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/kref.h>
9 #include <linux/list.h>
10 #include <linux/spinlock.h>
11 #include <linux/rbtree.h>
12 #include <linux/maple_tree.h>
13 #include <linux/rwsem.h>
14 #include <linux/completion.h>
15 #include <linux/cpumask.h>
16 #include <linux/uprobes.h>
17 #include <linux/rcupdate.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/workqueue.h>
20 #include <linux/seqlock.h>
21 #include <linux/percpu_counter.h>
25 #ifndef AT_VECTOR_SIZE_ARCH
26 #define AT_VECTOR_SIZE_ARCH 0
28 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
36 * Each physical page in the system has a struct page associated with
37 * it to keep track of whatever it is we are using the page for at the
38 * moment. Note that we have no way to track which tasks are using
39 * a page, though if it is a pagecache page, rmap structures can tell us
42 * If you allocate the page using alloc_pages(), you can use some of the
43 * space in struct page for your own purposes. The five words in the main
44 * union are available, except for bit 0 of the first word which must be
45 * kept clear. Many users use this word to store a pointer to an object
46 * which is guaranteed to be aligned. If you use the same storage as
47 * page->mapping, you must restore it to NULL before freeing the page.
49 * If your page will not be mapped to userspace, you can also use the four
50 * bytes in the mapcount union, but you must call page_mapcount_reset()
53 * If you want to use the refcount field, it must be used in such a way
54 * that other CPUs temporarily incrementing and then decrementing the
55 * refcount does not cause problems. On receiving the page from
56 * alloc_pages(), the refcount will be positive.
58 * If you allocate pages of order > 0, you can use some of the fields
59 * in each subpage, but you may need to restore some of their values
62 * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
63 * That requires that freelist & counters in struct slab be adjacent and
64 * double-word aligned. Because struct slab currently just reinterprets the
65 * bits of struct page, we align all struct pages to double-word boundaries,
66 * and ensure that 'freelist' is aligned within struct slab.
68 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
69 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
71 #define _struct_page_alignment __aligned(sizeof(unsigned long))
75 unsigned long flags; /* Atomic flags, some possibly
76 * updated asynchronously */
78 * Five words (20/40 bytes) are available in this union.
79 * WARNING: bit 0 of the first word is used for PageTail(). That
80 * means the other users of this union MUST NOT use the bit to
81 * avoid collision and false-positive PageTail().
84 struct { /* Page cache and anonymous pages */
86 * @lru: Pageout list, eg. active_list protected by
87 * lruvec->lru_lock. Sometimes used as a generic list
93 /* Or, for the Unevictable "LRU list" slot */
95 /* Always even, to negate PageTail */
97 /* Count page's or folio's mlocks */
98 unsigned int mlock_count;
102 struct list_head buddy_list;
103 struct list_head pcp_list;
105 /* See page-flags.h for PAGE_MAPPING_FLAGS */
106 struct address_space *mapping;
108 pgoff_t index; /* Our offset within mapping. */
109 unsigned long share; /* share count for fsdax */
112 * @private: Mapping-private opaque data.
113 * Usually used for buffer_heads if PagePrivate.
114 * Used for swp_entry_t if PageSwapCache.
115 * Indicates order in the buddy system if PageBuddy.
117 unsigned long private;
119 struct { /* page_pool used by netstack */
121 * @pp_magic: magic value to avoid recycling non
122 * page_pool allocated pages.
124 unsigned long pp_magic;
125 struct page_pool *pp;
126 unsigned long _pp_mapping_pad;
127 unsigned long dma_addr;
130 * dma_addr_upper: might require a 64-bit
131 * value on 32-bit architectures.
133 unsigned long dma_addr_upper;
135 * For frag page support, not supported in
136 * 32-bit architectures with 64-bit DMA.
138 atomic_long_t pp_frag_count;
141 struct { /* Tail pages of compound page */
142 unsigned long compound_head; /* Bit zero is set */
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;
194 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
198 unsigned long memcg_data;
202 * On machines where all RAM is mapped into kernel address space,
203 * we can simply calculate the virtual address. On machines with
204 * highmem some memory is mapped into kernel virtual memory
205 * dynamically, so we need a place to store that address.
206 * Note that this field could be 16 bits on x86 ... ;)
208 * Architectures with slow multiplication can define
209 * WANT_PAGE_VIRTUAL in asm/page.h
211 #if defined(WANT_PAGE_VIRTUAL)
212 void *virtual; /* Kernel virtual address (NULL if
213 not kmapped, ie. highmem) */
214 #endif /* WANT_PAGE_VIRTUAL */
218 * KMSAN metadata for this page:
219 * - shadow page: every bit indicates whether the corresponding
220 * bit of the original page is initialized (0) or not (1);
221 * - origin page: every 4 bytes contain an id of the stack trace
222 * where the uninitialized value was created.
224 struct page *kmsan_shadow;
225 struct page *kmsan_origin;
228 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
231 } _struct_page_alignment;
234 * struct encoded_page - a nonexistent type marking this pointer
236 * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
237 * with the low bits of the pointer indicating extra context-dependent
238 * information. Not super-common, but happens in mmu_gather and mlock
239 * handling, and this acts as a type system check on that use.
241 * We only really have two guaranteed bits in general, although you could
242 * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
245 * Use the supplied helper functions to endcode/decode the pointer and bits.
248 #define ENCODE_PAGE_BITS 3ul
249 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
251 BUILD_BUG_ON(flags > ENCODE_PAGE_BITS);
252 return (struct encoded_page *)(flags | (unsigned long)page);
255 static inline unsigned long encoded_page_flags(struct encoded_page *page)
257 return ENCODE_PAGE_BITS & (unsigned long)page;
260 static inline struct page *encoded_page_ptr(struct encoded_page *page)
262 return (struct page *)(~ENCODE_PAGE_BITS & (unsigned long)page);
266 * struct folio - Represents a contiguous set of bytes.
267 * @flags: Identical to the page flags.
268 * @lru: Least Recently Used list; tracks how recently this folio was used.
269 * @mlock_count: Number of times this folio has been pinned by mlock().
270 * @mapping: The file this page belongs to, or refers to the anon_vma for
272 * @index: Offset within the file, in units of pages. For anonymous memory,
273 * this is the index from the beginning of the mmap.
274 * @private: Filesystem per-folio data (see folio_attach_private()).
275 * Used for swp_entry_t if folio_test_swapcache().
276 * @_mapcount: Do not access this member directly. Use folio_mapcount() to
277 * find out how many times this folio is mapped by userspace.
278 * @_refcount: Do not access this member directly. Use folio_ref_count()
279 * to find how many references there are to this folio.
280 * @memcg_data: Memory Control Group data.
281 * @_folio_dtor: Which destructor to use for this folio.
282 * @_folio_order: Do not use directly, call folio_order().
283 * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
284 * @_nr_pages_mapped: Do not use directly, call folio_mapcount().
285 * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
286 * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
287 * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
288 * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
289 * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
290 * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
291 * @_deferred_list: Folios to be split under memory pressure.
293 * A folio is a physically, virtually and logically contiguous set
294 * of bytes. It is a power-of-two in size, and it is aligned to that
295 * same power-of-two. It is at least as large as %PAGE_SIZE. If it is
296 * in the page cache, it is at a file offset which is a multiple of that
297 * power-of-two. It may be mapped into userspace at an address which is
298 * at an arbitrary page offset, but its kernel virtual address is aligned
302 /* private: don't document the anon union */
308 struct list_head lru;
309 /* private: avoid cluttering the output */
313 unsigned int mlock_count;
318 struct address_space *mapping;
324 unsigned long memcg_data;
326 /* private: the union with struct page is transitional */
332 unsigned long _flags_1;
333 unsigned long _head_1;
335 unsigned char _folio_dtor;
336 unsigned char _folio_order;
337 atomic_t _entire_mapcount;
338 atomic_t _nr_pages_mapped;
341 unsigned int _folio_nr_pages;
343 /* private: the union with struct page is transitional */
345 struct page __page_1;
349 unsigned long _flags_2;
350 unsigned long _head_2;
352 void *_hugetlb_subpool;
353 void *_hugetlb_cgroup;
354 void *_hugetlb_cgroup_rsvd;
355 void *_hugetlb_hwpoison;
356 /* private: the union with struct page is transitional */
359 unsigned long _flags_2a;
360 unsigned long _head_2a;
362 struct list_head _deferred_list;
363 /* private: the union with struct page is transitional */
365 struct page __page_2;
369 #define FOLIO_MATCH(pg, fl) \
370 static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
371 FOLIO_MATCH(flags, flags);
372 FOLIO_MATCH(lru, lru);
373 FOLIO_MATCH(mapping, mapping);
374 FOLIO_MATCH(compound_head, lru);
375 FOLIO_MATCH(index, index);
376 FOLIO_MATCH(private, private);
377 FOLIO_MATCH(_mapcount, _mapcount);
378 FOLIO_MATCH(_refcount, _refcount);
380 FOLIO_MATCH(memcg_data, memcg_data);
383 #define FOLIO_MATCH(pg, fl) \
384 static_assert(offsetof(struct folio, fl) == \
385 offsetof(struct page, pg) + sizeof(struct page))
386 FOLIO_MATCH(flags, _flags_1);
387 FOLIO_MATCH(compound_head, _head_1);
389 #define FOLIO_MATCH(pg, fl) \
390 static_assert(offsetof(struct folio, fl) == \
391 offsetof(struct page, pg) + 2 * sizeof(struct page))
392 FOLIO_MATCH(flags, _flags_2);
393 FOLIO_MATCH(compound_head, _head_2);
397 * Used for sizing the vmemmap region on some architectures
399 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
401 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
402 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
405 * page_private can be used on tail pages. However, PagePrivate is only
406 * checked by the VM on the head page. So page_private on the tail pages
407 * should be used for data that's ancillary to the head page (eg attaching
408 * buffer heads to tail pages after attaching buffer heads to the head page)
410 #define page_private(page) ((page)->private)
412 static inline void set_page_private(struct page *page, unsigned long private)
414 page->private = private;
417 static inline void *folio_get_private(struct folio *folio)
419 return folio->private;
422 struct page_frag_cache {
424 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
430 /* we maintain a pagecount bias, so that we dont dirty cache line
431 * containing page->_refcount every time we allocate a fragment.
433 unsigned int pagecnt_bias;
437 typedef unsigned long vm_flags_t;
440 * A region containing a mapping of a non-memory backed file under NOMMU
441 * conditions. These are held in a global tree and are pinned by the VMAs that
445 struct rb_node vm_rb; /* link in global region tree */
446 vm_flags_t vm_flags; /* VMA vm_flags */
447 unsigned long vm_start; /* start address of region */
448 unsigned long vm_end; /* region initialised to here */
449 unsigned long vm_top; /* region allocated to here */
450 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
451 struct file *vm_file; /* the backing file or NULL */
453 int vm_usage; /* region usage count (access under nommu_region_sem) */
454 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
458 #ifdef CONFIG_USERFAULTFD
459 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
460 struct vm_userfaultfd_ctx {
461 struct userfaultfd_ctx *ctx;
463 #else /* CONFIG_USERFAULTFD */
464 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
465 struct vm_userfaultfd_ctx {};
466 #endif /* CONFIG_USERFAULTFD */
468 struct anon_vma_name {
470 /* The name needs to be at the end because it is dynamically sized. */
475 struct rw_semaphore lock;
478 struct vma_numab_state {
479 unsigned long next_scan;
480 unsigned long next_pid_reset;
481 unsigned long access_pids[2];
485 * This struct describes a virtual memory area. There is one of these
486 * per VM-area/task. A VM area is any part of the process virtual memory
487 * space that has a special rule for the page-fault handlers (ie a shared
488 * library, the executable area etc).
490 struct vm_area_struct {
491 /* The first cache line has the info for VMA tree walking. */
495 /* VMA covers [vm_start; vm_end) addresses within mm */
496 unsigned long vm_start;
497 unsigned long vm_end;
499 #ifdef CONFIG_PER_VMA_LOCK
500 struct rcu_head vm_rcu; /* Used for deferred freeing. */
504 struct mm_struct *vm_mm; /* The address space we belong to. */
505 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
509 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
512 const vm_flags_t vm_flags;
513 vm_flags_t __private __vm_flags;
516 #ifdef CONFIG_PER_VMA_LOCK
518 * Can only be written (using WRITE_ONCE()) while holding both:
519 * - mmap_lock (in write mode)
520 * - vm_lock->lock (in write mode)
521 * Can be read reliably while holding one of:
522 * - mmap_lock (in read or write mode)
523 * - vm_lock->lock (in read or write mode)
524 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
525 * while holding nothing (except RCU to keep the VMA struct allocated).
527 * This sequence counter is explicitly allowed to overflow; sequence
528 * counter reuse can only lead to occasional unnecessary use of the
532 struct vma_lock *vm_lock;
534 /* Flag to indicate areas detached from the mm->mm_mt tree */
539 * For areas with an address space and backing store,
540 * linkage into the address_space->i_mmap interval tree.
545 unsigned long rb_subtree_last;
549 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
550 * list, after a COW of one of the file pages. A MAP_SHARED vma
551 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
552 * or brk vma (with NULL file) can only be in an anon_vma list.
554 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
556 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
558 /* Function pointers to deal with this struct. */
559 const struct vm_operations_struct *vm_ops;
561 /* Information about our backing store: */
562 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
564 struct file * vm_file; /* File we map to (can be NULL). */
565 void * vm_private_data; /* was vm_pte (shared mem) */
567 #ifdef CONFIG_ANON_VMA_NAME
569 * For private and shared anonymous mappings, a pointer to a null
570 * terminated string containing the name given to the vma, or NULL if
571 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
573 struct anon_vma_name *anon_name;
576 atomic_long_t swap_readahead_info;
579 struct vm_region *vm_region; /* NOMMU mapping region */
582 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
584 #ifdef CONFIG_NUMA_BALANCING
585 struct vma_numab_state *numab_state; /* NUMA Balancing state */
587 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
588 } __randomize_layout;
590 #ifdef CONFIG_SCHED_MM_CID
601 * Fields which are often written to are placed in a separate
606 * @mm_count: The number of references to &struct
607 * mm_struct (@mm_users count as 1).
609 * Use mmgrab()/mmdrop() to modify. When this drops to
610 * 0, the &struct mm_struct is freed.
613 } ____cacheline_aligned_in_smp;
615 struct maple_tree mm_mt;
617 unsigned long (*get_unmapped_area) (struct file *filp,
618 unsigned long addr, unsigned long len,
619 unsigned long pgoff, unsigned long flags);
621 unsigned long mmap_base; /* base of mmap area */
622 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
623 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
624 /* Base addresses for compatible mmap() */
625 unsigned long mmap_compat_base;
626 unsigned long mmap_compat_legacy_base;
628 unsigned long task_size; /* size of task vm space */
631 #ifdef CONFIG_MEMBARRIER
633 * @membarrier_state: Flags controlling membarrier behavior.
635 * This field is close to @pgd to hopefully fit in the same
636 * cache-line, which needs to be touched by switch_mm().
638 atomic_t membarrier_state;
642 * @mm_users: The number of users including userspace.
644 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
645 * drops to 0 (i.e. when the task exits and there are no other
646 * temporary reference holders), we also release a reference on
647 * @mm_count (which may then free the &struct mm_struct if
648 * @mm_count also drops to 0).
652 #ifdef CONFIG_SCHED_MM_CID
654 * @pcpu_cid: Per-cpu current cid.
656 * Keep track of the currently allocated mm_cid for each cpu.
657 * The per-cpu mm_cid values are serialized by their respective
660 struct mm_cid __percpu *pcpu_cid;
662 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
664 * When the next mm_cid scan is due (in jiffies).
666 unsigned long mm_cid_next_scan;
669 atomic_long_t pgtables_bytes; /* size of all page tables */
671 int map_count; /* number of VMAs */
673 spinlock_t page_table_lock; /* Protects page tables and some
677 * With some kernel config, the current mmap_lock's offset
678 * inside 'mm_struct' is at 0x120, which is very optimal, as
679 * its two hot fields 'count' and 'owner' sit in 2 different
680 * cachelines, and when mmap_lock is highly contended, both
681 * of the 2 fields will be accessed frequently, current layout
682 * will help to reduce cache bouncing.
684 * So please be careful with adding new fields before
685 * mmap_lock, which can easily push the 2 fields into one
688 struct rw_semaphore mmap_lock;
690 struct list_head mmlist; /* List of maybe swapped mm's. These
691 * are globally strung together off
692 * init_mm.mmlist, and are protected
695 #ifdef CONFIG_PER_VMA_LOCK
697 * This field has lock-like semantics, meaning it is sometimes
698 * accessed with ACQUIRE/RELEASE semantics.
699 * Roughly speaking, incrementing the sequence number is
700 * equivalent to releasing locks on VMAs; reading the sequence
701 * number can be part of taking a read lock on a VMA.
703 * Can be modified under write mmap_lock using RELEASE
705 * Can be read with no other protection when holding write
707 * Can be read with ACQUIRE semantics if not holding write
714 unsigned long hiwater_rss; /* High-watermark of RSS usage */
715 unsigned long hiwater_vm; /* High-water virtual memory usage */
717 unsigned long total_vm; /* Total pages mapped */
718 unsigned long locked_vm; /* Pages that have PG_mlocked set */
719 atomic64_t pinned_vm; /* Refcount permanently increased */
720 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
721 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
722 unsigned long stack_vm; /* VM_STACK */
723 unsigned long def_flags;
726 * @write_protect_seq: Locked when any thread is write
727 * protecting pages mapped by this mm to enforce a later COW,
728 * for instance during page table copying for fork().
730 seqcount_t write_protect_seq;
732 spinlock_t arg_lock; /* protect the below fields */
734 unsigned long start_code, end_code, start_data, end_data;
735 unsigned long start_brk, brk, start_stack;
736 unsigned long arg_start, arg_end, env_start, env_end;
738 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
740 struct percpu_counter rss_stat[NR_MM_COUNTERS];
742 struct linux_binfmt *binfmt;
744 /* Architecture-specific MM context */
745 mm_context_t context;
747 unsigned long flags; /* Must use atomic bitops to access */
750 spinlock_t ioctx_lock;
751 struct kioctx_table __rcu *ioctx_table;
755 * "owner" points to a task that is regarded as the canonical
756 * user/owner of this mm. All of the following must be true in
757 * order for it to be changed:
759 * current == mm->owner
761 * new_owner->mm == mm
762 * new_owner->alloc_lock is held
764 struct task_struct __rcu *owner;
766 struct user_namespace *user_ns;
768 /* store ref to file /proc/<pid>/exe symlink points to */
769 struct file __rcu *exe_file;
770 #ifdef CONFIG_MMU_NOTIFIER
771 struct mmu_notifier_subscriptions *notifier_subscriptions;
773 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
774 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
776 #ifdef CONFIG_NUMA_BALANCING
778 * numa_next_scan is the next time that PTEs will be remapped
779 * PROT_NONE to trigger NUMA hinting faults; such faults gather
780 * statistics and migrate pages to new nodes if necessary.
782 unsigned long numa_next_scan;
784 /* Restart point for scanning and remapping PTEs. */
785 unsigned long numa_scan_offset;
787 /* numa_scan_seq prevents two threads remapping PTEs. */
791 * An operation with batched TLB flushing is going on. Anything
792 * that can move process memory needs to flush the TLB when
793 * moving a PROT_NONE mapped page.
795 atomic_t tlb_flush_pending;
796 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
797 /* See flush_tlb_batched_pending() */
798 atomic_t tlb_flush_batched;
800 struct uprobes_state uprobes_state;
801 #ifdef CONFIG_PREEMPT_RT
802 struct rcu_head delayed_drop;
804 #ifdef CONFIG_HUGETLB_PAGE
805 atomic_long_t hugetlb_usage;
807 struct work_struct async_put_work;
809 #ifdef CONFIG_IOMMU_SVA
814 * Represent how many pages of this process are involved in KSM
817 unsigned long ksm_merging_pages;
819 * Represent how many pages are checked for ksm merging
820 * including merged and not merged.
822 unsigned long ksm_rmap_items;
824 #ifdef CONFIG_LRU_GEN
826 /* this mm_struct is on lru_gen_mm_list */
827 struct list_head list;
829 * Set when switching to this mm_struct, as a hint of
830 * whether it has been used since the last time per-node
831 * page table walkers cleared the corresponding bits.
833 unsigned long bitmap;
835 /* points to the memcg of "owner" above */
836 struct mem_cgroup *memcg;
839 #endif /* CONFIG_LRU_GEN */
840 } __randomize_layout;
843 * The mm_cpumask needs to be at the end of mm_struct, because it
844 * is dynamically sized based on nr_cpu_ids.
846 unsigned long cpu_bitmap[];
849 #define MM_MT_FLAGS (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
851 extern struct mm_struct init_mm;
853 /* Pointer magic because the dynamic array size confuses some compilers. */
854 static inline void mm_init_cpumask(struct mm_struct *mm)
856 unsigned long cpu_bitmap = (unsigned long)mm;
858 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
859 cpumask_clear((struct cpumask *)cpu_bitmap);
862 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
863 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
865 return (struct cpumask *)&mm->cpu_bitmap;
868 #ifdef CONFIG_LRU_GEN
870 struct lru_gen_mm_list {
871 /* mm_struct list for page table walkers */
872 struct list_head fifo;
873 /* protects the list above */
877 void lru_gen_add_mm(struct mm_struct *mm);
878 void lru_gen_del_mm(struct mm_struct *mm);
880 void lru_gen_migrate_mm(struct mm_struct *mm);
883 static inline void lru_gen_init_mm(struct mm_struct *mm)
885 INIT_LIST_HEAD(&mm->lru_gen.list);
886 mm->lru_gen.bitmap = 0;
888 mm->lru_gen.memcg = NULL;
892 static inline void lru_gen_use_mm(struct mm_struct *mm)
895 * When the bitmap is set, page reclaim knows this mm_struct has been
896 * used since the last time it cleared the bitmap. So it might be worth
897 * walking the page tables of this mm_struct to clear the accessed bit.
899 WRITE_ONCE(mm->lru_gen.bitmap, -1);
902 #else /* !CONFIG_LRU_GEN */
904 static inline void lru_gen_add_mm(struct mm_struct *mm)
908 static inline void lru_gen_del_mm(struct mm_struct *mm)
913 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
918 static inline void lru_gen_init_mm(struct mm_struct *mm)
922 static inline void lru_gen_use_mm(struct mm_struct *mm)
926 #endif /* CONFIG_LRU_GEN */
928 struct vma_iterator {
932 #define VMA_ITERATOR(name, __mm, __addr) \
933 struct vma_iterator name = { \
935 .tree = &(__mm)->mm_mt, \
941 static inline void vma_iter_init(struct vma_iterator *vmi,
942 struct mm_struct *mm, unsigned long addr)
944 mas_init(&vmi->mas, &mm->mm_mt, addr);
947 #ifdef CONFIG_SCHED_MM_CID
950 MM_CID_UNSET = -1U, /* Unset state has lazy_put flag set. */
951 MM_CID_LAZY_PUT = (1U << 31),
954 static inline bool mm_cid_is_unset(int cid)
956 return cid == MM_CID_UNSET;
959 static inline bool mm_cid_is_lazy_put(int cid)
961 return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
964 static inline bool mm_cid_is_valid(int cid)
966 return !(cid & MM_CID_LAZY_PUT);
969 static inline int mm_cid_set_lazy_put(int cid)
971 return cid | MM_CID_LAZY_PUT;
974 static inline int mm_cid_clear_lazy_put(int cid)
976 return cid & ~MM_CID_LAZY_PUT;
979 /* Accessor for struct mm_struct's cidmask. */
980 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
982 unsigned long cid_bitmap = (unsigned long)mm;
984 cid_bitmap += offsetof(struct mm_struct, cpu_bitmap);
985 /* Skip cpu_bitmap */
986 cid_bitmap += cpumask_size();
987 return (struct cpumask *)cid_bitmap;
990 static inline void mm_init_cid(struct mm_struct *mm)
994 for_each_possible_cpu(i) {
995 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
997 pcpu_cid->cid = MM_CID_UNSET;
1000 cpumask_clear(mm_cidmask(mm));
1003 static inline int mm_alloc_cid(struct mm_struct *mm)
1005 mm->pcpu_cid = alloc_percpu(struct mm_cid);
1012 static inline void mm_destroy_cid(struct mm_struct *mm)
1014 free_percpu(mm->pcpu_cid);
1015 mm->pcpu_cid = NULL;
1018 static inline unsigned int mm_cid_size(void)
1020 return cpumask_size();
1022 #else /* CONFIG_SCHED_MM_CID */
1023 static inline void mm_init_cid(struct mm_struct *mm) { }
1024 static inline int mm_alloc_cid(struct mm_struct *mm) { return 0; }
1025 static inline void mm_destroy_cid(struct mm_struct *mm) { }
1026 static inline unsigned int mm_cid_size(void)
1030 #endif /* CONFIG_SCHED_MM_CID */
1033 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1034 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1035 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1040 * typedef vm_fault_t - Return type for page fault handlers.
1042 * Page fault handlers return a bitmask of %VM_FAULT values.
1044 typedef __bitwise unsigned int vm_fault_t;
1047 * enum vm_fault_reason - Page fault handlers return a bitmask of
1048 * these values to tell the core VM what happened when handling the
1049 * fault. Used to decide whether a process gets delivered SIGBUS or
1050 * just gets major/minor fault counters bumped up.
1052 * @VM_FAULT_OOM: Out Of Memory
1053 * @VM_FAULT_SIGBUS: Bad access
1054 * @VM_FAULT_MAJOR: Page read from storage
1055 * @VM_FAULT_HWPOISON: Hit poisoned small page
1056 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
1058 * @VM_FAULT_SIGSEGV: segmentation fault
1059 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
1060 * @VM_FAULT_LOCKED: ->fault locked the returned page
1061 * @VM_FAULT_RETRY: ->fault blocked, must retry
1062 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
1063 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
1064 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
1065 * fsync() to complete (for synchronous page faults
1067 * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released
1068 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
1071 enum vm_fault_reason {
1072 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
1073 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
1074 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
1075 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
1076 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1077 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
1078 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
1079 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
1080 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
1081 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
1082 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
1083 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
1084 VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000,
1085 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
1088 /* Encode hstate index for a hwpoisoned large page */
1089 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1090 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1092 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
1093 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
1094 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1096 #define VM_FAULT_RESULT_TRACE \
1097 { VM_FAULT_OOM, "OOM" }, \
1098 { VM_FAULT_SIGBUS, "SIGBUS" }, \
1099 { VM_FAULT_MAJOR, "MAJOR" }, \
1100 { VM_FAULT_HWPOISON, "HWPOISON" }, \
1101 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
1102 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
1103 { VM_FAULT_NOPAGE, "NOPAGE" }, \
1104 { VM_FAULT_LOCKED, "LOCKED" }, \
1105 { VM_FAULT_RETRY, "RETRY" }, \
1106 { VM_FAULT_FALLBACK, "FALLBACK" }, \
1107 { VM_FAULT_DONE_COW, "DONE_COW" }, \
1108 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }
1110 struct vm_special_mapping {
1111 const char *name; /* The name, e.g. "[vdso]". */
1114 * If .fault is not provided, this points to a
1115 * NULL-terminated array of pages that back the special mapping.
1117 * This must not be NULL unless .fault is provided.
1119 struct page **pages;
1122 * If non-NULL, then this is called to resolve page faults
1123 * on the special mapping. If used, .pages is not checked.
1125 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1126 struct vm_area_struct *vma,
1127 struct vm_fault *vmf);
1129 int (*mremap)(const struct vm_special_mapping *sm,
1130 struct vm_area_struct *new_vma);
1133 enum tlb_flush_reason {
1134 TLB_FLUSH_ON_TASK_SWITCH,
1135 TLB_REMOTE_SHOOTDOWN,
1136 TLB_LOCAL_SHOOTDOWN,
1137 TLB_LOCAL_MM_SHOOTDOWN,
1138 TLB_REMOTE_SEND_IPI,
1139 NR_TLB_FLUSH_REASONS,
1143 * A swap entry has to fit into a "unsigned long", as the entry is hidden
1144 * in the "index" field of the swapper address space.
1151 * enum fault_flag - Fault flag definitions.
1152 * @FAULT_FLAG_WRITE: Fault was a write fault.
1153 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1154 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1155 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1156 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1157 * @FAULT_FLAG_TRIED: The fault has been tried once.
1158 * @FAULT_FLAG_USER: The fault originated in userspace.
1159 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1160 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1161 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1162 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1163 * COW mapping, making sure that an exclusive anon page is
1164 * mapped after the fault.
1165 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1166 * We should only access orig_pte if this flag set.
1167 * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1169 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1170 * whether we would allow page faults to retry by specifying these two
1171 * fault flags correctly. Currently there can be three legal combinations:
1173 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and
1174 * this is the first try
1176 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and
1177 * we've already tried at least once
1179 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1181 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1182 * be used. Note that page faults can be allowed to retry for multiple times,
1183 * in which case we'll have an initial fault with flags (a) then later on
1184 * continuous faults with flags (b). We should always try to detect pending
1185 * signals before a retry to make sure the continuous page faults can still be
1186 * interrupted if necessary.
1188 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1189 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1190 * applied to mappings that are not COW mappings.
1193 FAULT_FLAG_WRITE = 1 << 0,
1194 FAULT_FLAG_MKWRITE = 1 << 1,
1195 FAULT_FLAG_ALLOW_RETRY = 1 << 2,
1196 FAULT_FLAG_RETRY_NOWAIT = 1 << 3,
1197 FAULT_FLAG_KILLABLE = 1 << 4,
1198 FAULT_FLAG_TRIED = 1 << 5,
1199 FAULT_FLAG_USER = 1 << 6,
1200 FAULT_FLAG_REMOTE = 1 << 7,
1201 FAULT_FLAG_INSTRUCTION = 1 << 8,
1202 FAULT_FLAG_INTERRUPTIBLE = 1 << 9,
1203 FAULT_FLAG_UNSHARE = 1 << 10,
1204 FAULT_FLAG_ORIG_PTE_VALID = 1 << 11,
1205 FAULT_FLAG_VMA_LOCK = 1 << 12,
1208 typedef unsigned int __bitwise zap_flags_t;
1211 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1212 * other. Here is what they mean, and how to use them:
1215 * FIXME: For pages which are part of a filesystem, mappings are subject to the
1216 * lifetime enforced by the filesystem and we need guarantees that longterm
1217 * users like RDMA and V4L2 only establish mappings which coordinate usage with
1218 * the filesystem. Ideas for this coordination include revoking the longterm
1219 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was
1220 * added after the problem with filesystems was found FS DAX VMAs are
1221 * specifically failed. Filesystem pages are still subject to bugs and use of
1222 * FOLL_LONGTERM should be avoided on those pages.
1224 * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1225 * that region. And so, CMA attempts to migrate the page before pinning, when
1226 * FOLL_LONGTERM is specified.
1228 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1229 * but an additional pin counting system) will be invoked. This is intended for
1230 * anything that gets a page reference and then touches page data (for example,
1231 * Direct IO). This lets the filesystem know that some non-file-system entity is
1232 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1233 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1234 * a call to unpin_user_page().
1236 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1237 * and separate refcounting mechanisms, however, and that means that each has
1238 * its own acquire and release mechanisms:
1240 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1242 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1244 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1245 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1246 * calls applied to them, and that's perfectly OK. This is a constraint on the
1247 * callers, not on the pages.)
1249 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1250 * directly by the caller. That's in order to help avoid mismatches when
1251 * releasing pages: get_user_pages*() pages must be released via put_page(),
1252 * while pin_user_pages*() pages must be released via unpin_user_page().
1254 * Please see Documentation/core-api/pin_user_pages.rst for more information.
1258 /* check pte is writable */
1259 FOLL_WRITE = 1 << 0,
1260 /* do get_page on page */
1262 /* give error on hole if it would be zero */
1264 /* get_user_pages read/write w/o permission */
1265 FOLL_FORCE = 1 << 3,
1267 * if a disk transfer is needed, start the IO and return without waiting
1270 FOLL_NOWAIT = 1 << 4,
1271 /* do not fault in pages */
1272 FOLL_NOFAULT = 1 << 5,
1273 /* check page is hwpoisoned */
1274 FOLL_HWPOISON = 1 << 6,
1275 /* don't do file mappings */
1278 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1279 * time period _often_ under userspace control. This is in contrast to
1280 * iov_iter_get_pages(), whose usages are transient.
1282 FOLL_LONGTERM = 1 << 8,
1283 /* split huge pmd before returning */
1284 FOLL_SPLIT_PMD = 1 << 9,
1285 /* allow returning PCI P2PDMA pages */
1286 FOLL_PCI_P2PDMA = 1 << 10,
1287 /* allow interrupts from generic signals */
1288 FOLL_INTERRUPTIBLE = 1 << 11,
1290 /* See also internal only FOLL flags in mm/internal.h */
1293 #endif /* _LINUX_MM_TYPES_H */