         };
 117                                         int units;      /* SLOB */
 118                                 };
 119                                 atomic_t _count;                /* Usage count, see below. */
 120                         };
 121                         unsigned int active;    /* SLAB */
 122                 };
 123         };
 124
 125         /* Third double word block */
 126         union {
 127                 struct list_head lru;   /* Pageout list, eg. active_list
 128                                          * protected by zone->lru_lock !
 129                                          * Can be used as a generic list
 130                                          * by the page owner.
 131                                          */
 132                 struct {                /* slub per cpu partial pages */
 133                         struct page *next;      /* Next partial slab */
 134 #ifdef CONFIG_64BIT
 135                         int pages;      /* Nr of partial slabs left */
 136                         int pobjects;   /* Approximate # of objects */
 137 #else
 138                         short int pages;
 139                         short int pobjects;
 140 #endif
 141                 };
 142
 143                 struct slab *slab_page; /* slab fields */
 144                 struct rcu_head rcu_head;       /* Used by SLAB
 145                                                  * when destroying via RCU
 146                                                  */
 147                 /* First tail page of compound page */
 148                 struct {
 149                         compound_page_dtor *compound_dtor;
 150                         unsigned long compound_order;
 151                 };
 152
 153 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
 154                 pgtable_t pmd_huge_pte; /* protected by page->ptl */
 155 #endif
 156         };
 157
 158         /* Remainder is not double word aligned */
 159         union {
 160                 unsigned long private;          /* Mapping-private opaque data:
 161                                                  * usually used for buffer_heads
 162                                                  * if PagePrivate set; used for
 163                                                  * swp_entry_t if PageSwapCache;
 164                                                  * indicates order in the buddy
 165                                                  * system if PG_buddy is set.
 166                                                  */
 167 #if USE_SPLIT_PTE_PTLOCKS
 168 #if ALLOC_SPLIT_PTLOCKS
 169                 spinlock_t *ptl;
 170 #else
 171                 spinlock_t ptl;
 172 #endif
 173 #endif
 174                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 175                 struct page *first_page;        /* Compound tail pages */
 176         };
 177
 178 #ifdef CONFIG_MEMCG
 179         struct mem_cgroup *mem_cgroup;
 180 #endif
 181
 182         /*
 183          * On machines where all RAM is mapped into kernel address space,
 184          * we can simply calculate the virtual address. On machines with
 185          * highmem some memory is mapped into kernel virtual memory
 186          * dynamically, so we need a place to store that address.
 187          * Note that this field could be 16 bits on x86 ... ;)
 188          *
 189          * Architectures with slow multiplication can define
 190          * WANT_PAGE_VIRTUAL in asm/page.h
 191          */
 192 #if defined(WANT_PAGE_VIRTUAL)
 193         void *virtual;                  /* Kernel virtual address (NULL if
 194                                            not kmapped, ie. highmem) */
 195 #endif /* WANT_PAGE_VIRTUAL */
 196
 197 #ifdef CONFIG_KMEMCHECK
 198         /*
 199          * kmemcheck wants to track the status of each byte in a page; this
 200          * is a pointer to such a status block. NULL if not tracked.
 201          */
 202         void *shadow;
 203 #endif
 204
 205 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 206         int _last_cpupid;
 207 #endif
 208 }
 209 /*
 210  * The struct page can be forced to be double word aligned so that atomic ops
 211  * on double words work. The SLUB allocator can make use of such a feature.
 212  */
 213 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 214         __aligned(2 * sizeof(unsigned long))
 215 #endif
 216 ;
 217
 218 struct page_frag {
 219         struct page *page;
 220 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 221         __u32 offset;
 222         __u32 size;
 223 #else
 224         __u16 offset;
 225         __u16 size;
 226 #endif
 227 };
 228
 229 #define PAGE_FRAG_CACHE_MAX_SIZE        __ALIGN_MASK(32768, ~PAGE_MASK)
 230 #define PAGE_FRAG_CACHE_MAX_ORDER       get_order(PAGE_FRAG_CACHE_MAX_SIZE)
 231
 232 struct page_frag_cache {
 233         void * va;
 234 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
 235         __u16 offset;
 236         __u16 size;
 237 #else
 238         __u32 offset;
 239 #endif
 240         /* we maintain a pagecount bias, so that we dont dirty cache line
 241          * containing page->_count every time we allocate a fragment.
 242          */
 243         unsigned int            pagecnt_bias;
 244         bool pfmemalloc;
 245 };
 246
 247 typedef unsigned long __nocast vm_flags_t;
 248
 249 /*
 250  * A region containing a mapping of a non-memory backed file under NOMMU
 251  * conditions.  These are held in a global tree and are pinned by the VMAs that
 252  * map parts of them.
 253  */
 254 struct vm_region {
 255         struct rb_node  vm_rb;          /* link in global region tree */
 256         vm_flags_t      vm_flags;       /* VMA vm_flags */
 257         unsigned long   vm_start;       /* start address of region */
 258         unsigned long   vm_end;         /* region initialised to here */
 259         unsigned long   vm_top;         /* region allocated to here */
 260         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 261         struct file     *vm_file;       /* the backing file or NULL */
 262
 263         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 264         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 265                                                 * this region */
 266 };
 267
 268 /*
 269  * This struct defines a memory VMM memory area. There is one of these
 270  * per VM-area/task.  A VM area is any part of the process virtual memory
 271  * space that has a special rule for the page-fault handlers (ie a shared
 272  * library, the executable area etc).
 273  */
 274 struct vm_area_struct {
 275         /* The first cache line has the info for VMA tree walking. */
 276
 277         unsigned long vm_start;         /* Our start address within vm_mm. */
 278         unsigned long vm_end;           /* The first byte after our end address
 279                                            within vm_mm. */
 280
 281         /* linked list of VM areas per task, sorted by address */
 282         struct vm_area_struct *vm_next, *vm_prev;
 283
 284         struct rb_node vm_rb;
 285
 286         /*
 287          * Largest free memory gap in bytes to the left of this VMA.
 288          * Either between this VMA and vma->vm_prev, or between one of the
 289          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 290          * get_unmapped_area find a free area of the right size.
 291          */
 292         unsigned long rb_subtree_gap;
 293
 294         /* Second cache line starts here. */
 295
 296         struct mm_struct *vm_mm;        /* The address space we belong to. */
 297         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 298         unsigned long vm_flags;         /* Flags, see mm.h. */
 299
 300         /*
 301          * For areas with an address space and backing store,
 302          * linkage into the address_space->i_mmap interval tree.
 303          */
 304         struct {
 305                 struct rb_node rb;
 306                 unsigned long rb_subtree_last;
 307         } shared;
 308
 309         /*
 310          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 311          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 312          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 313          * or brk vma (with NULL file) can only be in an anon_vma list.
 314          */
 315         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 316                                           * page_table_lock */
 317         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 318
 319         /* Function pointers to deal with this struct. */
 320         const struct vm_operations_struct *vm_ops;
 321
 322         /* Information about our backing store: */
 323         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 324                                            units, *not* PAGE_CACHE_SIZE */
 325         struct file * vm_file;          /* File we map to (can be NULL). */
 326         void * vm_private_data;         /* was vm_pte (shared mem) */
 327
 328 #ifndef CONFIG_MMU
 329         struct vm_region *vm_region;    /* NOMMU mapping region */
 330 #endif
 331 #ifdef CONFIG_NUMA
 332         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 333 #endif
 334 };
 335
 336 struct core_thread {
 337         struct task_struct *task;
 338         struct core_thread *next;
 339 };
 340
 341 struct core_state {
 342         atomic_t nr_threads;
 343         struct core_thread dumper;
 344         struct completion startup;
 345 };
 346
 347 enum {
 348         MM_FILEPAGES,
 349         MM_ANONPAGES,
 350         MM_SWAPENTS,
 351         NR_MM_COUNTERS
 352 };
 353
 354 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 355 #define SPLIT_RSS_COUNTING
 356 /* per-thread cached information, */
 357 struct task_rss_stat {
 358         int events;     /* for synchronization threshold */
 359         int count[NR_MM_COUNTERS];
 360 };
 361 #endif /* USE_SPLIT_PTE_PTLOCKS */
 362
 363 struct mm_rss_stat {
 364         atomic_long_t count[NR_MM_COUNTERS];
 365 };
 366
 367 struct kioctx_table;
 368 struct mm_struct {
 369         struct vm_area_struct *mmap;            /* list of VMAs */
 370         struct rb_root mm_rb;
 371         u32 vmacache_seqnum;                   /* per-thread vmacache */
 372 #ifdef CONFIG_MMU
 373         unsigned long (*get_unmapped_area) (struct file *filp,
 374                                 unsigned long addr, unsigned long len,
 375                                 unsigned long pgoff, unsigned long flags);
 376 #endif
 377         unsigned long mmap_base;                /* base of mmap area */
 378         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 379         unsigned long task_size;                /* size of task vm space */
 380         unsigned long highest_vm_end;           /* highest vma end address */
 381         pgd_t * pgd;
 382         atomic_t mm_users;                      /* How many users with user space? */
 383         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 384         atomic_long_t nr_ptes;                  /* PTE page table pages */
 385 #if CONFIG_PGTABLE_LEVELS > 2
 386         atomic_long_t nr_pmds;                  /* PMD page table pages */
 387 #endif
 388         int map_count;                          /* number of VMAs */
 389
 390         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 391         struct rw_semaphore mmap_sem;
 392
 393         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 394                                                  * together off init_mm.mmlist, and are protected
 395                                                  * by mmlist_lock
 396                                                  */
 397
 398
 399         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 400         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 401
 402         unsigned long total_vm;         /* Total pages mapped */
 403         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 404         unsigned long pinned_vm;        /* Refcount permanently increased */
 405         unsigned long shared_vm;        /* Shared pages (files) */
 406         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 407         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 408         unsigned long def_flags;
 409         unsigned long start_code, end_code, start_data, end_data;
 410         unsigned long start_brk, brk, start_stack;
 411         unsigned long arg_start, arg_end, env_start, env_end;
 412
 413         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 414
 415         /*
 416          * Special counters, in some configurations protected by the
 417          * page_table_lock, in other configurations by being atomic.
 418          */
 419         struct mm_rss_stat rss_stat;
 420
 421         struct linux_binfmt *binfmt;
 422
 423         cpumask_var_t cpu_vm_mask_var;
 424
 425         /* Architecture-specific MM context */
 426         mm_context_t context;
 427
 428         unsigned long flags; /* Must use atomic bitops to access the bits */
 429
 430         struct core_state *core_state; /* coredumping support */
 431 #ifdef CONFIG_AIO
 432         spinlock_t                      ioctx_lock;
 433         struct kioctx_table __rcu       *ioctx_table;
 434 #endif
 435 #ifdef CONFIG_MEMCG
 436         /*
 437          * "owner" points to a task that is regarded as the canonical
 438          * user/owner of this mm. All of the following must be true in
 439          * order for it to be changed:
 440          *
 441          * current == mm->owner
 442          * current->mm != mm
 443          * new_owner->mm == mm
 444          * new_owner->alloc_lock is held
 445          */
 446         struct task_struct __rcu *owner;
 447 #endif
 448
 449         /* store ref to file /proc/<pid>/exe symlink points to */
 450         struct file __rcu *exe_file;
 451 #ifdef CONFIG_MMU_NOTIFIER
 452         struct mmu_notifier_mm *mmu_notifier_mm;
 453 #endif
 454 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 455         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 456 #endif
 457 #ifdef CONFIG_CPUMASK_OFFSTACK
 458         struct cpumask cpumask_allocation;
 459 #endif
 460 #ifdef CONFIG_NUMA_BALANCING
 461         /*
 462          * numa_next_scan is the next time that the PTEs will be marked
 463          * pte_numa. NUMA hinting faults will gather statistics and migrate
 464          * pages to new nodes if necessary.
 465          */
 466         unsigned long numa_next_scan;
 467
 468         /* Restart point for scanning and setting pte_numa */
 469         unsigned long numa_scan_offset;
 470
 471         /* numa_scan_seq prevents two threads setting pte_numa */
 472         int numa_scan_seq;
 473 #endif
 474 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 475         /*
 476          * An operation with batched TLB flushing is going on. Anything that
 477          * can move process memory needs to flush the TLB when moving a
 478          * PROT_NONE or PROT_NUMA mapped page.
 479          */
 480         bool tlb_flush_pending;
 481 #endif
 482         struct uprobes_state uprobes_state;
 483 #ifdef CONFIG_X86_INTEL_MPX
 484         /* address of the bounds directory */
 485         void __user *bd_addr;
 486 #endif
 487 };
 488
 489 static inline void mm_init_cpumask(struct mm_struct *mm)
 490 {
 491 #ifdef CONFIG_CPUMASK_OFFSTACK
 492         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 493 #endif
 494         cpumask_clear(mm->cpu_vm_mask_var);
 495 }
 496
 497 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 498 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 499 {
 500         return mm->cpu_vm_mask_var;
 501 }
 502
 503 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 504 /*
 505  * Memory barriers to keep this state in sync are graciously provided by
 506  * the page table locks, outside of which no page table modifications happen.
 507  * The barriers below prevent the compiler from re-ordering the instructions
 508  * around the memory barriers that are already present in the code.
 509  */
 510 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 511 {
 512         barrier();
 513         return mm->tlb_flush_pending;
 514 }
 515 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 516 {
 517         mm->tlb_flush_pending = true;
 518
 519         /*
 520          * Guarantee that the tlb_flush_pending store does not leak into the
 521          * critical section updating the page tables
 522          */
 523         smp_mb__before_spinlock();
 524 }
 525 /* Clearing is done after a TLB flush, which also provides a barrier. */
 526 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 527 {
 528         barrier();
 529         mm->tlb_flush_pending = false;
 530 }
 531 #else
 532 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 533 {
 534         return false;
 535 }
 536 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 537 {
 538 }
 539 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 540 {
 541 }
 542 #endif
 543
 544 struct vm_special_mapping
 545 {
 546         const char *name;
 547         struct page **pages;
 548 };
 549
 550 enum tlb_flush_reason {
 551         TLB_FLUSH_ON_TASK_SWITCH,
 552         TLB_REMOTE_SHOOTDOWN,
 553         TLB_LOCAL_SHOOTDOWN,
 554         TLB_LOCAL_MM_SHOOTDOWN,
 555         NR_TLB_FLUSH_REASONS,
 556 };
 557
 558  /*
 559   * A swap entry has to fit into a "unsigned long", as the entry is hidden
 560   * in the "index" field of the swapper address space.
 561   */
 562 typedef struct {
 563         unsigned long val;
 564 } swp_entry_t;
 565
 566 #endif /* _LINUX_MM_TYPES_H */