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 */
67 bool pfmemalloc; /* If set by the page allocator,
68 * ALLOC_NO_WATERMARKS was set
69 * and the low watermark was not
70 * met implying that the system
71 * is under some pressure. The
72 * caller should try ensure
73 * this page is only used to
79 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
80 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
81 /* Used for cmpxchg_double in slub */
82 unsigned long counters;
85 * Keep _count separate from slub cmpxchg_double data.
86 * As the rest of the double word is protected by
87 * slab_lock but _count is not.
96 * Count of ptes mapped in
97 * mms, to show when page is
98 * mapped & limit reverse map
101 * Used also for tail pages
102 * refcounting instead of
103 * _count. Tail pages cannot
104 * be mapped and keeping the
105 * tail page _count zero at
106 * all times guarantees
107 * get_page_unless_zero() will
108 * never succeed on tail
118 int units; /* SLOB */
120 atomic_t _count; /* Usage count, see below. */
122 unsigned int active; /* SLAB */
126 /* Third double word block */
128 struct list_head lru; /* Pageout list, eg. active_list
129 * protected by zone->lru_lock !
130 * Can be used as a generic list
133 struct { /* slub per cpu partial pages */
134 struct page *next; /* Next partial slab */
136 int pages; /* Nr of partial slabs left */
137 int pobjects; /* Approximate # of objects */
144 struct slab *slab_page; /* slab fields */
145 struct rcu_head rcu_head; /* Used by SLAB
146 * when destroying via RCU
148 /* First tail page of compound page */
150 compound_page_dtor *compound_dtor;
151 unsigned long compound_order;
154 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
155 pgtable_t pmd_huge_pte; /* protected by page->ptl */
159 /* Remainder is not double word aligned */
161 unsigned long private; /* Mapping-private opaque data:
162 * usually used for buffer_heads
163 * if PagePrivate set; used for
164 * swp_entry_t if PageSwapCache;
165 * indicates order in the buddy
166 * system if PG_buddy is set.
168 #if USE_SPLIT_PTE_PTLOCKS
169 #if ALLOC_SPLIT_PTLOCKS
175 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
176 struct page *first_page; /* Compound tail pages */
180 struct mem_cgroup *mem_cgroup;
184 * On machines where all RAM is mapped into kernel address space,
185 * we can simply calculate the virtual address. On machines with
186 * highmem some memory is mapped into kernel virtual memory
187 * dynamically, so we need a place to store that address.
188 * Note that this field could be 16 bits on x86 ... ;)
190 * Architectures with slow multiplication can define
191 * WANT_PAGE_VIRTUAL in asm/page.h
193 #if defined(WANT_PAGE_VIRTUAL)
194 void *virtual; /* Kernel virtual address (NULL if
195 not kmapped, ie. highmem) */
196 #endif /* WANT_PAGE_VIRTUAL */
198 #ifdef CONFIG_KMEMCHECK
200 * kmemcheck wants to track the status of each byte in a page; this
201 * is a pointer to such a status block. NULL if not tracked.
206 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
211 * The struct page can be forced to be double word aligned so that atomic ops
212 * on double words work. The SLUB allocator can make use of such a feature.
214 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
215 __aligned(2 * sizeof(unsigned long))
221 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
230 typedef unsigned long __nocast vm_flags_t;
233 * A region containing a mapping of a non-memory backed file under NOMMU
234 * conditions. These are held in a global tree and are pinned by the VMAs that
238 struct rb_node vm_rb; /* link in global region tree */
239 vm_flags_t vm_flags; /* VMA vm_flags */
240 unsigned long vm_start; /* start address of region */
241 unsigned long vm_end; /* region initialised to here */
242 unsigned long vm_top; /* region allocated to here */
243 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
244 struct file *vm_file; /* the backing file or NULL */
246 int vm_usage; /* region usage count (access under nommu_region_sem) */
247 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
252 * This struct defines a memory VMM memory area. There is one of these
253 * per VM-area/task. A VM area is any part of the process virtual memory
254 * space that has a special rule for the page-fault handlers (ie a shared
255 * library, the executable area etc).
257 struct vm_area_struct {
258 /* The first cache line has the info for VMA tree walking. */
260 unsigned long vm_start; /* Our start address within vm_mm. */
261 unsigned long vm_end; /* The first byte after our end address
264 /* linked list of VM areas per task, sorted by address */
265 struct vm_area_struct *vm_next, *vm_prev;
267 struct rb_node vm_rb;
270 * Largest free memory gap in bytes to the left of this VMA.
271 * Either between this VMA and vma->vm_prev, or between one of the
272 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
273 * get_unmapped_area find a free area of the right size.
275 unsigned long rb_subtree_gap;
277 /* Second cache line starts here. */
279 struct mm_struct *vm_mm; /* The address space we belong to. */
280 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
281 unsigned long vm_flags; /* Flags, see mm.h. */
284 * For areas with an address space and backing store,
285 * linkage into the address_space->i_mmap interval tree.
289 unsigned long rb_subtree_last;
293 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
294 * list, after a COW of one of the file pages. A MAP_SHARED vma
295 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
296 * or brk vma (with NULL file) can only be in an anon_vma list.
298 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
300 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
302 /* Function pointers to deal with this struct. */
303 const struct vm_operations_struct *vm_ops;
305 /* Information about our backing store: */
306 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
307 units, *not* PAGE_CACHE_SIZE */
308 struct file * vm_file; /* File we map to (can be NULL). */
309 void * vm_private_data; /* was vm_pte (shared mem) */
312 struct vm_region *vm_region; /* NOMMU mapping region */
315 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
320 struct task_struct *task;
321 struct core_thread *next;
326 struct core_thread dumper;
327 struct completion startup;
337 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
338 #define SPLIT_RSS_COUNTING
339 /* per-thread cached information, */
340 struct task_rss_stat {
341 int events; /* for synchronization threshold */
342 int count[NR_MM_COUNTERS];
344 #endif /* USE_SPLIT_PTE_PTLOCKS */
347 atomic_long_t count[NR_MM_COUNTERS];
352 struct vm_area_struct *mmap; /* list of VMAs */
353 struct rb_root mm_rb;
354 u32 vmacache_seqnum; /* per-thread vmacache */
356 unsigned long (*get_unmapped_area) (struct file *filp,
357 unsigned long addr, unsigned long len,
358 unsigned long pgoff, unsigned long flags);
360 unsigned long mmap_base; /* base of mmap area */
361 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
362 unsigned long task_size; /* size of task vm space */
363 unsigned long highest_vm_end; /* highest vma end address */
365 atomic_t mm_users; /* How many users with user space? */
366 atomic_t mm_count; /* How many references to "struct mm_struct" (users count as 1) */
367 atomic_long_t nr_ptes; /* PTE page table pages */
368 #if CONFIG_PGTABLE_LEVELS > 2
369 atomic_long_t nr_pmds; /* PMD page table pages */
371 int map_count; /* number of VMAs */
373 spinlock_t page_table_lock; /* Protects page tables and some counters */
374 struct rw_semaphore mmap_sem;
376 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
377 * together off init_mm.mmlist, and are protected
382 unsigned long hiwater_rss; /* High-watermark of RSS usage */
383 unsigned long hiwater_vm; /* High-water virtual memory usage */
385 unsigned long total_vm; /* Total pages mapped */
386 unsigned long locked_vm; /* Pages that have PG_mlocked set */
387 unsigned long pinned_vm; /* Refcount permanently increased */
388 unsigned long shared_vm; /* Shared pages (files) */
389 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE */
390 unsigned long stack_vm; /* VM_GROWSUP/DOWN */
391 unsigned long def_flags;
392 unsigned long start_code, end_code, start_data, end_data;
393 unsigned long start_brk, brk, start_stack;
394 unsigned long arg_start, arg_end, env_start, env_end;
396 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
399 * Special counters, in some configurations protected by the
400 * page_table_lock, in other configurations by being atomic.
402 struct mm_rss_stat rss_stat;
404 struct linux_binfmt *binfmt;
406 cpumask_var_t cpu_vm_mask_var;
408 /* Architecture-specific MM context */
409 mm_context_t context;
411 unsigned long flags; /* Must use atomic bitops to access the bits */
413 struct core_state *core_state; /* coredumping support */
415 spinlock_t ioctx_lock;
416 struct kioctx_table __rcu *ioctx_table;
420 * "owner" points to a task that is regarded as the canonical
421 * user/owner of this mm. All of the following must be true in
422 * order for it to be changed:
424 * current == mm->owner
426 * new_owner->mm == mm
427 * new_owner->alloc_lock is held
429 struct task_struct __rcu *owner;
432 /* store ref to file /proc/<pid>/exe symlink points to */
433 struct file __rcu *exe_file;
434 #ifdef CONFIG_MMU_NOTIFIER
435 struct mmu_notifier_mm *mmu_notifier_mm;
437 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
438 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
440 #ifdef CONFIG_CPUMASK_OFFSTACK
441 struct cpumask cpumask_allocation;
443 #ifdef CONFIG_NUMA_BALANCING
445 * numa_next_scan is the next time that the PTEs will be marked
446 * pte_numa. NUMA hinting faults will gather statistics and migrate
447 * pages to new nodes if necessary.
449 unsigned long numa_next_scan;
451 /* Restart point for scanning and setting pte_numa */
452 unsigned long numa_scan_offset;
454 /* numa_scan_seq prevents two threads setting pte_numa */
457 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
459 * An operation with batched TLB flushing is going on. Anything that
460 * can move process memory needs to flush the TLB when moving a
461 * PROT_NONE or PROT_NUMA mapped page.
463 bool tlb_flush_pending;
465 struct uprobes_state uprobes_state;
466 #ifdef CONFIG_PREEMPT_RT_BASE
467 struct rcu_head delayed_drop;
469 #ifdef CONFIG_X86_INTEL_MPX
470 /* address of the bounds directory */
471 void __user *bd_addr;
475 static inline void mm_init_cpumask(struct mm_struct *mm)
477 #ifdef CONFIG_CPUMASK_OFFSTACK
478 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
480 cpumask_clear(mm->cpu_vm_mask_var);
483 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
484 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
486 return mm->cpu_vm_mask_var;
489 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
491 * Memory barriers to keep this state in sync are graciously provided by
492 * the page table locks, outside of which no page table modifications happen.
493 * The barriers below prevent the compiler from re-ordering the instructions
494 * around the memory barriers that are already present in the code.
496 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
499 return mm->tlb_flush_pending;
501 static inline void set_tlb_flush_pending(struct mm_struct *mm)
503 mm->tlb_flush_pending = true;
506 * Guarantee that the tlb_flush_pending store does not leak into the
507 * critical section updating the page tables
509 smp_mb__before_spinlock();
511 /* Clearing is done after a TLB flush, which also provides a barrier. */
512 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
515 mm->tlb_flush_pending = false;
518 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
522 static inline void set_tlb_flush_pending(struct mm_struct *mm)
525 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
530 struct vm_special_mapping
536 enum tlb_flush_reason {
537 TLB_FLUSH_ON_TASK_SWITCH,
538 TLB_REMOTE_SHOOTDOWN,
540 TLB_LOCAL_MM_SHOOTDOWN,
541 NR_TLB_FLUSH_REASONS,
545 * A swap entry has to fit into a "unsigned long", as the entry is hidden
546 * in the "index" field of the swapper address space.
552 #endif /* _LINUX_MM_TYPES_H */