1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
5 #include <linux/mm_types_task.h>
7 #include <linux/auxvec.h>
8 #include <linux/list.h>
9 #include <linux/spinlock.h>
10 #include <linux/rbtree.h>
11 #include <linux/rwsem.h>
12 #include <linux/completion.h>
13 #include <linux/cpumask.h>
14 #include <linux/uprobes.h>
15 #include <linux/page-flags-layout.h>
16 #include <linux/workqueue.h>
20 #ifndef AT_VECTOR_SIZE_ARCH
21 #define AT_VECTOR_SIZE_ARCH 0
23 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
30 * Each physical page in the system has a struct page associated with
31 * it to keep track of whatever it is we are using the page for at the
32 * moment. Note that we have no way to track which tasks are using
33 * a page, though if it is a pagecache page, rmap structures can tell us
36 * If you allocate the page using alloc_pages(), you can use some of the
37 * space in struct page for your own purposes. The five words in the main
38 * union are available, except for bit 0 of the first word which must be
39 * kept clear. Many users use this word to store a pointer to an object
40 * which is guaranteed to be aligned. If you use the same storage as
41 * page->mapping, you must restore it to NULL before freeing the page.
43 * If your page will not be mapped to userspace, you can also use the four
44 * bytes in the mapcount union, but you must call page_mapcount_reset()
47 * If you want to use the refcount field, it must be used in such a way
48 * that other CPUs temporarily incrementing and then decrementing the
49 * refcount does not cause problems. On receiving the page from
50 * alloc_pages(), the refcount will be positive.
52 * If you allocate pages of order > 0, you can use some of the fields
53 * in each subpage, but you may need to restore some of their values
56 * SLUB uses cmpxchg_double() to atomically update its freelist and
57 * counters. That requires that freelist & counters be adjacent and
58 * double-word aligned. We align all struct pages to double-word
59 * boundaries, and ensure that 'freelist' is aligned within the
62 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
63 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
65 #define _struct_page_alignment
69 unsigned long flags; /* Atomic flags, some possibly
70 * updated asynchronously */
72 * Five words (20/40 bytes) are available in this union.
73 * WARNING: bit 0 of the first word is used for PageTail(). That
74 * means the other users of this union MUST NOT use the bit to
75 * avoid collision and false-positive PageTail().
78 struct { /* Page cache and anonymous pages */
80 * @lru: Pageout list, eg. active_list protected by
81 * pgdat->lru_lock. Sometimes used as a generic list
85 /* See page-flags.h for PAGE_MAPPING_FLAGS */
86 struct address_space *mapping;
87 pgoff_t index; /* Our offset within mapping. */
89 * @private: Mapping-private opaque data.
90 * Usually used for buffer_heads if PagePrivate.
91 * Used for swp_entry_t if PageSwapCache.
92 * Indicates order in the buddy system if PageBuddy.
94 unsigned long private;
96 struct { /* page_pool used by netstack */
98 * @dma_addr: might require a 64-bit value even on
99 * 32-bit architectures.
103 struct { /* slab, slob and slub */
105 struct list_head slab_list;
106 struct { /* Partial pages */
109 int pages; /* Nr of pages left */
110 int pobjects; /* Approximate count */
117 struct kmem_cache *slab_cache; /* not slob */
118 /* Double-word boundary */
119 void *freelist; /* first free object */
121 void *s_mem; /* slab: first object */
122 unsigned long counters; /* SLUB */
130 struct { /* Tail pages of compound page */
131 unsigned long compound_head; /* Bit zero is set */
133 /* First tail page only */
134 unsigned char compound_dtor;
135 unsigned char compound_order;
136 atomic_t compound_mapcount;
138 struct { /* Second tail page of compound page */
139 unsigned long _compound_pad_1; /* compound_head */
140 unsigned long _compound_pad_2;
141 /* For both global and memcg */
142 struct list_head deferred_list;
144 struct { /* Page table pages */
145 unsigned long _pt_pad_1; /* compound_head */
146 pgtable_t pmd_huge_pte; /* protected by page->ptl */
147 unsigned long _pt_pad_2; /* mapping */
149 struct mm_struct *pt_mm; /* x86 pgds only */
150 atomic_t pt_frag_refcount; /* powerpc */
152 #if ALLOC_SPLIT_PTLOCKS
158 struct { /* ZONE_DEVICE pages */
159 /** @pgmap: Points to the hosting device page map. */
160 struct dev_pagemap *pgmap;
161 void *zone_device_data;
163 * ZONE_DEVICE private pages are counted as being
164 * mapped so the next 3 words hold the mapping, index,
165 * and private fields from the source anonymous or
166 * page cache page while the page is migrated to device
168 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
169 * use the mapping, index, and private fields when
170 * pmem backed DAX files are mapped.
174 /** @rcu_head: You can use this to free a page by RCU. */
175 struct rcu_head rcu_head;
178 union { /* This union is 4 bytes in size. */
180 * If the page can be mapped to userspace, encodes the number
181 * of times this page is referenced by a page table.
186 * If the page is neither PageSlab nor mappable to userspace,
187 * the value stored here may help determine what this page
188 * is used for. See page-flags.h for a list of page types
189 * which are currently stored here.
191 unsigned int page_type;
193 unsigned int active; /* SLAB */
194 int units; /* SLOB */
197 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
201 struct mem_cgroup *mem_cgroup;
205 * On machines where all RAM is mapped into kernel address space,
206 * we can simply calculate the virtual address. On machines with
207 * highmem some memory is mapped into kernel virtual memory
208 * dynamically, so we need a place to store that address.
209 * Note that this field could be 16 bits on x86 ... ;)
211 * Architectures with slow multiplication can define
212 * WANT_PAGE_VIRTUAL in asm/page.h
214 #if defined(WANT_PAGE_VIRTUAL)
215 void *virtual; /* Kernel virtual address (NULL if
216 not kmapped, ie. highmem) */
217 #endif /* WANT_PAGE_VIRTUAL */
219 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
222 } _struct_page_alignment;
225 * Used for sizing the vmemmap region on some architectures
227 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
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 #define page_private(page) ((page)->private)
233 #define set_page_private(page, v) ((page)->private = (v))
235 struct page_frag_cache {
237 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
243 /* we maintain a pagecount bias, so that we dont dirty cache line
244 * containing page->_refcount every time we allocate a fragment.
246 unsigned int pagecnt_bias;
250 typedef unsigned long vm_flags_t;
253 * A region containing a mapping of a non-memory backed file under NOMMU
254 * conditions. These are held in a global tree and are pinned by the VMAs that
258 struct rb_node vm_rb; /* link in global region tree */
259 vm_flags_t vm_flags; /* VMA vm_flags */
260 unsigned long vm_start; /* start address of region */
261 unsigned long vm_end; /* region initialised to here */
262 unsigned long vm_top; /* region allocated to here */
263 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
264 struct file *vm_file; /* the backing file or NULL */
266 int vm_usage; /* region usage count (access under nommu_region_sem) */
267 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
271 #ifdef CONFIG_USERFAULTFD
272 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
273 struct vm_userfaultfd_ctx {
274 struct userfaultfd_ctx *ctx;
276 #else /* CONFIG_USERFAULTFD */
277 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
278 struct vm_userfaultfd_ctx {};
279 #endif /* CONFIG_USERFAULTFD */
282 * This struct defines a memory VMM memory area. There is one of these
283 * per VM-area/task. A VM area is any part of the process virtual memory
284 * space that has a special rule for the page-fault handlers (ie a shared
285 * library, the executable area etc).
287 struct vm_area_struct {
288 /* The first cache line has the info for VMA tree walking. */
290 unsigned long vm_start; /* Our start address within vm_mm. */
291 unsigned long vm_end; /* The first byte after our end address
294 /* linked list of VM areas per task, sorted by address */
295 struct vm_area_struct *vm_next, *vm_prev;
297 struct rb_node vm_rb;
300 * Largest free memory gap in bytes to the left of this VMA.
301 * Either between this VMA and vma->vm_prev, or between one of the
302 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
303 * get_unmapped_area find a free area of the right size.
305 unsigned long rb_subtree_gap;
307 /* Second cache line starts here. */
309 struct mm_struct *vm_mm; /* The address space we belong to. */
310 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
311 unsigned long vm_flags; /* Flags, see mm.h. */
314 * For areas with an address space and backing store,
315 * linkage into the address_space->i_mmap interval tree.
319 unsigned long rb_subtree_last;
323 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
324 * list, after a COW of one of the file pages. A MAP_SHARED vma
325 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
326 * or brk vma (with NULL file) can only be in an anon_vma list.
328 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
330 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
332 /* Function pointers to deal with this struct. */
333 const struct vm_operations_struct *vm_ops;
335 /* Information about our backing store: */
336 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
338 struct file * vm_file; /* File we map to (can be NULL). */
339 void * vm_private_data; /* was vm_pte (shared mem) */
342 atomic_long_t swap_readahead_info;
345 struct vm_region *vm_region; /* NOMMU mapping region */
348 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
350 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
351 } __randomize_layout;
354 struct task_struct *task;
355 struct core_thread *next;
360 struct core_thread dumper;
361 struct completion startup;
367 struct vm_area_struct *mmap; /* list of VMAs */
368 struct rb_root mm_rb;
369 u64 vmacache_seqnum; /* per-thread vmacache */
371 unsigned long (*get_unmapped_area) (struct file *filp,
372 unsigned long addr, unsigned long len,
373 unsigned long pgoff, unsigned long flags);
375 unsigned long mmap_base; /* base of mmap area */
376 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
377 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
378 /* Base adresses for compatible mmap() */
379 unsigned long mmap_compat_base;
380 unsigned long mmap_compat_legacy_base;
382 unsigned long task_size; /* size of task vm space */
383 unsigned long highest_vm_end; /* highest vma end address */
387 * @mm_users: The number of users including userspace.
389 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
390 * drops to 0 (i.e. when the task exits and there are no other
391 * temporary reference holders), we also release a reference on
392 * @mm_count (which may then free the &struct mm_struct if
393 * @mm_count also drops to 0).
398 * @mm_count: The number of references to &struct mm_struct
399 * (@mm_users count as 1).
401 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
402 * &struct mm_struct is freed.
407 atomic_long_t pgtables_bytes; /* PTE page table pages */
409 int map_count; /* number of VMAs */
411 spinlock_t page_table_lock; /* Protects page tables and some
414 struct rw_semaphore mmap_sem;
416 struct list_head mmlist; /* List of maybe swapped mm's. These
417 * are globally strung together off
418 * init_mm.mmlist, and are protected
423 unsigned long hiwater_rss; /* High-watermark of RSS usage */
424 unsigned long hiwater_vm; /* High-water virtual memory usage */
426 unsigned long total_vm; /* Total pages mapped */
427 unsigned long locked_vm; /* Pages that have PG_mlocked set */
428 atomic64_t pinned_vm; /* Refcount permanently increased */
429 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
430 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
431 unsigned long stack_vm; /* VM_STACK */
432 unsigned long def_flags;
434 spinlock_t arg_lock; /* protect the below fields */
435 unsigned long start_code, end_code, start_data, end_data;
436 unsigned long start_brk, brk, start_stack;
437 unsigned long arg_start, arg_end, env_start, env_end;
439 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
442 * Special counters, in some configurations protected by the
443 * page_table_lock, in other configurations by being atomic.
445 struct mm_rss_stat rss_stat;
447 struct linux_binfmt *binfmt;
449 /* Architecture-specific MM context */
450 mm_context_t context;
452 unsigned long flags; /* Must use atomic bitops to access */
454 struct core_state *core_state; /* coredumping support */
455 #ifdef CONFIG_MEMBARRIER
456 atomic_t membarrier_state;
459 spinlock_t ioctx_lock;
460 struct kioctx_table __rcu *ioctx_table;
464 * "owner" points to a task that is regarded as the canonical
465 * user/owner of this mm. All of the following must be true in
466 * order for it to be changed:
468 * current == mm->owner
470 * new_owner->mm == mm
471 * new_owner->alloc_lock is held
473 struct task_struct __rcu *owner;
475 struct user_namespace *user_ns;
477 /* store ref to file /proc/<pid>/exe symlink points to */
478 struct file __rcu *exe_file;
479 #ifdef CONFIG_MMU_NOTIFIER
480 struct mmu_notifier_mm *mmu_notifier_mm;
482 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
483 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
485 #ifdef CONFIG_NUMA_BALANCING
487 * numa_next_scan is the next time that the PTEs will be marked
488 * pte_numa. NUMA hinting faults will gather statistics and
489 * migrate pages to new nodes if necessary.
491 unsigned long numa_next_scan;
493 /* Restart point for scanning and setting pte_numa */
494 unsigned long numa_scan_offset;
496 /* numa_scan_seq prevents two threads setting pte_numa */
500 * An operation with batched TLB flushing is going on. Anything
501 * that can move process memory needs to flush the TLB when
502 * moving a PROT_NONE or PROT_NUMA mapped page.
504 atomic_t tlb_flush_pending;
505 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
506 /* See flush_tlb_batched_pending() */
507 bool tlb_flush_batched;
509 struct uprobes_state uprobes_state;
510 #ifdef CONFIG_HUGETLB_PAGE
511 atomic_long_t hugetlb_usage;
513 struct work_struct async_put_work;
514 } __randomize_layout;
517 * The mm_cpumask needs to be at the end of mm_struct, because it
518 * is dynamically sized based on nr_cpu_ids.
520 unsigned long cpu_bitmap[];
523 extern struct mm_struct init_mm;
525 /* Pointer magic because the dynamic array size confuses some compilers. */
526 static inline void mm_init_cpumask(struct mm_struct *mm)
528 unsigned long cpu_bitmap = (unsigned long)mm;
530 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
531 cpumask_clear((struct cpumask *)cpu_bitmap);
534 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
535 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
537 return (struct cpumask *)&mm->cpu_bitmap;
541 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
542 unsigned long start, unsigned long end);
543 extern void tlb_finish_mmu(struct mmu_gather *tlb,
544 unsigned long start, unsigned long end);
546 static inline void init_tlb_flush_pending(struct mm_struct *mm)
548 atomic_set(&mm->tlb_flush_pending, 0);
551 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
553 atomic_inc(&mm->tlb_flush_pending);
555 * The only time this value is relevant is when there are indeed pages
556 * to flush. And we'll only flush pages after changing them, which
559 * So the ordering here is:
561 * atomic_inc(&mm->tlb_flush_pending);
568 * mm_tlb_flush_pending();
573 * atomic_dec(&mm->tlb_flush_pending);
575 * Where the increment if constrained by the PTL unlock, it thus
576 * ensures that the increment is visible if the PTE modification is
577 * visible. After all, if there is no PTE modification, nobody cares
578 * about TLB flushes either.
580 * This very much relies on users (mm_tlb_flush_pending() and
581 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
582 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
583 * locks (PPC) the unlock of one doesn't order against the lock of
586 * The decrement is ordered by the flush_tlb_range(), such that
587 * mm_tlb_flush_pending() will not return false unless all flushes have
592 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
595 * See inc_tlb_flush_pending().
597 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
598 * not order against TLB invalidate completion, which is what we need.
600 * Therefore we must rely on tlb_flush_*() to guarantee order.
602 atomic_dec(&mm->tlb_flush_pending);
605 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
608 * Must be called after having acquired the PTL; orders against that
609 * PTLs release and therefore ensures that if we observe the modified
610 * PTE we must also observe the increment from inc_tlb_flush_pending().
612 * That is, it only guarantees to return true if there is a flush
613 * pending for _this_ PTL.
615 return atomic_read(&mm->tlb_flush_pending);
618 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
621 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
622 * for which there is a TLB flush pending in order to guarantee
623 * we've seen both that PTE modification and the increment.
625 * (no requirement on actually still holding the PTL, that is irrelevant)
627 return atomic_read(&mm->tlb_flush_pending) > 1;
633 * typedef vm_fault_t - Return type for page fault handlers.
635 * Page fault handlers return a bitmask of %VM_FAULT values.
637 typedef __bitwise unsigned int vm_fault_t;
640 * enum vm_fault_reason - Page fault handlers return a bitmask of
641 * these values to tell the core VM what happened when handling the
642 * fault. Used to decide whether a process gets delivered SIGBUS or
643 * just gets major/minor fault counters bumped up.
645 * @VM_FAULT_OOM: Out Of Memory
646 * @VM_FAULT_SIGBUS: Bad access
647 * @VM_FAULT_MAJOR: Page read from storage
648 * @VM_FAULT_WRITE: Special case for get_user_pages
649 * @VM_FAULT_HWPOISON: Hit poisoned small page
650 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
652 * @VM_FAULT_SIGSEGV: segmentation fault
653 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
654 * @VM_FAULT_LOCKED: ->fault locked the returned page
655 * @VM_FAULT_RETRY: ->fault blocked, must retry
656 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
657 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
658 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
659 * fsync() to complete (for synchronous page faults
661 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
664 enum vm_fault_reason {
665 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
666 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
667 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
668 VM_FAULT_WRITE = (__force vm_fault_t)0x000008,
669 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
670 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
671 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
672 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
673 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
674 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
675 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
676 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
677 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
678 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
681 /* Encode hstate index for a hwpoisoned large page */
682 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
683 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
685 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
686 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
687 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
689 #define VM_FAULT_RESULT_TRACE \
690 { VM_FAULT_OOM, "OOM" }, \
691 { VM_FAULT_SIGBUS, "SIGBUS" }, \
692 { VM_FAULT_MAJOR, "MAJOR" }, \
693 { VM_FAULT_WRITE, "WRITE" }, \
694 { VM_FAULT_HWPOISON, "HWPOISON" }, \
695 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
696 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
697 { VM_FAULT_NOPAGE, "NOPAGE" }, \
698 { VM_FAULT_LOCKED, "LOCKED" }, \
699 { VM_FAULT_RETRY, "RETRY" }, \
700 { VM_FAULT_FALLBACK, "FALLBACK" }, \
701 { VM_FAULT_DONE_COW, "DONE_COW" }, \
702 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }
704 struct vm_special_mapping {
705 const char *name; /* The name, e.g. "[vdso]". */
708 * If .fault is not provided, this points to a
709 * NULL-terminated array of pages that back the special mapping.
711 * This must not be NULL unless .fault is provided.
716 * If non-NULL, then this is called to resolve page faults
717 * on the special mapping. If used, .pages is not checked.
719 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
720 struct vm_area_struct *vma,
721 struct vm_fault *vmf);
723 int (*mremap)(const struct vm_special_mapping *sm,
724 struct vm_area_struct *new_vma);
727 enum tlb_flush_reason {
728 TLB_FLUSH_ON_TASK_SWITCH,
729 TLB_REMOTE_SHOOTDOWN,
731 TLB_LOCAL_MM_SHOOTDOWN,
733 NR_TLB_FLUSH_REASONS,
737 * A swap entry has to fit into a "unsigned long", as the entry is hidden
738 * in the "index" field of the swapper address space.
744 #endif /* _LINUX_MM_TYPES_H */