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                 };
  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                         unsigned int active;    /* SLAB */
 113                 };
 114         };
 115
 116         /* Third double word block */
 117         union {
 118                 struct list_head lru;   /* Pageout list, eg. active_list
 119                                          * protected by zone->lru_lock !
 120                                          * Can be used as a generic list
 121                                          * by the page owner.
 122                                          */
 123                 struct {                /* slub per cpu partial pages */
 124                         struct page *next;      /* Next partial slab */
 125 #ifdef CONFIG_64BIT
 126                         int pages;      /* Nr of partial slabs left */
 127                         int pobjects;   /* Approximate # of objects */
 128 #else
 129                         short int pages;
 130                         short int pobjects;
 131 #endif
 132                 };
 133
 134                 struct slab *slab_page; /* slab fields */
 135                 struct rcu_head rcu_head;       /* Used by SLAB
 136                                                  * when destroying via RCU
 137                                                  */
 138                 /* First tail page of compound page */
 139                 struct {
 140                         compound_page_dtor *compound_dtor;
 141                         unsigned long compound_order;
 142                 };
 143
 144 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
 145                 pgtable_t pmd_huge_pte; /* protected by page->ptl */
 146 #endif
 147         };
 148
 149         /* Remainder is not double word aligned */
 150         union {
 151                 unsigned long private;          /* Mapping-private opaque data:
 152                                                  * usually used for buffer_heads
 153                                                  * if PagePrivate set; used for
 154                                                  * swp_entry_t if PageSwapCache;
 155                                                  * indicates order in the buddy
 156                                                  * system if PG_buddy is set.
 157                                                  */
 158 #if USE_SPLIT_PTE_PTLOCKS
 159 #if ALLOC_SPLIT_PTLOCKS
 160                 spinlock_t *ptl;
 161 #else
 162                 spinlock_t ptl;
 163 #endif
 164 #endif
 165                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 166                 struct page *first_page;        /* Compound tail pages */
 167         };
 168
 169 #ifdef CONFIG_MEMCG
 170         struct mem_cgroup *mem_cgroup;
 171 #endif
 172
 173         /*
 174          * On machines where all RAM is mapped into kernel address space,
 175          * we can simply calculate the virtual address. On machines with
 176          * highmem some memory is mapped into kernel virtual memory
 177          * dynamically, so we need a place to store that address.
 178          * Note that this field could be 16 bits on x86 ... ;)
 179          *
 180          * Architectures with slow multiplication can define
 181          * WANT_PAGE_VIRTUAL in asm/page.h
 182          */
 183 #if defined(WANT_PAGE_VIRTUAL)
 184         void *virtual;                  /* Kernel virtual address (NULL if
 185                                            not kmapped, ie. highmem) */
 186 #endif /* WANT_PAGE_VIRTUAL */
 187
 188 #ifdef CONFIG_KMEMCHECK
 189         /*
 190          * kmemcheck wants to track the status of each byte in a page; this
 191          * is a pointer to such a status block. NULL if not tracked.
 192          */
 193         void *shadow;
 194 #endif
 195
 196 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 197         int _last_cpupid;
 198 #endif
 199 }
 200 /*
 201  * The struct page can be forced to be double word aligned so that atomic ops
 202  * on double words work. The SLUB allocator can make use of such a feature.
 203  */
 204 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 205         __aligned(2 * sizeof(unsigned long))
 206 #endif
 207 ;
 208
 209 struct page_frag {
 210         struct page *page;
 211 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 212         __u32 offset;
 213         __u32 size;
 214 #else
 215         __u16 offset;
 216         __u16 size;
 217 #endif
 218 };
 219
 220 #define PAGE_FRAG_CACHE_MAX_SIZE        __ALIGN_MASK(32768, ~PAGE_MASK)
 221 #define PAGE_FRAG_CACHE_MAX_ORDER       get_order(PAGE_FRAG_CACHE_MAX_SIZE)
 222
 223 struct page_frag_cache {
 224         void * va;
 225 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
 226         __u16 offset;
 227         __u16 size;
 228 #else
 229         __u32 offset;
 230 #endif
 231         /* we maintain a pagecount bias, so that we dont dirty cache line
 232          * containing page->_count every time we allocate a fragment.
 233          */
 234         unsigned int            pagecnt_bias;
 235         bool pfmemalloc;
 236 };
 237
 238 typedef unsigned long __nocast vm_flags_t;
 239
 240 /*
 241  * A region containing a mapping of a non-memory backed file under NOMMU
 242  * conditions.  These are held in a global tree and are pinned by the VMAs that
 243  * map parts of them.
 244  */
 245 struct vm_region {
 246         struct rb_node  vm_rb;          /* link in global region tree */
 247         vm_flags_t      vm_flags;       /* VMA vm_flags */
 248         unsigned long   vm_start;       /* start address of region */
 249         unsigned long   vm_end;         /* region initialised to here */
 250         unsigned long   vm_top;         /* region allocated to here */
 251         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 252         struct file     *vm_file;       /* the backing file or NULL */
 253
 254         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 255         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 256                                                 * this region */
 257 };
 258
 259 #ifdef CONFIG_USERFAULTFD
 260 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
 261 struct vm_userfaultfd_ctx {
 262         struct userfaultfd_ctx *ctx;
 263 };
 264 #else /* CONFIG_USERFAULTFD */
 265 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
 266 struct vm_userfaultfd_ctx {};
 267 #endif /* CONFIG_USERFAULTFD */
 268
 269 /*
 270  * This struct defines a memory VMM memory area. There is one of these
 271  * per VM-area/task.  A VM area is any part of the process virtual memory
 272  * space that has a special rule for the page-fault handlers (ie a shared
 273  * library, the executable area etc).
 274  */
 275 struct vm_area_struct {
 276         /* The first cache line has the info for VMA tree walking. */
 277
 278         unsigned long vm_start;         /* Our start address within vm_mm. */
 279         unsigned long vm_end;           /* The first byte after our end address
 280                                            within vm_mm. */
 281
 282         /* linked list of VM areas per task, sorted by address */
 283         struct vm_area_struct *vm_next, *vm_prev;
 284
 285         struct rb_node vm_rb;
 286
 287         /*
 288          * Largest free memory gap in bytes to the left of this VMA.
 289          * Either between this VMA and vma->vm_prev, or between one of the
 290          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 291          * get_unmapped_area find a free area of the right size.
 292          */
 293         unsigned long rb_subtree_gap;
 294
 295         /* Second cache line starts here. */
 296
 297         struct mm_struct *vm_mm;        /* The address space we belong to. */
 298         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 299         unsigned long vm_flags;         /* Flags, see mm.h. */
 300
 301         /*
 302          * For areas with an address space and backing store,
 303          * linkage into the address_space->i_mmap interval tree.
 304          */
 305         struct {
 306                 struct rb_node rb;
 307                 unsigned long rb_subtree_last;
 308         } shared;
 309
 310         /*
 311          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 312          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 313          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 314          * or brk vma (with NULL file) can only be in an anon_vma list.
 315          */
 316         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 317                                           * page_table_lock */
 318         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 319
 320         /* Function pointers to deal with this struct. */
 321         const struct vm_operations_struct *vm_ops;
 322
 323         /* Information about our backing store: */
 324         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 325                                            units, *not* PAGE_CACHE_SIZE */
 326         struct file * vm_file;          /* File we map to (can be NULL). */
 327         void * vm_private_data;         /* was vm_pte (shared mem) */
 328
 329 #ifndef CONFIG_MMU
 330         struct vm_region *vm_region;    /* NOMMU mapping region */
 331 #endif
 332 #ifdef CONFIG_NUMA
 333         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 334 #endif
 335         struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 336 };
 337
 338 struct core_thread {
 339         struct task_struct *task;
 340         struct core_thread *next;
 341 };
 342
 343 struct core_state {
 344         atomic_t nr_threads;
 345         struct core_thread dumper;
 346         struct completion startup;
 347 };
 348
 349 enum {
 350         MM_FILEPAGES,
 351         MM_ANONPAGES,
 352         MM_SWAPENTS,
 353         NR_MM_COUNTERS
 354 };
 355
 356 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 357 #define SPLIT_RSS_COUNTING
 358 /* per-thread cached information, */
 359 struct task_rss_stat {
 360         int events;     /* for synchronization threshold */
 361         int count[NR_MM_COUNTERS];
 362 };
 363 #endif /* USE_SPLIT_PTE_PTLOCKS */
 364
 365 struct mm_rss_stat {
 366         atomic_long_t count[NR_MM_COUNTERS];
 367 };
 368
 369 struct kioctx_table;
 370 struct mm_struct {
 371         struct vm_area_struct *mmap;            /* list of VMAs */
 372         struct rb_root mm_rb;
 373         u32 vmacache_seqnum;                   /* per-thread vmacache */
 374 #ifdef CONFIG_MMU
 375         unsigned long (*get_unmapped_area) (struct file *filp,
 376                                 unsigned long addr, unsigned long len,
 377                                 unsigned long pgoff, unsigned long flags);
 378 #endif
 379         unsigned long mmap_base;                /* base of mmap area */
 380         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 381         unsigned long task_size;                /* size of task vm space */
 382         unsigned long highest_vm_end;           /* highest vma end address */
 383         pgd_t * pgd;
 384         atomic_t mm_users;                      /* How many users with user space? */
 385         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 386         atomic_long_t nr_ptes;                  /* PTE page table pages */
 387 #if CONFIG_PGTABLE_LEVELS > 2
 388         atomic_long_t nr_pmds;                  /* PMD page table pages */
 389 #endif
 390         int map_count;                          /* number of VMAs */
 391
 392         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 393         struct rw_semaphore mmap_sem;
 394
 395         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 396                                                  * together off init_mm.mmlist, and are protected
 397                                                  * by mmlist_lock
 398                                                  */
 399
 400
 401         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 402         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 403
 404         unsigned long total_vm;         /* Total pages mapped */
 405         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 406         unsigned long pinned_vm;        /* Refcount permanently increased */
 407         unsigned long shared_vm;        /* Shared pages (files) */
 408         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 409         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 410         unsigned long def_flags;
 411         unsigned long start_code, end_code, start_data, end_data;
 412         unsigned long start_brk, brk, start_stack;
 413         unsigned long arg_start, arg_end, env_start, env_end;
 414
 415         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 416
 417         /*
 418          * Special counters, in some configurations protected by the
 419          * page_table_lock, in other configurations by being atomic.
 420          */
 421         struct mm_rss_stat rss_stat;
 422
 423         struct linux_binfmt *binfmt;
 424
 425         cpumask_var_t cpu_vm_mask_var;
 426
 427         /* Architecture-specific MM context */
 428         mm_context_t context;
 429
 430         unsigned long flags; /* Must use atomic bitops to access the bits */
 431
 432         struct core_state *core_state; /* coredumping support */
 433 #ifdef CONFIG_AIO
 434         spinlock_t                      ioctx_lock;
 435         struct kioctx_table __rcu       *ioctx_table;
 436 #endif
 437 #ifdef CONFIG_MEMCG
 438         /*
 439          * "owner" points to a task that is regarded as the canonical
 440          * user/owner of this mm. All of the following must be true in
 441          * order for it to be changed:
 442          *
 443          * current == mm->owner
 444          * current->mm != mm
 445          * new_owner->mm == mm
 446          * new_owner->alloc_lock is held
 447          */
 448         struct task_struct __rcu *owner;
 449 #endif
 450
 451         /* store ref to file /proc/<pid>/exe symlink points to */
 452         struct file __rcu *exe_file;
 453 #ifdef CONFIG_MMU_NOTIFIER
 454         struct mmu_notifier_mm *mmu_notifier_mm;
 455 #endif
 456 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 457         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 458 #endif
 459 #ifdef CONFIG_CPUMASK_OFFSTACK
 460         struct cpumask cpumask_allocation;
 461 #endif
 462 #ifdef CONFIG_NUMA_BALANCING
 463         /*
 464          * numa_next_scan is the next time that the PTEs will be marked
 465          * pte_numa. NUMA hinting faults will gather statistics and migrate
 466          * pages to new nodes if necessary.
 467          */
 468         unsigned long numa_next_scan;
 469
 470         /* Restart point for scanning and setting pte_numa */
 471         unsigned long numa_scan_offset;
 472
 473         /* numa_scan_seq prevents two threads setting pte_numa */
 474         int numa_scan_seq;
 475 #endif
 476 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 477         /*
 478          * An operation with batched TLB flushing is going on. Anything that
 479          * can move process memory needs to flush the TLB when moving a
 480          * PROT_NONE or PROT_NUMA mapped page.
 481          */
 482         bool tlb_flush_pending;
 483 #endif
 484         struct uprobes_state uprobes_state;
 485 #ifdef CONFIG_X86_INTEL_MPX
 486         /* address of the bounds directory */
 487         void __user *bd_addr;
 488 #endif
 489 };
 490
 491 static inline void mm_init_cpumask(struct mm_struct *mm)
 492 {
 493 #ifdef CONFIG_CPUMASK_OFFSTACK
 494         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 495 #endif
 496         cpumask_clear(mm->cpu_vm_mask_var);
 497 }
 498
 499 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 500 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 501 {
 502         return mm->cpu_vm_mask_var;
 503 }
 504
 505 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 506 /*
 507  * Memory barriers to keep this state in sync are graciously provided by
 508  * the page table locks, outside of which no page table modifications happen.
 509  * The barriers below prevent the compiler from re-ordering the instructions
 510  * around the memory barriers that are already present in the code.
 511  */
 512 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 513 {
 514         barrier();
 515         return mm->tlb_flush_pending;
 516 }
 517 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 518 {
 519         mm->tlb_flush_pending = true;
 520
 521         /*
 522          * Guarantee that the tlb_flush_pending store does not leak into the
 523          * critical section updating the page tables
 524          */
 525         smp_mb__before_spinlock();
 526 }
 527 /* Clearing is done after a TLB flush, which also provides a barrier. */
 528 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 529 {
 530         barrier();
 531         mm->tlb_flush_pending = false;
 532 }
 533 #else
 534 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 535 {
 536         return false;
 537 }
 538 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 539 {
 540 }
 541 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 542 {
 543 }
 544 #endif
 545
 546 struct vm_special_mapping
 547 {
 548         const char *name;
 549         struct page **pages;
 550 };
 551
 552 enum tlb_flush_reason {
 553         TLB_FLUSH_ON_TASK_SWITCH,
 554         TLB_REMOTE_SHOOTDOWN,
 555         TLB_LOCAL_SHOOTDOWN,
 556         TLB_LOCAL_MM_SHOOTDOWN,
 557         TLB_REMOTE_SEND_IPI,
 558         NR_TLB_FLUSH_REASONS,
 559 };
 560
 561  /*
 562   * A swap entry has to fit into a "unsigned long", as the entry is hidden
 563   * in the "index" field of the swapper address space.
 564   */
 565 typedef struct {
 566         unsigned long val;
 567 } swp_entry_t;
 568
 569 #endif /* _LINUX_MM_TYPES_H */