include/linux/mm_types.h

   1 #ifndef _LINUX_MM_TYPES_H
   2 #define _LINUX_MM_TYPES_H
   3
   4 #include <linux/auxvec.h>
   5 #include <linux/types.h>
   6 #include <linux/threads.h>
   7 #include <linux/list.h>
   8 #include <linux/spinlock.h>
   9 #include <linux/rbtree.h>
  10 #include <linux/rwsem.h>
  11 #include <linux/completion.h>
  12 #include <linux/cpumask.h>
  13 #include <linux/page-debug-flags.h>
  14 #include <linux/uprobes.h>
  15 #include <linux/page-flags-layout.h>
  16 #include <asm/page.h>
  17 #include <asm/mmu.h>
  18
  19 #ifndef AT_VECTOR_SIZE_ARCH
  20 #define AT_VECTOR_SIZE_ARCH 0
  21 #endif
  22 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
  23
  24 struct address_space;
  25
  26 #define USE_SPLIT_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                                          */
 128                 struct {                /* slub per cpu partial pages */
 129                         struct page *next;      /* Next partial slab */
 130 #ifdef CONFIG_64BIT
 131                         int pages;      /* Nr of partial slabs left */
 132                         int pobjects;   /* Approximate # of objects */
 133 #else
 134                         short int pages;
 135                         short int pobjects;
 136 #endif
 137                 };
 138
 139                 struct list_head list;  /* slobs list of pages */
 140                 struct slab *slab_page; /* slab fields */
 141                 struct rcu_head rcu_head;       /* Used by SLAB
 142                                                  * when destroying via RCU
 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         /*
 170          * On machines where all RAM is mapped into kernel address space,
 171          * we can simply calculate the virtual address. On machines with
 172          * highmem some memory is mapped into kernel virtual memory
 173          * dynamically, so we need a place to store that address.
 174          * Note that this field could be 16 bits on x86 ... ;)
 175          *
 176          * Architectures with slow multiplication can define
 177          * WANT_PAGE_VIRTUAL in asm/page.h
 178          */
 179 #if defined(WANT_PAGE_VIRTUAL)
 180         void *virtual;                  /* Kernel virtual address (NULL if
 181                                            not kmapped, ie. highmem) */
 182 #endif /* WANT_PAGE_VIRTUAL */
 183 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
 184         unsigned long debug_flags;      /* Use atomic bitops on this */
 185 #endif
 186
 187 #ifdef CONFIG_KMEMCHECK
 188         /*
 189          * kmemcheck wants to track the status of each byte in a page; this
 190          * is a pointer to such a status block. NULL if not tracked.
 191          */
 192         void *shadow;
 193 #endif
 194
 195 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 196         int _last_cpupid;
 197 #endif
 198 }
 199 /*
 200  * The struct page can be forced to be double word aligned so that atomic ops
 201  * on double words work. The SLUB allocator can make use of such a feature.
 202  */
 203 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 204         __aligned(2 * sizeof(unsigned long))
 205 #endif
 206 ;
 207
 208 struct page_frag {
 209         struct page *page;
 210 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 211         __u32 offset;
 212         __u32 size;
 213 #else
 214         __u16 offset;
 215         __u16 size;
 216 #endif
 217 };
 218
 219 typedef unsigned long __nocast vm_flags_t;
 220
 221 /*
 222  * A region containing a mapping of a non-memory backed file under NOMMU
 223  * conditions.  These are held in a global tree and are pinned by the VMAs that
 224  * map parts of them.
 225  */
 226 struct vm_region {
 227         struct rb_node  vm_rb;          /* link in global region tree */
 228         vm_flags_t      vm_flags;       /* VMA vm_flags */
 229         unsigned long   vm_start;       /* start address of region */
 230         unsigned long   vm_end;         /* region initialised to here */
 231         unsigned long   vm_top;         /* region allocated to here */
 232         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 233         struct file     *vm_file;       /* the backing file or NULL */
 234
 235         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 236         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 237                                                 * this region */
 238 };
 239
 240 /*
 241  * This struct defines a memory VMM memory area. There is one of these
 242  * per VM-area/task.  A VM area is any part of the process virtual memory
 243  * space that has a special rule for the page-fault handlers (ie a shared
 244  * library, the executable area etc).
 245  */
 246 struct vm_area_struct {
 247         /* The first cache line has the info for VMA tree walking. */
 248
 249         unsigned long vm_start;         /* Our start address within vm_mm. */
 250         unsigned long vm_end;           /* The first byte after our end address
 251                                            within vm_mm. */
 252
 253         /* linked list of VM areas per task, sorted by address */
 254         struct vm_area_struct *vm_next, *vm_prev;
 255
 256         struct rb_node vm_rb;
 257
 258         /*
 259          * Largest free memory gap in bytes to the left of this VMA.
 260          * Either between this VMA and vma->vm_prev, or between one of the
 261          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 262          * get_unmapped_area find a free area of the right size.
 263          */
 264         unsigned long rb_subtree_gap;
 265
 266         /* Second cache line starts here. */
 267
 268         struct mm_struct *vm_mm;        /* The address space we belong to. */
 269         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 270         unsigned long vm_flags;         /* Flags, see mm.h. */
 271
 272         /*
 273          * For areas with an address space and backing store,
 274          * linkage into the address_space->i_mmap interval tree, or
 275          * linkage of vma in the address_space->i_mmap_nonlinear list.
 276          */
 277         union {
 278                 struct {
 279                         struct rb_node rb;
 280                         unsigned long rb_subtree_last;
 281                 } linear;
 282                 struct list_head nonlinear;
 283         } shared;
 284
 285         /*
 286          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 287          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 288          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 289          * or brk vma (with NULL file) can only be in an anon_vma list.
 290          */
 291         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 292                                           * page_table_lock */
 293         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 294
 295         /* Function pointers to deal with this struct. */
 296         const struct vm_operations_struct *vm_ops;
 297
 298         /* Information about our backing store: */
 299         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 300                                            units, *not* PAGE_CACHE_SIZE */
 301         struct file * vm_file;          /* File we map to (can be NULL). */
 302         void * vm_private_data;         /* was vm_pte (shared mem) */
 303
 304 #ifndef CONFIG_MMU
 305         struct vm_region *vm_region;    /* NOMMU mapping region */
 306 #endif
 307 #ifdef CONFIG_NUMA
 308         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 309 #endif
 310 };
 311
 312 struct core_thread {
 313         struct task_struct *task;
 314         struct core_thread *next;
 315 };
 316
 317 struct core_state {
 318         atomic_t nr_threads;
 319         struct core_thread dumper;
 320         struct completion startup;
 321 };
 322
 323 enum {
 324         MM_FILEPAGES,
 325         MM_ANONPAGES,
 326         MM_SWAPENTS,
 327         NR_MM_COUNTERS
 328 };
 329
 330 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 331 #define SPLIT_RSS_COUNTING
 332 /* per-thread cached information, */
 333 struct task_rss_stat {
 334         int events;     /* for synchronization threshold */
 335         int count[NR_MM_COUNTERS];
 336 };
 337 #endif /* USE_SPLIT_PTE_PTLOCKS */
 338
 339 struct mm_rss_stat {
 340         atomic_long_t count[NR_MM_COUNTERS];
 341 };
 342
 343 struct kioctx_table;
 344 struct mm_struct {
 345         struct vm_area_struct * mmap;           /* list of VMAs */
 346         struct rb_root mm_rb;
 347         struct vm_area_struct * mmap_cache;     /* last find_vma result */
 348 #ifdef CONFIG_MMU
 349         unsigned long (*get_unmapped_area) (struct file *filp,
 350                                 unsigned long addr, unsigned long len,
 351                                 unsigned long pgoff, unsigned long flags);
 352 #endif
 353         unsigned long mmap_base;                /* base of mmap area */
 354         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 355         unsigned long task_size;                /* size of task vm space */
 356         unsigned long highest_vm_end;           /* highest vma end address */
 357         pgd_t * pgd;
 358         atomic_t mm_users;                      /* How many users with user space? */
 359         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 360         atomic_long_t nr_ptes;                  /* Page table pages */
 361         int map_count;                          /* number of VMAs */
 362
 363         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 364         struct rw_semaphore mmap_sem;
 365
 366         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 367                                                  * together off init_mm.mmlist, and are protected
 368                                                  * by mmlist_lock
 369                                                  */
 370
 371
 372         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 373         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 374
 375         unsigned long total_vm;         /* Total pages mapped */
 376         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 377         unsigned long pinned_vm;        /* Refcount permanently increased */
 378         unsigned long shared_vm;        /* Shared pages (files) */
 379         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 380         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 381         unsigned long def_flags;
 382         unsigned long start_code, end_code, start_data, end_data;
 383         unsigned long start_brk, brk, start_stack;
 384         unsigned long arg_start, arg_end, env_start, env_end;
 385
 386         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 387
 388         /*
 389          * Special counters, in some configurations protected by the
 390          * page_table_lock, in other configurations by being atomic.
 391          */
 392         struct mm_rss_stat rss_stat;
 393
 394         struct linux_binfmt *binfmt;
 395
 396         cpumask_var_t cpu_vm_mask_var;
 397
 398         /* Architecture-specific MM context */
 399         mm_context_t context;
 400
 401         unsigned long flags; /* Must use atomic bitops to access the bits */
 402
 403         struct core_state *core_state; /* coredumping support */
 404 #ifdef CONFIG_AIO
 405         spinlock_t                      ioctx_lock;
 406         struct kioctx_table __rcu       *ioctx_table;
 407 #endif
 408 #ifdef CONFIG_MM_OWNER
 409         /*
 410          * "owner" points to a task that is regarded as the canonical
 411          * user/owner of this mm. All of the following must be true in
 412          * order for it to be changed:
 413          *
 414          * current == mm->owner
 415          * current->mm != mm
 416          * new_owner->mm == mm
 417          * new_owner->alloc_lock is held
 418          */
 419         struct task_struct __rcu *owner;
 420 #endif
 421
 422         /* store ref to file /proc/<pid>/exe symlink points to */
 423         struct file *exe_file;
 424 #ifdef CONFIG_MMU_NOTIFIER
 425         struct mmu_notifier_mm *mmu_notifier_mm;
 426 #endif
 427 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 428         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 429 #endif
 430 #ifdef CONFIG_CPUMASK_OFFSTACK
 431         struct cpumask cpumask_allocation;
 432 #endif
 433 #ifdef CONFIG_NUMA_BALANCING
 434         /*
 435          * numa_next_scan is the next time that the PTEs will be marked
 436          * pte_numa. NUMA hinting faults will gather statistics and migrate
 437          * pages to new nodes if necessary.
 438          */
 439         unsigned long numa_next_scan;
 440
 441         /* Restart point for scanning and setting pte_numa */
 442         unsigned long numa_scan_offset;
 443
 444         /* numa_scan_seq prevents two threads setting pte_numa */
 445         int numa_scan_seq;
 446 #endif
 447 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 448         /*
 449          * An operation with batched TLB flushing is going on. Anything that
 450          * can move process memory needs to flush the TLB when moving a
 451          * PROT_NONE or PROT_NUMA mapped page.
 452          */
 453         bool tlb_flush_pending;
 454 #endif
 455         struct uprobes_state uprobes_state;
 456 };
 457
 458 static inline void mm_init_cpumask(struct mm_struct *mm)
 459 {
 460 #ifdef CONFIG_CPUMASK_OFFSTACK
 461         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 462 #endif
 463 }
 464
 465 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 466 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 467 {
 468         return mm->cpu_vm_mask_var;
 469 }
 470
 471 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 472 /*
 473  * Memory barriers to keep this state in sync are graciously provided by
 474  * the page table locks, outside of which no page table modifications happen.
 475  * The barriers below prevent the compiler from re-ordering the instructions
 476  * around the memory barriers that are already present in the code.
 477  */
 478 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 479 {
 480         barrier();
 481         return mm->tlb_flush_pending;
 482 }
 483 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 484 {
 485         mm->tlb_flush_pending = true;
 486
 487         /*
 488          * Guarantee that the tlb_flush_pending store does not leak into the
 489          * critical section updating the page tables
 490          */
 491         smp_mb__before_spinlock();
 492 }
 493 /* Clearing is done after a TLB flush, which also provides a barrier. */
 494 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 495 {
 496         barrier();
 497         mm->tlb_flush_pending = false;
 498 }
 499 #else
 500 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 501 {
 502         return false;
 503 }
 504 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 505 {
 506 }
 507 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 508 {
 509 }
 510 #endif
 511
 512 #endif /* _LINUX_MM_TYPES_H */