xref: /linux-6.15/include/linux/mm_types.h (revision 6bef4c2f)
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 #include <linux/types.h>
23 
24 #include <asm/mmu.h>
25 
26 #ifndef AT_VECTOR_SIZE_ARCH
27 #define AT_VECTOR_SIZE_ARCH 0
28 #endif
29 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
30 
31 #define INIT_PASID	0
32 
33 struct address_space;
34 struct mem_cgroup;
35 
36 /*
37  * Each physical page in the system has a struct page associated with
38  * it to keep track of whatever it is we are using the page for at the
39  * moment. Note that we have no way to track which tasks are using
40  * a page, though if it is a pagecache page, rmap structures can tell us
41  * who is mapping it.
42  *
43  * If you allocate the page using alloc_pages(), you can use some of the
44  * space in struct page for your own purposes.  The five words in the main
45  * union are available, except for bit 0 of the first word which must be
46  * kept clear.  Many users use this word to store a pointer to an object
47  * which is guaranteed to be aligned.  If you use the same storage as
48  * page->mapping, you must restore it to NULL before freeing the page.
49  *
50  * The mapcount field must not be used for own purposes.
51  *
52  * If you want to use the refcount field, it must be used in such a way
53  * that other CPUs temporarily incrementing and then decrementing the
54  * refcount does not cause problems.  On receiving the page from
55  * alloc_pages(), the refcount will be positive.
56  *
57  * If you allocate pages of order > 0, you can use some of the fields
58  * in each subpage, but you may need to restore some of their values
59  * afterwards.
60  *
61  * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
62  * That requires that freelist & counters in struct slab be adjacent and
63  * double-word aligned. Because struct slab currently just reinterprets the
64  * bits of struct page, we align all struct pages to double-word boundaries,
65  * and ensure that 'freelist' is aligned within struct slab.
66  */
67 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
68 #define _struct_page_alignment	__aligned(2 * sizeof(unsigned long))
69 #else
70 #define _struct_page_alignment	__aligned(sizeof(unsigned long))
71 #endif
72 
73 struct page {
74 	unsigned long flags;		/* Atomic flags, some possibly
75 					 * updated asynchronously */
76 	/*
77 	 * Five words (20/40 bytes) are available in this union.
78 	 * WARNING: bit 0 of the first word is used for PageTail(). That
79 	 * means the other users of this union MUST NOT use the bit to
80 	 * avoid collision and false-positive PageTail().
81 	 */
82 	union {
83 		struct {	/* Page cache and anonymous pages */
84 			/**
85 			 * @lru: Pageout list, eg. active_list protected by
86 			 * lruvec->lru_lock.  Sometimes used as a generic list
87 			 * by the page owner.
88 			 */
89 			union {
90 				struct list_head lru;
91 
92 				/* Or, for the Unevictable "LRU list" slot */
93 				struct {
94 					/* Always even, to negate PageTail */
95 					void *__filler;
96 					/* Count page's or folio's mlocks */
97 					unsigned int mlock_count;
98 				};
99 
100 				/* Or, free page */
101 				struct list_head buddy_list;
102 				struct list_head pcp_list;
103 			};
104 			/* See page-flags.h for PAGE_MAPPING_FLAGS */
105 			struct address_space *mapping;
106 			union {
107 				pgoff_t index;		/* Our offset within mapping. */
108 				unsigned long share;	/* share count for fsdax */
109 			};
110 			/**
111 			 * @private: Mapping-private opaque data.
112 			 * Usually used for buffer_heads if PagePrivate.
113 			 * Used for swp_entry_t if swapcache flag set.
114 			 * Indicates order in the buddy system if PageBuddy.
115 			 */
116 			unsigned long private;
117 		};
118 		struct {	/* page_pool used by netstack */
119 			/**
120 			 * @pp_magic: magic value to avoid recycling non
121 			 * page_pool allocated pages.
122 			 */
123 			unsigned long pp_magic;
124 			struct page_pool *pp;
125 			unsigned long _pp_mapping_pad;
126 			unsigned long dma_addr;
127 			atomic_long_t pp_ref_count;
128 		};
129 		struct {	/* Tail pages of compound page */
130 			unsigned long compound_head;	/* Bit zero is set */
131 		};
132 		struct {	/* ZONE_DEVICE pages */
133 			/** @pgmap: Points to the hosting device page map. */
134 			struct dev_pagemap *pgmap;
135 			void *zone_device_data;
136 			/*
137 			 * ZONE_DEVICE private pages are counted as being
138 			 * mapped so the next 3 words hold the mapping, index,
139 			 * and private fields from the source anonymous or
140 			 * page cache page while the page is migrated to device
141 			 * private memory.
142 			 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
143 			 * use the mapping, index, and private fields when
144 			 * pmem backed DAX files are mapped.
145 			 */
146 		};
147 
148 		/** @rcu_head: You can use this to free a page by RCU. */
149 		struct rcu_head rcu_head;
150 	};
151 
152 	union {		/* This union is 4 bytes in size. */
153 		/*
154 		 * For head pages of typed folios, the value stored here
155 		 * allows for determining what this page is used for. The
156 		 * tail pages of typed folios will not store a type
157 		 * (page_type == _mapcount == -1).
158 		 *
159 		 * See page-flags.h for a list of page types which are currently
160 		 * stored here.
161 		 *
162 		 * Owners of typed folios may reuse the lower 16 bit of the
163 		 * head page page_type field after setting the page type,
164 		 * but must reset these 16 bit to -1 before clearing the
165 		 * page type.
166 		 */
167 		unsigned int page_type;
168 
169 		/*
170 		 * For pages that are part of non-typed folios for which mappings
171 		 * are tracked via the RMAP, encodes the number of times this page
172 		 * is directly referenced by a page table.
173 		 *
174 		 * Note that the mapcount is always initialized to -1, so that
175 		 * transitions both from it and to it can be tracked, using
176 		 * atomic_inc_and_test() and atomic_add_negative(-1).
177 		 */
178 		atomic_t _mapcount;
179 	};
180 
181 	/* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
182 	atomic_t _refcount;
183 
184 #ifdef CONFIG_MEMCG
185 	unsigned long memcg_data;
186 #elif defined(CONFIG_SLAB_OBJ_EXT)
187 	unsigned long _unused_slab_obj_exts;
188 #endif
189 
190 	/*
191 	 * On machines where all RAM is mapped into kernel address space,
192 	 * we can simply calculate the virtual address. On machines with
193 	 * highmem some memory is mapped into kernel virtual memory
194 	 * dynamically, so we need a place to store that address.
195 	 * Note that this field could be 16 bits on x86 ... ;)
196 	 *
197 	 * Architectures with slow multiplication can define
198 	 * WANT_PAGE_VIRTUAL in asm/page.h
199 	 */
200 #if defined(WANT_PAGE_VIRTUAL)
201 	void *virtual;			/* Kernel virtual address (NULL if
202 					   not kmapped, ie. highmem) */
203 #endif /* WANT_PAGE_VIRTUAL */
204 
205 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
206 	int _last_cpupid;
207 #endif
208 
209 #ifdef CONFIG_KMSAN
210 	/*
211 	 * KMSAN metadata for this page:
212 	 *  - shadow page: every bit indicates whether the corresponding
213 	 *    bit of the original page is initialized (0) or not (1);
214 	 *  - origin page: every 4 bytes contain an id of the stack trace
215 	 *    where the uninitialized value was created.
216 	 */
217 	struct page *kmsan_shadow;
218 	struct page *kmsan_origin;
219 #endif
220 } _struct_page_alignment;
221 
222 /*
223  * struct encoded_page - a nonexistent type marking this pointer
224  *
225  * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
226  * with the low bits of the pointer indicating extra context-dependent
227  * information. Only used in mmu_gather handling, and this acts as a type
228  * system check on that use.
229  *
230  * We only really have two guaranteed bits in general, although you could
231  * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
232  * for more.
233  *
234  * Use the supplied helper functions to endcode/decode the pointer and bits.
235  */
236 struct encoded_page;
237 
238 #define ENCODED_PAGE_BITS			3ul
239 
240 /* Perform rmap removal after we have flushed the TLB. */
241 #define ENCODED_PAGE_BIT_DELAY_RMAP		1ul
242 
243 /*
244  * The next item in an encoded_page array is the "nr_pages" argument, specifying
245  * the number of consecutive pages starting from this page, that all belong to
246  * the same folio. For example, "nr_pages" corresponds to the number of folio
247  * references that must be dropped. If this bit is not set, "nr_pages" is
248  * implicitly 1.
249  */
250 #define ENCODED_PAGE_BIT_NR_PAGES_NEXT		2ul
251 
252 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
253 {
254 	BUILD_BUG_ON(flags > ENCODED_PAGE_BITS);
255 	return (struct encoded_page *)(flags | (unsigned long)page);
256 }
257 
258 static inline unsigned long encoded_page_flags(struct encoded_page *page)
259 {
260 	return ENCODED_PAGE_BITS & (unsigned long)page;
261 }
262 
263 static inline struct page *encoded_page_ptr(struct encoded_page *page)
264 {
265 	return (struct page *)(~ENCODED_PAGE_BITS & (unsigned long)page);
266 }
267 
268 static __always_inline struct encoded_page *encode_nr_pages(unsigned long nr)
269 {
270 	VM_WARN_ON_ONCE((nr << 2) >> 2 != nr);
271 	return (struct encoded_page *)(nr << 2);
272 }
273 
274 static __always_inline unsigned long encoded_nr_pages(struct encoded_page *page)
275 {
276 	return ((unsigned long)page) >> 2;
277 }
278 
279 /*
280  * A swap entry has to fit into a "unsigned long", as the entry is hidden
281  * in the "index" field of the swapper address space.
282  */
283 typedef struct {
284 	unsigned long val;
285 } swp_entry_t;
286 
287 /**
288  * struct folio - Represents a contiguous set of bytes.
289  * @flags: Identical to the page flags.
290  * @lru: Least Recently Used list; tracks how recently this folio was used.
291  * @mlock_count: Number of times this folio has been pinned by mlock().
292  * @mapping: The file this page belongs to, or refers to the anon_vma for
293  *    anonymous memory.
294  * @index: Offset within the file, in units of pages.  For anonymous memory,
295  *    this is the index from the beginning of the mmap.
296  * @private: Filesystem per-folio data (see folio_attach_private()).
297  * @swap: Used for swp_entry_t if folio_test_swapcache().
298  * @_mapcount: Do not access this member directly.  Use folio_mapcount() to
299  *    find out how many times this folio is mapped by userspace.
300  * @_refcount: Do not access this member directly.  Use folio_ref_count()
301  *    to find how many references there are to this folio.
302  * @memcg_data: Memory Control Group data.
303  * @virtual: Virtual address in the kernel direct map.
304  * @_last_cpupid: IDs of last CPU and last process that accessed the folio.
305  * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
306  * @_large_mapcount: Do not use directly, call folio_mapcount().
307  * @_nr_pages_mapped: Do not use outside of rmap and debug code.
308  * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
309  * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
310  * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
311  * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
312  * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
313  * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
314  * @_deferred_list: Folios to be split under memory pressure.
315  * @_unused_slab_obj_exts: Placeholder to match obj_exts in struct slab.
316  *
317  * A folio is a physically, virtually and logically contiguous set
318  * of bytes.  It is a power-of-two in size, and it is aligned to that
319  * same power-of-two.  It is at least as large as %PAGE_SIZE.  If it is
320  * in the page cache, it is at a file offset which is a multiple of that
321  * power-of-two.  It may be mapped into userspace at an address which is
322  * at an arbitrary page offset, but its kernel virtual address is aligned
323  * to its size.
324  */
325 struct folio {
326 	/* private: don't document the anon union */
327 	union {
328 		struct {
329 	/* public: */
330 			unsigned long flags;
331 			union {
332 				struct list_head lru;
333 	/* private: avoid cluttering the output */
334 				struct {
335 					void *__filler;
336 	/* public: */
337 					unsigned int mlock_count;
338 	/* private: */
339 				};
340 	/* public: */
341 			};
342 			struct address_space *mapping;
343 			pgoff_t index;
344 			union {
345 				void *private;
346 				swp_entry_t swap;
347 			};
348 			atomic_t _mapcount;
349 			atomic_t _refcount;
350 #ifdef CONFIG_MEMCG
351 			unsigned long memcg_data;
352 #elif defined(CONFIG_SLAB_OBJ_EXT)
353 			unsigned long _unused_slab_obj_exts;
354 #endif
355 #if defined(WANT_PAGE_VIRTUAL)
356 			void *virtual;
357 #endif
358 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
359 			int _last_cpupid;
360 #endif
361 	/* private: the union with struct page is transitional */
362 		};
363 		struct page page;
364 	};
365 	union {
366 		struct {
367 			unsigned long _flags_1;
368 			unsigned long _head_1;
369 	/* public: */
370 			atomic_t _large_mapcount;
371 			atomic_t _entire_mapcount;
372 			atomic_t _nr_pages_mapped;
373 			atomic_t _pincount;
374 #ifdef CONFIG_64BIT
375 			unsigned int _folio_nr_pages;
376 #endif
377 	/* private: the union with struct page is transitional */
378 		};
379 		struct page __page_1;
380 	};
381 	union {
382 		struct {
383 			unsigned long _flags_2;
384 			unsigned long _head_2;
385 	/* public: */
386 			void *_hugetlb_subpool;
387 			void *_hugetlb_cgroup;
388 			void *_hugetlb_cgroup_rsvd;
389 			void *_hugetlb_hwpoison;
390 	/* private: the union with struct page is transitional */
391 		};
392 		struct {
393 			unsigned long _flags_2a;
394 			unsigned long _head_2a;
395 	/* public: */
396 			struct list_head _deferred_list;
397 	/* private: the union with struct page is transitional */
398 		};
399 		struct page __page_2;
400 	};
401 };
402 
403 #define FOLIO_MATCH(pg, fl)						\
404 	static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
405 FOLIO_MATCH(flags, flags);
406 FOLIO_MATCH(lru, lru);
407 FOLIO_MATCH(mapping, mapping);
408 FOLIO_MATCH(compound_head, lru);
409 FOLIO_MATCH(index, index);
410 FOLIO_MATCH(private, private);
411 FOLIO_MATCH(_mapcount, _mapcount);
412 FOLIO_MATCH(_refcount, _refcount);
413 #ifdef CONFIG_MEMCG
414 FOLIO_MATCH(memcg_data, memcg_data);
415 #endif
416 #if defined(WANT_PAGE_VIRTUAL)
417 FOLIO_MATCH(virtual, virtual);
418 #endif
419 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
420 FOLIO_MATCH(_last_cpupid, _last_cpupid);
421 #endif
422 #undef FOLIO_MATCH
423 #define FOLIO_MATCH(pg, fl)						\
424 	static_assert(offsetof(struct folio, fl) ==			\
425 			offsetof(struct page, pg) + sizeof(struct page))
426 FOLIO_MATCH(flags, _flags_1);
427 FOLIO_MATCH(compound_head, _head_1);
428 #undef FOLIO_MATCH
429 #define FOLIO_MATCH(pg, fl)						\
430 	static_assert(offsetof(struct folio, fl) ==			\
431 			offsetof(struct page, pg) + 2 * sizeof(struct page))
432 FOLIO_MATCH(flags, _flags_2);
433 FOLIO_MATCH(compound_head, _head_2);
434 FOLIO_MATCH(flags, _flags_2a);
435 FOLIO_MATCH(compound_head, _head_2a);
436 #undef FOLIO_MATCH
437 
438 /**
439  * struct ptdesc -    Memory descriptor for page tables.
440  * @__page_flags:     Same as page flags. Powerpc only.
441  * @pt_rcu_head:      For freeing page table pages.
442  * @pt_list:          List of used page tables. Used for s390 gmap shadow pages
443  *                    (which are not linked into the user page tables) and x86
444  *                    pgds.
445  * @_pt_pad_1:        Padding that aliases with page's compound head.
446  * @pmd_huge_pte:     Protected by ptdesc->ptl, used for THPs.
447  * @__page_mapping:   Aliases with page->mapping. Unused for page tables.
448  * @pt_index:         Used for s390 gmap.
449  * @pt_mm:            Used for x86 pgds.
450  * @pt_frag_refcount: For fragmented page table tracking. Powerpc only.
451  * @pt_share_count:   Used for HugeTLB PMD page table share count.
452  * @_pt_pad_2:        Padding to ensure proper alignment.
453  * @ptl:              Lock for the page table.
454  * @__page_type:      Same as page->page_type. Unused for page tables.
455  * @__page_refcount:  Same as page refcount.
456  * @pt_memcg_data:    Memcg data. Tracked for page tables here.
457  *
458  * This struct overlays struct page for now. Do not modify without a good
459  * understanding of the issues.
460  */
461 struct ptdesc {
462 	unsigned long __page_flags;
463 
464 	union {
465 		struct rcu_head pt_rcu_head;
466 		struct list_head pt_list;
467 		struct {
468 			unsigned long _pt_pad_1;
469 			pgtable_t pmd_huge_pte;
470 		};
471 	};
472 	unsigned long __page_mapping;
473 
474 	union {
475 		pgoff_t pt_index;
476 		struct mm_struct *pt_mm;
477 		atomic_t pt_frag_refcount;
478 #ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
479 		atomic_t pt_share_count;
480 #endif
481 	};
482 
483 	union {
484 		unsigned long _pt_pad_2;
485 #if ALLOC_SPLIT_PTLOCKS
486 		spinlock_t *ptl;
487 #else
488 		spinlock_t ptl;
489 #endif
490 	};
491 	unsigned int __page_type;
492 	atomic_t __page_refcount;
493 #ifdef CONFIG_MEMCG
494 	unsigned long pt_memcg_data;
495 #endif
496 };
497 
498 #define TABLE_MATCH(pg, pt)						\
499 	static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
500 TABLE_MATCH(flags, __page_flags);
501 TABLE_MATCH(compound_head, pt_list);
502 TABLE_MATCH(compound_head, _pt_pad_1);
503 TABLE_MATCH(mapping, __page_mapping);
504 TABLE_MATCH(index, pt_index);
505 TABLE_MATCH(rcu_head, pt_rcu_head);
506 TABLE_MATCH(page_type, __page_type);
507 TABLE_MATCH(_refcount, __page_refcount);
508 #ifdef CONFIG_MEMCG
509 TABLE_MATCH(memcg_data, pt_memcg_data);
510 #endif
511 #undef TABLE_MATCH
512 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
513 
514 #define ptdesc_page(pt)			(_Generic((pt),			\
515 	const struct ptdesc *:		(const struct page *)(pt),	\
516 	struct ptdesc *:		(struct page *)(pt)))
517 
518 #define ptdesc_folio(pt)		(_Generic((pt),			\
519 	const struct ptdesc *:		(const struct folio *)(pt),	\
520 	struct ptdesc *:		(struct folio *)(pt)))
521 
522 #define page_ptdesc(p)			(_Generic((p),			\
523 	const struct page *:		(const struct ptdesc *)(p),	\
524 	struct page *:			(struct ptdesc *)(p)))
525 
526 #ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
527 static inline void ptdesc_pmd_pts_init(struct ptdesc *ptdesc)
528 {
529 	atomic_set(&ptdesc->pt_share_count, 0);
530 }
531 
532 static inline void ptdesc_pmd_pts_inc(struct ptdesc *ptdesc)
533 {
534 	atomic_inc(&ptdesc->pt_share_count);
535 }
536 
537 static inline void ptdesc_pmd_pts_dec(struct ptdesc *ptdesc)
538 {
539 	atomic_dec(&ptdesc->pt_share_count);
540 }
541 
542 static inline int ptdesc_pmd_pts_count(struct ptdesc *ptdesc)
543 {
544 	return atomic_read(&ptdesc->pt_share_count);
545 }
546 #else
547 static inline void ptdesc_pmd_pts_init(struct ptdesc *ptdesc)
548 {
549 }
550 #endif
551 
552 /*
553  * Used for sizing the vmemmap region on some architectures
554  */
555 #define STRUCT_PAGE_MAX_SHIFT	(order_base_2(sizeof(struct page)))
556 
557 /*
558  * page_private can be used on tail pages.  However, PagePrivate is only
559  * checked by the VM on the head page.  So page_private on the tail pages
560  * should be used for data that's ancillary to the head page (eg attaching
561  * buffer heads to tail pages after attaching buffer heads to the head page)
562  */
563 #define page_private(page)		((page)->private)
564 
565 static inline void set_page_private(struct page *page, unsigned long private)
566 {
567 	page->private = private;
568 }
569 
570 static inline void *folio_get_private(struct folio *folio)
571 {
572 	return folio->private;
573 }
574 
575 typedef unsigned long vm_flags_t;
576 
577 /*
578  * A region containing a mapping of a non-memory backed file under NOMMU
579  * conditions.  These are held in a global tree and are pinned by the VMAs that
580  * map parts of them.
581  */
582 struct vm_region {
583 	struct rb_node	vm_rb;		/* link in global region tree */
584 	vm_flags_t	vm_flags;	/* VMA vm_flags */
585 	unsigned long	vm_start;	/* start address of region */
586 	unsigned long	vm_end;		/* region initialised to here */
587 	unsigned long	vm_top;		/* region allocated to here */
588 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
589 	struct file	*vm_file;	/* the backing file or NULL */
590 
591 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
592 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
593 						* this region */
594 };
595 
596 #ifdef CONFIG_USERFAULTFD
597 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
598 struct vm_userfaultfd_ctx {
599 	struct userfaultfd_ctx *ctx;
600 };
601 #else /* CONFIG_USERFAULTFD */
602 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
603 struct vm_userfaultfd_ctx {};
604 #endif /* CONFIG_USERFAULTFD */
605 
606 struct anon_vma_name {
607 	struct kref kref;
608 	/* The name needs to be at the end because it is dynamically sized. */
609 	char name[];
610 };
611 
612 #ifdef CONFIG_ANON_VMA_NAME
613 /*
614  * mmap_lock should be read-locked when calling anon_vma_name(). Caller should
615  * either keep holding the lock while using the returned pointer or it should
616  * raise anon_vma_name refcount before releasing the lock.
617  */
618 struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma);
619 struct anon_vma_name *anon_vma_name_alloc(const char *name);
620 void anon_vma_name_free(struct kref *kref);
621 #else /* CONFIG_ANON_VMA_NAME */
622 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
623 {
624 	return NULL;
625 }
626 
627 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
628 {
629 	return NULL;
630 }
631 #endif
632 
633 #define VMA_LOCK_OFFSET	0x40000000
634 #define VMA_REF_LIMIT	(VMA_LOCK_OFFSET - 1)
635 
636 struct vma_numab_state {
637 	/*
638 	 * Initialised as time in 'jiffies' after which VMA
639 	 * should be scanned.  Delays first scan of new VMA by at
640 	 * least sysctl_numa_balancing_scan_delay:
641 	 */
642 	unsigned long next_scan;
643 
644 	/*
645 	 * Time in jiffies when pids_active[] is reset to
646 	 * detect phase change behaviour:
647 	 */
648 	unsigned long pids_active_reset;
649 
650 	/*
651 	 * Approximate tracking of PIDs that trapped a NUMA hinting
652 	 * fault. May produce false positives due to hash collisions.
653 	 *
654 	 *   [0] Previous PID tracking
655 	 *   [1] Current PID tracking
656 	 *
657 	 * Window moves after next_pid_reset has expired approximately
658 	 * every VMA_PID_RESET_PERIOD jiffies:
659 	 */
660 	unsigned long pids_active[2];
661 
662 	/* MM scan sequence ID when scan first started after VMA creation */
663 	int start_scan_seq;
664 
665 	/*
666 	 * MM scan sequence ID when the VMA was last completely scanned.
667 	 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq
668 	 */
669 	int prev_scan_seq;
670 };
671 
672 /*
673  * This struct describes a virtual memory area. There is one of these
674  * per VM-area/task. A VM area is any part of the process virtual memory
675  * space that has a special rule for the page-fault handlers (ie a shared
676  * library, the executable area etc).
677  *
678  * Only explicitly marked struct members may be accessed by RCU readers before
679  * getting a stable reference.
680  */
681 struct vm_area_struct {
682 	/* The first cache line has the info for VMA tree walking. */
683 
684 	union {
685 		struct {
686 			/* VMA covers [vm_start; vm_end) addresses within mm */
687 			unsigned long vm_start;
688 			unsigned long vm_end;
689 		};
690 #ifdef CONFIG_PER_VMA_LOCK
691 		struct rcu_head vm_rcu;	/* Used for deferred freeing. */
692 #endif
693 	};
694 
695 	/*
696 	 * The address space we belong to.
697 	 * Unstable RCU readers are allowed to read this.
698 	 */
699 	struct mm_struct *vm_mm;
700 	pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
701 
702 	/*
703 	 * Flags, see mm.h.
704 	 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
705 	 */
706 	union {
707 		const vm_flags_t vm_flags;
708 		vm_flags_t __private __vm_flags;
709 	};
710 
711 #ifdef CONFIG_PER_VMA_LOCK
712 	/*
713 	 * Can only be written (using WRITE_ONCE()) while holding both:
714 	 *  - mmap_lock (in write mode)
715 	 *  - vm_refcnt bit at VMA_LOCK_OFFSET is set
716 	 * Can be read reliably while holding one of:
717 	 *  - mmap_lock (in read or write mode)
718 	 *  - vm_refcnt bit at VMA_LOCK_OFFSET is set or vm_refcnt > 1
719 	 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
720 	 * while holding nothing (except RCU to keep the VMA struct allocated).
721 	 *
722 	 * This sequence counter is explicitly allowed to overflow; sequence
723 	 * counter reuse can only lead to occasional unnecessary use of the
724 	 * slowpath.
725 	 */
726 	unsigned int vm_lock_seq;
727 #endif
728 	/*
729 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
730 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
731 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
732 	 * or brk vma (with NULL file) can only be in an anon_vma list.
733 	 */
734 	struct list_head anon_vma_chain; /* Serialized by mmap_lock &
735 					  * page_table_lock */
736 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
737 
738 	/* Function pointers to deal with this struct. */
739 	const struct vm_operations_struct *vm_ops;
740 
741 	/* Information about our backing store: */
742 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
743 					   units */
744 	struct file * vm_file;		/* File we map to (can be NULL). */
745 	void * vm_private_data;		/* was vm_pte (shared mem) */
746 
747 #ifdef CONFIG_SWAP
748 	atomic_long_t swap_readahead_info;
749 #endif
750 #ifndef CONFIG_MMU
751 	struct vm_region *vm_region;	/* NOMMU mapping region */
752 #endif
753 #ifdef CONFIG_NUMA
754 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
755 #endif
756 #ifdef CONFIG_NUMA_BALANCING
757 	struct vma_numab_state *numab_state;	/* NUMA Balancing state */
758 #endif
759 #ifdef CONFIG_PER_VMA_LOCK
760 	/* Unstable RCU readers are allowed to read this. */
761 	refcount_t vm_refcnt ____cacheline_aligned_in_smp;
762 #ifdef CONFIG_DEBUG_LOCK_ALLOC
763 	struct lockdep_map vmlock_dep_map;
764 #endif
765 #endif
766 	/*
767 	 * For areas with an address space and backing store,
768 	 * linkage into the address_space->i_mmap interval tree.
769 	 *
770 	 */
771 	struct {
772 		struct rb_node rb;
773 		unsigned long rb_subtree_last;
774 	} shared;
775 #ifdef CONFIG_ANON_VMA_NAME
776 	/*
777 	 * For private and shared anonymous mappings, a pointer to a null
778 	 * terminated string containing the name given to the vma, or NULL if
779 	 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
780 	 */
781 	struct anon_vma_name *anon_name;
782 #endif
783 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
784 } __randomize_layout;
785 
786 #ifdef CONFIG_NUMA
787 #define vma_policy(vma) ((vma)->vm_policy)
788 #else
789 #define vma_policy(vma) NULL
790 #endif
791 
792 #ifdef CONFIG_SCHED_MM_CID
793 struct mm_cid {
794 	u64 time;
795 	int cid;
796 	int recent_cid;
797 };
798 #endif
799 
800 struct kioctx_table;
801 struct iommu_mm_data;
802 struct mm_struct {
803 	struct {
804 		/*
805 		 * Fields which are often written to are placed in a separate
806 		 * cache line.
807 		 */
808 		struct {
809 			/**
810 			 * @mm_count: The number of references to &struct
811 			 * mm_struct (@mm_users count as 1).
812 			 *
813 			 * Use mmgrab()/mmdrop() to modify. When this drops to
814 			 * 0, the &struct mm_struct is freed.
815 			 */
816 			atomic_t mm_count;
817 		} ____cacheline_aligned_in_smp;
818 
819 		struct maple_tree mm_mt;
820 
821 		unsigned long mmap_base;	/* base of mmap area */
822 		unsigned long mmap_legacy_base;	/* base of mmap area in bottom-up allocations */
823 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
824 		/* Base addresses for compatible mmap() */
825 		unsigned long mmap_compat_base;
826 		unsigned long mmap_compat_legacy_base;
827 #endif
828 		unsigned long task_size;	/* size of task vm space */
829 		pgd_t * pgd;
830 
831 #ifdef CONFIG_MEMBARRIER
832 		/**
833 		 * @membarrier_state: Flags controlling membarrier behavior.
834 		 *
835 		 * This field is close to @pgd to hopefully fit in the same
836 		 * cache-line, which needs to be touched by switch_mm().
837 		 */
838 		atomic_t membarrier_state;
839 #endif
840 
841 		/**
842 		 * @mm_users: The number of users including userspace.
843 		 *
844 		 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
845 		 * drops to 0 (i.e. when the task exits and there are no other
846 		 * temporary reference holders), we also release a reference on
847 		 * @mm_count (which may then free the &struct mm_struct if
848 		 * @mm_count also drops to 0).
849 		 */
850 		atomic_t mm_users;
851 
852 #ifdef CONFIG_SCHED_MM_CID
853 		/**
854 		 * @pcpu_cid: Per-cpu current cid.
855 		 *
856 		 * Keep track of the currently allocated mm_cid for each cpu.
857 		 * The per-cpu mm_cid values are serialized by their respective
858 		 * runqueue locks.
859 		 */
860 		struct mm_cid __percpu *pcpu_cid;
861 		/*
862 		 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
863 		 *
864 		 * When the next mm_cid scan is due (in jiffies).
865 		 */
866 		unsigned long mm_cid_next_scan;
867 		/**
868 		 * @nr_cpus_allowed: Number of CPUs allowed for mm.
869 		 *
870 		 * Number of CPUs allowed in the union of all mm's
871 		 * threads allowed CPUs.
872 		 */
873 		unsigned int nr_cpus_allowed;
874 		/**
875 		 * @max_nr_cid: Maximum number of allowed concurrency
876 		 *              IDs allocated.
877 		 *
878 		 * Track the highest number of allowed concurrency IDs
879 		 * allocated for the mm.
880 		 */
881 		atomic_t max_nr_cid;
882 		/**
883 		 * @cpus_allowed_lock: Lock protecting mm cpus_allowed.
884 		 *
885 		 * Provide mutual exclusion for mm cpus_allowed and
886 		 * mm nr_cpus_allowed updates.
887 		 */
888 		raw_spinlock_t cpus_allowed_lock;
889 #endif
890 #ifdef CONFIG_MMU
891 		atomic_long_t pgtables_bytes;	/* size of all page tables */
892 #endif
893 		int map_count;			/* number of VMAs */
894 
895 		spinlock_t page_table_lock; /* Protects page tables and some
896 					     * counters
897 					     */
898 		/*
899 		 * With some kernel config, the current mmap_lock's offset
900 		 * inside 'mm_struct' is at 0x120, which is very optimal, as
901 		 * its two hot fields 'count' and 'owner' sit in 2 different
902 		 * cachelines,  and when mmap_lock is highly contended, both
903 		 * of the 2 fields will be accessed frequently, current layout
904 		 * will help to reduce cache bouncing.
905 		 *
906 		 * So please be careful with adding new fields before
907 		 * mmap_lock, which can easily push the 2 fields into one
908 		 * cacheline.
909 		 */
910 		struct rw_semaphore mmap_lock;
911 
912 		struct list_head mmlist; /* List of maybe swapped mm's.	These
913 					  * are globally strung together off
914 					  * init_mm.mmlist, and are protected
915 					  * by mmlist_lock
916 					  */
917 #ifdef CONFIG_PER_VMA_LOCK
918 		struct rcuwait vma_writer_wait;
919 		/*
920 		 * This field has lock-like semantics, meaning it is sometimes
921 		 * accessed with ACQUIRE/RELEASE semantics.
922 		 * Roughly speaking, incrementing the sequence number is
923 		 * equivalent to releasing locks on VMAs; reading the sequence
924 		 * number can be part of taking a read lock on a VMA.
925 		 * Incremented every time mmap_lock is write-locked/unlocked.
926 		 * Initialized to 0, therefore odd values indicate mmap_lock
927 		 * is write-locked and even values that it's released.
928 		 *
929 		 * Can be modified under write mmap_lock using RELEASE
930 		 * semantics.
931 		 * Can be read with no other protection when holding write
932 		 * mmap_lock.
933 		 * Can be read with ACQUIRE semantics if not holding write
934 		 * mmap_lock.
935 		 */
936 		seqcount_t mm_lock_seq;
937 #endif
938 
939 
940 		unsigned long hiwater_rss; /* High-watermark of RSS usage */
941 		unsigned long hiwater_vm;  /* High-water virtual memory usage */
942 
943 		unsigned long total_vm;	   /* Total pages mapped */
944 		unsigned long locked_vm;   /* Pages that have PG_mlocked set */
945 		atomic64_t    pinned_vm;   /* Refcount permanently increased */
946 		unsigned long data_vm;	   /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
947 		unsigned long exec_vm;	   /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
948 		unsigned long stack_vm;	   /* VM_STACK */
949 		unsigned long def_flags;
950 
951 		/**
952 		 * @write_protect_seq: Locked when any thread is write
953 		 * protecting pages mapped by this mm to enforce a later COW,
954 		 * for instance during page table copying for fork().
955 		 */
956 		seqcount_t write_protect_seq;
957 
958 		spinlock_t arg_lock; /* protect the below fields */
959 
960 		unsigned long start_code, end_code, start_data, end_data;
961 		unsigned long start_brk, brk, start_stack;
962 		unsigned long arg_start, arg_end, env_start, env_end;
963 
964 		unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
965 
966 		struct percpu_counter rss_stat[NR_MM_COUNTERS];
967 
968 		struct linux_binfmt *binfmt;
969 
970 		/* Architecture-specific MM context */
971 		mm_context_t context;
972 
973 		unsigned long flags; /* Must use atomic bitops to access */
974 
975 #ifdef CONFIG_AIO
976 		spinlock_t			ioctx_lock;
977 		struct kioctx_table __rcu	*ioctx_table;
978 #endif
979 #ifdef CONFIG_MEMCG
980 		/*
981 		 * "owner" points to a task that is regarded as the canonical
982 		 * user/owner of this mm. All of the following must be true in
983 		 * order for it to be changed:
984 		 *
985 		 * current == mm->owner
986 		 * current->mm != mm
987 		 * new_owner->mm == mm
988 		 * new_owner->alloc_lock is held
989 		 */
990 		struct task_struct __rcu *owner;
991 #endif
992 		struct user_namespace *user_ns;
993 
994 		/* store ref to file /proc/<pid>/exe symlink points to */
995 		struct file __rcu *exe_file;
996 #ifdef CONFIG_MMU_NOTIFIER
997 		struct mmu_notifier_subscriptions *notifier_subscriptions;
998 #endif
999 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1000 		pgtable_t pmd_huge_pte; /* protected by page_table_lock */
1001 #endif
1002 #ifdef CONFIG_NUMA_BALANCING
1003 		/*
1004 		 * numa_next_scan is the next time that PTEs will be remapped
1005 		 * PROT_NONE to trigger NUMA hinting faults; such faults gather
1006 		 * statistics and migrate pages to new nodes if necessary.
1007 		 */
1008 		unsigned long numa_next_scan;
1009 
1010 		/* Restart point for scanning and remapping PTEs. */
1011 		unsigned long numa_scan_offset;
1012 
1013 		/* numa_scan_seq prevents two threads remapping PTEs. */
1014 		int numa_scan_seq;
1015 #endif
1016 		/*
1017 		 * An operation with batched TLB flushing is going on. Anything
1018 		 * that can move process memory needs to flush the TLB when
1019 		 * moving a PROT_NONE mapped page.
1020 		 */
1021 		atomic_t tlb_flush_pending;
1022 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1023 		/* See flush_tlb_batched_pending() */
1024 		atomic_t tlb_flush_batched;
1025 #endif
1026 		struct uprobes_state uprobes_state;
1027 #ifdef CONFIG_PREEMPT_RT
1028 		struct rcu_head delayed_drop;
1029 #endif
1030 #ifdef CONFIG_HUGETLB_PAGE
1031 		atomic_long_t hugetlb_usage;
1032 #endif
1033 		struct work_struct async_put_work;
1034 
1035 #ifdef CONFIG_IOMMU_MM_DATA
1036 		struct iommu_mm_data *iommu_mm;
1037 #endif
1038 #ifdef CONFIG_KSM
1039 		/*
1040 		 * Represent how many pages of this process are involved in KSM
1041 		 * merging (not including ksm_zero_pages).
1042 		 */
1043 		unsigned long ksm_merging_pages;
1044 		/*
1045 		 * Represent how many pages are checked for ksm merging
1046 		 * including merged and not merged.
1047 		 */
1048 		unsigned long ksm_rmap_items;
1049 		/*
1050 		 * Represent how many empty pages are merged with kernel zero
1051 		 * pages when enabling KSM use_zero_pages.
1052 		 */
1053 		atomic_long_t ksm_zero_pages;
1054 #endif /* CONFIG_KSM */
1055 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1056 		struct {
1057 			/* this mm_struct is on lru_gen_mm_list */
1058 			struct list_head list;
1059 			/*
1060 			 * Set when switching to this mm_struct, as a hint of
1061 			 * whether it has been used since the last time per-node
1062 			 * page table walkers cleared the corresponding bits.
1063 			 */
1064 			unsigned long bitmap;
1065 #ifdef CONFIG_MEMCG
1066 			/* points to the memcg of "owner" above */
1067 			struct mem_cgroup *memcg;
1068 #endif
1069 		} lru_gen;
1070 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1071 	} __randomize_layout;
1072 
1073 	/*
1074 	 * The mm_cpumask needs to be at the end of mm_struct, because it
1075 	 * is dynamically sized based on nr_cpu_ids.
1076 	 */
1077 	unsigned long cpu_bitmap[];
1078 };
1079 
1080 #define MM_MT_FLAGS	(MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
1081 			 MT_FLAGS_USE_RCU)
1082 extern struct mm_struct init_mm;
1083 
1084 /* Pointer magic because the dynamic array size confuses some compilers. */
1085 static inline void mm_init_cpumask(struct mm_struct *mm)
1086 {
1087 	unsigned long cpu_bitmap = (unsigned long)mm;
1088 
1089 	cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1090 	cpumask_clear((struct cpumask *)cpu_bitmap);
1091 }
1092 
1093 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
1094 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
1095 {
1096 	return (struct cpumask *)&mm->cpu_bitmap;
1097 }
1098 
1099 #ifdef CONFIG_LRU_GEN
1100 
1101 struct lru_gen_mm_list {
1102 	/* mm_struct list for page table walkers */
1103 	struct list_head fifo;
1104 	/* protects the list above */
1105 	spinlock_t lock;
1106 };
1107 
1108 #endif /* CONFIG_LRU_GEN */
1109 
1110 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1111 
1112 void lru_gen_add_mm(struct mm_struct *mm);
1113 void lru_gen_del_mm(struct mm_struct *mm);
1114 void lru_gen_migrate_mm(struct mm_struct *mm);
1115 
1116 static inline void lru_gen_init_mm(struct mm_struct *mm)
1117 {
1118 	INIT_LIST_HEAD(&mm->lru_gen.list);
1119 	mm->lru_gen.bitmap = 0;
1120 #ifdef CONFIG_MEMCG
1121 	mm->lru_gen.memcg = NULL;
1122 #endif
1123 }
1124 
1125 static inline void lru_gen_use_mm(struct mm_struct *mm)
1126 {
1127 	/*
1128 	 * When the bitmap is set, page reclaim knows this mm_struct has been
1129 	 * used since the last time it cleared the bitmap. So it might be worth
1130 	 * walking the page tables of this mm_struct to clear the accessed bit.
1131 	 */
1132 	WRITE_ONCE(mm->lru_gen.bitmap, -1);
1133 }
1134 
1135 #else /* !CONFIG_LRU_GEN_WALKS_MMU */
1136 
1137 static inline void lru_gen_add_mm(struct mm_struct *mm)
1138 {
1139 }
1140 
1141 static inline void lru_gen_del_mm(struct mm_struct *mm)
1142 {
1143 }
1144 
1145 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
1146 {
1147 }
1148 
1149 static inline void lru_gen_init_mm(struct mm_struct *mm)
1150 {
1151 }
1152 
1153 static inline void lru_gen_use_mm(struct mm_struct *mm)
1154 {
1155 }
1156 
1157 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1158 
1159 struct vma_iterator {
1160 	struct ma_state mas;
1161 };
1162 
1163 #define VMA_ITERATOR(name, __mm, __addr)				\
1164 	struct vma_iterator name = {					\
1165 		.mas = {						\
1166 			.tree = &(__mm)->mm_mt,				\
1167 			.index = __addr,				\
1168 			.node = NULL,					\
1169 			.status = ma_start,				\
1170 		},							\
1171 	}
1172 
1173 static inline void vma_iter_init(struct vma_iterator *vmi,
1174 		struct mm_struct *mm, unsigned long addr)
1175 {
1176 	mas_init(&vmi->mas, &mm->mm_mt, addr);
1177 }
1178 
1179 #ifdef CONFIG_SCHED_MM_CID
1180 
1181 enum mm_cid_state {
1182 	MM_CID_UNSET = -1U,		/* Unset state has lazy_put flag set. */
1183 	MM_CID_LAZY_PUT = (1U << 31),
1184 };
1185 
1186 static inline bool mm_cid_is_unset(int cid)
1187 {
1188 	return cid == MM_CID_UNSET;
1189 }
1190 
1191 static inline bool mm_cid_is_lazy_put(int cid)
1192 {
1193 	return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1194 }
1195 
1196 static inline bool mm_cid_is_valid(int cid)
1197 {
1198 	return !(cid & MM_CID_LAZY_PUT);
1199 }
1200 
1201 static inline int mm_cid_set_lazy_put(int cid)
1202 {
1203 	return cid | MM_CID_LAZY_PUT;
1204 }
1205 
1206 static inline int mm_cid_clear_lazy_put(int cid)
1207 {
1208 	return cid & ~MM_CID_LAZY_PUT;
1209 }
1210 
1211 /*
1212  * mm_cpus_allowed: Union of all mm's threads allowed CPUs.
1213  */
1214 static inline cpumask_t *mm_cpus_allowed(struct mm_struct *mm)
1215 {
1216 	unsigned long bitmap = (unsigned long)mm;
1217 
1218 	bitmap += offsetof(struct mm_struct, cpu_bitmap);
1219 	/* Skip cpu_bitmap */
1220 	bitmap += cpumask_size();
1221 	return (struct cpumask *)bitmap;
1222 }
1223 
1224 /* Accessor for struct mm_struct's cidmask. */
1225 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1226 {
1227 	unsigned long cid_bitmap = (unsigned long)mm_cpus_allowed(mm);
1228 
1229 	/* Skip mm_cpus_allowed */
1230 	cid_bitmap += cpumask_size();
1231 	return (struct cpumask *)cid_bitmap;
1232 }
1233 
1234 static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p)
1235 {
1236 	int i;
1237 
1238 	for_each_possible_cpu(i) {
1239 		struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1240 
1241 		pcpu_cid->cid = MM_CID_UNSET;
1242 		pcpu_cid->recent_cid = MM_CID_UNSET;
1243 		pcpu_cid->time = 0;
1244 	}
1245 	mm->nr_cpus_allowed = p->nr_cpus_allowed;
1246 	atomic_set(&mm->max_nr_cid, 0);
1247 	raw_spin_lock_init(&mm->cpus_allowed_lock);
1248 	cpumask_copy(mm_cpus_allowed(mm), &p->cpus_mask);
1249 	cpumask_clear(mm_cidmask(mm));
1250 }
1251 
1252 static inline int mm_alloc_cid_noprof(struct mm_struct *mm, struct task_struct *p)
1253 {
1254 	mm->pcpu_cid = alloc_percpu_noprof(struct mm_cid);
1255 	if (!mm->pcpu_cid)
1256 		return -ENOMEM;
1257 	mm_init_cid(mm, p);
1258 	return 0;
1259 }
1260 #define mm_alloc_cid(...)	alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__))
1261 
1262 static inline void mm_destroy_cid(struct mm_struct *mm)
1263 {
1264 	free_percpu(mm->pcpu_cid);
1265 	mm->pcpu_cid = NULL;
1266 }
1267 
1268 static inline unsigned int mm_cid_size(void)
1269 {
1270 	return 2 * cpumask_size();	/* mm_cpus_allowed(), mm_cidmask(). */
1271 }
1272 
1273 static inline void mm_set_cpus_allowed(struct mm_struct *mm, const struct cpumask *cpumask)
1274 {
1275 	struct cpumask *mm_allowed = mm_cpus_allowed(mm);
1276 
1277 	if (!mm)
1278 		return;
1279 	/* The mm_cpus_allowed is the union of each thread allowed CPUs masks. */
1280 	raw_spin_lock(&mm->cpus_allowed_lock);
1281 	cpumask_or(mm_allowed, mm_allowed, cpumask);
1282 	WRITE_ONCE(mm->nr_cpus_allowed, cpumask_weight(mm_allowed));
1283 	raw_spin_unlock(&mm->cpus_allowed_lock);
1284 }
1285 #else /* CONFIG_SCHED_MM_CID */
1286 static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p) { }
1287 static inline int mm_alloc_cid(struct mm_struct *mm, struct task_struct *p) { return 0; }
1288 static inline void mm_destroy_cid(struct mm_struct *mm) { }
1289 
1290 static inline unsigned int mm_cid_size(void)
1291 {
1292 	return 0;
1293 }
1294 static inline void mm_set_cpus_allowed(struct mm_struct *mm, const struct cpumask *cpumask) { }
1295 #endif /* CONFIG_SCHED_MM_CID */
1296 
1297 struct mmu_gather;
1298 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1299 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1300 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1301 
1302 struct vm_fault;
1303 
1304 /**
1305  * typedef vm_fault_t - Return type for page fault handlers.
1306  *
1307  * Page fault handlers return a bitmask of %VM_FAULT values.
1308  */
1309 typedef __bitwise unsigned int vm_fault_t;
1310 
1311 /**
1312  * enum vm_fault_reason - Page fault handlers return a bitmask of
1313  * these values to tell the core VM what happened when handling the
1314  * fault. Used to decide whether a process gets delivered SIGBUS or
1315  * just gets major/minor fault counters bumped up.
1316  *
1317  * @VM_FAULT_OOM:		Out Of Memory
1318  * @VM_FAULT_SIGBUS:		Bad access
1319  * @VM_FAULT_MAJOR:		Page read from storage
1320  * @VM_FAULT_HWPOISON:		Hit poisoned small page
1321  * @VM_FAULT_HWPOISON_LARGE:	Hit poisoned large page. Index encoded
1322  *				in upper bits
1323  * @VM_FAULT_SIGSEGV:		segmentation fault
1324  * @VM_FAULT_NOPAGE:		->fault installed the pte, not return page
1325  * @VM_FAULT_LOCKED:		->fault locked the returned page
1326  * @VM_FAULT_RETRY:		->fault blocked, must retry
1327  * @VM_FAULT_FALLBACK:		huge page fault failed, fall back to small
1328  * @VM_FAULT_DONE_COW:		->fault has fully handled COW
1329  * @VM_FAULT_NEEDDSYNC:		->fault did not modify page tables and needs
1330  *				fsync() to complete (for synchronous page faults
1331  *				in DAX)
1332  * @VM_FAULT_COMPLETED:		->fault completed, meanwhile mmap lock released
1333  * @VM_FAULT_HINDEX_MASK:	mask HINDEX value
1334  *
1335  */
1336 enum vm_fault_reason {
1337 	VM_FAULT_OOM            = (__force vm_fault_t)0x000001,
1338 	VM_FAULT_SIGBUS         = (__force vm_fault_t)0x000002,
1339 	VM_FAULT_MAJOR          = (__force vm_fault_t)0x000004,
1340 	VM_FAULT_HWPOISON       = (__force vm_fault_t)0x000010,
1341 	VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1342 	VM_FAULT_SIGSEGV        = (__force vm_fault_t)0x000040,
1343 	VM_FAULT_NOPAGE         = (__force vm_fault_t)0x000100,
1344 	VM_FAULT_LOCKED         = (__force vm_fault_t)0x000200,
1345 	VM_FAULT_RETRY          = (__force vm_fault_t)0x000400,
1346 	VM_FAULT_FALLBACK       = (__force vm_fault_t)0x000800,
1347 	VM_FAULT_DONE_COW       = (__force vm_fault_t)0x001000,
1348 	VM_FAULT_NEEDDSYNC      = (__force vm_fault_t)0x002000,
1349 	VM_FAULT_COMPLETED      = (__force vm_fault_t)0x004000,
1350 	VM_FAULT_HINDEX_MASK    = (__force vm_fault_t)0x0f0000,
1351 };
1352 
1353 /* Encode hstate index for a hwpoisoned large page */
1354 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1355 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1356 
1357 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS |	\
1358 			VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON |	\
1359 			VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1360 
1361 #define VM_FAULT_RESULT_TRACE \
1362 	{ VM_FAULT_OOM,                 "OOM" },	\
1363 	{ VM_FAULT_SIGBUS,              "SIGBUS" },	\
1364 	{ VM_FAULT_MAJOR,               "MAJOR" },	\
1365 	{ VM_FAULT_HWPOISON,            "HWPOISON" },	\
1366 	{ VM_FAULT_HWPOISON_LARGE,      "HWPOISON_LARGE" },	\
1367 	{ VM_FAULT_SIGSEGV,             "SIGSEGV" },	\
1368 	{ VM_FAULT_NOPAGE,              "NOPAGE" },	\
1369 	{ VM_FAULT_LOCKED,              "LOCKED" },	\
1370 	{ VM_FAULT_RETRY,               "RETRY" },	\
1371 	{ VM_FAULT_FALLBACK,            "FALLBACK" },	\
1372 	{ VM_FAULT_DONE_COW,            "DONE_COW" },	\
1373 	{ VM_FAULT_NEEDDSYNC,           "NEEDDSYNC" },	\
1374 	{ VM_FAULT_COMPLETED,           "COMPLETED" }
1375 
1376 struct vm_special_mapping {
1377 	const char *name;	/* The name, e.g. "[vdso]". */
1378 
1379 	/*
1380 	 * If .fault is not provided, this points to a
1381 	 * NULL-terminated array of pages that back the special mapping.
1382 	 *
1383 	 * This must not be NULL unless .fault is provided.
1384 	 */
1385 	struct page **pages;
1386 
1387 	/*
1388 	 * If non-NULL, then this is called to resolve page faults
1389 	 * on the special mapping.  If used, .pages is not checked.
1390 	 */
1391 	vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1392 				struct vm_area_struct *vma,
1393 				struct vm_fault *vmf);
1394 
1395 	int (*mremap)(const struct vm_special_mapping *sm,
1396 		     struct vm_area_struct *new_vma);
1397 
1398 	void (*close)(const struct vm_special_mapping *sm,
1399 		      struct vm_area_struct *vma);
1400 };
1401 
1402 enum tlb_flush_reason {
1403 	TLB_FLUSH_ON_TASK_SWITCH,
1404 	TLB_REMOTE_SHOOTDOWN,
1405 	TLB_LOCAL_SHOOTDOWN,
1406 	TLB_LOCAL_MM_SHOOTDOWN,
1407 	TLB_REMOTE_SEND_IPI,
1408 	TLB_REMOTE_WRONG_CPU,
1409 	NR_TLB_FLUSH_REASONS,
1410 };
1411 
1412 /**
1413  * enum fault_flag - Fault flag definitions.
1414  * @FAULT_FLAG_WRITE: Fault was a write fault.
1415  * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1416  * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1417  * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1418  * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1419  * @FAULT_FLAG_TRIED: The fault has been tried once.
1420  * @FAULT_FLAG_USER: The fault originated in userspace.
1421  * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1422  * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1423  * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1424  * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1425  *                      COW mapping, making sure that an exclusive anon page is
1426  *                      mapped after the fault.
1427  * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1428  *                        We should only access orig_pte if this flag set.
1429  * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1430  *
1431  * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1432  * whether we would allow page faults to retry by specifying these two
1433  * fault flags correctly.  Currently there can be three legal combinations:
1434  *
1435  * (a) ALLOW_RETRY and !TRIED:  this means the page fault allows retry, and
1436  *                              this is the first try
1437  *
1438  * (b) ALLOW_RETRY and TRIED:   this means the page fault allows retry, and
1439  *                              we've already tried at least once
1440  *
1441  * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1442  *
1443  * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1444  * be used.  Note that page faults can be allowed to retry for multiple times,
1445  * in which case we'll have an initial fault with flags (a) then later on
1446  * continuous faults with flags (b).  We should always try to detect pending
1447  * signals before a retry to make sure the continuous page faults can still be
1448  * interrupted if necessary.
1449  *
1450  * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1451  * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1452  * applied to mappings that are not COW mappings.
1453  */
1454 enum fault_flag {
1455 	FAULT_FLAG_WRITE =		1 << 0,
1456 	FAULT_FLAG_MKWRITE =		1 << 1,
1457 	FAULT_FLAG_ALLOW_RETRY =	1 << 2,
1458 	FAULT_FLAG_RETRY_NOWAIT = 	1 << 3,
1459 	FAULT_FLAG_KILLABLE =		1 << 4,
1460 	FAULT_FLAG_TRIED = 		1 << 5,
1461 	FAULT_FLAG_USER =		1 << 6,
1462 	FAULT_FLAG_REMOTE =		1 << 7,
1463 	FAULT_FLAG_INSTRUCTION =	1 << 8,
1464 	FAULT_FLAG_INTERRUPTIBLE =	1 << 9,
1465 	FAULT_FLAG_UNSHARE =		1 << 10,
1466 	FAULT_FLAG_ORIG_PTE_VALID =	1 << 11,
1467 	FAULT_FLAG_VMA_LOCK =		1 << 12,
1468 };
1469 
1470 typedef unsigned int __bitwise zap_flags_t;
1471 
1472 /* Flags for clear_young_dirty_ptes(). */
1473 typedef int __bitwise cydp_t;
1474 
1475 /* Clear the access bit */
1476 #define CYDP_CLEAR_YOUNG		((__force cydp_t)BIT(0))
1477 
1478 /* Clear the dirty bit */
1479 #define CYDP_CLEAR_DIRTY		((__force cydp_t)BIT(1))
1480 
1481 /*
1482  * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1483  * other. Here is what they mean, and how to use them:
1484  *
1485  *
1486  * FIXME: For pages which are part of a filesystem, mappings are subject to the
1487  * lifetime enforced by the filesystem and we need guarantees that longterm
1488  * users like RDMA and V4L2 only establish mappings which coordinate usage with
1489  * the filesystem.  Ideas for this coordination include revoking the longterm
1490  * pin, delaying writeback, bounce buffer page writeback, etc.  As FS DAX was
1491  * added after the problem with filesystems was found FS DAX VMAs are
1492  * specifically failed.  Filesystem pages are still subject to bugs and use of
1493  * FOLL_LONGTERM should be avoided on those pages.
1494  *
1495  * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1496  * that region.  And so, CMA attempts to migrate the page before pinning, when
1497  * FOLL_LONGTERM is specified.
1498  *
1499  * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1500  * but an additional pin counting system) will be invoked. This is intended for
1501  * anything that gets a page reference and then touches page data (for example,
1502  * Direct IO). This lets the filesystem know that some non-file-system entity is
1503  * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1504  * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1505  * a call to unpin_user_page().
1506  *
1507  * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1508  * and separate refcounting mechanisms, however, and that means that each has
1509  * its own acquire and release mechanisms:
1510  *
1511  *     FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1512  *
1513  *     FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1514  *
1515  * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1516  * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1517  * calls applied to them, and that's perfectly OK. This is a constraint on the
1518  * callers, not on the pages.)
1519  *
1520  * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1521  * directly by the caller. That's in order to help avoid mismatches when
1522  * releasing pages: get_user_pages*() pages must be released via put_page(),
1523  * while pin_user_pages*() pages must be released via unpin_user_page().
1524  *
1525  * Please see Documentation/core-api/pin_user_pages.rst for more information.
1526  */
1527 
1528 enum {
1529 	/* check pte is writable */
1530 	FOLL_WRITE = 1 << 0,
1531 	/* do get_page on page */
1532 	FOLL_GET = 1 << 1,
1533 	/* give error on hole if it would be zero */
1534 	FOLL_DUMP = 1 << 2,
1535 	/* get_user_pages read/write w/o permission */
1536 	FOLL_FORCE = 1 << 3,
1537 	/*
1538 	 * if a disk transfer is needed, start the IO and return without waiting
1539 	 * upon it
1540 	 */
1541 	FOLL_NOWAIT = 1 << 4,
1542 	/* do not fault in pages */
1543 	FOLL_NOFAULT = 1 << 5,
1544 	/* check page is hwpoisoned */
1545 	FOLL_HWPOISON = 1 << 6,
1546 	/* don't do file mappings */
1547 	FOLL_ANON = 1 << 7,
1548 	/*
1549 	 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1550 	 * time period _often_ under userspace control.  This is in contrast to
1551 	 * iov_iter_get_pages(), whose usages are transient.
1552 	 */
1553 	FOLL_LONGTERM = 1 << 8,
1554 	/* split huge pmd before returning */
1555 	FOLL_SPLIT_PMD = 1 << 9,
1556 	/* allow returning PCI P2PDMA pages */
1557 	FOLL_PCI_P2PDMA = 1 << 10,
1558 	/* allow interrupts from generic signals */
1559 	FOLL_INTERRUPTIBLE = 1 << 11,
1560 	/*
1561 	 * Always honor (trigger) NUMA hinting faults.
1562 	 *
1563 	 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1564 	 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1565 	 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1566 	 * hinting faults.
1567 	 */
1568 	FOLL_HONOR_NUMA_FAULT = 1 << 12,
1569 
1570 	/* See also internal only FOLL flags in mm/internal.h */
1571 };
1572 
1573 /* mm flags */
1574 
1575 /*
1576  * The first two bits represent core dump modes for set-user-ID,
1577  * the modes are SUID_DUMP_* defined in linux/sched/coredump.h
1578  */
1579 #define MMF_DUMPABLE_BITS 2
1580 #define MMF_DUMPABLE_MASK ((1 << MMF_DUMPABLE_BITS) - 1)
1581 /* coredump filter bits */
1582 #define MMF_DUMP_ANON_PRIVATE	2
1583 #define MMF_DUMP_ANON_SHARED	3
1584 #define MMF_DUMP_MAPPED_PRIVATE	4
1585 #define MMF_DUMP_MAPPED_SHARED	5
1586 #define MMF_DUMP_ELF_HEADERS	6
1587 #define MMF_DUMP_HUGETLB_PRIVATE 7
1588 #define MMF_DUMP_HUGETLB_SHARED  8
1589 #define MMF_DUMP_DAX_PRIVATE	9
1590 #define MMF_DUMP_DAX_SHARED	10
1591 
1592 #define MMF_DUMP_FILTER_SHIFT	MMF_DUMPABLE_BITS
1593 #define MMF_DUMP_FILTER_BITS	9
1594 #define MMF_DUMP_FILTER_MASK \
1595 	(((1 << MMF_DUMP_FILTER_BITS) - 1) << MMF_DUMP_FILTER_SHIFT)
1596 #define MMF_DUMP_FILTER_DEFAULT \
1597 	((1 << MMF_DUMP_ANON_PRIVATE) |	(1 << MMF_DUMP_ANON_SHARED) |\
1598 	 (1 << MMF_DUMP_HUGETLB_PRIVATE) | MMF_DUMP_MASK_DEFAULT_ELF)
1599 
1600 #ifdef CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS
1601 # define MMF_DUMP_MASK_DEFAULT_ELF	(1 << MMF_DUMP_ELF_HEADERS)
1602 #else
1603 # define MMF_DUMP_MASK_DEFAULT_ELF	0
1604 #endif
1605 					/* leave room for more dump flags */
1606 #define MMF_VM_MERGEABLE	16	/* KSM may merge identical pages */
1607 #define MMF_VM_HUGEPAGE		17	/* set when mm is available for khugepaged */
1608 
1609 /*
1610  * This one-shot flag is dropped due to necessity of changing exe once again
1611  * on NFS restore
1612  */
1613 //#define MMF_EXE_FILE_CHANGED	18	/* see prctl_set_mm_exe_file() */
1614 
1615 #define MMF_HAS_UPROBES		19	/* has uprobes */
1616 #define MMF_RECALC_UPROBES	20	/* MMF_HAS_UPROBES can be wrong */
1617 #define MMF_OOM_SKIP		21	/* mm is of no interest for the OOM killer */
1618 #define MMF_UNSTABLE		22	/* mm is unstable for copy_from_user */
1619 #define MMF_HUGE_ZERO_PAGE	23      /* mm has ever used the global huge zero page */
1620 #define MMF_DISABLE_THP		24	/* disable THP for all VMAs */
1621 #define MMF_DISABLE_THP_MASK	(1 << MMF_DISABLE_THP)
1622 #define MMF_OOM_REAP_QUEUED	25	/* mm was queued for oom_reaper */
1623 #define MMF_MULTIPROCESS	26	/* mm is shared between processes */
1624 /*
1625  * MMF_HAS_PINNED: Whether this mm has pinned any pages.  This can be either
1626  * replaced in the future by mm.pinned_vm when it becomes stable, or grow into
1627  * a counter on its own. We're aggresive on this bit for now: even if the
1628  * pinned pages were unpinned later on, we'll still keep this bit set for the
1629  * lifecycle of this mm, just for simplicity.
1630  */
1631 #define MMF_HAS_PINNED		27	/* FOLL_PIN has run, never cleared */
1632 
1633 #define MMF_HAS_MDWE		28
1634 #define MMF_HAS_MDWE_MASK	(1 << MMF_HAS_MDWE)
1635 
1636 
1637 #define MMF_HAS_MDWE_NO_INHERIT	29
1638 
1639 #define MMF_VM_MERGE_ANY	30
1640 #define MMF_VM_MERGE_ANY_MASK	(1 << MMF_VM_MERGE_ANY)
1641 
1642 #define MMF_TOPDOWN		31	/* mm searches top down by default */
1643 #define MMF_TOPDOWN_MASK	(1 << MMF_TOPDOWN)
1644 
1645 #define MMF_INIT_MASK		(MMF_DUMPABLE_MASK | MMF_DUMP_FILTER_MASK |\
1646 				 MMF_DISABLE_THP_MASK | MMF_HAS_MDWE_MASK |\
1647 				 MMF_VM_MERGE_ANY_MASK | MMF_TOPDOWN_MASK)
1648 
1649 static inline unsigned long mmf_init_flags(unsigned long flags)
1650 {
1651 	if (flags & (1UL << MMF_HAS_MDWE_NO_INHERIT))
1652 		flags &= ~((1UL << MMF_HAS_MDWE) |
1653 			   (1UL << MMF_HAS_MDWE_NO_INHERIT));
1654 	return flags & MMF_INIT_MASK;
1655 }
1656 
1657 #endif /* _LINUX_MM_TYPES_H */
1658