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;
128 atomic_long_t pp_frag_count;
130 struct { /* Tail pages of compound page */
131 unsigned long compound_head; /* Bit zero is set */
133 struct { /* ZONE_DEVICE pages */
134 /** @pgmap: Points to the hosting device page map. */
135 struct dev_pagemap *pgmap;
136 void *zone_device_data;
138 * ZONE_DEVICE private pages are counted as being
139 * mapped so the next 3 words hold the mapping, index,
140 * and private fields from the source anonymous or
141 * page cache page while the page is migrated to device
143 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
144 * use the mapping, index, and private fields when
145 * pmem backed DAX files are mapped.
149 /** @rcu_head: You can use this to free a page by RCU. */
150 struct rcu_head rcu_head;
153 union { /* This union is 4 bytes in size. */
155 * If the page can be mapped to userspace, encodes the number
156 * of times this page is referenced by a page table.
161 * If the page is neither PageSlab nor mappable to userspace,
162 * the value stored here may help determine what this page
163 * is used for. See page-flags.h for a list of page types
164 * which are currently stored here.
166 unsigned int page_type;
169 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
173 unsigned long memcg_data;
177 * On machines where all RAM is mapped into kernel address space,
178 * we can simply calculate the virtual address. On machines with
179 * highmem some memory is mapped into kernel virtual memory
180 * dynamically, so we need a place to store that address.
181 * Note that this field could be 16 bits on x86 ... ;)
183 * Architectures with slow multiplication can define
184 * WANT_PAGE_VIRTUAL in asm/page.h
186 #if defined(WANT_PAGE_VIRTUAL)
187 void *virtual; /* Kernel virtual address (NULL if
188 not kmapped, ie. highmem) */
189 #endif /* WANT_PAGE_VIRTUAL */
193 * KMSAN metadata for this page:
194 * - shadow page: every bit indicates whether the corresponding
195 * bit of the original page is initialized (0) or not (1);
196 * - origin page: every 4 bytes contain an id of the stack trace
197 * where the uninitialized value was created.
199 struct page *kmsan_shadow;
200 struct page *kmsan_origin;
203 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
206 } _struct_page_alignment;
209 * struct encoded_page - a nonexistent type marking this pointer
211 * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
212 * with the low bits of the pointer indicating extra context-dependent
213 * information. Not super-common, but happens in mmu_gather and mlock
214 * handling, and this acts as a type system check on that use.
216 * We only really have two guaranteed bits in general, although you could
217 * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
220 * Use the supplied helper functions to endcode/decode the pointer and bits.
223 #define ENCODE_PAGE_BITS 3ul
224 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
226 BUILD_BUG_ON(flags > ENCODE_PAGE_BITS);
227 return (struct encoded_page *)(flags | (unsigned long)page);
230 static inline unsigned long encoded_page_flags(struct encoded_page *page)
232 return ENCODE_PAGE_BITS & (unsigned long)page;
235 static inline struct page *encoded_page_ptr(struct encoded_page *page)
237 return (struct page *)(~ENCODE_PAGE_BITS & (unsigned long)page);
241 * A swap entry has to fit into a "unsigned long", as the entry is hidden
242 * in the "index" field of the swapper address space.
249 * struct folio - Represents a contiguous set of bytes.
250 * @flags: Identical to the page flags.
251 * @lru: Least Recently Used list; tracks how recently this folio was used.
252 * @mlock_count: Number of times this folio has been pinned by mlock().
253 * @mapping: The file this page belongs to, or refers to the anon_vma for
255 * @index: Offset within the file, in units of pages. For anonymous memory,
256 * this is the index from the beginning of the mmap.
257 * @private: Filesystem per-folio data (see folio_attach_private()).
258 * @swap: Used for swp_entry_t if folio_test_swapcache().
259 * @_mapcount: Do not access this member directly. Use folio_mapcount() to
260 * find out how many times this folio is mapped by userspace.
261 * @_refcount: Do not access this member directly. Use folio_ref_count()
262 * to find how many references there are to this folio.
263 * @memcg_data: Memory Control Group data.
264 * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
265 * @_nr_pages_mapped: Do not use directly, call folio_mapcount().
266 * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
267 * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
268 * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
269 * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
270 * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
271 * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
272 * @_deferred_list: Folios to be split under memory pressure.
274 * A folio is a physically, virtually and logically contiguous set
275 * of bytes. It is a power-of-two in size, and it is aligned to that
276 * same power-of-two. It is at least as large as %PAGE_SIZE. If it is
277 * in the page cache, it is at a file offset which is a multiple of that
278 * power-of-two. It may be mapped into userspace at an address which is
279 * at an arbitrary page offset, but its kernel virtual address is aligned
283 /* private: don't document the anon union */
289 struct list_head lru;
290 /* private: avoid cluttering the output */
294 unsigned int mlock_count;
299 struct address_space *mapping;
308 unsigned long memcg_data;
310 /* private: the union with struct page is transitional */
316 unsigned long _flags_1;
317 unsigned long _head_1;
318 unsigned long _folio_avail;
320 atomic_t _entire_mapcount;
321 atomic_t _nr_pages_mapped;
324 unsigned int _folio_nr_pages;
326 /* private: the union with struct page is transitional */
328 struct page __page_1;
332 unsigned long _flags_2;
333 unsigned long _head_2;
335 void *_hugetlb_subpool;
336 void *_hugetlb_cgroup;
337 void *_hugetlb_cgroup_rsvd;
338 void *_hugetlb_hwpoison;
339 /* private: the union with struct page is transitional */
342 unsigned long _flags_2a;
343 unsigned long _head_2a;
345 struct list_head _deferred_list;
346 /* private: the union with struct page is transitional */
348 struct page __page_2;
352 #define FOLIO_MATCH(pg, fl) \
353 static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
354 FOLIO_MATCH(flags, flags);
355 FOLIO_MATCH(lru, lru);
356 FOLIO_MATCH(mapping, mapping);
357 FOLIO_MATCH(compound_head, lru);
358 FOLIO_MATCH(index, index);
359 FOLIO_MATCH(private, private);
360 FOLIO_MATCH(_mapcount, _mapcount);
361 FOLIO_MATCH(_refcount, _refcount);
363 FOLIO_MATCH(memcg_data, memcg_data);
366 #define FOLIO_MATCH(pg, fl) \
367 static_assert(offsetof(struct folio, fl) == \
368 offsetof(struct page, pg) + sizeof(struct page))
369 FOLIO_MATCH(flags, _flags_1);
370 FOLIO_MATCH(compound_head, _head_1);
372 #define FOLIO_MATCH(pg, fl) \
373 static_assert(offsetof(struct folio, fl) == \
374 offsetof(struct page, pg) + 2 * sizeof(struct page))
375 FOLIO_MATCH(flags, _flags_2);
376 FOLIO_MATCH(compound_head, _head_2);
377 FOLIO_MATCH(flags, _flags_2a);
378 FOLIO_MATCH(compound_head, _head_2a);
382 * struct ptdesc - Memory descriptor for page tables.
383 * @__page_flags: Same as page flags. Unused for page tables.
384 * @pt_rcu_head: For freeing page table pages.
385 * @pt_list: List of used page tables. Used for s390 and x86.
386 * @_pt_pad_1: Padding that aliases with page's compound head.
387 * @pmd_huge_pte: Protected by ptdesc->ptl, used for THPs.
388 * @__page_mapping: Aliases with page->mapping. Unused for page tables.
389 * @pt_mm: Used for x86 pgds.
390 * @pt_frag_refcount: For fragmented page table tracking. Powerpc and s390 only.
391 * @_pt_pad_2: Padding to ensure proper alignment.
392 * @ptl: Lock for the page table.
393 * @__page_type: Same as page->page_type. Unused for page tables.
394 * @_refcount: Same as page refcount. Used for s390 page tables.
395 * @pt_memcg_data: Memcg data. Tracked for page tables here.
397 * This struct overlays struct page for now. Do not modify without a good
398 * understanding of the issues.
401 unsigned long __page_flags;
404 struct rcu_head pt_rcu_head;
405 struct list_head pt_list;
407 unsigned long _pt_pad_1;
408 pgtable_t pmd_huge_pte;
411 unsigned long __page_mapping;
414 struct mm_struct *pt_mm;
415 atomic_t pt_frag_refcount;
419 unsigned long _pt_pad_2;
420 #if ALLOC_SPLIT_PTLOCKS
426 unsigned int __page_type;
429 unsigned long pt_memcg_data;
433 #define TABLE_MATCH(pg, pt) \
434 static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
435 TABLE_MATCH(flags, __page_flags);
436 TABLE_MATCH(compound_head, pt_list);
437 TABLE_MATCH(compound_head, _pt_pad_1);
438 TABLE_MATCH(mapping, __page_mapping);
439 TABLE_MATCH(rcu_head, pt_rcu_head);
440 TABLE_MATCH(page_type, __page_type);
441 TABLE_MATCH(_refcount, _refcount);
443 TABLE_MATCH(memcg_data, pt_memcg_data);
446 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
448 #define ptdesc_page(pt) (_Generic((pt), \
449 const struct ptdesc *: (const struct page *)(pt), \
450 struct ptdesc *: (struct page *)(pt)))
452 #define ptdesc_folio(pt) (_Generic((pt), \
453 const struct ptdesc *: (const struct folio *)(pt), \
454 struct ptdesc *: (struct folio *)(pt)))
456 #define page_ptdesc(p) (_Generic((p), \
457 const struct page *: (const struct ptdesc *)(p), \
458 struct page *: (struct ptdesc *)(p)))
461 * Used for sizing the vmemmap region on some architectures
463 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
465 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
466 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
469 * page_private can be used on tail pages. However, PagePrivate is only
470 * checked by the VM on the head page. So page_private on the tail pages
471 * should be used for data that's ancillary to the head page (eg attaching
472 * buffer heads to tail pages after attaching buffer heads to the head page)
474 #define page_private(page) ((page)->private)
476 static inline void set_page_private(struct page *page, unsigned long private)
478 page->private = private;
481 static inline void *folio_get_private(struct folio *folio)
483 return folio->private;
486 struct page_frag_cache {
488 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
494 /* we maintain a pagecount bias, so that we dont dirty cache line
495 * containing page->_refcount every time we allocate a fragment.
497 unsigned int pagecnt_bias;
501 typedef unsigned long vm_flags_t;
504 * A region containing a mapping of a non-memory backed file under NOMMU
505 * conditions. These are held in a global tree and are pinned by the VMAs that
509 struct rb_node vm_rb; /* link in global region tree */
510 vm_flags_t vm_flags; /* VMA vm_flags */
511 unsigned long vm_start; /* start address of region */
512 unsigned long vm_end; /* region initialised to here */
513 unsigned long vm_top; /* region allocated to here */
514 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
515 struct file *vm_file; /* the backing file or NULL */
517 int vm_usage; /* region usage count (access under nommu_region_sem) */
518 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
522 #ifdef CONFIG_USERFAULTFD
523 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
524 struct vm_userfaultfd_ctx {
525 struct userfaultfd_ctx *ctx;
527 #else /* CONFIG_USERFAULTFD */
528 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
529 struct vm_userfaultfd_ctx {};
530 #endif /* CONFIG_USERFAULTFD */
532 struct anon_vma_name {
534 /* The name needs to be at the end because it is dynamically sized. */
539 struct rw_semaphore lock;
542 struct vma_numab_state {
543 unsigned long next_scan;
544 unsigned long next_pid_reset;
545 unsigned long access_pids[2];
549 * This struct describes a virtual memory area. There is one of these
550 * per VM-area/task. A VM area is any part of the process virtual memory
551 * space that has a special rule for the page-fault handlers (ie a shared
552 * library, the executable area etc).
554 struct vm_area_struct {
555 /* The first cache line has the info for VMA tree walking. */
559 /* VMA covers [vm_start; vm_end) addresses within mm */
560 unsigned long vm_start;
561 unsigned long vm_end;
563 #ifdef CONFIG_PER_VMA_LOCK
564 struct rcu_head vm_rcu; /* Used for deferred freeing. */
568 struct mm_struct *vm_mm; /* The address space we belong to. */
569 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
573 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
576 const vm_flags_t vm_flags;
577 vm_flags_t __private __vm_flags;
580 #ifdef CONFIG_PER_VMA_LOCK
582 * Can only be written (using WRITE_ONCE()) while holding both:
583 * - mmap_lock (in write mode)
584 * - vm_lock->lock (in write mode)
585 * Can be read reliably while holding one of:
586 * - mmap_lock (in read or write mode)
587 * - vm_lock->lock (in read or write mode)
588 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
589 * while holding nothing (except RCU to keep the VMA struct allocated).
591 * This sequence counter is explicitly allowed to overflow; sequence
592 * counter reuse can only lead to occasional unnecessary use of the
596 struct vma_lock *vm_lock;
598 /* Flag to indicate areas detached from the mm->mm_mt tree */
603 * For areas with an address space and backing store,
604 * linkage into the address_space->i_mmap interval tree.
609 unsigned long rb_subtree_last;
613 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
614 * list, after a COW of one of the file pages. A MAP_SHARED vma
615 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
616 * or brk vma (with NULL file) can only be in an anon_vma list.
618 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
620 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
622 /* Function pointers to deal with this struct. */
623 const struct vm_operations_struct *vm_ops;
625 /* Information about our backing store: */
626 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
628 struct file * vm_file; /* File we map to (can be NULL). */
629 void * vm_private_data; /* was vm_pte (shared mem) */
631 #ifdef CONFIG_ANON_VMA_NAME
633 * For private and shared anonymous mappings, a pointer to a null
634 * terminated string containing the name given to the vma, or NULL if
635 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
637 struct anon_vma_name *anon_name;
640 atomic_long_t swap_readahead_info;
643 struct vm_region *vm_region; /* NOMMU mapping region */
646 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
648 #ifdef CONFIG_NUMA_BALANCING
649 struct vma_numab_state *numab_state; /* NUMA Balancing state */
651 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
652 } __randomize_layout;
654 #ifdef CONFIG_SCHED_MM_CID
665 * Fields which are often written to are placed in a separate
670 * @mm_count: The number of references to &struct
671 * mm_struct (@mm_users count as 1).
673 * Use mmgrab()/mmdrop() to modify. When this drops to
674 * 0, the &struct mm_struct is freed.
677 } ____cacheline_aligned_in_smp;
679 struct maple_tree mm_mt;
681 unsigned long (*get_unmapped_area) (struct file *filp,
682 unsigned long addr, unsigned long len,
683 unsigned long pgoff, unsigned long flags);
685 unsigned long mmap_base; /* base of mmap area */
686 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
687 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
688 /* Base addresses for compatible mmap() */
689 unsigned long mmap_compat_base;
690 unsigned long mmap_compat_legacy_base;
692 unsigned long task_size; /* size of task vm space */
695 #ifdef CONFIG_MEMBARRIER
697 * @membarrier_state: Flags controlling membarrier behavior.
699 * This field is close to @pgd to hopefully fit in the same
700 * cache-line, which needs to be touched by switch_mm().
702 atomic_t membarrier_state;
706 * @mm_users: The number of users including userspace.
708 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
709 * drops to 0 (i.e. when the task exits and there are no other
710 * temporary reference holders), we also release a reference on
711 * @mm_count (which may then free the &struct mm_struct if
712 * @mm_count also drops to 0).
716 #ifdef CONFIG_SCHED_MM_CID
718 * @pcpu_cid: Per-cpu current cid.
720 * Keep track of the currently allocated mm_cid for each cpu.
721 * The per-cpu mm_cid values are serialized by their respective
724 struct mm_cid __percpu *pcpu_cid;
726 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
728 * When the next mm_cid scan is due (in jiffies).
730 unsigned long mm_cid_next_scan;
733 atomic_long_t pgtables_bytes; /* size of all page tables */
735 int map_count; /* number of VMAs */
737 spinlock_t page_table_lock; /* Protects page tables and some
741 * With some kernel config, the current mmap_lock's offset
742 * inside 'mm_struct' is at 0x120, which is very optimal, as
743 * its two hot fields 'count' and 'owner' sit in 2 different
744 * cachelines, and when mmap_lock is highly contended, both
745 * of the 2 fields will be accessed frequently, current layout
746 * will help to reduce cache bouncing.
748 * So please be careful with adding new fields before
749 * mmap_lock, which can easily push the 2 fields into one
752 struct rw_semaphore mmap_lock;
754 struct list_head mmlist; /* List of maybe swapped mm's. These
755 * are globally strung together off
756 * init_mm.mmlist, and are protected
759 #ifdef CONFIG_PER_VMA_LOCK
761 * This field has lock-like semantics, meaning it is sometimes
762 * accessed with ACQUIRE/RELEASE semantics.
763 * Roughly speaking, incrementing the sequence number is
764 * equivalent to releasing locks on VMAs; reading the sequence
765 * number can be part of taking a read lock on a VMA.
767 * Can be modified under write mmap_lock using RELEASE
769 * Can be read with no other protection when holding write
771 * Can be read with ACQUIRE semantics if not holding write
778 unsigned long hiwater_rss; /* High-watermark of RSS usage */
779 unsigned long hiwater_vm; /* High-water virtual memory usage */
781 unsigned long total_vm; /* Total pages mapped */
782 unsigned long locked_vm; /* Pages that have PG_mlocked set */
783 atomic64_t pinned_vm; /* Refcount permanently increased */
784 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
785 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
786 unsigned long stack_vm; /* VM_STACK */
787 unsigned long def_flags;
790 * @write_protect_seq: Locked when any thread is write
791 * protecting pages mapped by this mm to enforce a later COW,
792 * for instance during page table copying for fork().
794 seqcount_t write_protect_seq;
796 spinlock_t arg_lock; /* protect the below fields */
798 unsigned long start_code, end_code, start_data, end_data;
799 unsigned long start_brk, brk, start_stack;
800 unsigned long arg_start, arg_end, env_start, env_end;
802 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
804 struct percpu_counter rss_stat[NR_MM_COUNTERS];
806 struct linux_binfmt *binfmt;
808 /* Architecture-specific MM context */
809 mm_context_t context;
811 unsigned long flags; /* Must use atomic bitops to access */
814 spinlock_t ioctx_lock;
815 struct kioctx_table __rcu *ioctx_table;
819 * "owner" points to a task that is regarded as the canonical
820 * user/owner of this mm. All of the following must be true in
821 * order for it to be changed:
823 * current == mm->owner
825 * new_owner->mm == mm
826 * new_owner->alloc_lock is held
828 struct task_struct __rcu *owner;
830 struct user_namespace *user_ns;
832 /* store ref to file /proc/<pid>/exe symlink points to */
833 struct file __rcu *exe_file;
834 #ifdef CONFIG_MMU_NOTIFIER
835 struct mmu_notifier_subscriptions *notifier_subscriptions;
837 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
838 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
840 #ifdef CONFIG_NUMA_BALANCING
842 * numa_next_scan is the next time that PTEs will be remapped
843 * PROT_NONE to trigger NUMA hinting faults; such faults gather
844 * statistics and migrate pages to new nodes if necessary.
846 unsigned long numa_next_scan;
848 /* Restart point for scanning and remapping PTEs. */
849 unsigned long numa_scan_offset;
851 /* numa_scan_seq prevents two threads remapping PTEs. */
855 * An operation with batched TLB flushing is going on. Anything
856 * that can move process memory needs to flush the TLB when
857 * moving a PROT_NONE mapped page.
859 atomic_t tlb_flush_pending;
860 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
861 /* See flush_tlb_batched_pending() */
862 atomic_t tlb_flush_batched;
864 struct uprobes_state uprobes_state;
865 #ifdef CONFIG_PREEMPT_RT
866 struct rcu_head delayed_drop;
868 #ifdef CONFIG_HUGETLB_PAGE
869 atomic_long_t hugetlb_usage;
871 struct work_struct async_put_work;
873 #ifdef CONFIG_IOMMU_SVA
878 * Represent how many pages of this process are involved in KSM
879 * merging (not including ksm_zero_pages).
881 unsigned long ksm_merging_pages;
883 * Represent how many pages are checked for ksm merging
884 * including merged and not merged.
886 unsigned long ksm_rmap_items;
888 * Represent how many empty pages are merged with kernel zero
889 * pages when enabling KSM use_zero_pages.
891 unsigned long ksm_zero_pages;
892 #endif /* CONFIG_KSM */
893 #ifdef CONFIG_LRU_GEN
895 /* this mm_struct is on lru_gen_mm_list */
896 struct list_head list;
898 * Set when switching to this mm_struct, as a hint of
899 * whether it has been used since the last time per-node
900 * page table walkers cleared the corresponding bits.
902 unsigned long bitmap;
904 /* points to the memcg of "owner" above */
905 struct mem_cgroup *memcg;
908 #endif /* CONFIG_LRU_GEN */
909 } __randomize_layout;
912 * The mm_cpumask needs to be at the end of mm_struct, because it
913 * is dynamically sized based on nr_cpu_ids.
915 unsigned long cpu_bitmap[];
918 #define MM_MT_FLAGS (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
920 extern struct mm_struct init_mm;
922 /* Pointer magic because the dynamic array size confuses some compilers. */
923 static inline void mm_init_cpumask(struct mm_struct *mm)
925 unsigned long cpu_bitmap = (unsigned long)mm;
927 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
928 cpumask_clear((struct cpumask *)cpu_bitmap);
931 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
932 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
934 return (struct cpumask *)&mm->cpu_bitmap;
937 #ifdef CONFIG_LRU_GEN
939 struct lru_gen_mm_list {
940 /* mm_struct list for page table walkers */
941 struct list_head fifo;
942 /* protects the list above */
946 void lru_gen_add_mm(struct mm_struct *mm);
947 void lru_gen_del_mm(struct mm_struct *mm);
949 void lru_gen_migrate_mm(struct mm_struct *mm);
952 static inline void lru_gen_init_mm(struct mm_struct *mm)
954 INIT_LIST_HEAD(&mm->lru_gen.list);
955 mm->lru_gen.bitmap = 0;
957 mm->lru_gen.memcg = NULL;
961 static inline void lru_gen_use_mm(struct mm_struct *mm)
964 * When the bitmap is set, page reclaim knows this mm_struct has been
965 * used since the last time it cleared the bitmap. So it might be worth
966 * walking the page tables of this mm_struct to clear the accessed bit.
968 WRITE_ONCE(mm->lru_gen.bitmap, -1);
971 #else /* !CONFIG_LRU_GEN */
973 static inline void lru_gen_add_mm(struct mm_struct *mm)
977 static inline void lru_gen_del_mm(struct mm_struct *mm)
982 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
987 static inline void lru_gen_init_mm(struct mm_struct *mm)
991 static inline void lru_gen_use_mm(struct mm_struct *mm)
995 #endif /* CONFIG_LRU_GEN */
997 struct vma_iterator {
1001 #define VMA_ITERATOR(name, __mm, __addr) \
1002 struct vma_iterator name = { \
1004 .tree = &(__mm)->mm_mt, \
1006 .node = MAS_START, \
1010 static inline void vma_iter_init(struct vma_iterator *vmi,
1011 struct mm_struct *mm, unsigned long addr)
1013 mas_init(&vmi->mas, &mm->mm_mt, addr);
1016 #ifdef CONFIG_SCHED_MM_CID
1019 MM_CID_UNSET = -1U, /* Unset state has lazy_put flag set. */
1020 MM_CID_LAZY_PUT = (1U << 31),
1023 static inline bool mm_cid_is_unset(int cid)
1025 return cid == MM_CID_UNSET;
1028 static inline bool mm_cid_is_lazy_put(int cid)
1030 return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1033 static inline bool mm_cid_is_valid(int cid)
1035 return !(cid & MM_CID_LAZY_PUT);
1038 static inline int mm_cid_set_lazy_put(int cid)
1040 return cid | MM_CID_LAZY_PUT;
1043 static inline int mm_cid_clear_lazy_put(int cid)
1045 return cid & ~MM_CID_LAZY_PUT;
1048 /* Accessor for struct mm_struct's cidmask. */
1049 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1051 unsigned long cid_bitmap = (unsigned long)mm;
1053 cid_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1054 /* Skip cpu_bitmap */
1055 cid_bitmap += cpumask_size();
1056 return (struct cpumask *)cid_bitmap;
1059 static inline void mm_init_cid(struct mm_struct *mm)
1063 for_each_possible_cpu(i) {
1064 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1066 pcpu_cid->cid = MM_CID_UNSET;
1069 cpumask_clear(mm_cidmask(mm));
1072 static inline int mm_alloc_cid(struct mm_struct *mm)
1074 mm->pcpu_cid = alloc_percpu(struct mm_cid);
1081 static inline void mm_destroy_cid(struct mm_struct *mm)
1083 free_percpu(mm->pcpu_cid);
1084 mm->pcpu_cid = NULL;
1087 static inline unsigned int mm_cid_size(void)
1089 return cpumask_size();
1091 #else /* CONFIG_SCHED_MM_CID */
1092 static inline void mm_init_cid(struct mm_struct *mm) { }
1093 static inline int mm_alloc_cid(struct mm_struct *mm) { return 0; }
1094 static inline void mm_destroy_cid(struct mm_struct *mm) { }
1095 static inline unsigned int mm_cid_size(void)
1099 #endif /* CONFIG_SCHED_MM_CID */
1102 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1103 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1104 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1109 * typedef vm_fault_t - Return type for page fault handlers.
1111 * Page fault handlers return a bitmask of %VM_FAULT values.
1113 typedef __bitwise unsigned int vm_fault_t;
1116 * enum vm_fault_reason - Page fault handlers return a bitmask of
1117 * these values to tell the core VM what happened when handling the
1118 * fault. Used to decide whether a process gets delivered SIGBUS or
1119 * just gets major/minor fault counters bumped up.
1121 * @VM_FAULT_OOM: Out Of Memory
1122 * @VM_FAULT_SIGBUS: Bad access
1123 * @VM_FAULT_MAJOR: Page read from storage
1124 * @VM_FAULT_HWPOISON: Hit poisoned small page
1125 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
1127 * @VM_FAULT_SIGSEGV: segmentation fault
1128 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
1129 * @VM_FAULT_LOCKED: ->fault locked the returned page
1130 * @VM_FAULT_RETRY: ->fault blocked, must retry
1131 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
1132 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
1133 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
1134 * fsync() to complete (for synchronous page faults
1136 * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released
1137 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
1140 enum vm_fault_reason {
1141 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
1142 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
1143 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
1144 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
1145 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1146 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
1147 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
1148 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
1149 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
1150 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
1151 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
1152 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
1153 VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000,
1154 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
1157 /* Encode hstate index for a hwpoisoned large page */
1158 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1159 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1161 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
1162 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
1163 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1165 #define VM_FAULT_RESULT_TRACE \
1166 { VM_FAULT_OOM, "OOM" }, \
1167 { VM_FAULT_SIGBUS, "SIGBUS" }, \
1168 { VM_FAULT_MAJOR, "MAJOR" }, \
1169 { VM_FAULT_HWPOISON, "HWPOISON" }, \
1170 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
1171 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
1172 { VM_FAULT_NOPAGE, "NOPAGE" }, \
1173 { VM_FAULT_LOCKED, "LOCKED" }, \
1174 { VM_FAULT_RETRY, "RETRY" }, \
1175 { VM_FAULT_FALLBACK, "FALLBACK" }, \
1176 { VM_FAULT_DONE_COW, "DONE_COW" }, \
1177 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }, \
1178 { VM_FAULT_COMPLETED, "COMPLETED" }
1180 struct vm_special_mapping {
1181 const char *name; /* The name, e.g. "[vdso]". */
1184 * If .fault is not provided, this points to a
1185 * NULL-terminated array of pages that back the special mapping.
1187 * This must not be NULL unless .fault is provided.
1189 struct page **pages;
1192 * If non-NULL, then this is called to resolve page faults
1193 * on the special mapping. If used, .pages is not checked.
1195 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1196 struct vm_area_struct *vma,
1197 struct vm_fault *vmf);
1199 int (*mremap)(const struct vm_special_mapping *sm,
1200 struct vm_area_struct *new_vma);
1203 enum tlb_flush_reason {
1204 TLB_FLUSH_ON_TASK_SWITCH,
1205 TLB_REMOTE_SHOOTDOWN,
1206 TLB_LOCAL_SHOOTDOWN,
1207 TLB_LOCAL_MM_SHOOTDOWN,
1208 TLB_REMOTE_SEND_IPI,
1209 NR_TLB_FLUSH_REASONS,
1213 * enum fault_flag - Fault flag definitions.
1214 * @FAULT_FLAG_WRITE: Fault was a write fault.
1215 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1216 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1217 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1218 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1219 * @FAULT_FLAG_TRIED: The fault has been tried once.
1220 * @FAULT_FLAG_USER: The fault originated in userspace.
1221 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1222 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1223 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1224 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1225 * COW mapping, making sure that an exclusive anon page is
1226 * mapped after the fault.
1227 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1228 * We should only access orig_pte if this flag set.
1229 * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1231 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1232 * whether we would allow page faults to retry by specifying these two
1233 * fault flags correctly. Currently there can be three legal combinations:
1235 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and
1236 * this is the first try
1238 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and
1239 * we've already tried at least once
1241 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1243 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1244 * be used. Note that page faults can be allowed to retry for multiple times,
1245 * in which case we'll have an initial fault with flags (a) then later on
1246 * continuous faults with flags (b). We should always try to detect pending
1247 * signals before a retry to make sure the continuous page faults can still be
1248 * interrupted if necessary.
1250 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1251 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1252 * applied to mappings that are not COW mappings.
1255 FAULT_FLAG_WRITE = 1 << 0,
1256 FAULT_FLAG_MKWRITE = 1 << 1,
1257 FAULT_FLAG_ALLOW_RETRY = 1 << 2,
1258 FAULT_FLAG_RETRY_NOWAIT = 1 << 3,
1259 FAULT_FLAG_KILLABLE = 1 << 4,
1260 FAULT_FLAG_TRIED = 1 << 5,
1261 FAULT_FLAG_USER = 1 << 6,
1262 FAULT_FLAG_REMOTE = 1 << 7,
1263 FAULT_FLAG_INSTRUCTION = 1 << 8,
1264 FAULT_FLAG_INTERRUPTIBLE = 1 << 9,
1265 FAULT_FLAG_UNSHARE = 1 << 10,
1266 FAULT_FLAG_ORIG_PTE_VALID = 1 << 11,
1267 FAULT_FLAG_VMA_LOCK = 1 << 12,
1270 typedef unsigned int __bitwise zap_flags_t;
1273 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1274 * other. Here is what they mean, and how to use them:
1277 * FIXME: For pages which are part of a filesystem, mappings are subject to the
1278 * lifetime enforced by the filesystem and we need guarantees that longterm
1279 * users like RDMA and V4L2 only establish mappings which coordinate usage with
1280 * the filesystem. Ideas for this coordination include revoking the longterm
1281 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was
1282 * added after the problem with filesystems was found FS DAX VMAs are
1283 * specifically failed. Filesystem pages are still subject to bugs and use of
1284 * FOLL_LONGTERM should be avoided on those pages.
1286 * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1287 * that region. And so, CMA attempts to migrate the page before pinning, when
1288 * FOLL_LONGTERM is specified.
1290 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1291 * but an additional pin counting system) will be invoked. This is intended for
1292 * anything that gets a page reference and then touches page data (for example,
1293 * Direct IO). This lets the filesystem know that some non-file-system entity is
1294 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1295 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1296 * a call to unpin_user_page().
1298 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1299 * and separate refcounting mechanisms, however, and that means that each has
1300 * its own acquire and release mechanisms:
1302 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1304 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1306 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1307 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1308 * calls applied to them, and that's perfectly OK. This is a constraint on the
1309 * callers, not on the pages.)
1311 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1312 * directly by the caller. That's in order to help avoid mismatches when
1313 * releasing pages: get_user_pages*() pages must be released via put_page(),
1314 * while pin_user_pages*() pages must be released via unpin_user_page().
1316 * Please see Documentation/core-api/pin_user_pages.rst for more information.
1320 /* check pte is writable */
1321 FOLL_WRITE = 1 << 0,
1322 /* do get_page on page */
1324 /* give error on hole if it would be zero */
1326 /* get_user_pages read/write w/o permission */
1327 FOLL_FORCE = 1 << 3,
1329 * if a disk transfer is needed, start the IO and return without waiting
1332 FOLL_NOWAIT = 1 << 4,
1333 /* do not fault in pages */
1334 FOLL_NOFAULT = 1 << 5,
1335 /* check page is hwpoisoned */
1336 FOLL_HWPOISON = 1 << 6,
1337 /* don't do file mappings */
1340 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1341 * time period _often_ under userspace control. This is in contrast to
1342 * iov_iter_get_pages(), whose usages are transient.
1344 FOLL_LONGTERM = 1 << 8,
1345 /* split huge pmd before returning */
1346 FOLL_SPLIT_PMD = 1 << 9,
1347 /* allow returning PCI P2PDMA pages */
1348 FOLL_PCI_P2PDMA = 1 << 10,
1349 /* allow interrupts from generic signals */
1350 FOLL_INTERRUPTIBLE = 1 << 11,
1352 * Always honor (trigger) NUMA hinting faults.
1354 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1355 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1356 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1359 FOLL_HONOR_NUMA_FAULT = 1 << 12,
1361 /* See also internal only FOLL flags in mm/internal.h */
1364 #endif /* _LINUX_MM_TYPES_H */