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/rcupdate.h>
15 #include <linux/page-flags-layout.h>
19 #ifndef AT_VECTOR_SIZE_ARCH
20 #define AT_VECTOR_SIZE_ARCH 0
22 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
27 #define USE_SPLIT_PTE_PTLOCKS (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
28 #define USE_SPLIT_PMD_PTLOCKS (USE_SPLIT_PTE_PTLOCKS && \
29 IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
30 #define ALLOC_SPLIT_PTLOCKS (SPINLOCK_SIZE > BITS_PER_LONG/8)
32 typedef void compound_page_dtor(struct page *);
35 * Each physical page in the system has a struct page associated with
36 * it to keep track of whatever it is we are using the page for at the
37 * moment. Note that we have no way to track which tasks are using
38 * a page, though if it is a pagecache page, rmap structures can tell us
41 * The objects in struct page are organized in double word blocks in
42 * order to allows us to use atomic double word operations on portions
43 * of struct page. That is currently only used by slub but the arrangement
44 * allows the use of atomic double word operations on the flags/mapping
45 * and lru list pointers also.
48 /* First double word block */
49 unsigned long flags; /* Atomic flags, some possibly
50 * updated asynchronously */
52 struct address_space *mapping; /* If low bit clear, points to
53 * inode address_space, or NULL.
54 * If page mapped as anonymous
55 * memory, low bit is set, and
56 * it points to anon_vma object:
57 * see PAGE_MAPPING_ANON below.
59 void *s_mem; /* slab first object */
62 /* Second double word */
65 pgoff_t index; /* Our offset within mapping. */
66 void *freelist; /* sl[aou]b first free object */
70 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
71 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
72 /* Used for cmpxchg_double in slub */
73 unsigned long counters;
76 * Keep _count separate from slub cmpxchg_double data.
77 * As the rest of the double word is protected by
78 * slab_lock but _count is not.
87 * Count of ptes mapped in
88 * mms, to show when page is
89 * mapped & limit reverse map
92 * Used also for tail pages
93 * refcounting instead of
94 * _count. Tail pages cannot
95 * be mapped and keeping the
96 * tail page _count zero at
97 * all times guarantees
98 * get_page_unless_zero() will
99 * never succeed on tail
109 int units; /* SLOB */
111 atomic_t _count; /* Usage count, see below. */
113 unsigned int active; /* SLAB */
117 /* Third double word block */
119 struct list_head lru; /* Pageout list, eg. active_list
120 * protected by zone->lru_lock !
121 * Can be used as a generic list
124 struct { /* slub per cpu partial pages */
125 struct page *next; /* Next partial slab */
127 int pages; /* Nr of partial slabs left */
128 int pobjects; /* Approximate # of objects */
135 struct slab *slab_page; /* slab fields */
136 struct rcu_head rcu_head; /* Used by SLAB
137 * when destroying via RCU
139 /* First tail page of compound page */
141 compound_page_dtor *compound_dtor;
142 unsigned long compound_order;
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; /* PTE page table pages */
359 #if CONFIG_PGTABLE_LEVELS > 2
360 atomic_long_t nr_pmds; /* PMD page table pages */
362 int map_count; /* number of VMAs */
364 spinlock_t page_table_lock; /* Protects page tables and some counters */
365 struct rw_semaphore mmap_sem;
367 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
368 * together off init_mm.mmlist, and are protected
373 unsigned long hiwater_rss; /* High-watermark of RSS usage */
374 unsigned long hiwater_vm; /* High-water virtual memory usage */
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;
387 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
390 * Special counters, in some configurations protected by the
391 * page_table_lock, in other configurations by being atomic.
393 struct mm_rss_stat rss_stat;
395 struct linux_binfmt *binfmt;
397 cpumask_var_t cpu_vm_mask_var;
399 /* Architecture-specific MM context */
400 mm_context_t context;
402 unsigned long flags; /* Must use atomic bitops to access the bits */
404 struct core_state *core_state; /* coredumping support */
406 spinlock_t ioctx_lock;
407 struct kioctx_table __rcu *ioctx_table;
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:
415 * current == mm->owner
417 * new_owner->mm == mm
418 * new_owner->alloc_lock is held
420 struct task_struct __rcu *owner;
423 /* store ref to file /proc/<pid>/exe symlink points to */
424 struct file __rcu *exe_file;
425 #ifdef CONFIG_MMU_NOTIFIER
426 struct mmu_notifier_mm *mmu_notifier_mm;
428 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
429 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
431 #ifdef CONFIG_CPUMASK_OFFSTACK
432 struct cpumask cpumask_allocation;
434 #ifdef CONFIG_NUMA_BALANCING
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.
440 unsigned long numa_next_scan;
442 /* Restart point for scanning and setting pte_numa */
443 unsigned long numa_scan_offset;
445 /* numa_scan_seq prevents two threads setting pte_numa */
448 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
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.
454 bool tlb_flush_pending;
456 struct uprobes_state uprobes_state;
457 #ifdef CONFIG_PREEMPT_RT_BASE
458 struct rcu_head delayed_drop;
460 #ifdef CONFIG_X86_INTEL_MPX
461 /* address of the bounds directory */
462 void __user *bd_addr;
466 static inline void mm_init_cpumask(struct mm_struct *mm)
468 #ifdef CONFIG_CPUMASK_OFFSTACK
469 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
471 cpumask_clear(mm->cpu_vm_mask_var);
474 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
475 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
477 return mm->cpu_vm_mask_var;
480 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
482 * Memory barriers to keep this state in sync are graciously provided by
483 * the page table locks, outside of which no page table modifications happen.
484 * The barriers below prevent the compiler from re-ordering the instructions
485 * around the memory barriers that are already present in the code.
487 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
490 return mm->tlb_flush_pending;
492 static inline void set_tlb_flush_pending(struct mm_struct *mm)
494 mm->tlb_flush_pending = true;
497 * Guarantee that the tlb_flush_pending store does not leak into the
498 * critical section updating the page tables
500 smp_mb__before_spinlock();
502 /* Clearing is done after a TLB flush, which also provides a barrier. */
503 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
506 mm->tlb_flush_pending = false;
509 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
513 static inline void set_tlb_flush_pending(struct mm_struct *mm)
516 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
521 struct vm_special_mapping
527 enum tlb_flush_reason {
528 TLB_FLUSH_ON_TASK_SWITCH,
529 TLB_REMOTE_SHOOTDOWN,
531 TLB_LOCAL_MM_SHOOTDOWN,
532 NR_TLB_FLUSH_REASONS,
536 * A swap entry has to fit into a "unsigned long", as the entry is hidden
537 * in the "index" field of the swapper address space.
543 #endif /* _LINUX_MM_TYPES_H */