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 struct mem_cgroup;
  26
  27 #define USE_SPLIT_PTE_PTLOCKS   (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
  28 #define USE_SPLIT_PMD_PTLOCKS   (USE_SPLIT_PTE_PTLOCKS && \
  29                 IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
  30 #define ALLOC_SPLIT_PTLOCKS     (SPINLOCK_SIZE > BITS_PER_LONG/8)
  31
  32 /*
  33  * Each physical page in the system has a struct page associated with
  34  * it to keep track of whatever it is we are using the page for at the
  35  * moment. Note that we have no way to track which tasks are using
  36  * a page, though if it is a pagecache page, rmap structures can tell us
  37  * who is mapping it.
  38  *
  39  * The objects in struct page are organized in double word blocks in
  40  * order to allows us to use atomic double word operations on portions
  41  * of struct page. That is currently only used by slub but the arrangement
  42  * allows the use of atomic double word operations on the flags/mapping
  43  * and lru list pointers also.
  44  */
  45 struct page {
  46         /* First double word block */
  47         unsigned long flags;            /* Atomic flags, some possibly
  48                                          * updated asynchronously */
  49         union {
  50                 struct address_space *mapping;  /* If low bit clear, points to
  51                                                  * inode address_space, or NULL.
  52                                                  * If page mapped as anonymous
  53                                                  * memory, low bit is set, and
  54                                                  * it points to anon_vma object:
  55                                                  * see PAGE_MAPPING_ANON below.
  56                                                  */
  57                 void *s_mem;                    /* slab first object */
  58         };
  59
  60         /* Second double word */
  61         struct {
  62                 union {
  63                         pgoff_t index;          /* Our offset within mapping. */
  64                         void *freelist;         /* sl[aou]b first free object */
  65                         bool pfmemalloc;        /* If set by the page allocator,
  66                                                  * ALLOC_NO_WATERMARKS was set
  67                                                  * and the low watermark was not
  68                                                  * met implying that the system
  69                                                  * is under some pressure. The
  70                                                  * caller should try ensure
  71                                                  * this page is only used to
  72                                                  * free other pages.
  73                                                  */
  74                 };
  75
  76                 union {
  77 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  78         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  79                         /* Used for cmpxchg_double in slub */
  80                         unsigned long counters;
  81 #else
  82                         /*
  83                          * Keep _count separate from slub cmpxchg_double data.
  84                          * As the rest of the double word is protected by
  85                          * slab_lock but _count is not.
  86                          */
  87                         unsigned counters;
  88 #endif
  89
  90                         struct {
  91
  92                                 union {
  93                                         /*
  94                                          * Count of ptes mapped in
  95                                          * mms, to show when page is
  96                                          * mapped & limit reverse map
  97                                          * searches.
  98                                          *
  99                                          * Used also for tail pages
 100                                          * refcounting instead of
 101                                          * _count. Tail pages cannot
 102                                          * be mapped and keeping the
 103                                          * tail page _count zero at
 104                                          * all times guarantees
 105                                          * get_page_unless_zero() will
 106                                          * never succeed on tail
 107                                          * pages.
 108                                          */
 109                                         atomic_t _mapcount;
 110
 111                                         struct { /* SLUB */
 112                                                 unsigned inuse:16;
 113                                                 unsigned objects:15;
 114                                                 unsigned frozen:1;
 115                                         };
 116                                         int units;      /* SLOB */
 117                                 };
 118                                 atomic_t _count;                /* Usage count, see below. */
 119                         };
 120                         unsigned int active;    /* SLAB */
 121                 };
 122         };
 123
 124         /* Third double word block */
 125         union {
 126                 struct list_head lru;   /* Pageout list, eg. active_list
 127                                          * protected by zone->lru_lock !
 128                                          * Can be used as a generic list
 129                                          * by the page owner.
 130                                          */
 131                 struct {                /* slub per cpu partial pages */
 132                         struct page *next;      /* Next partial slab */
 133 #ifdef CONFIG_64BIT
 134                         int pages;      /* Nr of partial slabs left */
 135                         int pobjects;   /* Approximate # of objects */
 136 #else
 137                         short int pages;
 138                         short int pobjects;
 139 #endif
 140                 };
 141
 142                 struct slab *slab_page; /* slab fields */
 143                 struct rcu_head rcu_head;       /* Used by SLAB
 144                                                  * when destroying via RCU
 145                                                  */
 146 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
 147                 pgtable_t pmd_huge_pte; /* protected by page->ptl */
 148 #endif
 149         };
 150
 151         /* Remainder is not double word aligned */
 152         union {
 153                 unsigned long private;          /* Mapping-private opaque data:
 154                                                  * usually used for buffer_heads
 155                                                  * if PagePrivate set; used for
 156                                                  * swp_entry_t if PageSwapCache;
 157                                                  * indicates order in the buddy
 158                                                  * system if PG_buddy is set.
 159                                                  */
 160 #if USE_SPLIT_PTE_PTLOCKS
 161 #if ALLOC_SPLIT_PTLOCKS
 162                 spinlock_t *ptl;
 163 #else
 164                 spinlock_t ptl;
 165 #endif
 166 #endif
 167                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 168                 struct page *first_page;        /* Compound tail pages */
 169         };
 170
 171 #ifdef CONFIG_MEMCG
 172         struct mem_cgroup *mem_cgroup;
 173 #endif
 174
 175         /*
 176          * On machines where all RAM is mapped into kernel address space,
 177          * we can simply calculate the virtual address. On machines with
 178          * highmem some memory is mapped into kernel virtual memory
 179          * dynamically, so we need a place to store that address.
 180          * Note that this field could be 16 bits on x86 ... ;)
 181          *
 182          * Architectures with slow multiplication can define
 183          * WANT_PAGE_VIRTUAL in asm/page.h
 184          */
 185 #if defined(WANT_PAGE_VIRTUAL)
 186         void *virtual;                  /* Kernel virtual address (NULL if
 187                                            not kmapped, ie. highmem) */
 188 #endif /* WANT_PAGE_VIRTUAL */
 189 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
 190         unsigned long debug_flags;      /* Use atomic bitops on this */
 191 #endif
 192
 193 #ifdef CONFIG_KMEMCHECK
 194         /*
 195          * kmemcheck wants to track the status of each byte in a page; this
 196          * is a pointer to such a status block. NULL if not tracked.
 197          */
 198         void *shadow;
 199 #endif
 200
 201 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 202         int _last_cpupid;
 203 #endif
 204 }
 205 /*
 206  * The struct page can be forced to be double word aligned so that atomic ops
 207  * on double words work. The SLUB allocator can make use of such a feature.
 208  */
 209 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 210         __aligned(2 * sizeof(unsigned long))
 211 #endif
 212 ;
 213
 214 struct page_frag {
 215         struct page *page;
 216 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 217         __u32 offset;
 218         __u32 size;
 219 #else
 220         __u16 offset;
 221         __u16 size;
 222 #endif
 223 };
 224
 225 typedef unsigned long __nocast vm_flags_t;
 226
 227 /*
 228  * A region containing a mapping of a non-memory backed file under NOMMU
 229  * conditions.  These are held in a global tree and are pinned by the VMAs that
 230  * map parts of them.
 231  */
 232 struct vm_region {
 233         struct rb_node  vm_rb;          /* link in global region tree */
 234         vm_flags_t      vm_flags;       /* VMA vm_flags */
 235         unsigned long   vm_start;       /* start address of region */
 236         unsigned long   vm_end;         /* region initialised to here */
 237         unsigned long   vm_top;         /* region allocated to here */
 238         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 239         struct file     *vm_file;       /* the backing file or NULL */
 240
 241         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 242         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 243                                                 * this region */
 244 };
 245
 246 /*
 247  * This struct defines a memory VMM memory area. There is one of these
 248  * per VM-area/task.  A VM area is any part of the process virtual memory
 249  * space that has a special rule for the page-fault handlers (ie a shared
 250  * library, the executable area etc).
 251  */
 252 struct vm_area_struct {
 253         /* The first cache line has the info for VMA tree walking. */
 254
 255         unsigned long vm_start;         /* Our start address within vm_mm. */
 256         unsigned long vm_end;           /* The first byte after our end address
 257                                            within vm_mm. */
 258
 259         /* linked list of VM areas per task, sorted by address */
 260         struct vm_area_struct *vm_next, *vm_prev;
 261
 262         struct rb_node vm_rb;
 263
 264         /*
 265          * Largest free memory gap in bytes to the left of this VMA.
 266          * Either between this VMA and vma->vm_prev, or between one of the
 267          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 268          * get_unmapped_area find a free area of the right size.
 269          */
 270         unsigned long rb_subtree_gap;
 271
 272         /* Second cache line starts here. */
 273
 274         struct mm_struct *vm_mm;        /* The address space we belong to. */
 275         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 276         unsigned long vm_flags;         /* Flags, see mm.h. */
 277
 278         /*
 279          * For areas with an address space and backing store,
 280          * linkage into the address_space->i_mmap interval tree, or
 281          * linkage of vma in the address_space->i_mmap_nonlinear list.
 282          */
 283         union {
 284                 struct {
 285                         struct rb_node rb;
 286                         unsigned long rb_subtree_last;
 287                 } linear;
 288                 struct list_head nonlinear;
 289         } shared;
 290
 291         /*
 292          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 293          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 294          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 295          * or brk vma (with NULL file) can only be in an anon_vma list.
 296          */
 297         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 298                                           * page_table_lock */
 299         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 300
 301         /* Function pointers to deal with this struct. */
 302         const struct vm_operations_struct *vm_ops;
 303
 304         /* Information about our backing store: */
 305         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 306                                            units, *not* PAGE_CACHE_SIZE */
 307         struct file * vm_file;          /* File we map to (can be NULL). */
 308         void * vm_private_data;         /* was vm_pte (shared mem) */
 309
 310 #ifndef CONFIG_MMU
 311         struct vm_region *vm_region;    /* NOMMU mapping region */
 312 #endif
 313 #ifdef CONFIG_NUMA
 314         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 315 #endif
 316 };
 317
 318 struct core_thread {
 319         struct task_struct *task;
 320         struct core_thread *next;
 321 };
 322
 323 struct core_state {
 324         atomic_t nr_threads;
 325         struct core_thread dumper;
 326         struct completion startup;
 327 };
 328
 329 enum {
 330         MM_FILEPAGES,
 331         MM_ANONPAGES,
 332         MM_SWAPENTS,
 333         NR_MM_COUNTERS
 334 };
 335
 336 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 337 #define SPLIT_RSS_COUNTING
 338 /* per-thread cached information, */
 339 struct task_rss_stat {
 340         int events;     /* for synchronization threshold */
 341         int count[NR_MM_COUNTERS];
 342 };
 343 #endif /* USE_SPLIT_PTE_PTLOCKS */
 344
 345 struct mm_rss_stat {
 346         atomic_long_t count[NR_MM_COUNTERS];
 347 };
 348
 349 struct kioctx_table;
 350 struct mm_struct {
 351         struct vm_area_struct *mmap;            /* list of VMAs */
 352         struct rb_root mm_rb;
 353         u32 vmacache_seqnum;                   /* per-thread vmacache */
 354 #ifdef CONFIG_MMU
 355         unsigned long (*get_unmapped_area) (struct file *filp,
 356                                 unsigned long addr, unsigned long len,
 357                                 unsigned long pgoff, unsigned long flags);
 358 #endif
 359         unsigned long mmap_base;                /* base of mmap area */
 360         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 361         unsigned long task_size;                /* size of task vm space */
 362         unsigned long highest_vm_end;           /* highest vma end address */
 363         pgd_t * pgd;
 364         atomic_t mm_users;                      /* How many users with user space? */
 365         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 366         atomic_long_t nr_ptes;                  /* Page table pages */
 367         int map_count;                          /* number of VMAs */
 368
 369         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 370         struct rw_semaphore mmap_sem;
 371
 372         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 373                                                  * together off init_mm.mmlist, and are protected
 374                                                  * by mmlist_lock
 375                                                  */
 376
 377
 378         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 379         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 380
 381         unsigned long total_vm;         /* Total pages mapped */
 382         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 383         unsigned long pinned_vm;        /* Refcount permanently increased */
 384         unsigned long shared_vm;        /* Shared pages (files) */
 385         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 386         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 387         unsigned long def_flags;
 388         unsigned long start_code, end_code, start_data, end_data;
 389         unsigned long start_brk, brk, start_stack;
 390         unsigned long arg_start, arg_end, env_start, env_end;
 391
 392         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 393
 394         /*
 395          * Special counters, in some configurations protected by the
 396          * page_table_lock, in other configurations by being atomic.
 397          */
 398         struct mm_rss_stat rss_stat;
 399
 400         struct linux_binfmt *binfmt;
 401
 402         cpumask_var_t cpu_vm_mask_var;
 403
 404         /* Architecture-specific MM context */
 405         mm_context_t context;
 406
 407         unsigned long flags; /* Must use atomic bitops to access the bits */
 408
 409         struct core_state *core_state; /* coredumping support */
 410 #ifdef CONFIG_AIO
 411         spinlock_t                      ioctx_lock;
 412         struct kioctx_table __rcu       *ioctx_table;
 413 #endif
 414 #ifdef CONFIG_MEMCG
 415         /*
 416          * "owner" points to a task that is regarded as the canonical
 417          * user/owner of this mm. All of the following must be true in
 418          * order for it to be changed:
 419          *
 420          * current == mm->owner
 421          * current->mm != mm
 422          * new_owner->mm == mm
 423          * new_owner->alloc_lock is held
 424          */
 425         struct task_struct __rcu *owner;
 426 #endif
 427
 428         /* store ref to file /proc/<pid>/exe symlink points to */
 429         struct file *exe_file;
 430 #ifdef CONFIG_MMU_NOTIFIER
 431         struct mmu_notifier_mm *mmu_notifier_mm;
 432 #endif
 433 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 434         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 435 #endif
 436 #ifdef CONFIG_CPUMASK_OFFSTACK
 437         struct cpumask cpumask_allocation;
 438 #endif
 439 #ifdef CONFIG_NUMA_BALANCING
 440         /*
 441          * numa_next_scan is the next time that the PTEs will be marked
 442          * pte_numa. NUMA hinting faults will gather statistics and migrate
 443          * pages to new nodes if necessary.
 444          */
 445         unsigned long numa_next_scan;
 446
 447         /* Restart point for scanning and setting pte_numa */
 448         unsigned long numa_scan_offset;
 449
 450         /* numa_scan_seq prevents two threads setting pte_numa */
 451         int numa_scan_seq;
 452 #endif
 453 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 454         /*
 455          * An operation with batched TLB flushing is going on. Anything that
 456          * can move process memory needs to flush the TLB when moving a
 457          * PROT_NONE or PROT_NUMA mapped page.
 458          */
 459         bool tlb_flush_pending;
 460 #endif
 461         struct uprobes_state uprobes_state;
 462 #ifdef CONFIG_X86_INTEL_MPX
 463         /* address of the bounds directory */
 464         void __user *bd_addr;
 465 #endif
 466 };
 467
 468 static inline void mm_init_cpumask(struct mm_struct *mm)
 469 {
 470 #ifdef CONFIG_CPUMASK_OFFSTACK
 471         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 472 #endif
 473         cpumask_clear(mm->cpu_vm_mask_var);
 474 }
 475
 476 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 477 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 478 {
 479         return mm->cpu_vm_mask_var;
 480 }
 481
 482 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 483 /*
 484  * Memory barriers to keep this state in sync are graciously provided by
 485  * the page table locks, outside of which no page table modifications happen.
 486  * The barriers below prevent the compiler from re-ordering the instructions
 487  * around the memory barriers that are already present in the code.
 488  */
 489 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 490 {
 491         barrier();
 492         return mm->tlb_flush_pending;
 493 }
 494 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 495 {
 496         mm->tlb_flush_pending = true;
 497
 498         /*
 499          * Guarantee that the tlb_flush_pending store does not leak into the
 500          * critical section updating the page tables
 501          */
 502         smp_mb__before_spinlock();
 503 }
 504 /* Clearing is done after a TLB flush, which also provides a barrier. */
 505 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 506 {
 507         barrier();
 508         mm->tlb_flush_pending = false;
 509 }
 510 #else
 511 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 512 {
 513         return false;
 514 }
 515 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 516 {
 517 }
 518 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 519 {
 520 }
 521 #endif
 522
 523 struct vm_special_mapping
 524 {
 525         const char *name;
 526         struct page **pages;
 527 };
 528
 529 enum tlb_flush_reason {
 530         TLB_FLUSH_ON_TASK_SWITCH,
 531         TLB_REMOTE_SHOOTDOWN,
 532         TLB_LOCAL_SHOOTDOWN,
 533         TLB_LOCAL_MM_SHOOTDOWN,
 534         NR_TLB_FLUSH_REASONS,
 535 };
 536
 537 #endif /* _LINUX_MM_TYPES_H */