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))
31 * Each physical page in the system has a struct page associated with
32 * it to keep track of whatever it is we are using the page for at the
33 * moment. Note that we have no way to track which tasks are using
34 * a page, though if it is a pagecache page, rmap structures can tell us
37 * If you allocate the page using alloc_pages(), you can use some of the
38 * space in struct page for your own purposes. The five words in the main
39 * union are available, except for bit 0 of the first word which must be
40 * kept clear. Many users use this word to store a pointer to an object
41 * which is guaranteed to be aligned. If you use the same storage as
42 * page->mapping, you must restore it to NULL before freeing the page.
44 * If your page will not be mapped to userspace, you can also use the four
45 * bytes in the mapcount union, but you must call page_mapcount_reset()
48 * If you want to use the refcount field, it must be used in such a way
49 * that other CPUs temporarily incrementing and then decrementing the
50 * refcount does not cause problems. On receiving the page from
51 * alloc_pages(), the refcount will be positive.
53 * If you allocate pages of order > 0, you can use some of the fields
54 * in each subpage, but you may need to restore some of their values
57 * SLUB uses cmpxchg_double() to atomically update its freelist and
58 * counters. That requires that freelist & counters be adjacent and
59 * double-word aligned. We align all struct pages to double-word
60 * boundaries, and ensure that 'freelist' is aligned within the
63 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
64 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
66 #define _struct_page_alignment
70 unsigned long flags; /* Atomic flags, some possibly
71 * updated asynchronously */
73 * Five words (20/40 bytes) are available in this union.
74 * WARNING: bit 0 of the first word is used for PageTail(). That
75 * means the other users of this union MUST NOT use the bit to
76 * avoid collision and false-positive PageTail().
79 struct { /* Page cache and anonymous pages */
81 * @lru: Pageout list, eg. active_list protected by
82 * pgdat->lru_lock. Sometimes used as a generic list
86 /* See page-flags.h for PAGE_MAPPING_FLAGS */
87 struct address_space *mapping;
88 pgoff_t index; /* Our offset within mapping. */
90 * @private: Mapping-private opaque data.
91 * Usually used for buffer_heads if PagePrivate.
92 * Used for swp_entry_t if PageSwapCache.
93 * Indicates order in the buddy system if PageBuddy.
95 unsigned long private;
97 struct { /* page_pool used by netstack */
99 * @dma_addr: might require a 64-bit value even on
100 * 32-bit architectures.
104 struct { /* slab, slob and slub */
106 struct list_head slab_list; /* uses lru */
107 struct { /* Partial pages */
110 int pages; /* Nr of pages left */
111 int pobjects; /* Approximate count */
118 struct kmem_cache *slab_cache; /* not slob */
119 /* Double-word boundary */
120 void *freelist; /* first free object */
122 void *s_mem; /* slab: first object */
123 unsigned long counters; /* SLUB */
131 struct { /* Tail pages of compound page */
132 unsigned long compound_head; /* Bit zero is set */
134 /* First tail page only */
135 unsigned char compound_dtor;
136 unsigned char compound_order;
137 atomic_t compound_mapcount;
139 struct { /* Second tail page of compound page */
140 unsigned long _compound_pad_1; /* compound_head */
141 unsigned long _compound_pad_2;
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 unsigned long hmm_data;
162 unsigned long _zd_pad_1; /* uses mapping */
165 /** @rcu_head: You can use this to free a page by RCU. */
166 struct rcu_head rcu_head;
169 union { /* This union is 4 bytes in size. */
171 * If the page can be mapped to userspace, encodes the number
172 * of times this page is referenced by a page table.
177 * If the page is neither PageSlab nor mappable to userspace,
178 * the value stored here may help determine what this page
179 * is used for. See page-flags.h for a list of page types
180 * which are currently stored here.
182 unsigned int page_type;
184 unsigned int active; /* SLAB */
185 int units; /* SLOB */
188 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
192 struct mem_cgroup *mem_cgroup;
196 * On machines where all RAM is mapped into kernel address space,
197 * we can simply calculate the virtual address. On machines with
198 * highmem some memory is mapped into kernel virtual memory
199 * dynamically, so we need a place to store that address.
200 * Note that this field could be 16 bits on x86 ... ;)
202 * Architectures with slow multiplication can define
203 * WANT_PAGE_VIRTUAL in asm/page.h
205 #if defined(WANT_PAGE_VIRTUAL)
206 void *virtual; /* Kernel virtual address (NULL if
207 not kmapped, ie. highmem) */
208 #endif /* WANT_PAGE_VIRTUAL */
210 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
213 } _struct_page_alignment;
216 * Used for sizing the vmemmap region on some architectures
218 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
220 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
221 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
223 struct page_frag_cache {
225 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
231 /* we maintain a pagecount bias, so that we dont dirty cache line
232 * containing page->_refcount every time we allocate a fragment.
234 unsigned int pagecnt_bias;
238 typedef unsigned long vm_flags_t;
241 * A region containing a mapping of a non-memory backed file under NOMMU
242 * conditions. These are held in a global tree and are pinned by the VMAs that
246 struct rb_node vm_rb; /* link in global region tree */
247 vm_flags_t vm_flags; /* VMA vm_flags */
248 unsigned long vm_start; /* start address of region */
249 unsigned long vm_end; /* region initialised to here */
250 unsigned long vm_top; /* region allocated to here */
251 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
252 struct file *vm_file; /* the backing file or NULL */
254 int vm_usage; /* region usage count (access under nommu_region_sem) */
255 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
259 #ifdef CONFIG_USERFAULTFD
260 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
261 struct vm_userfaultfd_ctx {
262 struct userfaultfd_ctx *ctx;
264 #else /* CONFIG_USERFAULTFD */
265 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
266 struct vm_userfaultfd_ctx {};
267 #endif /* CONFIG_USERFAULTFD */
270 * This struct defines a memory VMM memory area. There is one of these
271 * per VM-area/task. A VM area is any part of the process virtual memory
272 * space that has a special rule for the page-fault handlers (ie a shared
273 * library, the executable area etc).
275 struct vm_area_struct {
276 /* The first cache line has the info for VMA tree walking. */
278 unsigned long vm_start; /* Our start address within vm_mm. */
279 unsigned long vm_end; /* The first byte after our end address
282 /* linked list of VM areas per task, sorted by address */
283 struct vm_area_struct *vm_next, *vm_prev;
285 struct rb_node vm_rb;
288 * Largest free memory gap in bytes to the left of this VMA.
289 * Either between this VMA and vma->vm_prev, or between one of the
290 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
291 * get_unmapped_area find a free area of the right size.
293 unsigned long rb_subtree_gap;
295 /* Second cache line starts here. */
297 struct mm_struct *vm_mm; /* The address space we belong to. */
298 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
299 unsigned long vm_flags; /* Flags, see mm.h. */
302 * For areas with an address space and backing store,
303 * linkage into the address_space->i_mmap interval tree.
307 unsigned long rb_subtree_last;
311 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
312 * list, after a COW of one of the file pages. A MAP_SHARED vma
313 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
314 * or brk vma (with NULL file) can only be in an anon_vma list.
316 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
318 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
320 /* Function pointers to deal with this struct. */
321 const struct vm_operations_struct *vm_ops;
323 /* Information about our backing store: */
324 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
326 struct file * vm_file; /* File we map to (can be NULL). */
327 void * vm_private_data; /* was vm_pte (shared mem) */
329 atomic_long_t swap_readahead_info;
331 struct vm_region *vm_region; /* NOMMU mapping region */
334 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
336 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
337 } __randomize_layout;
340 struct task_struct *task;
341 struct core_thread *next;
346 struct core_thread dumper;
347 struct completion startup;
353 struct vm_area_struct *mmap; /* list of VMAs */
354 struct rb_root mm_rb;
355 u64 vmacache_seqnum; /* per-thread vmacache */
357 unsigned long (*get_unmapped_area) (struct file *filp,
358 unsigned long addr, unsigned long len,
359 unsigned long pgoff, unsigned long flags);
361 unsigned long mmap_base; /* base of mmap area */
362 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
363 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
364 /* Base adresses for compatible mmap() */
365 unsigned long mmap_compat_base;
366 unsigned long mmap_compat_legacy_base;
368 unsigned long task_size; /* size of task vm space */
369 unsigned long highest_vm_end; /* highest vma end address */
373 * @mm_users: The number of users including userspace.
375 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
376 * drops to 0 (i.e. when the task exits and there are no other
377 * temporary reference holders), we also release a reference on
378 * @mm_count (which may then free the &struct mm_struct if
379 * @mm_count also drops to 0).
384 * @mm_count: The number of references to &struct mm_struct
385 * (@mm_users count as 1).
387 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
388 * &struct mm_struct is freed.
393 atomic_long_t pgtables_bytes; /* PTE page table pages */
395 int map_count; /* number of VMAs */
397 spinlock_t page_table_lock; /* Protects page tables and some
400 struct rw_semaphore mmap_sem;
402 struct list_head mmlist; /* List of maybe swapped mm's. These
403 * are globally strung together off
404 * init_mm.mmlist, and are protected
409 unsigned long hiwater_rss; /* High-watermark of RSS usage */
410 unsigned long hiwater_vm; /* High-water virtual memory usage */
412 unsigned long total_vm; /* Total pages mapped */
413 unsigned long locked_vm; /* Pages that have PG_mlocked set */
414 unsigned long pinned_vm; /* Refcount permanently increased */
415 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
416 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
417 unsigned long stack_vm; /* VM_STACK */
418 unsigned long def_flags;
420 spinlock_t arg_lock; /* protect the below fields */
421 unsigned long start_code, end_code, start_data, end_data;
422 unsigned long start_brk, brk, start_stack;
423 unsigned long arg_start, arg_end, env_start, env_end;
425 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
428 * Special counters, in some configurations protected by the
429 * page_table_lock, in other configurations by being atomic.
431 struct mm_rss_stat rss_stat;
433 struct linux_binfmt *binfmt;
435 /* Architecture-specific MM context */
436 mm_context_t context;
438 unsigned long flags; /* Must use atomic bitops to access */
440 struct core_state *core_state; /* coredumping support */
441 #ifdef CONFIG_MEMBARRIER
442 atomic_t membarrier_state;
445 spinlock_t ioctx_lock;
446 struct kioctx_table __rcu *ioctx_table;
450 * "owner" points to a task that is regarded as the canonical
451 * user/owner of this mm. All of the following must be true in
452 * order for it to be changed:
454 * current == mm->owner
456 * new_owner->mm == mm
457 * new_owner->alloc_lock is held
459 struct task_struct __rcu *owner;
461 struct user_namespace *user_ns;
463 /* store ref to file /proc/<pid>/exe symlink points to */
464 struct file __rcu *exe_file;
465 #ifdef CONFIG_MMU_NOTIFIER
466 struct mmu_notifier_mm *mmu_notifier_mm;
468 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
469 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
471 #ifdef CONFIG_NUMA_BALANCING
473 * numa_next_scan is the next time that the PTEs will be marked
474 * pte_numa. NUMA hinting faults will gather statistics and
475 * migrate pages to new nodes if necessary.
477 unsigned long numa_next_scan;
479 /* Restart point for scanning and setting pte_numa */
480 unsigned long numa_scan_offset;
482 /* numa_scan_seq prevents two threads setting pte_numa */
486 * An operation with batched TLB flushing is going on. Anything
487 * that can move process memory needs to flush the TLB when
488 * moving a PROT_NONE or PROT_NUMA mapped page.
490 atomic_t tlb_flush_pending;
491 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
492 /* See flush_tlb_batched_pending() */
493 bool tlb_flush_batched;
495 struct uprobes_state uprobes_state;
496 #ifdef CONFIG_HUGETLB_PAGE
497 atomic_long_t hugetlb_usage;
499 struct work_struct async_put_work;
501 #if IS_ENABLED(CONFIG_HMM)
502 /* HMM needs to track a few things per mm */
505 } __randomize_layout;
508 * The mm_cpumask needs to be at the end of mm_struct, because it
509 * is dynamically sized based on nr_cpu_ids.
511 unsigned long cpu_bitmap[];
514 extern struct mm_struct init_mm;
516 /* Pointer magic because the dynamic array size confuses some compilers. */
517 static inline void mm_init_cpumask(struct mm_struct *mm)
519 unsigned long cpu_bitmap = (unsigned long)mm;
521 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
522 cpumask_clear((struct cpumask *)cpu_bitmap);
525 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
526 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
528 return (struct cpumask *)&mm->cpu_bitmap;
532 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
533 unsigned long start, unsigned long end);
534 extern void tlb_finish_mmu(struct mmu_gather *tlb,
535 unsigned long start, unsigned long end);
537 static inline void init_tlb_flush_pending(struct mm_struct *mm)
539 atomic_set(&mm->tlb_flush_pending, 0);
542 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
544 atomic_inc(&mm->tlb_flush_pending);
546 * The only time this value is relevant is when there are indeed pages
547 * to flush. And we'll only flush pages after changing them, which
550 * So the ordering here is:
552 * atomic_inc(&mm->tlb_flush_pending);
559 * mm_tlb_flush_pending();
564 * atomic_dec(&mm->tlb_flush_pending);
566 * Where the increment if constrained by the PTL unlock, it thus
567 * ensures that the increment is visible if the PTE modification is
568 * visible. After all, if there is no PTE modification, nobody cares
569 * about TLB flushes either.
571 * This very much relies on users (mm_tlb_flush_pending() and
572 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
573 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
574 * locks (PPC) the unlock of one doesn't order against the lock of
577 * The decrement is ordered by the flush_tlb_range(), such that
578 * mm_tlb_flush_pending() will not return false unless all flushes have
583 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
586 * See inc_tlb_flush_pending().
588 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
589 * not order against TLB invalidate completion, which is what we need.
591 * Therefore we must rely on tlb_flush_*() to guarantee order.
593 atomic_dec(&mm->tlb_flush_pending);
596 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
599 * Must be called after having acquired the PTL; orders against that
600 * PTLs release and therefore ensures that if we observe the modified
601 * PTE we must also observe the increment from inc_tlb_flush_pending().
603 * That is, it only guarantees to return true if there is a flush
604 * pending for _this_ PTL.
606 return atomic_read(&mm->tlb_flush_pending);
609 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
612 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
613 * for which there is a TLB flush pending in order to guarantee
614 * we've seen both that PTE modification and the increment.
616 * (no requirement on actually still holding the PTL, that is irrelevant)
618 return atomic_read(&mm->tlb_flush_pending) > 1;
624 * typedef vm_fault_t - Return type for page fault handlers.
626 * Page fault handlers return a bitmask of %VM_FAULT values.
628 typedef __bitwise unsigned int vm_fault_t;
631 * enum vm_fault_reason - Page fault handlers return a bitmask of
632 * these values to tell the core VM what happened when handling the
633 * fault. Used to decide whether a process gets delivered SIGBUS or
634 * just gets major/minor fault counters bumped up.
636 * @VM_FAULT_OOM: Out Of Memory
637 * @VM_FAULT_SIGBUS: Bad access
638 * @VM_FAULT_MAJOR: Page read from storage
639 * @VM_FAULT_WRITE: Special case for get_user_pages
640 * @VM_FAULT_HWPOISON: Hit poisoned small page
641 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
643 * @VM_FAULT_SIGSEGV: segmentation fault
644 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
645 * @VM_FAULT_LOCKED: ->fault locked the returned page
646 * @VM_FAULT_RETRY: ->fault blocked, must retry
647 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
648 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
649 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
650 * fsync() to complete (for synchronous page faults
652 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
655 enum vm_fault_reason {
656 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
657 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
658 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
659 VM_FAULT_WRITE = (__force vm_fault_t)0x000008,
660 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
661 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
662 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
663 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
664 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
665 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
666 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
667 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
668 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
669 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
672 /* Encode hstate index for a hwpoisoned large page */
673 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
674 #define VM_FAULT_GET_HINDEX(x) (((x) >> 16) & 0xf)
676 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
677 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
678 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
680 #define VM_FAULT_RESULT_TRACE \
681 { VM_FAULT_OOM, "OOM" }, \
682 { VM_FAULT_SIGBUS, "SIGBUS" }, \
683 { VM_FAULT_MAJOR, "MAJOR" }, \
684 { VM_FAULT_WRITE, "WRITE" }, \
685 { VM_FAULT_HWPOISON, "HWPOISON" }, \
686 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
687 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
688 { VM_FAULT_NOPAGE, "NOPAGE" }, \
689 { VM_FAULT_LOCKED, "LOCKED" }, \
690 { VM_FAULT_RETRY, "RETRY" }, \
691 { VM_FAULT_FALLBACK, "FALLBACK" }, \
692 { VM_FAULT_DONE_COW, "DONE_COW" }, \
693 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }
695 struct vm_special_mapping {
696 const char *name; /* The name, e.g. "[vdso]". */
699 * If .fault is not provided, this points to a
700 * NULL-terminated array of pages that back the special mapping.
702 * This must not be NULL unless .fault is provided.
707 * If non-NULL, then this is called to resolve page faults
708 * on the special mapping. If used, .pages is not checked.
710 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
711 struct vm_area_struct *vma,
712 struct vm_fault *vmf);
714 int (*mremap)(const struct vm_special_mapping *sm,
715 struct vm_area_struct *new_vma);
718 enum tlb_flush_reason {
719 TLB_FLUSH_ON_TASK_SWITCH,
720 TLB_REMOTE_SHOOTDOWN,
722 TLB_LOCAL_MM_SHOOTDOWN,
724 NR_TLB_FLUSH_REASONS,
728 * A swap entry has to fit into a "unsigned long", as the entry is hidden
729 * in the "index" field of the swapper address space.
735 #endif /* _LINUX_MM_TYPES_H */