1 #ifndef _LINUX_MM_TYPES_H
2 #define _LINUX_MM_TYPES_H
4 #include <linux/mm_types_task.h>
6 #include <linux/auxvec.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))
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
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.
41 /* First double word block */
42 unsigned long flags; /* Atomic flags, some possibly
43 * updated asynchronously */
45 struct address_space *mapping; /* If low bit clear, points to
46 * inode address_space, or NULL.
47 * If page mapped as anonymous
48 * memory, low bit is set, and
49 * it points to anon_vma object:
50 * see PAGE_MAPPING_ANON below.
52 void *s_mem; /* slab first object */
53 atomic_t compound_mapcount; /* first tail page */
54 /* page_deferred_list().next -- second tail page */
57 /* Second double word */
59 pgoff_t index; /* Our offset within mapping. */
60 void *freelist; /* sl[aou]b first free object */
61 /* page_deferred_list().prev -- second tail page */
65 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
66 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
67 /* Used for cmpxchg_double in slub */
68 unsigned long counters;
71 * Keep _refcount separate from slub cmpxchg_double data.
72 * As the rest of the double word is protected by slab_lock
73 * but _refcount is not.
81 * Count of ptes mapped in mms, to show when
82 * page is mapped & limit reverse map searches.
84 * Extra information about page type may be
85 * stored here for pages that are never mapped,
86 * in which case the value MUST BE <= -2.
87 * See page-flags.h for more details.
91 unsigned int active; /* SLAB */
100 * Usage count, *USE WRAPPER FUNCTION* when manual
101 * accounting. See page_ref.h
108 * Third double word block
110 * WARNING: bit 0 of the first word encode PageTail(). That means
111 * the rest users of the storage space MUST NOT use the bit to
112 * avoid collision and false-positive PageTail().
115 struct list_head lru; /* Pageout list, eg. active_list
116 * protected by zone_lru_lock !
117 * Can be used as a generic list
120 struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an
121 * lru or handled by a slab
122 * allocator, this points to the
123 * hosting device page map.
125 struct { /* slub per cpu partial pages */
126 struct page *next; /* Next partial slab */
128 int pages; /* Nr of partial slabs left */
129 int pobjects; /* Approximate # of objects */
136 struct rcu_head rcu_head; /* Used by SLAB
137 * when destroying via RCU
139 /* Tail pages of compound page */
141 unsigned long compound_head; /* If bit zero is set */
143 /* First tail page only */
146 * On 64 bit system we have enough space in struct page
147 * to encode compound_dtor and compound_order with
148 * unsigned int. It can help compiler generate better or
149 * smaller code on some archtectures.
151 unsigned int compound_dtor;
152 unsigned int compound_order;
154 unsigned short int compound_dtor;
155 unsigned short int compound_order;
159 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
161 unsigned long __pad; /* do not overlay pmd_huge_pte
162 * with compound_head to avoid
163 * possible bit 0 collision.
165 pgtable_t pmd_huge_pte; /* protected by page->ptl */
170 /* Remainder is not double word aligned */
172 unsigned long private; /* Mapping-private opaque data:
173 * usually used for buffer_heads
174 * if PagePrivate set; used for
175 * swp_entry_t if PageSwapCache;
176 * indicates order in the buddy
177 * system if PG_buddy is set.
179 #if USE_SPLIT_PTE_PTLOCKS
180 #if ALLOC_SPLIT_PTLOCKS
186 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
190 struct mem_cgroup *mem_cgroup;
194 * On machines where all RAM is mapped into kernel address space,
195 * we can simply calculate the virtual address. On machines with
196 * highmem some memory is mapped into kernel virtual memory
197 * dynamically, so we need a place to store that address.
198 * Note that this field could be 16 bits on x86 ... ;)
200 * Architectures with slow multiplication can define
201 * WANT_PAGE_VIRTUAL in asm/page.h
203 #if defined(WANT_PAGE_VIRTUAL)
204 void *virtual; /* Kernel virtual address (NULL if
205 not kmapped, ie. highmem) */
206 #endif /* WANT_PAGE_VIRTUAL */
208 #ifdef CONFIG_KMEMCHECK
210 * kmemcheck wants to track the status of each byte in a page; this
211 * is a pointer to such a status block. NULL if not tracked.
216 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
221 * The struct page can be forced to be double word aligned so that atomic ops
222 * on double words work. The SLUB allocator can make use of such a feature.
224 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
225 __aligned(2 * sizeof(unsigned long))
229 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
230 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
232 struct page_frag_cache {
234 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
240 /* we maintain a pagecount bias, so that we dont dirty cache line
241 * containing page->_refcount every time we allocate a fragment.
243 unsigned int pagecnt_bias;
247 typedef unsigned long vm_flags_t;
250 * A region containing a mapping of a non-memory backed file under NOMMU
251 * conditions. These are held in a global tree and are pinned by the VMAs that
255 struct rb_node vm_rb; /* link in global region tree */
256 vm_flags_t vm_flags; /* VMA vm_flags */
257 unsigned long vm_start; /* start address of region */
258 unsigned long vm_end; /* region initialised to here */
259 unsigned long vm_top; /* region allocated to here */
260 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
261 struct file *vm_file; /* the backing file or NULL */
263 int vm_usage; /* region usage count (access under nommu_region_sem) */
264 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
268 #ifdef CONFIG_USERFAULTFD
269 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
270 struct vm_userfaultfd_ctx {
271 struct userfaultfd_ctx *ctx;
273 #else /* CONFIG_USERFAULTFD */
274 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
275 struct vm_userfaultfd_ctx {};
276 #endif /* CONFIG_USERFAULTFD */
279 * This struct defines a memory VMM memory area. There is one of these
280 * per VM-area/task. A VM area is any part of the process virtual memory
281 * space that has a special rule for the page-fault handlers (ie a shared
282 * library, the executable area etc).
284 struct vm_area_struct {
285 /* The first cache line has the info for VMA tree walking. */
287 unsigned long vm_start; /* Our start address within vm_mm. */
288 unsigned long vm_end; /* The first byte after our end address
291 /* linked list of VM areas per task, sorted by address */
292 struct vm_area_struct *vm_next, *vm_prev;
294 struct rb_node vm_rb;
297 * Largest free memory gap in bytes to the left of this VMA.
298 * Either between this VMA and vma->vm_prev, or between one of the
299 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
300 * get_unmapped_area find a free area of the right size.
302 unsigned long rb_subtree_gap;
304 /* Second cache line starts here. */
306 struct mm_struct *vm_mm; /* The address space we belong to. */
307 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
308 unsigned long vm_flags; /* Flags, see mm.h. */
311 * For areas with an address space and backing store,
312 * linkage into the address_space->i_mmap interval tree.
316 unsigned long rb_subtree_last;
320 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
321 * list, after a COW of one of the file pages. A MAP_SHARED vma
322 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
323 * or brk vma (with NULL file) can only be in an anon_vma list.
325 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
327 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
329 /* Function pointers to deal with this struct. */
330 const struct vm_operations_struct *vm_ops;
332 /* Information about our backing store: */
333 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
335 struct file * vm_file; /* File we map to (can be NULL). */
336 void * vm_private_data; /* was vm_pte (shared mem) */
338 atomic_long_t swap_readahead_info;
340 struct vm_region *vm_region; /* NOMMU mapping region */
343 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
345 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
346 } __randomize_layout;
349 struct task_struct *task;
350 struct core_thread *next;
355 struct core_thread dumper;
356 struct completion startup;
361 struct vm_area_struct *mmap; /* list of VMAs */
362 struct rb_root mm_rb;
363 u32 vmacache_seqnum; /* per-thread vmacache */
365 unsigned long (*get_unmapped_area) (struct file *filp,
366 unsigned long addr, unsigned long len,
367 unsigned long pgoff, unsigned long flags);
369 unsigned long mmap_base; /* base of mmap area */
370 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
371 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
372 /* Base adresses for compatible mmap() */
373 unsigned long mmap_compat_base;
374 unsigned long mmap_compat_legacy_base;
376 unsigned long task_size; /* size of task vm space */
377 unsigned long highest_vm_end; /* highest vma end address */
381 * @mm_users: The number of users including userspace.
383 * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops
384 * to 0 (i.e. when the task exits and there are no other temporary
385 * reference holders), we also release a reference on @mm_count
386 * (which may then free the &struct mm_struct if @mm_count also
392 * @mm_count: The number of references to &struct mm_struct
393 * (@mm_users count as 1).
395 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
396 * &struct mm_struct is freed.
400 atomic_long_t nr_ptes; /* PTE page table pages */
401 #if CONFIG_PGTABLE_LEVELS > 2
402 atomic_long_t nr_pmds; /* PMD page table pages */
404 int map_count; /* number of VMAs */
406 spinlock_t page_table_lock; /* Protects page tables and some counters */
407 struct rw_semaphore mmap_sem;
409 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
410 * together off init_mm.mmlist, and are protected
415 unsigned long hiwater_rss; /* High-watermark of RSS usage */
416 unsigned long hiwater_vm; /* High-water virtual memory usage */
418 unsigned long total_vm; /* Total pages mapped */
419 unsigned long locked_vm; /* Pages that have PG_mlocked set */
420 unsigned long pinned_vm; /* Refcount permanently increased */
421 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
422 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
423 unsigned long stack_vm; /* VM_STACK */
424 unsigned long def_flags;
425 unsigned long start_code, end_code, start_data, end_data;
426 unsigned long start_brk, brk, start_stack;
427 unsigned long arg_start, arg_end, env_start, env_end;
429 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
432 * Special counters, in some configurations protected by the
433 * page_table_lock, in other configurations by being atomic.
435 struct mm_rss_stat rss_stat;
437 struct linux_binfmt *binfmt;
439 cpumask_var_t cpu_vm_mask_var;
441 /* Architecture-specific MM context */
442 mm_context_t context;
444 unsigned long flags; /* Must use atomic bitops to access the bits */
446 struct core_state *core_state; /* coredumping support */
448 spinlock_t ioctx_lock;
449 struct kioctx_table __rcu *ioctx_table;
453 * "owner" points to a task that is regarded as the canonical
454 * user/owner of this mm. All of the following must be true in
455 * order for it to be changed:
457 * current == mm->owner
459 * new_owner->mm == mm
460 * new_owner->alloc_lock is held
462 struct task_struct __rcu *owner;
464 struct user_namespace *user_ns;
466 /* store ref to file /proc/<pid>/exe symlink points to */
467 struct file __rcu *exe_file;
468 #ifdef CONFIG_MMU_NOTIFIER
469 struct mmu_notifier_mm *mmu_notifier_mm;
471 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
472 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
474 #ifdef CONFIG_CPUMASK_OFFSTACK
475 struct cpumask cpumask_allocation;
477 #ifdef CONFIG_NUMA_BALANCING
479 * numa_next_scan is the next time that the PTEs will be marked
480 * pte_numa. NUMA hinting faults will gather statistics and migrate
481 * pages to new nodes if necessary.
483 unsigned long numa_next_scan;
485 /* Restart point for scanning and setting pte_numa */
486 unsigned long numa_scan_offset;
488 /* numa_scan_seq prevents two threads setting pte_numa */
492 * An operation with batched TLB flushing is going on. Anything that
493 * can move process memory needs to flush the TLB when moving a
494 * PROT_NONE or PROT_NUMA mapped page.
496 atomic_t tlb_flush_pending;
497 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
498 /* See flush_tlb_batched_pending() */
499 bool tlb_flush_batched;
501 struct uprobes_state uprobes_state;
502 #ifdef CONFIG_HUGETLB_PAGE
503 atomic_long_t hugetlb_usage;
505 struct work_struct async_put_work;
506 } __randomize_layout;
508 extern struct mm_struct init_mm;
510 static inline void mm_init_cpumask(struct mm_struct *mm)
512 #ifdef CONFIG_CPUMASK_OFFSTACK
513 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
515 cpumask_clear(mm->cpu_vm_mask_var);
518 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
519 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
521 return mm->cpu_vm_mask_var;
525 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
526 unsigned long start, unsigned long end);
527 extern void tlb_finish_mmu(struct mmu_gather *tlb,
528 unsigned long start, unsigned long end);
530 static inline void init_tlb_flush_pending(struct mm_struct *mm)
532 atomic_set(&mm->tlb_flush_pending, 0);
535 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
537 atomic_inc(&mm->tlb_flush_pending);
539 * The only time this value is relevant is when there are indeed pages
540 * to flush. And we'll only flush pages after changing them, which
543 * So the ordering here is:
545 * atomic_inc(&mm->tlb_flush_pending);
552 * mm_tlb_flush_pending();
557 * atomic_dec(&mm->tlb_flush_pending);
559 * Where the increment if constrained by the PTL unlock, it thus
560 * ensures that the increment is visible if the PTE modification is
561 * visible. After all, if there is no PTE modification, nobody cares
562 * about TLB flushes either.
564 * This very much relies on users (mm_tlb_flush_pending() and
565 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
566 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
567 * locks (PPC) the unlock of one doesn't order against the lock of
570 * The decrement is ordered by the flush_tlb_range(), such that
571 * mm_tlb_flush_pending() will not return false unless all flushes have
576 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
579 * See inc_tlb_flush_pending().
581 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
582 * not order against TLB invalidate completion, which is what we need.
584 * Therefore we must rely on tlb_flush_*() to guarantee order.
586 atomic_dec(&mm->tlb_flush_pending);
589 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
592 * Must be called after having acquired the PTL; orders against that
593 * PTLs release and therefore ensures that if we observe the modified
594 * PTE we must also observe the increment from inc_tlb_flush_pending().
596 * That is, it only guarantees to return true if there is a flush
597 * pending for _this_ PTL.
599 return atomic_read(&mm->tlb_flush_pending);
602 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
605 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
606 * for which there is a TLB flush pending in order to guarantee
607 * we've seen both that PTE modification and the increment.
609 * (no requirement on actually still holding the PTL, that is irrelevant)
611 return atomic_read(&mm->tlb_flush_pending) > 1;
616 struct vm_special_mapping {
617 const char *name; /* The name, e.g. "[vdso]". */
620 * If .fault is not provided, this points to a
621 * NULL-terminated array of pages that back the special mapping.
623 * This must not be NULL unless .fault is provided.
628 * If non-NULL, then this is called to resolve page faults
629 * on the special mapping. If used, .pages is not checked.
631 int (*fault)(const struct vm_special_mapping *sm,
632 struct vm_area_struct *vma,
633 struct vm_fault *vmf);
635 int (*mremap)(const struct vm_special_mapping *sm,
636 struct vm_area_struct *new_vma);
639 enum tlb_flush_reason {
640 TLB_FLUSH_ON_TASK_SWITCH,
641 TLB_REMOTE_SHOOTDOWN,
643 TLB_LOCAL_MM_SHOOTDOWN,
645 NR_TLB_FLUSH_REASONS,
649 * A swap entry has to fit into a "unsigned long", as the entry is hidden
650 * in the "index" field of the swapper address space.
656 #endif /* _LINUX_MM_TYPES_H */