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))
29 * Each physical page in the system has a struct page associated with
30 * it to keep track of whatever it is we are using the page for at the
31 * moment. Note that we have no way to track which tasks are using
32 * a page, though if it is a pagecache page, rmap structures can tell us
35 * The objects in struct page are organized in double word blocks in
36 * order to allows us to use atomic double word operations on portions
37 * of struct page. That is currently only used by slub but the arrangement
38 * allows the use of atomic double word operations on the flags/mapping
39 * and lru list pointers also.
42 /* First double word block */
43 unsigned long flags; /* Atomic flags, some possibly
44 * updated asynchronously */
46 struct address_space *mapping; /* If low bit clear, points to
47 * inode address_space, or NULL.
48 * If page mapped as anonymous
49 * memory, low bit is set, and
50 * it points to anon_vma object:
51 * see PAGE_MAPPING_ANON below.
53 void *s_mem; /* slab first object */
54 atomic_t compound_mapcount; /* first tail page */
55 /* page_deferred_list().next -- second tail page */
58 /* Second double word */
60 pgoff_t index; /* Our offset within mapping. */
61 void *freelist; /* sl[aou]b first free object */
62 /* page_deferred_list().prev -- second tail page */
66 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
67 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
68 /* Used for cmpxchg_double in slub */
69 unsigned long counters;
72 * Keep _refcount separate from slub cmpxchg_double data.
73 * As the rest of the double word is protected by slab_lock
74 * but _refcount is not.
82 * Count of ptes mapped in mms, to show when
83 * page is mapped & limit reverse map searches.
85 * Extra information about page type may be
86 * stored here for pages that are never mapped,
87 * in which case the value MUST BE <= -2.
88 * See page-flags.h for more details.
92 unsigned int active; /* SLAB */
101 * Usage count, *USE WRAPPER FUNCTION* when manual
102 * accounting. See page_ref.h
109 * Third double word block
111 * WARNING: bit 0 of the first word encode PageTail(). That means
112 * the rest users of the storage space MUST NOT use the bit to
113 * avoid collision and false-positive PageTail().
116 struct list_head lru; /* Pageout list, eg. active_list
117 * protected by zone_lru_lock !
118 * Can be used as a generic list
121 struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an
122 * lru or handled by a slab
123 * allocator, this points to the
124 * hosting device page map.
126 struct { /* slub per cpu partial pages */
127 struct page *next; /* Next partial slab */
129 int pages; /* Nr of partial slabs left */
130 int pobjects; /* Approximate # of objects */
137 struct rcu_head rcu_head; /* Used by SLAB
138 * when destroying via RCU
140 /* Tail pages of compound page */
142 unsigned long compound_head; /* If bit zero is set */
144 /* First tail page only */
147 * On 64 bit system we have enough space in struct page
148 * to encode compound_dtor and compound_order with
149 * unsigned int. It can help compiler generate better or
150 * smaller code on some archtectures.
152 unsigned int compound_dtor;
153 unsigned int compound_order;
155 unsigned short int compound_dtor;
156 unsigned short int compound_order;
160 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
162 unsigned long __pad; /* do not overlay pmd_huge_pte
163 * with compound_head to avoid
164 * possible bit 0 collision.
166 pgtable_t pmd_huge_pte; /* protected by page->ptl */
171 /* Remainder is not double word aligned */
173 unsigned long private; /* Mapping-private opaque data:
174 * usually used for buffer_heads
175 * if PagePrivate set; used for
176 * swp_entry_t if PageSwapCache;
177 * indicates order in the buddy
178 * system if PG_buddy is set.
180 #if USE_SPLIT_PTE_PTLOCKS
181 #if ALLOC_SPLIT_PTLOCKS
187 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
191 struct mem_cgroup *mem_cgroup;
195 * On machines where all RAM is mapped into kernel address space,
196 * we can simply calculate the virtual address. On machines with
197 * highmem some memory is mapped into kernel virtual memory
198 * dynamically, so we need a place to store that address.
199 * Note that this field could be 16 bits on x86 ... ;)
201 * Architectures with slow multiplication can define
202 * WANT_PAGE_VIRTUAL in asm/page.h
204 #if defined(WANT_PAGE_VIRTUAL)
205 void *virtual; /* Kernel virtual address (NULL if
206 not kmapped, ie. highmem) */
207 #endif /* WANT_PAGE_VIRTUAL */
209 #ifdef CONFIG_KMEMCHECK
211 * kmemcheck wants to track the status of each byte in a page; this
212 * is a pointer to such a status block. NULL if not tracked.
217 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
222 * The struct page can be forced to be double word aligned so that atomic ops
223 * on double words work. The SLUB allocator can make use of such a feature.
225 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
226 __aligned(2 * sizeof(unsigned long))
230 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
231 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
233 struct page_frag_cache {
235 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
241 /* we maintain a pagecount bias, so that we dont dirty cache line
242 * containing page->_refcount every time we allocate a fragment.
244 unsigned int pagecnt_bias;
248 typedef unsigned long vm_flags_t;
251 * A region containing a mapping of a non-memory backed file under NOMMU
252 * conditions. These are held in a global tree and are pinned by the VMAs that
256 struct rb_node vm_rb; /* link in global region tree */
257 vm_flags_t vm_flags; /* VMA vm_flags */
258 unsigned long vm_start; /* start address of region */
259 unsigned long vm_end; /* region initialised to here */
260 unsigned long vm_top; /* region allocated to here */
261 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
262 struct file *vm_file; /* the backing file or NULL */
264 int vm_usage; /* region usage count (access under nommu_region_sem) */
265 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
269 #ifdef CONFIG_USERFAULTFD
270 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
271 struct vm_userfaultfd_ctx {
272 struct userfaultfd_ctx *ctx;
274 #else /* CONFIG_USERFAULTFD */
275 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
276 struct vm_userfaultfd_ctx {};
277 #endif /* CONFIG_USERFAULTFD */
280 * This struct defines a memory VMM memory area. There is one of these
281 * per VM-area/task. A VM area is any part of the process virtual memory
282 * space that has a special rule for the page-fault handlers (ie a shared
283 * library, the executable area etc).
285 struct vm_area_struct {
286 /* The first cache line has the info for VMA tree walking. */
288 unsigned long vm_start; /* Our start address within vm_mm. */
289 unsigned long vm_end; /* The first byte after our end address
292 /* linked list of VM areas per task, sorted by address */
293 struct vm_area_struct *vm_next, *vm_prev;
295 struct rb_node vm_rb;
298 * Largest free memory gap in bytes to the left of this VMA.
299 * Either between this VMA and vma->vm_prev, or between one of the
300 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
301 * get_unmapped_area find a free area of the right size.
303 unsigned long rb_subtree_gap;
305 /* Second cache line starts here. */
307 struct mm_struct *vm_mm; /* The address space we belong to. */
308 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
309 unsigned long vm_flags; /* Flags, see mm.h. */
312 * For areas with an address space and backing store,
313 * linkage into the address_space->i_mmap interval tree.
317 unsigned long rb_subtree_last;
321 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
322 * list, after a COW of one of the file pages. A MAP_SHARED vma
323 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
324 * or brk vma (with NULL file) can only be in an anon_vma list.
326 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
328 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
330 /* Function pointers to deal with this struct. */
331 const struct vm_operations_struct *vm_ops;
333 /* Information about our backing store: */
334 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
336 struct file * vm_file; /* File we map to (can be NULL). */
337 void * vm_private_data; /* was vm_pte (shared mem) */
339 atomic_long_t swap_readahead_info;
341 struct vm_region *vm_region; /* NOMMU mapping region */
344 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
346 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
347 } __randomize_layout;
350 struct task_struct *task;
351 struct core_thread *next;
356 struct core_thread dumper;
357 struct completion startup;
362 struct vm_area_struct *mmap; /* list of VMAs */
363 struct rb_root mm_rb;
364 u32 vmacache_seqnum; /* per-thread vmacache */
366 unsigned long (*get_unmapped_area) (struct file *filp,
367 unsigned long addr, unsigned long len,
368 unsigned long pgoff, unsigned long flags);
370 unsigned long mmap_base; /* base of mmap area */
371 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
372 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
373 /* Base adresses for compatible mmap() */
374 unsigned long mmap_compat_base;
375 unsigned long mmap_compat_legacy_base;
377 unsigned long task_size; /* size of task vm space */
378 unsigned long highest_vm_end; /* highest vma end address */
382 * @mm_users: The number of users including userspace.
384 * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops
385 * to 0 (i.e. when the task exits and there are no other temporary
386 * reference holders), we also release a reference on @mm_count
387 * (which may then free the &struct mm_struct if @mm_count also
393 * @mm_count: The number of references to &struct mm_struct
394 * (@mm_users count as 1).
396 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
397 * &struct mm_struct is freed.
401 atomic_long_t nr_ptes; /* PTE page table pages */
402 #if CONFIG_PGTABLE_LEVELS > 2
403 atomic_long_t nr_pmds; /* PMD page table pages */
405 int map_count; /* number of VMAs */
407 spinlock_t page_table_lock; /* Protects page tables and some counters */
408 struct rw_semaphore mmap_sem;
410 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
411 * together off init_mm.mmlist, and are protected
416 unsigned long hiwater_rss; /* High-watermark of RSS usage */
417 unsigned long hiwater_vm; /* High-water virtual memory usage */
419 unsigned long total_vm; /* Total pages mapped */
420 unsigned long locked_vm; /* Pages that have PG_mlocked set */
421 unsigned long pinned_vm; /* Refcount permanently increased */
422 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
423 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
424 unsigned long stack_vm; /* VM_STACK */
425 unsigned long def_flags;
426 unsigned long start_code, end_code, start_data, end_data;
427 unsigned long start_brk, brk, start_stack;
428 unsigned long arg_start, arg_end, env_start, env_end;
430 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
433 * Special counters, in some configurations protected by the
434 * page_table_lock, in other configurations by being atomic.
436 struct mm_rss_stat rss_stat;
438 struct linux_binfmt *binfmt;
440 cpumask_var_t cpu_vm_mask_var;
442 /* Architecture-specific MM context */
443 mm_context_t context;
445 unsigned long flags; /* Must use atomic bitops to access the bits */
447 struct core_state *core_state; /* coredumping support */
449 spinlock_t ioctx_lock;
450 struct kioctx_table __rcu *ioctx_table;
454 * "owner" points to a task that is regarded as the canonical
455 * user/owner of this mm. All of the following must be true in
456 * order for it to be changed:
458 * current == mm->owner
460 * new_owner->mm == mm
461 * new_owner->alloc_lock is held
463 struct task_struct __rcu *owner;
465 struct user_namespace *user_ns;
467 /* store ref to file /proc/<pid>/exe symlink points to */
468 struct file __rcu *exe_file;
469 #ifdef CONFIG_MMU_NOTIFIER
470 struct mmu_notifier_mm *mmu_notifier_mm;
472 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
473 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
475 #ifdef CONFIG_CPUMASK_OFFSTACK
476 struct cpumask cpumask_allocation;
478 #ifdef CONFIG_NUMA_BALANCING
480 * numa_next_scan is the next time that the PTEs will be marked
481 * pte_numa. NUMA hinting faults will gather statistics and migrate
482 * pages to new nodes if necessary.
484 unsigned long numa_next_scan;
486 /* Restart point for scanning and setting pte_numa */
487 unsigned long numa_scan_offset;
489 /* numa_scan_seq prevents two threads setting pte_numa */
493 * An operation with batched TLB flushing is going on. Anything that
494 * can move process memory needs to flush the TLB when moving a
495 * PROT_NONE or PROT_NUMA mapped page.
497 atomic_t tlb_flush_pending;
498 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
499 /* See flush_tlb_batched_pending() */
500 bool tlb_flush_batched;
502 struct uprobes_state uprobes_state;
503 #ifdef CONFIG_HUGETLB_PAGE
504 atomic_long_t hugetlb_usage;
506 struct work_struct async_put_work;
508 #if IS_ENABLED(CONFIG_HMM)
509 /* HMM needs to track a few things per mm */
512 } __randomize_layout;
514 extern struct mm_struct init_mm;
516 static inline void mm_init_cpumask(struct mm_struct *mm)
518 #ifdef CONFIG_CPUMASK_OFFSTACK
519 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
521 cpumask_clear(mm->cpu_vm_mask_var);
524 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
525 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
527 return mm->cpu_vm_mask_var;
531 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
532 unsigned long start, unsigned long end);
533 extern void tlb_finish_mmu(struct mmu_gather *tlb,
534 unsigned long start, unsigned long end);
536 static inline void init_tlb_flush_pending(struct mm_struct *mm)
538 atomic_set(&mm->tlb_flush_pending, 0);
541 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
543 atomic_inc(&mm->tlb_flush_pending);
545 * The only time this value is relevant is when there are indeed pages
546 * to flush. And we'll only flush pages after changing them, which
549 * So the ordering here is:
551 * atomic_inc(&mm->tlb_flush_pending);
558 * mm_tlb_flush_pending();
563 * atomic_dec(&mm->tlb_flush_pending);
565 * Where the increment if constrained by the PTL unlock, it thus
566 * ensures that the increment is visible if the PTE modification is
567 * visible. After all, if there is no PTE modification, nobody cares
568 * about TLB flushes either.
570 * This very much relies on users (mm_tlb_flush_pending() and
571 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
572 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
573 * locks (PPC) the unlock of one doesn't order against the lock of
576 * The decrement is ordered by the flush_tlb_range(), such that
577 * mm_tlb_flush_pending() will not return false unless all flushes have
582 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
585 * See inc_tlb_flush_pending().
587 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
588 * not order against TLB invalidate completion, which is what we need.
590 * Therefore we must rely on tlb_flush_*() to guarantee order.
592 atomic_dec(&mm->tlb_flush_pending);
595 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
598 * Must be called after having acquired the PTL; orders against that
599 * PTLs release and therefore ensures that if we observe the modified
600 * PTE we must also observe the increment from inc_tlb_flush_pending().
602 * That is, it only guarantees to return true if there is a flush
603 * pending for _this_ PTL.
605 return atomic_read(&mm->tlb_flush_pending);
608 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
611 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
612 * for which there is a TLB flush pending in order to guarantee
613 * we've seen both that PTE modification and the increment.
615 * (no requirement on actually still holding the PTL, that is irrelevant)
617 return atomic_read(&mm->tlb_flush_pending) > 1;
622 struct vm_special_mapping {
623 const char *name; /* The name, e.g. "[vdso]". */
626 * If .fault is not provided, this points to a
627 * NULL-terminated array of pages that back the special mapping.
629 * This must not be NULL unless .fault is provided.
634 * If non-NULL, then this is called to resolve page faults
635 * on the special mapping. If used, .pages is not checked.
637 int (*fault)(const struct vm_special_mapping *sm,
638 struct vm_area_struct *vma,
639 struct vm_fault *vmf);
641 int (*mremap)(const struct vm_special_mapping *sm,
642 struct vm_area_struct *new_vma);
645 enum tlb_flush_reason {
646 TLB_FLUSH_ON_TASK_SWITCH,
647 TLB_REMOTE_SHOOTDOWN,
649 TLB_LOCAL_MM_SHOOTDOWN,
651 NR_TLB_FLUSH_REASONS,
655 * A swap entry has to fit into a "unsigned long", as the entry is hidden
656 * in the "index" field of the swapper address space.
662 #endif /* _LINUX_MM_TYPES_H */