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