1 #ifndef _LINUX_MM_TYPES_H
2 #define _LINUX_MM_TYPES_H
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>
18 #ifndef AT_VECTOR_SIZE_ARCH
19 #define AT_VECTOR_SIZE_ARCH 0
21 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
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)
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
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.
45 /* First double word block */
46 unsigned long flags; /* Atomic flags, some possibly
47 * updated asynchronously */
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.
56 void *s_mem; /* slab first object */
59 /* Second double word */
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
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;
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.
93 * Count of ptes mapped in
94 * mms, to show when page is
95 * mapped & limit reverse map
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
115 int units; /* SLOB */
117 atomic_t _count; /* Usage count, see below. */
119 unsigned int active; /* SLAB */
123 /* Third double word block */
125 struct list_head lru; /* Pageout list, eg. active_list
126 * protected by zone->lru_lock !
127 * Can be used as a generic list
130 struct { /* slub per cpu partial pages */
131 struct page *next; /* Next partial slab */
133 int pages; /* Nr of partial slabs left */
134 int pobjects; /* Approximate # of objects */
141 struct slab *slab_page; /* slab fields */
142 struct rcu_head rcu_head; /* Used by SLAB
143 * when destroying via RCU
145 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
146 pgtable_t pmd_huge_pte; /* protected by page->ptl */
150 /* Remainder is not double word aligned */
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.
159 #if USE_SPLIT_PTE_PTLOCKS
160 #if ALLOC_SPLIT_PTLOCKS
166 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
167 struct page *first_page; /* Compound tail pages */
171 struct mem_cgroup *mem_cgroup;
175 * On machines where all RAM is mapped into kernel address space,
176 * we can simply calculate the virtual address. On machines with
177 * highmem some memory is mapped into kernel virtual memory
178 * dynamically, so we need a place to store that address.
179 * Note that this field could be 16 bits on x86 ... ;)
181 * Architectures with slow multiplication can define
182 * WANT_PAGE_VIRTUAL in asm/page.h
184 #if defined(WANT_PAGE_VIRTUAL)
185 void *virtual; /* Kernel virtual address (NULL if
186 not kmapped, ie. highmem) */
187 #endif /* WANT_PAGE_VIRTUAL */
189 #ifdef CONFIG_KMEMCHECK
191 * kmemcheck wants to track the status of each byte in a page; this
192 * is a pointer to such a status block. NULL if not tracked.
197 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
202 * The struct page can be forced to be double word aligned so that atomic ops
203 * on double words work. The SLUB allocator can make use of such a feature.
205 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
206 __aligned(2 * sizeof(unsigned long))
212 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
221 typedef unsigned long __nocast vm_flags_t;
224 * A region containing a mapping of a non-memory backed file under NOMMU
225 * conditions. These are held in a global tree and are pinned by the VMAs that
229 struct rb_node vm_rb; /* link in global region tree */
230 vm_flags_t vm_flags; /* VMA vm_flags */
231 unsigned long vm_start; /* start address of region */
232 unsigned long vm_end; /* region initialised to here */
233 unsigned long vm_top; /* region allocated to here */
234 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
235 struct file *vm_file; /* the backing file or NULL */
237 int vm_usage; /* region usage count (access under nommu_region_sem) */
238 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
243 * This struct defines a memory VMM memory area. There is one of these
244 * per VM-area/task. A VM area is any part of the process virtual memory
245 * space that has a special rule for the page-fault handlers (ie a shared
246 * library, the executable area etc).
248 struct vm_area_struct {
249 /* The first cache line has the info for VMA tree walking. */
251 unsigned long vm_start; /* Our start address within vm_mm. */
252 unsigned long vm_end; /* The first byte after our end address
255 /* linked list of VM areas per task, sorted by address */
256 struct vm_area_struct *vm_next, *vm_prev;
258 struct rb_node vm_rb;
261 * Largest free memory gap in bytes to the left of this VMA.
262 * Either between this VMA and vma->vm_prev, or between one of the
263 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
264 * get_unmapped_area find a free area of the right size.
266 unsigned long rb_subtree_gap;
268 /* Second cache line starts here. */
270 struct mm_struct *vm_mm; /* The address space we belong to. */
271 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
272 unsigned long vm_flags; /* Flags, see mm.h. */
275 * For areas with an address space and backing store,
276 * linkage into the address_space->i_mmap interval tree.
280 unsigned long rb_subtree_last;
284 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
285 * list, after a COW of one of the file pages. A MAP_SHARED vma
286 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
287 * or brk vma (with NULL file) can only be in an anon_vma list.
289 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
291 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
293 /* Function pointers to deal with this struct. */
294 const struct vm_operations_struct *vm_ops;
296 /* Information about our backing store: */
297 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
298 units, *not* PAGE_CACHE_SIZE */
299 struct file * vm_file; /* File we map to (can be NULL). */
300 void * vm_private_data; /* was vm_pte (shared mem) */
303 struct vm_region *vm_region; /* NOMMU mapping region */
306 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
311 struct task_struct *task;
312 struct core_thread *next;
317 struct core_thread dumper;
318 struct completion startup;
328 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
329 #define SPLIT_RSS_COUNTING
330 /* per-thread cached information, */
331 struct task_rss_stat {
332 int events; /* for synchronization threshold */
333 int count[NR_MM_COUNTERS];
335 #endif /* USE_SPLIT_PTE_PTLOCKS */
338 atomic_long_t count[NR_MM_COUNTERS];
343 struct vm_area_struct *mmap; /* list of VMAs */
344 struct rb_root mm_rb;
345 u32 vmacache_seqnum; /* per-thread vmacache */
347 unsigned long (*get_unmapped_area) (struct file *filp,
348 unsigned long addr, unsigned long len,
349 unsigned long pgoff, unsigned long flags);
351 unsigned long mmap_base; /* base of mmap area */
352 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
353 unsigned long task_size; /* size of task vm space */
354 unsigned long highest_vm_end; /* highest vma end address */
356 atomic_t mm_users; /* How many users with user space? */
357 atomic_t mm_count; /* How many references to "struct mm_struct" (users count as 1) */
358 atomic_long_t nr_ptes; /* Page table pages */
359 int map_count; /* number of VMAs */
361 spinlock_t page_table_lock; /* Protects page tables and some counters */
362 struct rw_semaphore mmap_sem;
364 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
365 * together off init_mm.mmlist, and are protected
370 unsigned long hiwater_rss; /* High-watermark of RSS usage */
371 unsigned long hiwater_vm; /* High-water virtual memory usage */
373 unsigned long total_vm; /* Total pages mapped */
374 unsigned long locked_vm; /* Pages that have PG_mlocked set */
375 unsigned long pinned_vm; /* Refcount permanently increased */
376 unsigned long shared_vm; /* Shared pages (files) */
377 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE */
378 unsigned long stack_vm; /* VM_GROWSUP/DOWN */
379 unsigned long def_flags;
380 unsigned long start_code, end_code, start_data, end_data;
381 unsigned long start_brk, brk, start_stack;
382 unsigned long arg_start, arg_end, env_start, env_end;
384 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
387 * Special counters, in some configurations protected by the
388 * page_table_lock, in other configurations by being atomic.
390 struct mm_rss_stat rss_stat;
392 struct linux_binfmt *binfmt;
394 cpumask_var_t cpu_vm_mask_var;
396 /* Architecture-specific MM context */
397 mm_context_t context;
399 unsigned long flags; /* Must use atomic bitops to access the bits */
401 struct core_state *core_state; /* coredumping support */
403 spinlock_t ioctx_lock;
404 struct kioctx_table __rcu *ioctx_table;
408 * "owner" points to a task that is regarded as the canonical
409 * user/owner of this mm. All of the following must be true in
410 * order for it to be changed:
412 * current == mm->owner
414 * new_owner->mm == mm
415 * new_owner->alloc_lock is held
417 struct task_struct __rcu *owner;
420 /* store ref to file /proc/<pid>/exe symlink points to */
421 struct file *exe_file;
422 #ifdef CONFIG_MMU_NOTIFIER
423 struct mmu_notifier_mm *mmu_notifier_mm;
425 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
426 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
428 #ifdef CONFIG_CPUMASK_OFFSTACK
429 struct cpumask cpumask_allocation;
431 #ifdef CONFIG_NUMA_BALANCING
433 * numa_next_scan is the next time that the PTEs will be marked
434 * pte_numa. NUMA hinting faults will gather statistics and migrate
435 * pages to new nodes if necessary.
437 unsigned long numa_next_scan;
439 /* Restart point for scanning and setting pte_numa */
440 unsigned long numa_scan_offset;
442 /* numa_scan_seq prevents two threads setting pte_numa */
445 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
447 * An operation with batched TLB flushing is going on. Anything that
448 * can move process memory needs to flush the TLB when moving a
449 * PROT_NONE or PROT_NUMA mapped page.
451 bool tlb_flush_pending;
453 struct uprobes_state uprobes_state;
454 #ifdef CONFIG_X86_INTEL_MPX
455 /* address of the bounds directory */
456 void __user *bd_addr;
460 static inline void mm_init_cpumask(struct mm_struct *mm)
462 #ifdef CONFIG_CPUMASK_OFFSTACK
463 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
465 cpumask_clear(mm->cpu_vm_mask_var);
468 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
469 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
471 return mm->cpu_vm_mask_var;
474 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
476 * Memory barriers to keep this state in sync are graciously provided by
477 * the page table locks, outside of which no page table modifications happen.
478 * The barriers below prevent the compiler from re-ordering the instructions
479 * around the memory barriers that are already present in the code.
481 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
484 return mm->tlb_flush_pending;
486 static inline void set_tlb_flush_pending(struct mm_struct *mm)
488 mm->tlb_flush_pending = true;
491 * Guarantee that the tlb_flush_pending store does not leak into the
492 * critical section updating the page tables
494 smp_mb__before_spinlock();
496 /* Clearing is done after a TLB flush, which also provides a barrier. */
497 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
500 mm->tlb_flush_pending = false;
503 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
507 static inline void set_tlb_flush_pending(struct mm_struct *mm)
510 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
515 struct vm_special_mapping
521 enum tlb_flush_reason {
522 TLB_FLUSH_ON_TASK_SWITCH,
523 TLB_REMOTE_SHOOTDOWN,
525 TLB_LOCAL_MM_SHOOTDOWN,
526 NR_TLB_FLUSH_REASONS,
530 * A swap entry has to fit into a "unsigned long", as the entry is hidden
531 * in the "index" field of the swapper address space.
537 #endif /* _LINUX_MM_TYPES_H */