xref: /linux-6.15/include/linux/mm_types.h (revision ea8a163e)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
4 
5 #include <linux/mm_types_task.h>
6 
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/rcupdate.h>
16 #include <linux/page-flags-layout.h>
17 #include <linux/workqueue.h>
18 #include <linux/seqlock.h>
19 
20 #include <asm/mmu.h>
21 
22 #ifndef AT_VECTOR_SIZE_ARCH
23 #define AT_VECTOR_SIZE_ARCH 0
24 #endif
25 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
26 
27 #define INIT_PASID	0
28 
29 struct address_space;
30 struct mem_cgroup;
31 
32 /*
33  * Each physical page in the system has a struct page associated with
34  * it to keep track of whatever it is we are using the page for at the
35  * moment. Note that we have no way to track which tasks are using
36  * a page, though if it is a pagecache page, rmap structures can tell us
37  * who is mapping it.
38  *
39  * If you allocate the page using alloc_pages(), you can use some of the
40  * space in struct page for your own purposes.  The five words in the main
41  * union are available, except for bit 0 of the first word which must be
42  * kept clear.  Many users use this word to store a pointer to an object
43  * which is guaranteed to be aligned.  If you use the same storage as
44  * page->mapping, you must restore it to NULL before freeing the page.
45  *
46  * If your page will not be mapped to userspace, you can also use the four
47  * bytes in the mapcount union, but you must call page_mapcount_reset()
48  * before freeing it.
49  *
50  * If you want to use the refcount field, it must be used in such a way
51  * that other CPUs temporarily incrementing and then decrementing the
52  * refcount does not cause problems.  On receiving the page from
53  * alloc_pages(), the refcount will be positive.
54  *
55  * If you allocate pages of order > 0, you can use some of the fields
56  * in each subpage, but you may need to restore some of their values
57  * afterwards.
58  *
59  * SLUB uses cmpxchg_double() to atomically update its freelist and
60  * counters.  That requires that freelist & counters be adjacent and
61  * double-word aligned.  We align all struct pages to double-word
62  * boundaries, and ensure that 'freelist' is aligned within the
63  * struct.
64  */
65 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
66 #define _struct_page_alignment	__aligned(2 * sizeof(unsigned long))
67 #else
68 #define _struct_page_alignment
69 #endif
70 
71 struct page {
72 	unsigned long flags;		/* Atomic flags, some possibly
73 					 * updated asynchronously */
74 	/*
75 	 * Five words (20/40 bytes) are available in this union.
76 	 * WARNING: bit 0 of the first word is used for PageTail(). That
77 	 * means the other users of this union MUST NOT use the bit to
78 	 * avoid collision and false-positive PageTail().
79 	 */
80 	union {
81 		struct {	/* Page cache and anonymous pages */
82 			/**
83 			 * @lru: Pageout list, eg. active_list protected by
84 			 * lruvec->lru_lock.  Sometimes used as a generic list
85 			 * by the page owner.
86 			 */
87 			struct list_head lru;
88 			/* See page-flags.h for PAGE_MAPPING_FLAGS */
89 			struct address_space *mapping;
90 			pgoff_t index;		/* Our offset within mapping. */
91 			/**
92 			 * @private: Mapping-private opaque data.
93 			 * Usually used for buffer_heads if PagePrivate.
94 			 * Used for swp_entry_t if PageSwapCache.
95 			 * Indicates order in the buddy system if PageBuddy.
96 			 */
97 			unsigned long private;
98 		};
99 		struct {	/* page_pool used by netstack */
100 			/**
101 			 * @pp_magic: magic value to avoid recycling non
102 			 * page_pool allocated pages.
103 			 */
104 			unsigned long pp_magic;
105 			struct page_pool *pp;
106 			unsigned long _pp_mapping_pad;
107 			unsigned long dma_addr;
108 			atomic_long_t pp_frag_count;
109 		};
110 		struct {	/* slab, slob and slub */
111 			union {
112 				struct list_head slab_list;
113 				struct {	/* Partial pages */
114 					struct page *next;
115 #ifdef CONFIG_64BIT
116 					int pages;	/* Nr of pages left */
117 #else
118 					short int pages;
119 #endif
120 				};
121 			};
122 			struct kmem_cache *slab_cache; /* not slob */
123 			/* Double-word boundary */
124 			void *freelist;		/* first free object */
125 			union {
126 				void *s_mem;	/* slab: first object */
127 				unsigned long counters;		/* SLUB */
128 				struct {			/* SLUB */
129 					unsigned inuse:16;
130 					unsigned objects:15;
131 					unsigned frozen:1;
132 				};
133 			};
134 		};
135 		struct {	/* Tail pages of compound page */
136 			unsigned long compound_head;	/* Bit zero is set */
137 
138 			/* First tail page only */
139 			unsigned char compound_dtor;
140 			unsigned char compound_order;
141 			atomic_t compound_mapcount;
142 			unsigned int compound_nr; /* 1 << compound_order */
143 		};
144 		struct {	/* Second tail page of compound page */
145 			unsigned long _compound_pad_1;	/* compound_head */
146 			atomic_t hpage_pinned_refcount;
147 			/* For both global and memcg */
148 			struct list_head deferred_list;
149 		};
150 		struct {	/* Page table pages */
151 			unsigned long _pt_pad_1;	/* compound_head */
152 			pgtable_t pmd_huge_pte; /* protected by page->ptl */
153 			unsigned long _pt_pad_2;	/* mapping */
154 			union {
155 				struct mm_struct *pt_mm; /* x86 pgds only */
156 				atomic_t pt_frag_refcount; /* powerpc */
157 			};
158 #if ALLOC_SPLIT_PTLOCKS
159 			spinlock_t *ptl;
160 #else
161 			spinlock_t ptl;
162 #endif
163 		};
164 		struct {	/* ZONE_DEVICE pages */
165 			/** @pgmap: Points to the hosting device page map. */
166 			struct dev_pagemap *pgmap;
167 			void *zone_device_data;
168 			/*
169 			 * ZONE_DEVICE private pages are counted as being
170 			 * mapped so the next 3 words hold the mapping, index,
171 			 * and private fields from the source anonymous or
172 			 * page cache page while the page is migrated to device
173 			 * private memory.
174 			 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
175 			 * use the mapping, index, and private fields when
176 			 * pmem backed DAX files are mapped.
177 			 */
178 		};
179 
180 		/** @rcu_head: You can use this to free a page by RCU. */
181 		struct rcu_head rcu_head;
182 	};
183 
184 	union {		/* This union is 4 bytes in size. */
185 		/*
186 		 * If the page can be mapped to userspace, encodes the number
187 		 * of times this page is referenced by a page table.
188 		 */
189 		atomic_t _mapcount;
190 
191 		/*
192 		 * If the page is neither PageSlab nor mappable to userspace,
193 		 * the value stored here may help determine what this page
194 		 * is used for.  See page-flags.h for a list of page types
195 		 * which are currently stored here.
196 		 */
197 		unsigned int page_type;
198 
199 		unsigned int active;		/* SLAB */
200 		int units;			/* SLOB */
201 	};
202 
203 	/* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
204 	atomic_t _refcount;
205 
206 #ifdef CONFIG_MEMCG
207 	unsigned long memcg_data;
208 #endif
209 
210 	/*
211 	 * On machines where all RAM is mapped into kernel address space,
212 	 * we can simply calculate the virtual address. On machines with
213 	 * highmem some memory is mapped into kernel virtual memory
214 	 * dynamically, so we need a place to store that address.
215 	 * Note that this field could be 16 bits on x86 ... ;)
216 	 *
217 	 * Architectures with slow multiplication can define
218 	 * WANT_PAGE_VIRTUAL in asm/page.h
219 	 */
220 #if defined(WANT_PAGE_VIRTUAL)
221 	void *virtual;			/* Kernel virtual address (NULL if
222 					   not kmapped, ie. highmem) */
223 #endif /* WANT_PAGE_VIRTUAL */
224 
225 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
226 	int _last_cpupid;
227 #endif
228 } _struct_page_alignment;
229 
230 /**
231  * struct folio - Represents a contiguous set of bytes.
232  * @flags: Identical to the page flags.
233  * @lru: Least Recently Used list; tracks how recently this folio was used.
234  * @mapping: The file this page belongs to, or refers to the anon_vma for
235  *    anonymous memory.
236  * @index: Offset within the file, in units of pages.  For anonymous memory,
237  *    this is the index from the beginning of the mmap.
238  * @private: Filesystem per-folio data (see folio_attach_private()).
239  *    Used for swp_entry_t if folio_test_swapcache().
240  * @_mapcount: Do not access this member directly.  Use folio_mapcount() to
241  *    find out how many times this folio is mapped by userspace.
242  * @_refcount: Do not access this member directly.  Use folio_ref_count()
243  *    to find how many references there are to this folio.
244  * @memcg_data: Memory Control Group data.
245  *
246  * A folio is a physically, virtually and logically contiguous set
247  * of bytes.  It is a power-of-two in size, and it is aligned to that
248  * same power-of-two.  It is at least as large as %PAGE_SIZE.  If it is
249  * in the page cache, it is at a file offset which is a multiple of that
250  * power-of-two.  It may be mapped into userspace at an address which is
251  * at an arbitrary page offset, but its kernel virtual address is aligned
252  * to its size.
253  */
254 struct folio {
255 	/* private: don't document the anon union */
256 	union {
257 		struct {
258 	/* public: */
259 			unsigned long flags;
260 			struct list_head lru;
261 			struct address_space *mapping;
262 			pgoff_t index;
263 			void *private;
264 			atomic_t _mapcount;
265 			atomic_t _refcount;
266 #ifdef CONFIG_MEMCG
267 			unsigned long memcg_data;
268 #endif
269 	/* private: the union with struct page is transitional */
270 		};
271 		struct page page;
272 	};
273 };
274 
275 static_assert(sizeof(struct page) == sizeof(struct folio));
276 #define FOLIO_MATCH(pg, fl)						\
277 	static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
278 FOLIO_MATCH(flags, flags);
279 FOLIO_MATCH(lru, lru);
280 FOLIO_MATCH(compound_head, lru);
281 FOLIO_MATCH(index, index);
282 FOLIO_MATCH(private, private);
283 FOLIO_MATCH(_mapcount, _mapcount);
284 FOLIO_MATCH(_refcount, _refcount);
285 #ifdef CONFIG_MEMCG
286 FOLIO_MATCH(memcg_data, memcg_data);
287 #endif
288 #undef FOLIO_MATCH
289 
290 static inline atomic_t *folio_mapcount_ptr(struct folio *folio)
291 {
292 	struct page *tail = &folio->page + 1;
293 	return &tail->compound_mapcount;
294 }
295 
296 static inline atomic_t *compound_mapcount_ptr(struct page *page)
297 {
298 	return &page[1].compound_mapcount;
299 }
300 
301 static inline atomic_t *compound_pincount_ptr(struct page *page)
302 {
303 	return &page[2].hpage_pinned_refcount;
304 }
305 
306 /*
307  * Used for sizing the vmemmap region on some architectures
308  */
309 #define STRUCT_PAGE_MAX_SHIFT	(order_base_2(sizeof(struct page)))
310 
311 #define PAGE_FRAG_CACHE_MAX_SIZE	__ALIGN_MASK(32768, ~PAGE_MASK)
312 #define PAGE_FRAG_CACHE_MAX_ORDER	get_order(PAGE_FRAG_CACHE_MAX_SIZE)
313 
314 /*
315  * page_private can be used on tail pages.  However, PagePrivate is only
316  * checked by the VM on the head page.  So page_private on the tail pages
317  * should be used for data that's ancillary to the head page (eg attaching
318  * buffer heads to tail pages after attaching buffer heads to the head page)
319  */
320 #define page_private(page)		((page)->private)
321 
322 static inline void set_page_private(struct page *page, unsigned long private)
323 {
324 	page->private = private;
325 }
326 
327 static inline void *folio_get_private(struct folio *folio)
328 {
329 	return folio->private;
330 }
331 
332 struct page_frag_cache {
333 	void * va;
334 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
335 	__u16 offset;
336 	__u16 size;
337 #else
338 	__u32 offset;
339 #endif
340 	/* we maintain a pagecount bias, so that we dont dirty cache line
341 	 * containing page->_refcount every time we allocate a fragment.
342 	 */
343 	unsigned int		pagecnt_bias;
344 	bool pfmemalloc;
345 };
346 
347 typedef unsigned long vm_flags_t;
348 
349 /*
350  * A region containing a mapping of a non-memory backed file under NOMMU
351  * conditions.  These are held in a global tree and are pinned by the VMAs that
352  * map parts of them.
353  */
354 struct vm_region {
355 	struct rb_node	vm_rb;		/* link in global region tree */
356 	vm_flags_t	vm_flags;	/* VMA vm_flags */
357 	unsigned long	vm_start;	/* start address of region */
358 	unsigned long	vm_end;		/* region initialised to here */
359 	unsigned long	vm_top;		/* region allocated to here */
360 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
361 	struct file	*vm_file;	/* the backing file or NULL */
362 
363 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
364 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
365 						* this region */
366 };
367 
368 #ifdef CONFIG_USERFAULTFD
369 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
370 struct vm_userfaultfd_ctx {
371 	struct userfaultfd_ctx *ctx;
372 };
373 #else /* CONFIG_USERFAULTFD */
374 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
375 struct vm_userfaultfd_ctx {};
376 #endif /* CONFIG_USERFAULTFD */
377 
378 /*
379  * This struct describes a virtual memory area. There is one of these
380  * per VM-area/task. A VM area is any part of the process virtual memory
381  * space that has a special rule for the page-fault handlers (ie a shared
382  * library, the executable area etc).
383  */
384 struct vm_area_struct {
385 	/* The first cache line has the info for VMA tree walking. */
386 
387 	unsigned long vm_start;		/* Our start address within vm_mm. */
388 	unsigned long vm_end;		/* The first byte after our end address
389 					   within vm_mm. */
390 
391 	/* linked list of VM areas per task, sorted by address */
392 	struct vm_area_struct *vm_next, *vm_prev;
393 
394 	struct rb_node vm_rb;
395 
396 	/*
397 	 * Largest free memory gap in bytes to the left of this VMA.
398 	 * Either between this VMA and vma->vm_prev, or between one of the
399 	 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
400 	 * get_unmapped_area find a free area of the right size.
401 	 */
402 	unsigned long rb_subtree_gap;
403 
404 	/* Second cache line starts here. */
405 
406 	struct mm_struct *vm_mm;	/* The address space we belong to. */
407 
408 	/*
409 	 * Access permissions of this VMA.
410 	 * See vmf_insert_mixed_prot() for discussion.
411 	 */
412 	pgprot_t vm_page_prot;
413 	unsigned long vm_flags;		/* Flags, see mm.h. */
414 
415 	/*
416 	 * For areas with an address space and backing store,
417 	 * linkage into the address_space->i_mmap interval tree.
418 	 */
419 	struct {
420 		struct rb_node rb;
421 		unsigned long rb_subtree_last;
422 	} shared;
423 
424 	/*
425 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
426 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
427 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
428 	 * or brk vma (with NULL file) can only be in an anon_vma list.
429 	 */
430 	struct list_head anon_vma_chain; /* Serialized by mmap_lock &
431 					  * page_table_lock */
432 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
433 
434 	/* Function pointers to deal with this struct. */
435 	const struct vm_operations_struct *vm_ops;
436 
437 	/* Information about our backing store: */
438 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
439 					   units */
440 	struct file * vm_file;		/* File we map to (can be NULL). */
441 	void * vm_private_data;		/* was vm_pte (shared mem) */
442 
443 #ifdef CONFIG_SWAP
444 	atomic_long_t swap_readahead_info;
445 #endif
446 #ifndef CONFIG_MMU
447 	struct vm_region *vm_region;	/* NOMMU mapping region */
448 #endif
449 #ifdef CONFIG_NUMA
450 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
451 #endif
452 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
453 } __randomize_layout;
454 
455 struct kioctx_table;
456 struct mm_struct {
457 	struct {
458 		struct vm_area_struct *mmap;		/* list of VMAs */
459 		struct rb_root mm_rb;
460 		u64 vmacache_seqnum;                   /* per-thread vmacache */
461 #ifdef CONFIG_MMU
462 		unsigned long (*get_unmapped_area) (struct file *filp,
463 				unsigned long addr, unsigned long len,
464 				unsigned long pgoff, unsigned long flags);
465 #endif
466 		unsigned long mmap_base;	/* base of mmap area */
467 		unsigned long mmap_legacy_base;	/* base of mmap area in bottom-up allocations */
468 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
469 		/* Base addresses for compatible mmap() */
470 		unsigned long mmap_compat_base;
471 		unsigned long mmap_compat_legacy_base;
472 #endif
473 		unsigned long task_size;	/* size of task vm space */
474 		unsigned long highest_vm_end;	/* highest vma end address */
475 		pgd_t * pgd;
476 
477 #ifdef CONFIG_MEMBARRIER
478 		/**
479 		 * @membarrier_state: Flags controlling membarrier behavior.
480 		 *
481 		 * This field is close to @pgd to hopefully fit in the same
482 		 * cache-line, which needs to be touched by switch_mm().
483 		 */
484 		atomic_t membarrier_state;
485 #endif
486 
487 		/**
488 		 * @mm_users: The number of users including userspace.
489 		 *
490 		 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
491 		 * drops to 0 (i.e. when the task exits and there are no other
492 		 * temporary reference holders), we also release a reference on
493 		 * @mm_count (which may then free the &struct mm_struct if
494 		 * @mm_count also drops to 0).
495 		 */
496 		atomic_t mm_users;
497 
498 		/**
499 		 * @mm_count: The number of references to &struct mm_struct
500 		 * (@mm_users count as 1).
501 		 *
502 		 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
503 		 * &struct mm_struct is freed.
504 		 */
505 		atomic_t mm_count;
506 
507 #ifdef CONFIG_MMU
508 		atomic_long_t pgtables_bytes;	/* PTE page table pages */
509 #endif
510 		int map_count;			/* number of VMAs */
511 
512 		spinlock_t page_table_lock; /* Protects page tables and some
513 					     * counters
514 					     */
515 		/*
516 		 * With some kernel config, the current mmap_lock's offset
517 		 * inside 'mm_struct' is at 0x120, which is very optimal, as
518 		 * its two hot fields 'count' and 'owner' sit in 2 different
519 		 * cachelines,  and when mmap_lock is highly contended, both
520 		 * of the 2 fields will be accessed frequently, current layout
521 		 * will help to reduce cache bouncing.
522 		 *
523 		 * So please be careful with adding new fields before
524 		 * mmap_lock, which can easily push the 2 fields into one
525 		 * cacheline.
526 		 */
527 		struct rw_semaphore mmap_lock;
528 
529 		struct list_head mmlist; /* List of maybe swapped mm's.	These
530 					  * are globally strung together off
531 					  * init_mm.mmlist, and are protected
532 					  * by mmlist_lock
533 					  */
534 
535 
536 		unsigned long hiwater_rss; /* High-watermark of RSS usage */
537 		unsigned long hiwater_vm;  /* High-water virtual memory usage */
538 
539 		unsigned long total_vm;	   /* Total pages mapped */
540 		unsigned long locked_vm;   /* Pages that have PG_mlocked set */
541 		atomic64_t    pinned_vm;   /* Refcount permanently increased */
542 		unsigned long data_vm;	   /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
543 		unsigned long exec_vm;	   /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
544 		unsigned long stack_vm;	   /* VM_STACK */
545 		unsigned long def_flags;
546 
547 		/**
548 		 * @write_protect_seq: Locked when any thread is write
549 		 * protecting pages mapped by this mm to enforce a later COW,
550 		 * for instance during page table copying for fork().
551 		 */
552 		seqcount_t write_protect_seq;
553 
554 		spinlock_t arg_lock; /* protect the below fields */
555 
556 		unsigned long start_code, end_code, start_data, end_data;
557 		unsigned long start_brk, brk, start_stack;
558 		unsigned long arg_start, arg_end, env_start, env_end;
559 
560 		unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
561 
562 		/*
563 		 * Special counters, in some configurations protected by the
564 		 * page_table_lock, in other configurations by being atomic.
565 		 */
566 		struct mm_rss_stat rss_stat;
567 
568 		struct linux_binfmt *binfmt;
569 
570 		/* Architecture-specific MM context */
571 		mm_context_t context;
572 
573 		unsigned long flags; /* Must use atomic bitops to access */
574 
575 #ifdef CONFIG_AIO
576 		spinlock_t			ioctx_lock;
577 		struct kioctx_table __rcu	*ioctx_table;
578 #endif
579 #ifdef CONFIG_MEMCG
580 		/*
581 		 * "owner" points to a task that is regarded as the canonical
582 		 * user/owner of this mm. All of the following must be true in
583 		 * order for it to be changed:
584 		 *
585 		 * current == mm->owner
586 		 * current->mm != mm
587 		 * new_owner->mm == mm
588 		 * new_owner->alloc_lock is held
589 		 */
590 		struct task_struct __rcu *owner;
591 #endif
592 		struct user_namespace *user_ns;
593 
594 		/* store ref to file /proc/<pid>/exe symlink points to */
595 		struct file __rcu *exe_file;
596 #ifdef CONFIG_MMU_NOTIFIER
597 		struct mmu_notifier_subscriptions *notifier_subscriptions;
598 #endif
599 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
600 		pgtable_t pmd_huge_pte; /* protected by page_table_lock */
601 #endif
602 #ifdef CONFIG_NUMA_BALANCING
603 		/*
604 		 * numa_next_scan is the next time that the PTEs will be marked
605 		 * pte_numa. NUMA hinting faults will gather statistics and
606 		 * migrate pages to new nodes if necessary.
607 		 */
608 		unsigned long numa_next_scan;
609 
610 		/* Restart point for scanning and setting pte_numa */
611 		unsigned long numa_scan_offset;
612 
613 		/* numa_scan_seq prevents two threads setting pte_numa */
614 		int numa_scan_seq;
615 #endif
616 		/*
617 		 * An operation with batched TLB flushing is going on. Anything
618 		 * that can move process memory needs to flush the TLB when
619 		 * moving a PROT_NONE or PROT_NUMA mapped page.
620 		 */
621 		atomic_t tlb_flush_pending;
622 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
623 		/* See flush_tlb_batched_pending() */
624 		bool tlb_flush_batched;
625 #endif
626 		struct uprobes_state uprobes_state;
627 #ifdef CONFIG_PREEMPT_RT
628 		struct rcu_head delayed_drop;
629 #endif
630 #ifdef CONFIG_HUGETLB_PAGE
631 		atomic_long_t hugetlb_usage;
632 #endif
633 		struct work_struct async_put_work;
634 
635 #ifdef CONFIG_IOMMU_SUPPORT
636 		u32 pasid;
637 #endif
638 	} __randomize_layout;
639 
640 	/*
641 	 * The mm_cpumask needs to be at the end of mm_struct, because it
642 	 * is dynamically sized based on nr_cpu_ids.
643 	 */
644 	unsigned long cpu_bitmap[];
645 };
646 
647 extern struct mm_struct init_mm;
648 
649 /* Pointer magic because the dynamic array size confuses some compilers. */
650 static inline void mm_init_cpumask(struct mm_struct *mm)
651 {
652 	unsigned long cpu_bitmap = (unsigned long)mm;
653 
654 	cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
655 	cpumask_clear((struct cpumask *)cpu_bitmap);
656 }
657 
658 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
659 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
660 {
661 	return (struct cpumask *)&mm->cpu_bitmap;
662 }
663 
664 struct mmu_gather;
665 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
666 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
667 extern void tlb_finish_mmu(struct mmu_gather *tlb);
668 
669 static inline void init_tlb_flush_pending(struct mm_struct *mm)
670 {
671 	atomic_set(&mm->tlb_flush_pending, 0);
672 }
673 
674 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
675 {
676 	atomic_inc(&mm->tlb_flush_pending);
677 	/*
678 	 * The only time this value is relevant is when there are indeed pages
679 	 * to flush. And we'll only flush pages after changing them, which
680 	 * requires the PTL.
681 	 *
682 	 * So the ordering here is:
683 	 *
684 	 *	atomic_inc(&mm->tlb_flush_pending);
685 	 *	spin_lock(&ptl);
686 	 *	...
687 	 *	set_pte_at();
688 	 *	spin_unlock(&ptl);
689 	 *
690 	 *				spin_lock(&ptl)
691 	 *				mm_tlb_flush_pending();
692 	 *				....
693 	 *				spin_unlock(&ptl);
694 	 *
695 	 *	flush_tlb_range();
696 	 *	atomic_dec(&mm->tlb_flush_pending);
697 	 *
698 	 * Where the increment if constrained by the PTL unlock, it thus
699 	 * ensures that the increment is visible if the PTE modification is
700 	 * visible. After all, if there is no PTE modification, nobody cares
701 	 * about TLB flushes either.
702 	 *
703 	 * This very much relies on users (mm_tlb_flush_pending() and
704 	 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
705 	 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
706 	 * locks (PPC) the unlock of one doesn't order against the lock of
707 	 * another PTL.
708 	 *
709 	 * The decrement is ordered by the flush_tlb_range(), such that
710 	 * mm_tlb_flush_pending() will not return false unless all flushes have
711 	 * completed.
712 	 */
713 }
714 
715 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
716 {
717 	/*
718 	 * See inc_tlb_flush_pending().
719 	 *
720 	 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
721 	 * not order against TLB invalidate completion, which is what we need.
722 	 *
723 	 * Therefore we must rely on tlb_flush_*() to guarantee order.
724 	 */
725 	atomic_dec(&mm->tlb_flush_pending);
726 }
727 
728 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
729 {
730 	/*
731 	 * Must be called after having acquired the PTL; orders against that
732 	 * PTLs release and therefore ensures that if we observe the modified
733 	 * PTE we must also observe the increment from inc_tlb_flush_pending().
734 	 *
735 	 * That is, it only guarantees to return true if there is a flush
736 	 * pending for _this_ PTL.
737 	 */
738 	return atomic_read(&mm->tlb_flush_pending);
739 }
740 
741 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
742 {
743 	/*
744 	 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
745 	 * for which there is a TLB flush pending in order to guarantee
746 	 * we've seen both that PTE modification and the increment.
747 	 *
748 	 * (no requirement on actually still holding the PTL, that is irrelevant)
749 	 */
750 	return atomic_read(&mm->tlb_flush_pending) > 1;
751 }
752 
753 struct vm_fault;
754 
755 /**
756  * typedef vm_fault_t - Return type for page fault handlers.
757  *
758  * Page fault handlers return a bitmask of %VM_FAULT values.
759  */
760 typedef __bitwise unsigned int vm_fault_t;
761 
762 /**
763  * enum vm_fault_reason - Page fault handlers return a bitmask of
764  * these values to tell the core VM what happened when handling the
765  * fault. Used to decide whether a process gets delivered SIGBUS or
766  * just gets major/minor fault counters bumped up.
767  *
768  * @VM_FAULT_OOM:		Out Of Memory
769  * @VM_FAULT_SIGBUS:		Bad access
770  * @VM_FAULT_MAJOR:		Page read from storage
771  * @VM_FAULT_WRITE:		Special case for get_user_pages
772  * @VM_FAULT_HWPOISON:		Hit poisoned small page
773  * @VM_FAULT_HWPOISON_LARGE:	Hit poisoned large page. Index encoded
774  *				in upper bits
775  * @VM_FAULT_SIGSEGV:		segmentation fault
776  * @VM_FAULT_NOPAGE:		->fault installed the pte, not return page
777  * @VM_FAULT_LOCKED:		->fault locked the returned page
778  * @VM_FAULT_RETRY:		->fault blocked, must retry
779  * @VM_FAULT_FALLBACK:		huge page fault failed, fall back to small
780  * @VM_FAULT_DONE_COW:		->fault has fully handled COW
781  * @VM_FAULT_NEEDDSYNC:		->fault did not modify page tables and needs
782  *				fsync() to complete (for synchronous page faults
783  *				in DAX)
784  * @VM_FAULT_HINDEX_MASK:	mask HINDEX value
785  *
786  */
787 enum vm_fault_reason {
788 	VM_FAULT_OOM            = (__force vm_fault_t)0x000001,
789 	VM_FAULT_SIGBUS         = (__force vm_fault_t)0x000002,
790 	VM_FAULT_MAJOR          = (__force vm_fault_t)0x000004,
791 	VM_FAULT_WRITE          = (__force vm_fault_t)0x000008,
792 	VM_FAULT_HWPOISON       = (__force vm_fault_t)0x000010,
793 	VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
794 	VM_FAULT_SIGSEGV        = (__force vm_fault_t)0x000040,
795 	VM_FAULT_NOPAGE         = (__force vm_fault_t)0x000100,
796 	VM_FAULT_LOCKED         = (__force vm_fault_t)0x000200,
797 	VM_FAULT_RETRY          = (__force vm_fault_t)0x000400,
798 	VM_FAULT_FALLBACK       = (__force vm_fault_t)0x000800,
799 	VM_FAULT_DONE_COW       = (__force vm_fault_t)0x001000,
800 	VM_FAULT_NEEDDSYNC      = (__force vm_fault_t)0x002000,
801 	VM_FAULT_HINDEX_MASK    = (__force vm_fault_t)0x0f0000,
802 };
803 
804 /* Encode hstate index for a hwpoisoned large page */
805 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
806 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
807 
808 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS |	\
809 			VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON |	\
810 			VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
811 
812 #define VM_FAULT_RESULT_TRACE \
813 	{ VM_FAULT_OOM,                 "OOM" },	\
814 	{ VM_FAULT_SIGBUS,              "SIGBUS" },	\
815 	{ VM_FAULT_MAJOR,               "MAJOR" },	\
816 	{ VM_FAULT_WRITE,               "WRITE" },	\
817 	{ VM_FAULT_HWPOISON,            "HWPOISON" },	\
818 	{ VM_FAULT_HWPOISON_LARGE,      "HWPOISON_LARGE" },	\
819 	{ VM_FAULT_SIGSEGV,             "SIGSEGV" },	\
820 	{ VM_FAULT_NOPAGE,              "NOPAGE" },	\
821 	{ VM_FAULT_LOCKED,              "LOCKED" },	\
822 	{ VM_FAULT_RETRY,               "RETRY" },	\
823 	{ VM_FAULT_FALLBACK,            "FALLBACK" },	\
824 	{ VM_FAULT_DONE_COW,            "DONE_COW" },	\
825 	{ VM_FAULT_NEEDDSYNC,           "NEEDDSYNC" }
826 
827 struct vm_special_mapping {
828 	const char *name;	/* The name, e.g. "[vdso]". */
829 
830 	/*
831 	 * If .fault is not provided, this points to a
832 	 * NULL-terminated array of pages that back the special mapping.
833 	 *
834 	 * This must not be NULL unless .fault is provided.
835 	 */
836 	struct page **pages;
837 
838 	/*
839 	 * If non-NULL, then this is called to resolve page faults
840 	 * on the special mapping.  If used, .pages is not checked.
841 	 */
842 	vm_fault_t (*fault)(const struct vm_special_mapping *sm,
843 				struct vm_area_struct *vma,
844 				struct vm_fault *vmf);
845 
846 	int (*mremap)(const struct vm_special_mapping *sm,
847 		     struct vm_area_struct *new_vma);
848 };
849 
850 enum tlb_flush_reason {
851 	TLB_FLUSH_ON_TASK_SWITCH,
852 	TLB_REMOTE_SHOOTDOWN,
853 	TLB_LOCAL_SHOOTDOWN,
854 	TLB_LOCAL_MM_SHOOTDOWN,
855 	TLB_REMOTE_SEND_IPI,
856 	NR_TLB_FLUSH_REASONS,
857 };
858 
859  /*
860   * A swap entry has to fit into a "unsigned long", as the entry is hidden
861   * in the "index" field of the swapper address space.
862   */
863 typedef struct {
864 	unsigned long val;
865 } swp_entry_t;
866 
867 #endif /* _LINUX_MM_TYPES_H */
868