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>
15 #include <linux/workqueue.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)
33 * Each physical page in the system has a struct page associated with
34 * it to keep track of whatever it is we are using the page for at the
35 * moment. Note that we have no way to track which tasks are using
36 * a page, though if it is a pagecache page, rmap structures can tell us
39 * The objects in struct page are organized in double word blocks in
40 * order to allows us to use atomic double word operations on portions
41 * of struct page. That is currently only used by slub but the arrangement
42 * allows the use of atomic double word operations on the flags/mapping
43 * and lru list pointers also.
46 /* First double word block */
47 unsigned long flags; /* Atomic flags, some possibly
48 * updated asynchronously */
50 struct address_space *mapping; /* If low bit clear, points to
51 * inode address_space, or NULL.
52 * If page mapped as anonymous
53 * memory, low bit is set, and
54 * it points to anon_vma object:
55 * see PAGE_MAPPING_ANON below.
57 void *s_mem; /* slab first object */
60 /* Second double word */
63 pgoff_t index; /* Our offset within mapping. */
64 void *freelist; /* sl[aou]b first free object */
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;
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.
85 * Count of ptes mapped in
86 * mms, to show when page is
87 * mapped & limit reverse map
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
107 int units; /* SLOB */
109 atomic_t _count; /* Usage count, see below. */
111 unsigned int active; /* SLAB */
116 * Third double word block
118 * WARNING: bit 0 of the first word encode PageTail(). That means
119 * the rest users of the storage space MUST NOT use the bit to
120 * avoid collision and false-positive PageTail().
123 struct list_head lru; /* Pageout list, eg. active_list
124 * protected by zone->lru_lock !
125 * Can be used as a generic list
128 struct { /* slub per cpu partial pages */
129 struct page *next; /* Next partial slab */
131 int pages; /* Nr of partial slabs left */
132 int pobjects; /* Approximate # of objects */
139 struct rcu_head rcu_head; /* Used by SLAB
140 * when destroying via RCU
142 /* Tail pages of compound page */
144 unsigned long compound_head; /* If bit zero is set */
146 /* First tail page only */
149 * On 64 bit system we have enough space in struct page
150 * to encode compound_dtor and compound_order with
151 * unsigned int. It can help compiler generate better or
152 * smaller code on some archtectures.
154 unsigned int compound_dtor;
155 unsigned int compound_order;
157 unsigned short int compound_dtor;
158 unsigned short int compound_order;
162 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
164 unsigned long __pad; /* do not overlay pmd_huge_pte
165 * with compound_head to avoid
166 * possible bit 0 collision.
168 pgtable_t pmd_huge_pte; /* protected by page->ptl */
173 /* Remainder is not double word aligned */
175 unsigned long private; /* Mapping-private opaque data:
176 * usually used for buffer_heads
177 * if PagePrivate set; used for
178 * swp_entry_t if PageSwapCache;
179 * indicates order in the buddy
180 * system if PG_buddy is set.
182 #if USE_SPLIT_PTE_PTLOCKS
183 #if ALLOC_SPLIT_PTLOCKS
189 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
193 struct mem_cgroup *mem_cgroup;
197 * On machines where all RAM is mapped into kernel address space,
198 * we can simply calculate the virtual address. On machines with
199 * highmem some memory is mapped into kernel virtual memory
200 * dynamically, so we need a place to store that address.
201 * Note that this field could be 16 bits on x86 ... ;)
203 * Architectures with slow multiplication can define
204 * WANT_PAGE_VIRTUAL in asm/page.h
206 #if defined(WANT_PAGE_VIRTUAL)
207 void *virtual; /* Kernel virtual address (NULL if
208 not kmapped, ie. highmem) */
209 #endif /* WANT_PAGE_VIRTUAL */
211 #ifdef CONFIG_BLK_DEV_IO_TRACE
212 struct task_struct *tsk_dirty; /* task that sets this page dirty */
215 #ifdef CONFIG_KMEMCHECK
217 * kmemcheck wants to track the status of each byte in a page; this
218 * is a pointer to such a status block. NULL if not tracked.
223 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
228 * The struct page can be forced to be double word aligned so that atomic ops
229 * on double words work. The SLUB allocator can make use of such a feature.
231 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
232 __aligned(2 * sizeof(unsigned long))
238 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
247 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
248 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
250 struct page_frag_cache {
252 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
258 /* we maintain a pagecount bias, so that we dont dirty cache line
259 * containing page->_count every time we allocate a fragment.
261 unsigned int pagecnt_bias;
265 typedef unsigned long vm_flags_t;
268 * A region containing a mapping of a non-memory backed file under NOMMU
269 * conditions. These are held in a global tree and are pinned by the VMAs that
273 struct rb_node vm_rb; /* link in global region tree */
274 vm_flags_t vm_flags; /* VMA vm_flags */
275 unsigned long vm_start; /* start address of region */
276 unsigned long vm_end; /* region initialised to here */
277 unsigned long vm_top; /* region allocated to here */
278 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
279 struct file *vm_file; /* the backing file or NULL */
281 int vm_usage; /* region usage count (access under nommu_region_sem) */
282 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
286 #ifdef CONFIG_USERFAULTFD
287 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
288 struct vm_userfaultfd_ctx {
289 struct userfaultfd_ctx *ctx;
291 #else /* CONFIG_USERFAULTFD */
292 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
293 struct vm_userfaultfd_ctx {};
294 #endif /* CONFIG_USERFAULTFD */
297 * This struct defines a memory VMM memory area. There is one of these
298 * per VM-area/task. A VM area is any part of the process virtual memory
299 * space that has a special rule for the page-fault handlers (ie a shared
300 * library, the executable area etc).
302 struct vm_area_struct {
303 /* The first cache line has the info for VMA tree walking. */
305 unsigned long vm_start; /* Our start address within vm_mm. */
306 unsigned long vm_end; /* The first byte after our end address
309 /* linked list of VM areas per task, sorted by address */
310 struct vm_area_struct *vm_next, *vm_prev;
312 struct rb_node vm_rb;
315 * Largest free memory gap in bytes to the left of this VMA.
316 * Either between this VMA and vma->vm_prev, or between one of the
317 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
318 * get_unmapped_area find a free area of the right size.
320 unsigned long rb_subtree_gap;
322 /* Second cache line starts here. */
324 struct mm_struct *vm_mm; /* The address space we belong to. */
325 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
326 unsigned long vm_flags; /* Flags, see mm.h. */
329 * For areas with an address space and backing store,
330 * linkage into the address_space->i_mmap interval tree.
332 * For private anonymous mappings, a pointer to a null terminated string
333 * in the user process containing the name given to the vma, or NULL
339 unsigned long rb_subtree_last;
341 const char __user *anon_name;
345 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
346 * list, after a COW of one of the file pages. A MAP_SHARED vma
347 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
348 * or brk vma (with NULL file) can only be in an anon_vma list.
350 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
352 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
354 /* Function pointers to deal with this struct. */
355 const struct vm_operations_struct *vm_ops;
357 /* Information about our backing store: */
358 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
359 units, *not* PAGE_CACHE_SIZE */
360 struct file * vm_file; /* File we map to (can be NULL). */
361 void * vm_private_data; /* was vm_pte (shared mem) */
364 struct vm_region *vm_region; /* NOMMU mapping region */
367 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
369 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
370 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
371 seqcount_t vm_sequence;
372 atomic_t vm_ref_count; /* see vma_get(), vma_put() */
377 struct task_struct *task;
378 struct core_thread *next;
383 struct core_thread dumper;
384 struct completion startup;
394 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
395 #define SPLIT_RSS_COUNTING
396 /* per-thread cached information, */
397 struct task_rss_stat {
398 int events; /* for synchronization threshold */
399 int count[NR_MM_COUNTERS];
401 #endif /* USE_SPLIT_PTE_PTLOCKS */
404 atomic_long_t count[NR_MM_COUNTERS];
409 struct vm_area_struct *mmap; /* list of VMAs */
410 struct rb_root mm_rb;
411 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
414 u64 vmacache_seqnum; /* per-thread vmacache */
416 unsigned long (*get_unmapped_area) (struct file *filp,
417 unsigned long addr, unsigned long len,
418 unsigned long pgoff, unsigned long flags);
420 unsigned long mmap_base; /* base of mmap area */
421 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
422 unsigned long task_size; /* size of task vm space */
423 unsigned long highest_vm_end; /* highest vma end address */
425 atomic_t mm_users; /* How many users with user space? */
426 atomic_t mm_count; /* How many references to "struct mm_struct" (users count as 1) */
427 atomic_long_t nr_ptes; /* PTE page table pages */
428 #if CONFIG_PGTABLE_LEVELS > 2
429 atomic_long_t nr_pmds; /* PMD page table pages */
431 int map_count; /* number of VMAs */
433 spinlock_t page_table_lock; /* Protects page tables and some counters */
434 struct rw_semaphore mmap_sem;
436 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
437 * together off init_mm.mmlist, and are protected
442 unsigned long hiwater_rss; /* High-watermark of RSS usage */
443 unsigned long hiwater_vm; /* High-water virtual memory usage */
445 unsigned long total_vm; /* Total pages mapped */
446 unsigned long locked_vm; /* Pages that have PG_mlocked set */
447 unsigned long pinned_vm; /* Refcount permanently increased */
448 unsigned long shared_vm; /* Shared pages (files) */
449 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE */
450 unsigned long stack_vm; /* VM_GROWSUP/DOWN */
451 unsigned long def_flags;
452 unsigned long start_code, end_code, start_data, end_data;
453 unsigned long start_brk, brk, start_stack;
454 unsigned long arg_start, arg_end, env_start, env_end;
456 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
459 * Special counters, in some configurations protected by the
460 * page_table_lock, in other configurations by being atomic.
462 struct mm_rss_stat rss_stat;
464 struct linux_binfmt *binfmt;
466 cpumask_var_t cpu_vm_mask_var;
468 /* Architecture-specific MM context */
469 mm_context_t context;
471 unsigned long flags; /* Must use atomic bitops to access the bits */
473 struct core_state *core_state; /* coredumping support */
475 spinlock_t ioctx_lock;
476 struct kioctx_table __rcu *ioctx_table;
480 * "owner" points to a task that is regarded as the canonical
481 * user/owner of this mm. All of the following must be true in
482 * order for it to be changed:
484 * current == mm->owner
486 * new_owner->mm == mm
487 * new_owner->alloc_lock is held
489 struct task_struct __rcu *owner;
491 struct user_namespace *user_ns;
493 /* store ref to file /proc/<pid>/exe symlink points to */
494 struct file __rcu *exe_file;
495 #ifdef CONFIG_MMU_NOTIFIER
496 struct mmu_notifier_mm *mmu_notifier_mm;
498 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
499 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
501 #ifdef CONFIG_CPUMASK_OFFSTACK
502 struct cpumask cpumask_allocation;
504 #ifdef CONFIG_NUMA_BALANCING
506 * numa_next_scan is the next time that the PTEs will be marked
507 * pte_numa. NUMA hinting faults will gather statistics and migrate
508 * pages to new nodes if necessary.
510 unsigned long numa_next_scan;
512 /* Restart point for scanning and setting pte_numa */
513 unsigned long numa_scan_offset;
515 /* numa_scan_seq prevents two threads setting pte_numa */
518 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
520 * An operation with batched TLB flushing is going on. Anything that
521 * can move process memory needs to flush the TLB when moving a
522 * PROT_NONE or PROT_NUMA mapped page.
524 bool tlb_flush_pending;
526 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
527 /* See flush_tlb_batched_pending() */
528 bool tlb_flush_batched;
530 struct uprobes_state uprobes_state;
531 #ifdef CONFIG_X86_INTEL_MPX
532 /* address of the bounds directory */
533 void __user *bd_addr;
535 #ifdef CONFIG_HUGETLB_PAGE
536 atomic_long_t hugetlb_usage;
538 #ifdef CONFIG_MSM_APP_SETTINGS
542 struct work_struct async_put_work;
545 static inline void mm_init_cpumask(struct mm_struct *mm)
547 #ifdef CONFIG_CPUMASK_OFFSTACK
548 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
550 cpumask_clear(mm->cpu_vm_mask_var);
553 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
554 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
556 return mm->cpu_vm_mask_var;
559 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
561 * Memory barriers to keep this state in sync are graciously provided by
562 * the page table locks, outside of which no page table modifications happen.
563 * The barriers below prevent the compiler from re-ordering the instructions
564 * around the memory barriers that are already present in the code.
566 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
569 return mm->tlb_flush_pending;
571 static inline void set_tlb_flush_pending(struct mm_struct *mm)
573 mm->tlb_flush_pending = true;
576 * Guarantee that the tlb_flush_pending store does not leak into the
577 * critical section updating the page tables
579 smp_mb__before_spinlock();
581 /* Clearing is done after a TLB flush, which also provides a barrier. */
582 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
585 mm->tlb_flush_pending = false;
588 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
592 static inline void set_tlb_flush_pending(struct mm_struct *mm)
595 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
600 struct vm_special_mapping
606 enum tlb_flush_reason {
607 TLB_FLUSH_ON_TASK_SWITCH,
608 TLB_REMOTE_SHOOTDOWN,
610 TLB_LOCAL_MM_SHOOTDOWN,
612 NR_TLB_FLUSH_REASONS,
616 * A swap entry has to fit into a "unsigned long", as the entry is hidden
617 * in the "index" field of the swapper address space.
623 /* Return the name for an anonymous mapping or NULL for a file-backed mapping */
624 static inline const char __user *vma_get_anon_name(struct vm_area_struct *vma)
629 return vma->anon_name;
632 #endif /* _LINUX_MM_TYPES_H */