xref: /openbmc/linux/include/linux/mm_types.h (revision e3a2d3f6)
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/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>
22 
23 #include <asm/mmu.h>
24 
25 #ifndef AT_VECTOR_SIZE_ARCH
26 #define AT_VECTOR_SIZE_ARCH 0
27 #endif
28 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
29 
30 #define INIT_PASID	0
31 
32 struct address_space;
33 struct mem_cgroup;
34 
35 /*
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
40  * who is mapping it.
41  *
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.
48  *
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()
51  * before freeing it.
52  *
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.
57  *
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
60  * afterwards.
61  *
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.
67  */
68 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
69 #define _struct_page_alignment	__aligned(2 * sizeof(unsigned long))
70 #else
71 #define _struct_page_alignment	__aligned(sizeof(unsigned long))
72 #endif
73 
74 struct page {
75 	unsigned long flags;		/* Atomic flags, some possibly
76 					 * updated asynchronously */
77 	/*
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().
82 	 */
83 	union {
84 		struct {	/* Page cache and anonymous pages */
85 			/**
86 			 * @lru: Pageout list, eg. active_list protected by
87 			 * lruvec->lru_lock.  Sometimes used as a generic list
88 			 * by the page owner.
89 			 */
90 			union {
91 				struct list_head lru;
92 
93 				/* Or, for the Unevictable "LRU list" slot */
94 				struct {
95 					/* Always even, to negate PageTail */
96 					void *__filler;
97 					/* Count page's or folio's mlocks */
98 					unsigned int mlock_count;
99 				};
100 
101 				/* Or, free page */
102 				struct list_head buddy_list;
103 				struct list_head pcp_list;
104 			};
105 			/* See page-flags.h for PAGE_MAPPING_FLAGS */
106 			struct address_space *mapping;
107 			union {
108 				pgoff_t index;		/* Our offset within mapping. */
109 				unsigned long share;	/* share count for fsdax */
110 			};
111 			/**
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.
116 			 */
117 			unsigned long private;
118 		};
119 		struct {	/* page_pool used by netstack */
120 			/**
121 			 * @pp_magic: magic value to avoid recycling non
122 			 * page_pool allocated pages.
123 			 */
124 			unsigned long pp_magic;
125 			struct page_pool *pp;
126 			unsigned long _pp_mapping_pad;
127 			unsigned long dma_addr;
128 			union {
129 				/**
130 				 * dma_addr_upper: might require a 64-bit
131 				 * value on 32-bit architectures.
132 				 */
133 				unsigned long dma_addr_upper;
134 				/**
135 				 * For frag page support, not supported in
136 				 * 32-bit architectures with 64-bit DMA.
137 				 */
138 				atomic_long_t pp_frag_count;
139 			};
140 		};
141 		struct {	/* Tail pages of compound page */
142 			unsigned long compound_head;	/* Bit zero is set */
143 		};
144 		struct {	/* ZONE_DEVICE pages */
145 			/** @pgmap: Points to the hosting device page map. */
146 			struct dev_pagemap *pgmap;
147 			void *zone_device_data;
148 			/*
149 			 * ZONE_DEVICE private pages are counted as being
150 			 * mapped so the next 3 words hold the mapping, index,
151 			 * and private fields from the source anonymous or
152 			 * page cache page while the page is migrated to device
153 			 * private memory.
154 			 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
155 			 * use the mapping, index, and private fields when
156 			 * pmem backed DAX files are mapped.
157 			 */
158 		};
159 
160 		/** @rcu_head: You can use this to free a page by RCU. */
161 		struct rcu_head rcu_head;
162 	};
163 
164 	union {		/* This union is 4 bytes in size. */
165 		/*
166 		 * If the page can be mapped to userspace, encodes the number
167 		 * of times this page is referenced by a page table.
168 		 */
169 		atomic_t _mapcount;
170 
171 		/*
172 		 * If the page is neither PageSlab nor mappable to userspace,
173 		 * the value stored here may help determine what this page
174 		 * is used for.  See page-flags.h for a list of page types
175 		 * which are currently stored here.
176 		 */
177 		unsigned int page_type;
178 	};
179 
180 	/* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
181 	atomic_t _refcount;
182 
183 #ifdef CONFIG_MEMCG
184 	unsigned long memcg_data;
185 #endif
186 
187 	/*
188 	 * On machines where all RAM is mapped into kernel address space,
189 	 * we can simply calculate the virtual address. On machines with
190 	 * highmem some memory is mapped into kernel virtual memory
191 	 * dynamically, so we need a place to store that address.
192 	 * Note that this field could be 16 bits on x86 ... ;)
193 	 *
194 	 * Architectures with slow multiplication can define
195 	 * WANT_PAGE_VIRTUAL in asm/page.h
196 	 */
197 #if defined(WANT_PAGE_VIRTUAL)
198 	void *virtual;			/* Kernel virtual address (NULL if
199 					   not kmapped, ie. highmem) */
200 #endif /* WANT_PAGE_VIRTUAL */
201 
202 #ifdef CONFIG_KMSAN
203 	/*
204 	 * KMSAN metadata for this page:
205 	 *  - shadow page: every bit indicates whether the corresponding
206 	 *    bit of the original page is initialized (0) or not (1);
207 	 *  - origin page: every 4 bytes contain an id of the stack trace
208 	 *    where the uninitialized value was created.
209 	 */
210 	struct page *kmsan_shadow;
211 	struct page *kmsan_origin;
212 #endif
213 
214 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
215 	int _last_cpupid;
216 #endif
217 } _struct_page_alignment;
218 
219 /*
220  * struct encoded_page - a nonexistent type marking this pointer
221  *
222  * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
223  * with the low bits of the pointer indicating extra context-dependent
224  * information. Not super-common, but happens in mmu_gather and mlock
225  * handling, and this acts as a type system check on that use.
226  *
227  * We only really have two guaranteed bits in general, although you could
228  * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
229  * for more.
230  *
231  * Use the supplied helper functions to endcode/decode the pointer and bits.
232  */
233 struct encoded_page;
234 #define ENCODE_PAGE_BITS 3ul
encode_page(struct page * page,unsigned long flags)235 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
236 {
237 	BUILD_BUG_ON(flags > ENCODE_PAGE_BITS);
238 	return (struct encoded_page *)(flags | (unsigned long)page);
239 }
240 
encoded_page_flags(struct encoded_page * page)241 static inline unsigned long encoded_page_flags(struct encoded_page *page)
242 {
243 	return ENCODE_PAGE_BITS & (unsigned long)page;
244 }
245 
encoded_page_ptr(struct encoded_page * page)246 static inline struct page *encoded_page_ptr(struct encoded_page *page)
247 {
248 	return (struct page *)(~ENCODE_PAGE_BITS & (unsigned long)page);
249 }
250 
251 /*
252  * A swap entry has to fit into a "unsigned long", as the entry is hidden
253  * in the "index" field of the swapper address space.
254  */
255 typedef struct {
256 	unsigned long val;
257 } swp_entry_t;
258 
259 /**
260  * struct folio - Represents a contiguous set of bytes.
261  * @flags: Identical to the page flags.
262  * @lru: Least Recently Used list; tracks how recently this folio was used.
263  * @mlock_count: Number of times this folio has been pinned by mlock().
264  * @mapping: The file this page belongs to, or refers to the anon_vma for
265  *    anonymous memory.
266  * @index: Offset within the file, in units of pages.  For anonymous memory,
267  *    this is the index from the beginning of the mmap.
268  * @private: Filesystem per-folio data (see folio_attach_private()).
269  * @swap: Used for swp_entry_t if folio_test_swapcache().
270  * @_mapcount: Do not access this member directly.  Use folio_mapcount() to
271  *    find out how many times this folio is mapped by userspace.
272  * @_refcount: Do not access this member directly.  Use folio_ref_count()
273  *    to find how many references there are to this folio.
274  * @memcg_data: Memory Control Group data.
275  * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
276  * @_nr_pages_mapped: Do not use directly, call folio_mapcount().
277  * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
278  * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
279  * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
280  * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
281  * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
282  * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
283  * @_deferred_list: Folios to be split under memory pressure.
284  *
285  * A folio is a physically, virtually and logically contiguous set
286  * of bytes.  It is a power-of-two in size, and it is aligned to that
287  * same power-of-two.  It is at least as large as %PAGE_SIZE.  If it is
288  * in the page cache, it is at a file offset which is a multiple of that
289  * power-of-two.  It may be mapped into userspace at an address which is
290  * at an arbitrary page offset, but its kernel virtual address is aligned
291  * to its size.
292  */
293 struct folio {
294 	/* private: don't document the anon union */
295 	union {
296 		struct {
297 	/* public: */
298 			unsigned long flags;
299 			union {
300 				struct list_head lru;
301 	/* private: avoid cluttering the output */
302 				struct {
303 					void *__filler;
304 	/* public: */
305 					unsigned int mlock_count;
306 	/* private: */
307 				};
308 	/* public: */
309 			};
310 			struct address_space *mapping;
311 			pgoff_t index;
312 			union {
313 				void *private;
314 				swp_entry_t swap;
315 			};
316 			atomic_t _mapcount;
317 			atomic_t _refcount;
318 #ifdef CONFIG_MEMCG
319 			unsigned long memcg_data;
320 #endif
321 	/* private: the union with struct page is transitional */
322 		};
323 		struct page page;
324 	};
325 	union {
326 		struct {
327 			unsigned long _flags_1;
328 			unsigned long _head_1;
329 			unsigned long _folio_avail;
330 	/* public: */
331 			atomic_t _entire_mapcount;
332 			atomic_t _nr_pages_mapped;
333 			atomic_t _pincount;
334 #ifdef CONFIG_64BIT
335 			unsigned int _folio_nr_pages;
336 #endif
337 	/* private: the union with struct page is transitional */
338 		};
339 		struct page __page_1;
340 	};
341 	union {
342 		struct {
343 			unsigned long _flags_2;
344 			unsigned long _head_2;
345 	/* public: */
346 			void *_hugetlb_subpool;
347 			void *_hugetlb_cgroup;
348 			void *_hugetlb_cgroup_rsvd;
349 			void *_hugetlb_hwpoison;
350 	/* private: the union with struct page is transitional */
351 		};
352 		struct {
353 			unsigned long _flags_2a;
354 			unsigned long _head_2a;
355 	/* public: */
356 			struct list_head _deferred_list;
357 	/* private: the union with struct page is transitional */
358 		};
359 		struct page __page_2;
360 	};
361 };
362 
363 #define FOLIO_MATCH(pg, fl)						\
364 	static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
365 FOLIO_MATCH(flags, flags);
366 FOLIO_MATCH(lru, lru);
367 FOLIO_MATCH(mapping, mapping);
368 FOLIO_MATCH(compound_head, lru);
369 FOLIO_MATCH(index, index);
370 FOLIO_MATCH(private, private);
371 FOLIO_MATCH(_mapcount, _mapcount);
372 FOLIO_MATCH(_refcount, _refcount);
373 #ifdef CONFIG_MEMCG
374 FOLIO_MATCH(memcg_data, memcg_data);
375 #endif
376 #undef FOLIO_MATCH
377 #define FOLIO_MATCH(pg, fl)						\
378 	static_assert(offsetof(struct folio, fl) ==			\
379 			offsetof(struct page, pg) + sizeof(struct page))
380 FOLIO_MATCH(flags, _flags_1);
381 FOLIO_MATCH(compound_head, _head_1);
382 #undef FOLIO_MATCH
383 #define FOLIO_MATCH(pg, fl)						\
384 	static_assert(offsetof(struct folio, fl) ==			\
385 			offsetof(struct page, pg) + 2 * sizeof(struct page))
386 FOLIO_MATCH(flags, _flags_2);
387 FOLIO_MATCH(compound_head, _head_2);
388 FOLIO_MATCH(flags, _flags_2a);
389 FOLIO_MATCH(compound_head, _head_2a);
390 #undef FOLIO_MATCH
391 
392 /**
393  * struct ptdesc -    Memory descriptor for page tables.
394  * @__page_flags:     Same as page flags. Unused for page tables.
395  * @pt_rcu_head:      For freeing page table pages.
396  * @pt_list:          List of used page tables. Used for s390 and x86.
397  * @_pt_pad_1:        Padding that aliases with page's compound head.
398  * @pmd_huge_pte:     Protected by ptdesc->ptl, used for THPs.
399  * @__page_mapping:   Aliases with page->mapping. Unused for page tables.
400  * @pt_mm:            Used for x86 pgds.
401  * @pt_frag_refcount: For fragmented page table tracking. Powerpc and s390 only.
402  * @_pt_pad_2:        Padding to ensure proper alignment.
403  * @ptl:              Lock for the page table.
404  * @__page_type:      Same as page->page_type. Unused for page tables.
405  * @_refcount:        Same as page refcount. Used for s390 page tables.
406  * @pt_memcg_data:    Memcg data. Tracked for page tables here.
407  *
408  * This struct overlays struct page for now. Do not modify without a good
409  * understanding of the issues.
410  */
411 struct ptdesc {
412 	unsigned long __page_flags;
413 
414 	union {
415 		struct rcu_head pt_rcu_head;
416 		struct list_head pt_list;
417 		struct {
418 			unsigned long _pt_pad_1;
419 			pgtable_t pmd_huge_pte;
420 		};
421 	};
422 	unsigned long __page_mapping;
423 
424 	union {
425 		struct mm_struct *pt_mm;
426 		atomic_t pt_frag_refcount;
427 	};
428 
429 	union {
430 		unsigned long _pt_pad_2;
431 #if ALLOC_SPLIT_PTLOCKS
432 		spinlock_t *ptl;
433 #else
434 		spinlock_t ptl;
435 #endif
436 	};
437 	unsigned int __page_type;
438 	atomic_t _refcount;
439 #ifdef CONFIG_MEMCG
440 	unsigned long pt_memcg_data;
441 #endif
442 };
443 
444 #define TABLE_MATCH(pg, pt)						\
445 	static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
446 TABLE_MATCH(flags, __page_flags);
447 TABLE_MATCH(compound_head, pt_list);
448 TABLE_MATCH(compound_head, _pt_pad_1);
449 TABLE_MATCH(mapping, __page_mapping);
450 TABLE_MATCH(rcu_head, pt_rcu_head);
451 TABLE_MATCH(page_type, __page_type);
452 TABLE_MATCH(_refcount, _refcount);
453 #ifdef CONFIG_MEMCG
454 TABLE_MATCH(memcg_data, pt_memcg_data);
455 #endif
456 #undef TABLE_MATCH
457 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
458 
459 #define ptdesc_page(pt)			(_Generic((pt),			\
460 	const struct ptdesc *:		(const struct page *)(pt),	\
461 	struct ptdesc *:		(struct page *)(pt)))
462 
463 #define ptdesc_folio(pt)		(_Generic((pt),			\
464 	const struct ptdesc *:		(const struct folio *)(pt),	\
465 	struct ptdesc *:		(struct folio *)(pt)))
466 
467 #define page_ptdesc(p)			(_Generic((p),			\
468 	const struct page *:		(const struct ptdesc *)(p),	\
469 	struct page *:			(struct ptdesc *)(p)))
470 
471 /*
472  * Used for sizing the vmemmap region on some architectures
473  */
474 #define STRUCT_PAGE_MAX_SHIFT	(order_base_2(sizeof(struct page)))
475 
476 #define PAGE_FRAG_CACHE_MAX_SIZE	__ALIGN_MASK(32768, ~PAGE_MASK)
477 #define PAGE_FRAG_CACHE_MAX_ORDER	get_order(PAGE_FRAG_CACHE_MAX_SIZE)
478 
479 /*
480  * page_private can be used on tail pages.  However, PagePrivate is only
481  * checked by the VM on the head page.  So page_private on the tail pages
482  * should be used for data that's ancillary to the head page (eg attaching
483  * buffer heads to tail pages after attaching buffer heads to the head page)
484  */
485 #define page_private(page)		((page)->private)
486 
set_page_private(struct page * page,unsigned long private)487 static inline void set_page_private(struct page *page, unsigned long private)
488 {
489 	page->private = private;
490 }
491 
folio_get_private(struct folio * folio)492 static inline void *folio_get_private(struct folio *folio)
493 {
494 	return folio->private;
495 }
496 
497 struct page_frag_cache {
498 	void * va;
499 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
500 	__u16 offset;
501 	__u16 size;
502 #else
503 	__u32 offset;
504 #endif
505 	/* we maintain a pagecount bias, so that we dont dirty cache line
506 	 * containing page->_refcount every time we allocate a fragment.
507 	 */
508 	unsigned int		pagecnt_bias;
509 	bool pfmemalloc;
510 };
511 
512 typedef unsigned long vm_flags_t;
513 
514 /*
515  * A region containing a mapping of a non-memory backed file under NOMMU
516  * conditions.  These are held in a global tree and are pinned by the VMAs that
517  * map parts of them.
518  */
519 struct vm_region {
520 	struct rb_node	vm_rb;		/* link in global region tree */
521 	vm_flags_t	vm_flags;	/* VMA vm_flags */
522 	unsigned long	vm_start;	/* start address of region */
523 	unsigned long	vm_end;		/* region initialised to here */
524 	unsigned long	vm_top;		/* region allocated to here */
525 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
526 	struct file	*vm_file;	/* the backing file or NULL */
527 
528 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
529 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
530 						* this region */
531 };
532 
533 #ifdef CONFIG_USERFAULTFD
534 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
535 struct vm_userfaultfd_ctx {
536 	struct userfaultfd_ctx *ctx;
537 };
538 #else /* CONFIG_USERFAULTFD */
539 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
540 struct vm_userfaultfd_ctx {};
541 #endif /* CONFIG_USERFAULTFD */
542 
543 struct anon_vma_name {
544 	struct kref kref;
545 	/* The name needs to be at the end because it is dynamically sized. */
546 	char name[];
547 };
548 
549 struct vma_lock {
550 	struct rw_semaphore lock;
551 };
552 
553 struct vma_numab_state {
554 	/*
555 	 * Initialised as time in 'jiffies' after which VMA
556 	 * should be scanned.  Delays first scan of new VMA by at
557 	 * least sysctl_numa_balancing_scan_delay:
558 	 */
559 	unsigned long next_scan;
560 
561 	/*
562 	 * Time in jiffies when pids_active[] is reset to
563 	 * detect phase change behaviour:
564 	 */
565 	unsigned long pids_active_reset;
566 
567 	/*
568 	 * Approximate tracking of PIDs that trapped a NUMA hinting
569 	 * fault. May produce false positives due to hash collisions.
570 	 *
571 	 *   [0] Previous PID tracking
572 	 *   [1] Current PID tracking
573 	 *
574 	 * Window moves after next_pid_reset has expired approximately
575 	 * every VMA_PID_RESET_PERIOD jiffies:
576 	 */
577 	unsigned long pids_active[2];
578 
579 	/*
580 	 * MM scan sequence ID when the VMA was last completely scanned.
581 	 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq
582 	 */
583 	int prev_scan_seq;
584 };
585 
586 /*
587  * This struct describes a virtual memory area. There is one of these
588  * per VM-area/task. A VM area is any part of the process virtual memory
589  * space that has a special rule for the page-fault handlers (ie a shared
590  * library, the executable area etc).
591  */
592 struct vm_area_struct {
593 	/* The first cache line has the info for VMA tree walking. */
594 
595 	union {
596 		struct {
597 			/* VMA covers [vm_start; vm_end) addresses within mm */
598 			unsigned long vm_start;
599 			unsigned long vm_end;
600 		};
601 #ifdef CONFIG_PER_VMA_LOCK
602 		struct rcu_head vm_rcu;	/* Used for deferred freeing. */
603 #endif
604 	};
605 
606 	struct mm_struct *vm_mm;	/* The address space we belong to. */
607 	pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
608 
609 	/*
610 	 * Flags, see mm.h.
611 	 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
612 	 */
613 	union {
614 		const vm_flags_t vm_flags;
615 		vm_flags_t __private __vm_flags;
616 	};
617 
618 #ifdef CONFIG_PER_VMA_LOCK
619 	/*
620 	 * Can only be written (using WRITE_ONCE()) while holding both:
621 	 *  - mmap_lock (in write mode)
622 	 *  - vm_lock->lock (in write mode)
623 	 * Can be read reliably while holding one of:
624 	 *  - mmap_lock (in read or write mode)
625 	 *  - vm_lock->lock (in read or write mode)
626 	 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
627 	 * while holding nothing (except RCU to keep the VMA struct allocated).
628 	 *
629 	 * This sequence counter is explicitly allowed to overflow; sequence
630 	 * counter reuse can only lead to occasional unnecessary use of the
631 	 * slowpath.
632 	 */
633 	int vm_lock_seq;
634 	struct vma_lock *vm_lock;
635 
636 	/* Flag to indicate areas detached from the mm->mm_mt tree */
637 	bool detached;
638 #endif
639 
640 	/*
641 	 * For areas with an address space and backing store,
642 	 * linkage into the address_space->i_mmap interval tree.
643 	 *
644 	 */
645 	struct {
646 		struct rb_node rb;
647 		unsigned long rb_subtree_last;
648 	} shared;
649 
650 	/*
651 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
652 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
653 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
654 	 * or brk vma (with NULL file) can only be in an anon_vma list.
655 	 */
656 	struct list_head anon_vma_chain; /* Serialized by mmap_lock &
657 					  * page_table_lock */
658 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
659 
660 	/* Function pointers to deal with this struct. */
661 	const struct vm_operations_struct *vm_ops;
662 
663 	/* Information about our backing store: */
664 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
665 					   units */
666 	struct file * vm_file;		/* File we map to (can be NULL). */
667 	void * vm_private_data;		/* was vm_pte (shared mem) */
668 
669 #ifdef CONFIG_ANON_VMA_NAME
670 	/*
671 	 * For private and shared anonymous mappings, a pointer to a null
672 	 * terminated string containing the name given to the vma, or NULL if
673 	 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
674 	 */
675 	struct anon_vma_name *anon_name;
676 #endif
677 #ifdef CONFIG_SWAP
678 	atomic_long_t swap_readahead_info;
679 #endif
680 #ifndef CONFIG_MMU
681 	struct vm_region *vm_region;	/* NOMMU mapping region */
682 #endif
683 #ifdef CONFIG_NUMA
684 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
685 #endif
686 #ifdef CONFIG_NUMA_BALANCING
687 	struct vma_numab_state *numab_state;	/* NUMA Balancing state */
688 #endif
689 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
690 } __randomize_layout;
691 
692 #ifdef CONFIG_SCHED_MM_CID
693 struct mm_cid {
694 	u64 time;
695 	int cid;
696 };
697 #endif
698 
699 struct kioctx_table;
700 struct mm_struct {
701 	struct {
702 		/*
703 		 * Fields which are often written to are placed in a separate
704 		 * cache line.
705 		 */
706 		struct {
707 			/**
708 			 * @mm_count: The number of references to &struct
709 			 * mm_struct (@mm_users count as 1).
710 			 *
711 			 * Use mmgrab()/mmdrop() to modify. When this drops to
712 			 * 0, the &struct mm_struct is freed.
713 			 */
714 			atomic_t mm_count;
715 		} ____cacheline_aligned_in_smp;
716 
717 		struct maple_tree mm_mt;
718 #ifdef CONFIG_MMU
719 		unsigned long (*get_unmapped_area) (struct file *filp,
720 				unsigned long addr, unsigned long len,
721 				unsigned long pgoff, unsigned long flags);
722 #endif
723 		unsigned long mmap_base;	/* base of mmap area */
724 		unsigned long mmap_legacy_base;	/* base of mmap area in bottom-up allocations */
725 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
726 		/* Base addresses for compatible mmap() */
727 		unsigned long mmap_compat_base;
728 		unsigned long mmap_compat_legacy_base;
729 #endif
730 		unsigned long task_size;	/* size of task vm space */
731 		pgd_t * pgd;
732 
733 #ifdef CONFIG_MEMBARRIER
734 		/**
735 		 * @membarrier_state: Flags controlling membarrier behavior.
736 		 *
737 		 * This field is close to @pgd to hopefully fit in the same
738 		 * cache-line, which needs to be touched by switch_mm().
739 		 */
740 		atomic_t membarrier_state;
741 #endif
742 
743 		/**
744 		 * @mm_users: The number of users including userspace.
745 		 *
746 		 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
747 		 * drops to 0 (i.e. when the task exits and there are no other
748 		 * temporary reference holders), we also release a reference on
749 		 * @mm_count (which may then free the &struct mm_struct if
750 		 * @mm_count also drops to 0).
751 		 */
752 		atomic_t mm_users;
753 
754 #ifdef CONFIG_SCHED_MM_CID
755 		/**
756 		 * @pcpu_cid: Per-cpu current cid.
757 		 *
758 		 * Keep track of the currently allocated mm_cid for each cpu.
759 		 * The per-cpu mm_cid values are serialized by their respective
760 		 * runqueue locks.
761 		 */
762 		struct mm_cid __percpu *pcpu_cid;
763 		/*
764 		 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
765 		 *
766 		 * When the next mm_cid scan is due (in jiffies).
767 		 */
768 		unsigned long mm_cid_next_scan;
769 #endif
770 #ifdef CONFIG_MMU
771 		atomic_long_t pgtables_bytes;	/* size of all page tables */
772 #endif
773 		int map_count;			/* number of VMAs */
774 
775 		spinlock_t page_table_lock; /* Protects page tables and some
776 					     * counters
777 					     */
778 		/*
779 		 * With some kernel config, the current mmap_lock's offset
780 		 * inside 'mm_struct' is at 0x120, which is very optimal, as
781 		 * its two hot fields 'count' and 'owner' sit in 2 different
782 		 * cachelines,  and when mmap_lock is highly contended, both
783 		 * of the 2 fields will be accessed frequently, current layout
784 		 * will help to reduce cache bouncing.
785 		 *
786 		 * So please be careful with adding new fields before
787 		 * mmap_lock, which can easily push the 2 fields into one
788 		 * cacheline.
789 		 */
790 		struct rw_semaphore mmap_lock;
791 
792 		struct list_head mmlist; /* List of maybe swapped mm's.	These
793 					  * are globally strung together off
794 					  * init_mm.mmlist, and are protected
795 					  * by mmlist_lock
796 					  */
797 #ifdef CONFIG_PER_VMA_LOCK
798 		/*
799 		 * This field has lock-like semantics, meaning it is sometimes
800 		 * accessed with ACQUIRE/RELEASE semantics.
801 		 * Roughly speaking, incrementing the sequence number is
802 		 * equivalent to releasing locks on VMAs; reading the sequence
803 		 * number can be part of taking a read lock on a VMA.
804 		 *
805 		 * Can be modified under write mmap_lock using RELEASE
806 		 * semantics.
807 		 * Can be read with no other protection when holding write
808 		 * mmap_lock.
809 		 * Can be read with ACQUIRE semantics if not holding write
810 		 * mmap_lock.
811 		 */
812 		int mm_lock_seq;
813 #endif
814 
815 
816 		unsigned long hiwater_rss; /* High-watermark of RSS usage */
817 		unsigned long hiwater_vm;  /* High-water virtual memory usage */
818 
819 		unsigned long total_vm;	   /* Total pages mapped */
820 		unsigned long locked_vm;   /* Pages that have PG_mlocked set */
821 		atomic64_t    pinned_vm;   /* Refcount permanently increased */
822 		unsigned long data_vm;	   /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
823 		unsigned long exec_vm;	   /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
824 		unsigned long stack_vm;	   /* VM_STACK */
825 		unsigned long def_flags;
826 
827 		/**
828 		 * @write_protect_seq: Locked when any thread is write
829 		 * protecting pages mapped by this mm to enforce a later COW,
830 		 * for instance during page table copying for fork().
831 		 */
832 		seqcount_t write_protect_seq;
833 
834 		spinlock_t arg_lock; /* protect the below fields */
835 
836 		unsigned long start_code, end_code, start_data, end_data;
837 		unsigned long start_brk, brk, start_stack;
838 		unsigned long arg_start, arg_end, env_start, env_end;
839 
840 		unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
841 
842 		struct percpu_counter rss_stat[NR_MM_COUNTERS];
843 
844 		struct linux_binfmt *binfmt;
845 
846 		/* Architecture-specific MM context */
847 		mm_context_t context;
848 
849 		unsigned long flags; /* Must use atomic bitops to access */
850 
851 #ifdef CONFIG_AIO
852 		spinlock_t			ioctx_lock;
853 		struct kioctx_table __rcu	*ioctx_table;
854 #endif
855 #ifdef CONFIG_MEMCG
856 		/*
857 		 * "owner" points to a task that is regarded as the canonical
858 		 * user/owner of this mm. All of the following must be true in
859 		 * order for it to be changed:
860 		 *
861 		 * current == mm->owner
862 		 * current->mm != mm
863 		 * new_owner->mm == mm
864 		 * new_owner->alloc_lock is held
865 		 */
866 		struct task_struct __rcu *owner;
867 #endif
868 		struct user_namespace *user_ns;
869 
870 		/* store ref to file /proc/<pid>/exe symlink points to */
871 		struct file __rcu *exe_file;
872 #ifdef CONFIG_MMU_NOTIFIER
873 		struct mmu_notifier_subscriptions *notifier_subscriptions;
874 #endif
875 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
876 		pgtable_t pmd_huge_pte; /* protected by page_table_lock */
877 #endif
878 #ifdef CONFIG_NUMA_BALANCING
879 		/*
880 		 * numa_next_scan is the next time that PTEs will be remapped
881 		 * PROT_NONE to trigger NUMA hinting faults; such faults gather
882 		 * statistics and migrate pages to new nodes if necessary.
883 		 */
884 		unsigned long numa_next_scan;
885 
886 		/* Restart point for scanning and remapping PTEs. */
887 		unsigned long numa_scan_offset;
888 
889 		/* numa_scan_seq prevents two threads remapping PTEs. */
890 		int numa_scan_seq;
891 #endif
892 		/*
893 		 * An operation with batched TLB flushing is going on. Anything
894 		 * that can move process memory needs to flush the TLB when
895 		 * moving a PROT_NONE mapped page.
896 		 */
897 		atomic_t tlb_flush_pending;
898 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
899 		/* See flush_tlb_batched_pending() */
900 		atomic_t tlb_flush_batched;
901 #endif
902 		struct uprobes_state uprobes_state;
903 #ifdef CONFIG_PREEMPT_RT
904 		struct rcu_head delayed_drop;
905 #endif
906 #ifdef CONFIG_HUGETLB_PAGE
907 		atomic_long_t hugetlb_usage;
908 #endif
909 		struct work_struct async_put_work;
910 
911 #ifdef CONFIG_IOMMU_SVA
912 		u32 pasid;
913 #endif
914 #ifdef CONFIG_KSM
915 		/*
916 		 * Represent how many pages of this process are involved in KSM
917 		 * merging (not including ksm_zero_pages).
918 		 */
919 		unsigned long ksm_merging_pages;
920 		/*
921 		 * Represent how many pages are checked for ksm merging
922 		 * including merged and not merged.
923 		 */
924 		unsigned long ksm_rmap_items;
925 		/*
926 		 * Represent how many empty pages are merged with kernel zero
927 		 * pages when enabling KSM use_zero_pages.
928 		 */
929 		atomic_long_t ksm_zero_pages;
930 #endif /* CONFIG_KSM */
931 #ifdef CONFIG_LRU_GEN
932 		struct {
933 			/* this mm_struct is on lru_gen_mm_list */
934 			struct list_head list;
935 			/*
936 			 * Set when switching to this mm_struct, as a hint of
937 			 * whether it has been used since the last time per-node
938 			 * page table walkers cleared the corresponding bits.
939 			 */
940 			unsigned long bitmap;
941 #ifdef CONFIG_MEMCG
942 			/* points to the memcg of "owner" above */
943 			struct mem_cgroup *memcg;
944 #endif
945 		} lru_gen;
946 #endif /* CONFIG_LRU_GEN */
947 	} __randomize_layout;
948 
949 	/*
950 	 * The mm_cpumask needs to be at the end of mm_struct, because it
951 	 * is dynamically sized based on nr_cpu_ids.
952 	 */
953 	unsigned long cpu_bitmap[];
954 };
955 
956 #define MM_MT_FLAGS	(MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
957 			 MT_FLAGS_USE_RCU)
958 extern struct mm_struct init_mm;
959 
960 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)961 static inline void mm_init_cpumask(struct mm_struct *mm)
962 {
963 	unsigned long cpu_bitmap = (unsigned long)mm;
964 
965 	cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
966 	cpumask_clear((struct cpumask *)cpu_bitmap);
967 }
968 
969 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)970 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
971 {
972 	return (struct cpumask *)&mm->cpu_bitmap;
973 }
974 
975 #ifdef CONFIG_LRU_GEN
976 
977 struct lru_gen_mm_list {
978 	/* mm_struct list for page table walkers */
979 	struct list_head fifo;
980 	/* protects the list above */
981 	spinlock_t lock;
982 };
983 
984 void lru_gen_add_mm(struct mm_struct *mm);
985 void lru_gen_del_mm(struct mm_struct *mm);
986 #ifdef CONFIG_MEMCG
987 void lru_gen_migrate_mm(struct mm_struct *mm);
988 #endif
989 
lru_gen_init_mm(struct mm_struct * mm)990 static inline void lru_gen_init_mm(struct mm_struct *mm)
991 {
992 	INIT_LIST_HEAD(&mm->lru_gen.list);
993 	mm->lru_gen.bitmap = 0;
994 #ifdef CONFIG_MEMCG
995 	mm->lru_gen.memcg = NULL;
996 #endif
997 }
998 
lru_gen_use_mm(struct mm_struct * mm)999 static inline void lru_gen_use_mm(struct mm_struct *mm)
1000 {
1001 	/*
1002 	 * When the bitmap is set, page reclaim knows this mm_struct has been
1003 	 * used since the last time it cleared the bitmap. So it might be worth
1004 	 * walking the page tables of this mm_struct to clear the accessed bit.
1005 	 */
1006 	WRITE_ONCE(mm->lru_gen.bitmap, -1);
1007 }
1008 
1009 #else /* !CONFIG_LRU_GEN */
1010 
lru_gen_add_mm(struct mm_struct * mm)1011 static inline void lru_gen_add_mm(struct mm_struct *mm)
1012 {
1013 }
1014 
lru_gen_del_mm(struct mm_struct * mm)1015 static inline void lru_gen_del_mm(struct mm_struct *mm)
1016 {
1017 }
1018 
1019 #ifdef CONFIG_MEMCG
lru_gen_migrate_mm(struct mm_struct * mm)1020 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
1021 {
1022 }
1023 #endif
1024 
lru_gen_init_mm(struct mm_struct * mm)1025 static inline void lru_gen_init_mm(struct mm_struct *mm)
1026 {
1027 }
1028 
lru_gen_use_mm(struct mm_struct * mm)1029 static inline void lru_gen_use_mm(struct mm_struct *mm)
1030 {
1031 }
1032 
1033 #endif /* CONFIG_LRU_GEN */
1034 
1035 struct vma_iterator {
1036 	struct ma_state mas;
1037 };
1038 
1039 #define VMA_ITERATOR(name, __mm, __addr)				\
1040 	struct vma_iterator name = {					\
1041 		.mas = {						\
1042 			.tree = &(__mm)->mm_mt,				\
1043 			.index = __addr,				\
1044 			.node = MAS_START,				\
1045 		},							\
1046 	}
1047 
vma_iter_init(struct vma_iterator * vmi,struct mm_struct * mm,unsigned long addr)1048 static inline void vma_iter_init(struct vma_iterator *vmi,
1049 		struct mm_struct *mm, unsigned long addr)
1050 {
1051 	mas_init(&vmi->mas, &mm->mm_mt, addr);
1052 }
1053 
1054 #ifdef CONFIG_SCHED_MM_CID
1055 
1056 enum mm_cid_state {
1057 	MM_CID_UNSET = -1U,		/* Unset state has lazy_put flag set. */
1058 	MM_CID_LAZY_PUT = (1U << 31),
1059 };
1060 
mm_cid_is_unset(int cid)1061 static inline bool mm_cid_is_unset(int cid)
1062 {
1063 	return cid == MM_CID_UNSET;
1064 }
1065 
mm_cid_is_lazy_put(int cid)1066 static inline bool mm_cid_is_lazy_put(int cid)
1067 {
1068 	return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1069 }
1070 
mm_cid_is_valid(int cid)1071 static inline bool mm_cid_is_valid(int cid)
1072 {
1073 	return !(cid & MM_CID_LAZY_PUT);
1074 }
1075 
mm_cid_set_lazy_put(int cid)1076 static inline int mm_cid_set_lazy_put(int cid)
1077 {
1078 	return cid | MM_CID_LAZY_PUT;
1079 }
1080 
mm_cid_clear_lazy_put(int cid)1081 static inline int mm_cid_clear_lazy_put(int cid)
1082 {
1083 	return cid & ~MM_CID_LAZY_PUT;
1084 }
1085 
1086 /* Accessor for struct mm_struct's cidmask. */
mm_cidmask(struct mm_struct * mm)1087 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1088 {
1089 	unsigned long cid_bitmap = (unsigned long)mm;
1090 
1091 	cid_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1092 	/* Skip cpu_bitmap */
1093 	cid_bitmap += cpumask_size();
1094 	return (struct cpumask *)cid_bitmap;
1095 }
1096 
mm_init_cid(struct mm_struct * mm)1097 static inline void mm_init_cid(struct mm_struct *mm)
1098 {
1099 	int i;
1100 
1101 	for_each_possible_cpu(i) {
1102 		struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1103 
1104 		pcpu_cid->cid = MM_CID_UNSET;
1105 		pcpu_cid->time = 0;
1106 	}
1107 	cpumask_clear(mm_cidmask(mm));
1108 }
1109 
mm_alloc_cid(struct mm_struct * mm)1110 static inline int mm_alloc_cid(struct mm_struct *mm)
1111 {
1112 	mm->pcpu_cid = alloc_percpu(struct mm_cid);
1113 	if (!mm->pcpu_cid)
1114 		return -ENOMEM;
1115 	mm_init_cid(mm);
1116 	return 0;
1117 }
1118 
mm_destroy_cid(struct mm_struct * mm)1119 static inline void mm_destroy_cid(struct mm_struct *mm)
1120 {
1121 	free_percpu(mm->pcpu_cid);
1122 	mm->pcpu_cid = NULL;
1123 }
1124 
mm_cid_size(void)1125 static inline unsigned int mm_cid_size(void)
1126 {
1127 	return cpumask_size();
1128 }
1129 #else /* CONFIG_SCHED_MM_CID */
mm_init_cid(struct mm_struct * mm)1130 static inline void mm_init_cid(struct mm_struct *mm) { }
mm_alloc_cid(struct mm_struct * mm)1131 static inline int mm_alloc_cid(struct mm_struct *mm) { return 0; }
mm_destroy_cid(struct mm_struct * mm)1132 static inline void mm_destroy_cid(struct mm_struct *mm) { }
mm_cid_size(void)1133 static inline unsigned int mm_cid_size(void)
1134 {
1135 	return 0;
1136 }
1137 #endif /* CONFIG_SCHED_MM_CID */
1138 
1139 struct mmu_gather;
1140 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1141 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1142 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1143 
1144 struct vm_fault;
1145 
1146 /**
1147  * typedef vm_fault_t - Return type for page fault handlers.
1148  *
1149  * Page fault handlers return a bitmask of %VM_FAULT values.
1150  */
1151 typedef __bitwise unsigned int vm_fault_t;
1152 
1153 /**
1154  * enum vm_fault_reason - Page fault handlers return a bitmask of
1155  * these values to tell the core VM what happened when handling the
1156  * fault. Used to decide whether a process gets delivered SIGBUS or
1157  * just gets major/minor fault counters bumped up.
1158  *
1159  * @VM_FAULT_OOM:		Out Of Memory
1160  * @VM_FAULT_SIGBUS:		Bad access
1161  * @VM_FAULT_MAJOR:		Page read from storage
1162  * @VM_FAULT_HWPOISON:		Hit poisoned small page
1163  * @VM_FAULT_HWPOISON_LARGE:	Hit poisoned large page. Index encoded
1164  *				in upper bits
1165  * @VM_FAULT_SIGSEGV:		segmentation fault
1166  * @VM_FAULT_NOPAGE:		->fault installed the pte, not return page
1167  * @VM_FAULT_LOCKED:		->fault locked the returned page
1168  * @VM_FAULT_RETRY:		->fault blocked, must retry
1169  * @VM_FAULT_FALLBACK:		huge page fault failed, fall back to small
1170  * @VM_FAULT_DONE_COW:		->fault has fully handled COW
1171  * @VM_FAULT_NEEDDSYNC:		->fault did not modify page tables and needs
1172  *				fsync() to complete (for synchronous page faults
1173  *				in DAX)
1174  * @VM_FAULT_COMPLETED:		->fault completed, meanwhile mmap lock released
1175  * @VM_FAULT_HINDEX_MASK:	mask HINDEX value
1176  *
1177  */
1178 enum vm_fault_reason {
1179 	VM_FAULT_OOM            = (__force vm_fault_t)0x000001,
1180 	VM_FAULT_SIGBUS         = (__force vm_fault_t)0x000002,
1181 	VM_FAULT_MAJOR          = (__force vm_fault_t)0x000004,
1182 	VM_FAULT_HWPOISON       = (__force vm_fault_t)0x000010,
1183 	VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1184 	VM_FAULT_SIGSEGV        = (__force vm_fault_t)0x000040,
1185 	VM_FAULT_NOPAGE         = (__force vm_fault_t)0x000100,
1186 	VM_FAULT_LOCKED         = (__force vm_fault_t)0x000200,
1187 	VM_FAULT_RETRY          = (__force vm_fault_t)0x000400,
1188 	VM_FAULT_FALLBACK       = (__force vm_fault_t)0x000800,
1189 	VM_FAULT_DONE_COW       = (__force vm_fault_t)0x001000,
1190 	VM_FAULT_NEEDDSYNC      = (__force vm_fault_t)0x002000,
1191 	VM_FAULT_COMPLETED      = (__force vm_fault_t)0x004000,
1192 	VM_FAULT_HINDEX_MASK    = (__force vm_fault_t)0x0f0000,
1193 };
1194 
1195 /* Encode hstate index for a hwpoisoned large page */
1196 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1197 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1198 
1199 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS |	\
1200 			VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON |	\
1201 			VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1202 
1203 #define VM_FAULT_RESULT_TRACE \
1204 	{ VM_FAULT_OOM,                 "OOM" },	\
1205 	{ VM_FAULT_SIGBUS,              "SIGBUS" },	\
1206 	{ VM_FAULT_MAJOR,               "MAJOR" },	\
1207 	{ VM_FAULT_HWPOISON,            "HWPOISON" },	\
1208 	{ VM_FAULT_HWPOISON_LARGE,      "HWPOISON_LARGE" },	\
1209 	{ VM_FAULT_SIGSEGV,             "SIGSEGV" },	\
1210 	{ VM_FAULT_NOPAGE,              "NOPAGE" },	\
1211 	{ VM_FAULT_LOCKED,              "LOCKED" },	\
1212 	{ VM_FAULT_RETRY,               "RETRY" },	\
1213 	{ VM_FAULT_FALLBACK,            "FALLBACK" },	\
1214 	{ VM_FAULT_DONE_COW,            "DONE_COW" },	\
1215 	{ VM_FAULT_NEEDDSYNC,           "NEEDDSYNC" },	\
1216 	{ VM_FAULT_COMPLETED,           "COMPLETED" }
1217 
1218 struct vm_special_mapping {
1219 	const char *name;	/* The name, e.g. "[vdso]". */
1220 
1221 	/*
1222 	 * If .fault is not provided, this points to a
1223 	 * NULL-terminated array of pages that back the special mapping.
1224 	 *
1225 	 * This must not be NULL unless .fault is provided.
1226 	 */
1227 	struct page **pages;
1228 
1229 	/*
1230 	 * If non-NULL, then this is called to resolve page faults
1231 	 * on the special mapping.  If used, .pages is not checked.
1232 	 */
1233 	vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1234 				struct vm_area_struct *vma,
1235 				struct vm_fault *vmf);
1236 
1237 	int (*mremap)(const struct vm_special_mapping *sm,
1238 		     struct vm_area_struct *new_vma);
1239 };
1240 
1241 enum tlb_flush_reason {
1242 	TLB_FLUSH_ON_TASK_SWITCH,
1243 	TLB_REMOTE_SHOOTDOWN,
1244 	TLB_LOCAL_SHOOTDOWN,
1245 	TLB_LOCAL_MM_SHOOTDOWN,
1246 	TLB_REMOTE_SEND_IPI,
1247 	NR_TLB_FLUSH_REASONS,
1248 };
1249 
1250 /**
1251  * enum fault_flag - Fault flag definitions.
1252  * @FAULT_FLAG_WRITE: Fault was a write fault.
1253  * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1254  * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1255  * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1256  * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1257  * @FAULT_FLAG_TRIED: The fault has been tried once.
1258  * @FAULT_FLAG_USER: The fault originated in userspace.
1259  * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1260  * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1261  * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1262  * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1263  *                      COW mapping, making sure that an exclusive anon page is
1264  *                      mapped after the fault.
1265  * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1266  *                        We should only access orig_pte if this flag set.
1267  * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1268  *
1269  * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1270  * whether we would allow page faults to retry by specifying these two
1271  * fault flags correctly.  Currently there can be three legal combinations:
1272  *
1273  * (a) ALLOW_RETRY and !TRIED:  this means the page fault allows retry, and
1274  *                              this is the first try
1275  *
1276  * (b) ALLOW_RETRY and TRIED:   this means the page fault allows retry, and
1277  *                              we've already tried at least once
1278  *
1279  * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1280  *
1281  * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1282  * be used.  Note that page faults can be allowed to retry for multiple times,
1283  * in which case we'll have an initial fault with flags (a) then later on
1284  * continuous faults with flags (b).  We should always try to detect pending
1285  * signals before a retry to make sure the continuous page faults can still be
1286  * interrupted if necessary.
1287  *
1288  * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1289  * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1290  * applied to mappings that are not COW mappings.
1291  */
1292 enum fault_flag {
1293 	FAULT_FLAG_WRITE =		1 << 0,
1294 	FAULT_FLAG_MKWRITE =		1 << 1,
1295 	FAULT_FLAG_ALLOW_RETRY =	1 << 2,
1296 	FAULT_FLAG_RETRY_NOWAIT = 	1 << 3,
1297 	FAULT_FLAG_KILLABLE =		1 << 4,
1298 	FAULT_FLAG_TRIED = 		1 << 5,
1299 	FAULT_FLAG_USER =		1 << 6,
1300 	FAULT_FLAG_REMOTE =		1 << 7,
1301 	FAULT_FLAG_INSTRUCTION =	1 << 8,
1302 	FAULT_FLAG_INTERRUPTIBLE =	1 << 9,
1303 	FAULT_FLAG_UNSHARE =		1 << 10,
1304 	FAULT_FLAG_ORIG_PTE_VALID =	1 << 11,
1305 	FAULT_FLAG_VMA_LOCK =		1 << 12,
1306 };
1307 
1308 typedef unsigned int __bitwise zap_flags_t;
1309 
1310 /*
1311  * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1312  * other. Here is what they mean, and how to use them:
1313  *
1314  *
1315  * FIXME: For pages which are part of a filesystem, mappings are subject to the
1316  * lifetime enforced by the filesystem and we need guarantees that longterm
1317  * users like RDMA and V4L2 only establish mappings which coordinate usage with
1318  * the filesystem.  Ideas for this coordination include revoking the longterm
1319  * pin, delaying writeback, bounce buffer page writeback, etc.  As FS DAX was
1320  * added after the problem with filesystems was found FS DAX VMAs are
1321  * specifically failed.  Filesystem pages are still subject to bugs and use of
1322  * FOLL_LONGTERM should be avoided on those pages.
1323  *
1324  * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1325  * that region.  And so, CMA attempts to migrate the page before pinning, when
1326  * FOLL_LONGTERM is specified.
1327  *
1328  * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1329  * but an additional pin counting system) will be invoked. This is intended for
1330  * anything that gets a page reference and then touches page data (for example,
1331  * Direct IO). This lets the filesystem know that some non-file-system entity is
1332  * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1333  * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1334  * a call to unpin_user_page().
1335  *
1336  * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1337  * and separate refcounting mechanisms, however, and that means that each has
1338  * its own acquire and release mechanisms:
1339  *
1340  *     FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1341  *
1342  *     FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1343  *
1344  * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1345  * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1346  * calls applied to them, and that's perfectly OK. This is a constraint on the
1347  * callers, not on the pages.)
1348  *
1349  * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1350  * directly by the caller. That's in order to help avoid mismatches when
1351  * releasing pages: get_user_pages*() pages must be released via put_page(),
1352  * while pin_user_pages*() pages must be released via unpin_user_page().
1353  *
1354  * Please see Documentation/core-api/pin_user_pages.rst for more information.
1355  */
1356 
1357 enum {
1358 	/* check pte is writable */
1359 	FOLL_WRITE = 1 << 0,
1360 	/* do get_page on page */
1361 	FOLL_GET = 1 << 1,
1362 	/* give error on hole if it would be zero */
1363 	FOLL_DUMP = 1 << 2,
1364 	/* get_user_pages read/write w/o permission */
1365 	FOLL_FORCE = 1 << 3,
1366 	/*
1367 	 * if a disk transfer is needed, start the IO and return without waiting
1368 	 * upon it
1369 	 */
1370 	FOLL_NOWAIT = 1 << 4,
1371 	/* do not fault in pages */
1372 	FOLL_NOFAULT = 1 << 5,
1373 	/* check page is hwpoisoned */
1374 	FOLL_HWPOISON = 1 << 6,
1375 	/* don't do file mappings */
1376 	FOLL_ANON = 1 << 7,
1377 	/*
1378 	 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1379 	 * time period _often_ under userspace control.  This is in contrast to
1380 	 * iov_iter_get_pages(), whose usages are transient.
1381 	 */
1382 	FOLL_LONGTERM = 1 << 8,
1383 	/* split huge pmd before returning */
1384 	FOLL_SPLIT_PMD = 1 << 9,
1385 	/* allow returning PCI P2PDMA pages */
1386 	FOLL_PCI_P2PDMA = 1 << 10,
1387 	/* allow interrupts from generic signals */
1388 	FOLL_INTERRUPTIBLE = 1 << 11,
1389 	/*
1390 	 * Always honor (trigger) NUMA hinting faults.
1391 	 *
1392 	 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1393 	 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1394 	 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1395 	 * hinting faults.
1396 	 */
1397 	FOLL_HONOR_NUMA_FAULT = 1 << 12,
1398 
1399 	/* See also internal only FOLL flags in mm/internal.h */
1400 };
1401 
1402 #endif /* _LINUX_MM_TYPES_H */
1403