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 <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 23struct address_space; 24 25#define USE_SPLIT_PTLOCKS (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS) 26 27/* 28 * Each physical page in the system has a struct page associated with 29 * it to keep track of whatever it is we are using the page for at the 30 * moment. Note that we have no way to track which tasks are using 31 * a page, though if it is a pagecache page, rmap structures can tell us 32 * who is mapping it. 33 * 34 * The objects in struct page are organized in double word blocks in 35 * order to allows us to use atomic double word operations on portions 36 * of struct page. That is currently only used by slub but the arrangement 37 * allows the use of atomic double word operations on the flags/mapping 38 * and lru list pointers also. 39 */ 40struct page { 41 /* First double word block */ 42 unsigned long flags; /* Atomic flags, some possibly 43 * updated asynchronously */ 44 struct address_space *mapping; /* If low bit clear, points to 45 * inode address_space, or NULL. 46 * If page mapped as anonymous 47 * memory, low bit is set, and 48 * it points to anon_vma object: 49 * see PAGE_MAPPING_ANON below. 50 */ 51 /* Second double word */ 52 struct { 53 union { 54 pgoff_t index; /* Our offset within mapping. */ 55 void *freelist; /* slub/slob first free object */ 56 bool pfmemalloc; /* If set by the page allocator, 57 * ALLOC_NO_WATERMARKS was set 58 * and the low watermark was not 59 * met implying that the system 60 * is under some pressure. The 61 * caller should try ensure 62 * this page is only used to 63 * free other pages. 64 */ 65 }; 66 67 union { 68#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ 69 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) 70 /* Used for cmpxchg_double in slub */ 71 unsigned long counters; 72#else 73 /* 74 * Keep _count separate from slub cmpxchg_double data. 75 * As the rest of the double word is protected by 76 * slab_lock but _count is not. 77 */ 78 unsigned counters; 79#endif 80 81 struct { 82 83 union { 84 /* 85 * Count of ptes mapped in 86 * mms, to show when page is 87 * mapped & limit reverse map 88 * searches. 89 * 90 * Used also for tail pages 91 * refcounting instead of 92 * _count. Tail pages cannot 93 * be mapped and keeping the 94 * tail page _count zero at 95 * all times guarantees 96 * get_page_unless_zero() will 97 * never succeed on tail 98 * pages. 99 */ 100 atomic_t _mapcount; 101 102 struct { /* SLUB */ 103 unsigned inuse:16; 104 unsigned objects:15; 105 unsigned frozen:1; 106 }; 107 int units; /* SLOB */ 108 }; 109 atomic_t _count; /* Usage count, see below. */ 110 }; 111 }; 112 }; 113 114 /* Third double word block */ 115 union { 116 struct list_head lru; /* Pageout list, eg. active_list 117 * protected by zone->lru_lock ! 118 */ 119 struct { /* slub per cpu partial pages */ 120 struct page *next; /* Next partial slab */ 121#ifdef CONFIG_64BIT 122 int pages; /* Nr of partial slabs left */ 123 int pobjects; /* Approximate # of objects */ 124#else 125 short int pages; 126 short int pobjects; 127#endif 128 }; 129 130 struct list_head list; /* slobs list of pages */ 131 struct slab *slab_page; /* slab fields */ 132 }; 133 134 /* Remainder is not double word aligned */ 135 union { 136 unsigned long private; /* Mapping-private opaque data: 137 * usually used for buffer_heads 138 * if PagePrivate set; used for 139 * swp_entry_t if PageSwapCache; 140 * indicates order in the buddy 141 * system if PG_buddy is set. 142 */ 143#if USE_SPLIT_PTLOCKS 144 spinlock_t ptl; 145#endif 146 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */ 147 struct page *first_page; /* Compound tail pages */ 148 }; 149 150 /* 151 * On machines where all RAM is mapped into kernel address space, 152 * we can simply calculate the virtual address. On machines with 153 * highmem some memory is mapped into kernel virtual memory 154 * dynamically, so we need a place to store that address. 155 * Note that this field could be 16 bits on x86 ... ;) 156 * 157 * Architectures with slow multiplication can define 158 * WANT_PAGE_VIRTUAL in asm/page.h 159 */ 160#if defined(WANT_PAGE_VIRTUAL) 161 void *virtual; /* Kernel virtual address (NULL if 162 not kmapped, ie. highmem) */ 163#endif /* WANT_PAGE_VIRTUAL */ 164#ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS 165 unsigned long debug_flags; /* Use atomic bitops on this */ 166#endif 167 168#ifdef CONFIG_KMEMCHECK 169 /* 170 * kmemcheck wants to track the status of each byte in a page; this 171 * is a pointer to such a status block. NULL if not tracked. 172 */ 173 void *shadow; 174#endif 175 176#ifdef CONFIG_NUMA_BALANCING 177 int _last_nid; 178#endif 179} 180/* 181 * The struct page can be forced to be double word aligned so that atomic ops 182 * on double words work. The SLUB allocator can make use of such a feature. 183 */ 184#ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE 185 __aligned(2 * sizeof(unsigned long)) 186#endif 187; 188 189struct page_frag { 190 struct page *page; 191#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 192 __u32 offset; 193 __u32 size; 194#else 195 __u16 offset; 196 __u16 size; 197#endif 198}; 199 200typedef unsigned long __nocast vm_flags_t; 201 202/* 203 * A region containing a mapping of a non-memory backed file under NOMMU 204 * conditions. These are held in a global tree and are pinned by the VMAs that 205 * map parts of them. 206 */ 207struct vm_region { 208 struct rb_node vm_rb; /* link in global region tree */ 209 vm_flags_t vm_flags; /* VMA vm_flags */ 210 unsigned long vm_start; /* start address of region */ 211 unsigned long vm_end; /* region initialised to here */ 212 unsigned long vm_top; /* region allocated to here */ 213 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */ 214 struct file *vm_file; /* the backing file or NULL */ 215 216 int vm_usage; /* region usage count (access under nommu_region_sem) */ 217 bool vm_icache_flushed : 1; /* true if the icache has been flushed for 218 * this region */ 219}; 220 221/* 222 * This struct defines a memory VMM memory area. There is one of these 223 * per VM-area/task. A VM area is any part of the process virtual memory 224 * space that has a special rule for the page-fault handlers (ie a shared 225 * library, the executable area etc). 226 */ 227struct vm_area_struct { 228 /* The first cache line has the info for VMA tree walking. */ 229 230 unsigned long vm_start; /* Our start address within vm_mm. */ 231 unsigned long vm_end; /* The first byte after our end address 232 within vm_mm. */ 233 234 /* linked list of VM areas per task, sorted by address */ 235 struct vm_area_struct *vm_next, *vm_prev; 236 237 struct rb_node vm_rb; 238 239 /* 240 * Largest free memory gap in bytes to the left of this VMA. 241 * Either between this VMA and vma->vm_prev, or between one of the 242 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps 243 * get_unmapped_area find a free area of the right size. 244 */ 245 unsigned long rb_subtree_gap; 246 247 /* Second cache line starts here. */ 248 249 struct mm_struct *vm_mm; /* The address space we belong to. */ 250 pgprot_t vm_page_prot; /* Access permissions of this VMA. */ 251 unsigned long vm_flags; /* Flags, see mm.h. */ 252 253 /* 254 * For areas with an address space and backing store, 255 * linkage into the address_space->i_mmap interval tree, or 256 * linkage of vma in the address_space->i_mmap_nonlinear list. 257 */ 258 union { 259 struct { 260 struct rb_node rb; 261 unsigned long rb_subtree_last; 262 } linear; 263 struct list_head nonlinear; 264 } shared; 265 266 /* 267 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma 268 * list, after a COW of one of the file pages. A MAP_SHARED vma 269 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack 270 * or brk vma (with NULL file) can only be in an anon_vma list. 271 */ 272 struct list_head anon_vma_chain; /* Serialized by mmap_sem & 273 * page_table_lock */ 274 struct anon_vma *anon_vma; /* Serialized by page_table_lock */ 275 276 /* Function pointers to deal with this struct. */ 277 const struct vm_operations_struct *vm_ops; 278 279 /* Information about our backing store: */ 280 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE 281 units, *not* PAGE_CACHE_SIZE */ 282 struct file * vm_file; /* File we map to (can be NULL). */ 283 void * vm_private_data; /* was vm_pte (shared mem) */ 284 285#ifndef CONFIG_MMU 286 struct vm_region *vm_region; /* NOMMU mapping region */ 287#endif 288#ifdef CONFIG_NUMA 289 struct mempolicy *vm_policy; /* NUMA policy for the VMA */ 290#endif 291}; 292 293struct core_thread { 294 struct task_struct *task; 295 struct core_thread *next; 296}; 297 298struct core_state { 299 atomic_t nr_threads; 300 struct core_thread dumper; 301 struct completion startup; 302}; 303 304enum { 305 MM_FILEPAGES, 306 MM_ANONPAGES, 307 MM_SWAPENTS, 308 NR_MM_COUNTERS 309}; 310 311#if USE_SPLIT_PTLOCKS && defined(CONFIG_MMU) 312#define SPLIT_RSS_COUNTING 313/* per-thread cached information, */ 314struct task_rss_stat { 315 int events; /* for synchronization threshold */ 316 int count[NR_MM_COUNTERS]; 317}; 318#endif /* USE_SPLIT_PTLOCKS */ 319 320struct mm_rss_stat { 321 atomic_long_t count[NR_MM_COUNTERS]; 322}; 323 324struct mm_struct { 325 struct vm_area_struct * mmap; /* list of VMAs */ 326 struct rb_root mm_rb; 327 struct vm_area_struct * mmap_cache; /* last find_vma result */ 328#ifdef CONFIG_MMU 329 unsigned long (*get_unmapped_area) (struct file *filp, 330 unsigned long addr, unsigned long len, 331 unsigned long pgoff, unsigned long flags); 332 void (*unmap_area) (struct mm_struct *mm, unsigned long addr); 333#endif 334 unsigned long mmap_base; /* base of mmap area */ 335 unsigned long task_size; /* size of task vm space */ 336 unsigned long cached_hole_size; /* if non-zero, the largest hole below free_area_cache */ 337 unsigned long free_area_cache; /* first hole of size cached_hole_size or larger */ 338 unsigned long highest_vm_end; /* highest vma end address */ 339 pgd_t * pgd; 340 atomic_t mm_users; /* How many users with user space? */ 341 atomic_t mm_count; /* How many references to "struct mm_struct" (users count as 1) */ 342 int map_count; /* number of VMAs */ 343 344 spinlock_t page_table_lock; /* Protects page tables and some counters */ 345 struct rw_semaphore mmap_sem; 346 347 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung 348 * together off init_mm.mmlist, and are protected 349 * by mmlist_lock 350 */ 351 352 353 unsigned long hiwater_rss; /* High-watermark of RSS usage */ 354 unsigned long hiwater_vm; /* High-water virtual memory usage */ 355 356 unsigned long total_vm; /* Total pages mapped */ 357 unsigned long locked_vm; /* Pages that have PG_mlocked set */ 358 unsigned long pinned_vm; /* Refcount permanently increased */ 359 unsigned long shared_vm; /* Shared pages (files) */ 360 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE */ 361 unsigned long stack_vm; /* VM_GROWSUP/DOWN */ 362 unsigned long def_flags; 363 unsigned long nr_ptes; /* Page table pages */ 364 unsigned long start_code, end_code, start_data, end_data; 365 unsigned long start_brk, brk, start_stack; 366 unsigned long arg_start, arg_end, env_start, env_end; 367 368 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */ 369 370 /* 371 * Special counters, in some configurations protected by the 372 * page_table_lock, in other configurations by being atomic. 373 */ 374 struct mm_rss_stat rss_stat; 375 376 struct linux_binfmt *binfmt; 377 378 cpumask_var_t cpu_vm_mask_var; 379 380 /* Architecture-specific MM context */ 381 mm_context_t context; 382 383 unsigned long flags; /* Must use atomic bitops to access the bits */ 384 385 struct core_state *core_state; /* coredumping support */ 386#ifdef CONFIG_AIO 387 spinlock_t ioctx_lock; 388 struct hlist_head ioctx_list; 389#endif 390#ifdef CONFIG_MM_OWNER 391 /* 392 * "owner" points to a task that is regarded as the canonical 393 * user/owner of this mm. All of the following must be true in 394 * order for it to be changed: 395 * 396 * current == mm->owner 397 * current->mm != mm 398 * new_owner->mm == mm 399 * new_owner->alloc_lock is held 400 */ 401 struct task_struct __rcu *owner; 402#endif 403 404 /* store ref to file /proc/<pid>/exe symlink points to */ 405 struct file *exe_file; 406#ifdef CONFIG_MMU_NOTIFIER 407 struct mmu_notifier_mm *mmu_notifier_mm; 408#endif 409#ifdef CONFIG_TRANSPARENT_HUGEPAGE 410 pgtable_t pmd_huge_pte; /* protected by page_table_lock */ 411#endif 412#ifdef CONFIG_CPUMASK_OFFSTACK 413 struct cpumask cpumask_allocation; 414#endif 415#ifdef CONFIG_NUMA_BALANCING 416 /* 417 * numa_next_scan is the next time when the PTEs will me marked 418 * pte_numa to gather statistics and migrate pages to new nodes 419 * if necessary 420 */ 421 unsigned long numa_next_scan; 422 423 /* numa_next_reset is when the PTE scanner period will be reset */ 424 unsigned long numa_next_reset; 425 426 /* Restart point for scanning and setting pte_numa */ 427 unsigned long numa_scan_offset; 428 429 /* numa_scan_seq prevents two threads setting pte_numa */ 430 int numa_scan_seq; 431 432 /* 433 * The first node a task was scheduled on. If a task runs on 434 * a different node than Make PTE Scan Go Now. 435 */ 436 int first_nid; 437#endif 438 struct uprobes_state uprobes_state; 439}; 440 441/* first nid will either be a valid NID or one of these values */ 442#define NUMA_PTE_SCAN_INIT -1 443#define NUMA_PTE_SCAN_ACTIVE -2 444 445static inline void mm_init_cpumask(struct mm_struct *mm) 446{ 447#ifdef CONFIG_CPUMASK_OFFSTACK 448 mm->cpu_vm_mask_var = &mm->cpumask_allocation; 449#endif 450} 451 452/* Future-safe accessor for struct mm_struct's cpu_vm_mask. */ 453static inline cpumask_t *mm_cpumask(struct mm_struct *mm) 454{ 455 return mm->cpu_vm_mask_var; 456} 457 458#endif /* _LINUX_MM_TYPES_H */ 459