xref: /linux-6.15/include/linux/mm_types.h (revision 024bfd2e)
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  * The mapcount field must not be used for own purposes.
50  *
51  * If you want to use the refcount field, it must be used in such a way
52  * that other CPUs temporarily incrementing and then decrementing the
53  * refcount does not cause problems.  On receiving the page from
54  * alloc_pages(), the refcount will be positive.
55  *
56  * If you allocate pages of order > 0, you can use some of the fields
57  * in each subpage, but you may need to restore some of their values
58  * afterwards.
59  *
60  * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
61  * That requires that freelist & counters in struct slab be adjacent and
62  * double-word aligned. Because struct slab currently just reinterprets the
63  * bits of struct page, we align all struct pages to double-word boundaries,
64  * and ensure that 'freelist' is aligned within struct slab.
65  */
66 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
67 #define _struct_page_alignment	__aligned(2 * sizeof(unsigned long))
68 #else
69 #define _struct_page_alignment	__aligned(sizeof(unsigned long))
70 #endif
71 
72 struct page {
73 	unsigned long flags;		/* Atomic flags, some possibly
74 					 * updated asynchronously */
75 	/*
76 	 * Five words (20/40 bytes) are available in this union.
77 	 * WARNING: bit 0 of the first word is used for PageTail(). That
78 	 * means the other users of this union MUST NOT use the bit to
79 	 * avoid collision and false-positive PageTail().
80 	 */
81 	union {
82 		struct {	/* Page cache and anonymous pages */
83 			/**
84 			 * @lru: Pageout list, eg. active_list protected by
85 			 * lruvec->lru_lock.  Sometimes used as a generic list
86 			 * by the page owner.
87 			 */
88 			union {
89 				struct list_head lru;
90 
91 				/* Or, for the Unevictable "LRU list" slot */
92 				struct {
93 					/* Always even, to negate PageTail */
94 					void *__filler;
95 					/* Count page's or folio's mlocks */
96 					unsigned int mlock_count;
97 				};
98 
99 				/* Or, free page */
100 				struct list_head buddy_list;
101 				struct list_head pcp_list;
102 			};
103 			/* See page-flags.h for PAGE_MAPPING_FLAGS */
104 			struct address_space *mapping;
105 			union {
106 				pgoff_t index;		/* Our offset within mapping. */
107 				unsigned long share;	/* share count for fsdax */
108 			};
109 			/**
110 			 * @private: Mapping-private opaque data.
111 			 * Usually used for buffer_heads if PagePrivate.
112 			 * Used for swp_entry_t if swapcache flag set.
113 			 * Indicates order in the buddy system if PageBuddy.
114 			 */
115 			unsigned long private;
116 		};
117 		struct {	/* page_pool used by netstack */
118 			/**
119 			 * @pp_magic: magic value to avoid recycling non
120 			 * page_pool allocated pages.
121 			 */
122 			unsigned long pp_magic;
123 			struct page_pool *pp;
124 			unsigned long _pp_mapping_pad;
125 			unsigned long dma_addr;
126 			atomic_long_t pp_ref_count;
127 		};
128 		struct {	/* Tail pages of compound page */
129 			unsigned long compound_head;	/* Bit zero is set */
130 		};
131 		struct {	/* ZONE_DEVICE pages */
132 			/** @pgmap: Points to the hosting device page map. */
133 			struct dev_pagemap *pgmap;
134 			void *zone_device_data;
135 			/*
136 			 * ZONE_DEVICE private pages are counted as being
137 			 * mapped so the next 3 words hold the mapping, index,
138 			 * and private fields from the source anonymous or
139 			 * page cache page while the page is migrated to device
140 			 * private memory.
141 			 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
142 			 * use the mapping, index, and private fields when
143 			 * pmem backed DAX files are mapped.
144 			 */
145 		};
146 
147 		/** @rcu_head: You can use this to free a page by RCU. */
148 		struct rcu_head rcu_head;
149 	};
150 
151 	union {		/* This union is 4 bytes in size. */
152 		/*
153 		 * For head pages of typed folios, the value stored here
154 		 * allows for determining what this page is used for. The
155 		 * tail pages of typed folios will not store a type
156 		 * (page_type == _mapcount == -1).
157 		 *
158 		 * See page-flags.h for a list of page types which are currently
159 		 * stored here.
160 		 *
161 		 * Owners of typed folios may reuse the lower 16 bit of the
162 		 * head page page_type field after setting the page type,
163 		 * but must reset these 16 bit to -1 before clearing the
164 		 * page type.
165 		 */
166 		unsigned int page_type;
167 
168 		/*
169 		 * For pages that are part of non-typed folios for which mappings
170 		 * are tracked via the RMAP, encodes the number of times this page
171 		 * is directly referenced by a page table.
172 		 *
173 		 * Note that the mapcount is always initialized to -1, so that
174 		 * transitions both from it and to it can be tracked, using
175 		 * atomic_inc_and_test() and atomic_add_negative(-1).
176 		 */
177 		atomic_t _mapcount;
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 #elif defined(CONFIG_SLAB_OBJ_EXT)
186 	unsigned long _unused_slab_obj_exts;
187 #endif
188 
189 	/*
190 	 * On machines where all RAM is mapped into kernel address space,
191 	 * we can simply calculate the virtual address. On machines with
192 	 * highmem some memory is mapped into kernel virtual memory
193 	 * dynamically, so we need a place to store that address.
194 	 * Note that this field could be 16 bits on x86 ... ;)
195 	 *
196 	 * Architectures with slow multiplication can define
197 	 * WANT_PAGE_VIRTUAL in asm/page.h
198 	 */
199 #if defined(WANT_PAGE_VIRTUAL)
200 	void *virtual;			/* Kernel virtual address (NULL if
201 					   not kmapped, ie. highmem) */
202 #endif /* WANT_PAGE_VIRTUAL */
203 
204 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
205 	int _last_cpupid;
206 #endif
207 
208 #ifdef CONFIG_KMSAN
209 	/*
210 	 * KMSAN metadata for this page:
211 	 *  - shadow page: every bit indicates whether the corresponding
212 	 *    bit of the original page is initialized (0) or not (1);
213 	 *  - origin page: every 4 bytes contain an id of the stack trace
214 	 *    where the uninitialized value was created.
215 	 */
216 	struct page *kmsan_shadow;
217 	struct page *kmsan_origin;
218 #endif
219 } _struct_page_alignment;
220 
221 /*
222  * struct encoded_page - a nonexistent type marking this pointer
223  *
224  * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
225  * with the low bits of the pointer indicating extra context-dependent
226  * information. Only used in mmu_gather handling, and this acts as a type
227  * system check on that use.
228  *
229  * We only really have two guaranteed bits in general, although you could
230  * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
231  * for more.
232  *
233  * Use the supplied helper functions to endcode/decode the pointer and bits.
234  */
235 struct encoded_page;
236 
237 #define ENCODED_PAGE_BITS			3ul
238 
239 /* Perform rmap removal after we have flushed the TLB. */
240 #define ENCODED_PAGE_BIT_DELAY_RMAP		1ul
241 
242 /*
243  * The next item in an encoded_page array is the "nr_pages" argument, specifying
244  * the number of consecutive pages starting from this page, that all belong to
245  * the same folio. For example, "nr_pages" corresponds to the number of folio
246  * references that must be dropped. If this bit is not set, "nr_pages" is
247  * implicitly 1.
248  */
249 #define ENCODED_PAGE_BIT_NR_PAGES_NEXT		2ul
250 
251 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
252 {
253 	BUILD_BUG_ON(flags > ENCODED_PAGE_BITS);
254 	return (struct encoded_page *)(flags | (unsigned long)page);
255 }
256 
257 static inline unsigned long encoded_page_flags(struct encoded_page *page)
258 {
259 	return ENCODED_PAGE_BITS & (unsigned long)page;
260 }
261 
262 static inline struct page *encoded_page_ptr(struct encoded_page *page)
263 {
264 	return (struct page *)(~ENCODED_PAGE_BITS & (unsigned long)page);
265 }
266 
267 static __always_inline struct encoded_page *encode_nr_pages(unsigned long nr)
268 {
269 	VM_WARN_ON_ONCE((nr << 2) >> 2 != nr);
270 	return (struct encoded_page *)(nr << 2);
271 }
272 
273 static __always_inline unsigned long encoded_nr_pages(struct encoded_page *page)
274 {
275 	return ((unsigned long)page) >> 2;
276 }
277 
278 /*
279  * A swap entry has to fit into a "unsigned long", as the entry is hidden
280  * in the "index" field of the swapper address space.
281  */
282 typedef struct {
283 	unsigned long val;
284 } swp_entry_t;
285 
286 /**
287  * struct folio - Represents a contiguous set of bytes.
288  * @flags: Identical to the page flags.
289  * @lru: Least Recently Used list; tracks how recently this folio was used.
290  * @mlock_count: Number of times this folio has been pinned by mlock().
291  * @mapping: The file this page belongs to, or refers to the anon_vma for
292  *    anonymous memory.
293  * @index: Offset within the file, in units of pages.  For anonymous memory,
294  *    this is the index from the beginning of the mmap.
295  * @private: Filesystem per-folio data (see folio_attach_private()).
296  * @swap: Used for swp_entry_t if folio_test_swapcache().
297  * @_mapcount: Do not access this member directly.  Use folio_mapcount() to
298  *    find out how many times this folio is mapped by userspace.
299  * @_refcount: Do not access this member directly.  Use folio_ref_count()
300  *    to find how many references there are to this folio.
301  * @memcg_data: Memory Control Group data.
302  * @virtual: Virtual address in the kernel direct map.
303  * @_last_cpupid: IDs of last CPU and last process that accessed the folio.
304  * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
305  * @_large_mapcount: Do not use directly, call folio_mapcount().
306  * @_nr_pages_mapped: Do not use outside of rmap and debug code.
307  * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
308  * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
309  * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
310  * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
311  * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
312  * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
313  * @_deferred_list: Folios to be split under memory pressure.
314  * @_unused_slab_obj_exts: Placeholder to match obj_exts in struct slab.
315  *
316  * A folio is a physically, virtually and logically contiguous set
317  * of bytes.  It is a power-of-two in size, and it is aligned to that
318  * same power-of-two.  It is at least as large as %PAGE_SIZE.  If it is
319  * in the page cache, it is at a file offset which is a multiple of that
320  * power-of-two.  It may be mapped into userspace at an address which is
321  * at an arbitrary page offset, but its kernel virtual address is aligned
322  * to its size.
323  */
324 struct folio {
325 	/* private: don't document the anon union */
326 	union {
327 		struct {
328 	/* public: */
329 			unsigned long flags;
330 			union {
331 				struct list_head lru;
332 	/* private: avoid cluttering the output */
333 				struct {
334 					void *__filler;
335 	/* public: */
336 					unsigned int mlock_count;
337 	/* private: */
338 				};
339 	/* public: */
340 			};
341 			struct address_space *mapping;
342 			pgoff_t index;
343 			union {
344 				void *private;
345 				swp_entry_t swap;
346 			};
347 			atomic_t _mapcount;
348 			atomic_t _refcount;
349 #ifdef CONFIG_MEMCG
350 			unsigned long memcg_data;
351 #elif defined(CONFIG_SLAB_OBJ_EXT)
352 			unsigned long _unused_slab_obj_exts;
353 #endif
354 #if defined(WANT_PAGE_VIRTUAL)
355 			void *virtual;
356 #endif
357 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
358 			int _last_cpupid;
359 #endif
360 	/* private: the union with struct page is transitional */
361 		};
362 		struct page page;
363 	};
364 	union {
365 		struct {
366 			unsigned long _flags_1;
367 			unsigned long _head_1;
368 	/* public: */
369 			atomic_t _large_mapcount;
370 			atomic_t _entire_mapcount;
371 			atomic_t _nr_pages_mapped;
372 			atomic_t _pincount;
373 #ifdef CONFIG_64BIT
374 			unsigned int _folio_nr_pages;
375 #endif
376 	/* private: the union with struct page is transitional */
377 		};
378 		struct page __page_1;
379 	};
380 	union {
381 		struct {
382 			unsigned long _flags_2;
383 			unsigned long _head_2;
384 	/* public: */
385 			void *_hugetlb_subpool;
386 			void *_hugetlb_cgroup;
387 			void *_hugetlb_cgroup_rsvd;
388 			void *_hugetlb_hwpoison;
389 	/* private: the union with struct page is transitional */
390 		};
391 		struct {
392 			unsigned long _flags_2a;
393 			unsigned long _head_2a;
394 	/* public: */
395 			struct list_head _deferred_list;
396 	/* private: the union with struct page is transitional */
397 		};
398 		struct page __page_2;
399 	};
400 };
401 
402 #define FOLIO_MATCH(pg, fl)						\
403 	static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
404 FOLIO_MATCH(flags, flags);
405 FOLIO_MATCH(lru, lru);
406 FOLIO_MATCH(mapping, mapping);
407 FOLIO_MATCH(compound_head, lru);
408 FOLIO_MATCH(index, index);
409 FOLIO_MATCH(private, private);
410 FOLIO_MATCH(_mapcount, _mapcount);
411 FOLIO_MATCH(_refcount, _refcount);
412 #ifdef CONFIG_MEMCG
413 FOLIO_MATCH(memcg_data, memcg_data);
414 #endif
415 #if defined(WANT_PAGE_VIRTUAL)
416 FOLIO_MATCH(virtual, virtual);
417 #endif
418 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
419 FOLIO_MATCH(_last_cpupid, _last_cpupid);
420 #endif
421 #undef FOLIO_MATCH
422 #define FOLIO_MATCH(pg, fl)						\
423 	static_assert(offsetof(struct folio, fl) ==			\
424 			offsetof(struct page, pg) + sizeof(struct page))
425 FOLIO_MATCH(flags, _flags_1);
426 FOLIO_MATCH(compound_head, _head_1);
427 #undef FOLIO_MATCH
428 #define FOLIO_MATCH(pg, fl)						\
429 	static_assert(offsetof(struct folio, fl) ==			\
430 			offsetof(struct page, pg) + 2 * sizeof(struct page))
431 FOLIO_MATCH(flags, _flags_2);
432 FOLIO_MATCH(compound_head, _head_2);
433 FOLIO_MATCH(flags, _flags_2a);
434 FOLIO_MATCH(compound_head, _head_2a);
435 #undef FOLIO_MATCH
436 
437 /**
438  * struct ptdesc -    Memory descriptor for page tables.
439  * @__page_flags:     Same as page flags. Powerpc only.
440  * @pt_rcu_head:      For freeing page table pages.
441  * @pt_list:          List of used page tables. Used for s390 and x86.
442  * @_pt_pad_1:        Padding that aliases with page's compound head.
443  * @pmd_huge_pte:     Protected by ptdesc->ptl, used for THPs.
444  * @__page_mapping:   Aliases with page->mapping. Unused for page tables.
445  * @pt_index:         Used for s390 gmap.
446  * @pt_mm:            Used for x86 pgds.
447  * @pt_frag_refcount: For fragmented page table tracking. Powerpc only.
448  * @_pt_pad_2:        Padding to ensure proper alignment.
449  * @ptl:              Lock for the page table.
450  * @__page_type:      Same as page->page_type. Unused for page tables.
451  * @__page_refcount:  Same as page refcount.
452  * @pt_memcg_data:    Memcg data. Tracked for page tables here.
453  *
454  * This struct overlays struct page for now. Do not modify without a good
455  * understanding of the issues.
456  */
457 struct ptdesc {
458 	unsigned long __page_flags;
459 
460 	union {
461 		struct rcu_head pt_rcu_head;
462 		struct list_head pt_list;
463 		struct {
464 			unsigned long _pt_pad_1;
465 			pgtable_t pmd_huge_pte;
466 		};
467 	};
468 	unsigned long __page_mapping;
469 
470 	union {
471 		pgoff_t pt_index;
472 		struct mm_struct *pt_mm;
473 		atomic_t pt_frag_refcount;
474 	};
475 
476 	union {
477 		unsigned long _pt_pad_2;
478 #if ALLOC_SPLIT_PTLOCKS
479 		spinlock_t *ptl;
480 #else
481 		spinlock_t ptl;
482 #endif
483 	};
484 	unsigned int __page_type;
485 	atomic_t __page_refcount;
486 #ifdef CONFIG_MEMCG
487 	unsigned long pt_memcg_data;
488 #endif
489 };
490 
491 #define TABLE_MATCH(pg, pt)						\
492 	static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
493 TABLE_MATCH(flags, __page_flags);
494 TABLE_MATCH(compound_head, pt_list);
495 TABLE_MATCH(compound_head, _pt_pad_1);
496 TABLE_MATCH(mapping, __page_mapping);
497 TABLE_MATCH(index, pt_index);
498 TABLE_MATCH(rcu_head, pt_rcu_head);
499 TABLE_MATCH(page_type, __page_type);
500 TABLE_MATCH(_refcount, __page_refcount);
501 #ifdef CONFIG_MEMCG
502 TABLE_MATCH(memcg_data, pt_memcg_data);
503 #endif
504 #undef TABLE_MATCH
505 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
506 
507 #define ptdesc_page(pt)			(_Generic((pt),			\
508 	const struct ptdesc *:		(const struct page *)(pt),	\
509 	struct ptdesc *:		(struct page *)(pt)))
510 
511 #define ptdesc_folio(pt)		(_Generic((pt),			\
512 	const struct ptdesc *:		(const struct folio *)(pt),	\
513 	struct ptdesc *:		(struct folio *)(pt)))
514 
515 #define page_ptdesc(p)			(_Generic((p),			\
516 	const struct page *:		(const struct ptdesc *)(p),	\
517 	struct page *:			(struct ptdesc *)(p)))
518 
519 /*
520  * Used for sizing the vmemmap region on some architectures
521  */
522 #define STRUCT_PAGE_MAX_SHIFT	(order_base_2(sizeof(struct page)))
523 
524 /*
525  * page_private can be used on tail pages.  However, PagePrivate is only
526  * checked by the VM on the head page.  So page_private on the tail pages
527  * should be used for data that's ancillary to the head page (eg attaching
528  * buffer heads to tail pages after attaching buffer heads to the head page)
529  */
530 #define page_private(page)		((page)->private)
531 
532 static inline void set_page_private(struct page *page, unsigned long private)
533 {
534 	page->private = private;
535 }
536 
537 static inline void *folio_get_private(struct folio *folio)
538 {
539 	return folio->private;
540 }
541 
542 typedef unsigned long vm_flags_t;
543 
544 /*
545  * A region containing a mapping of a non-memory backed file under NOMMU
546  * conditions.  These are held in a global tree and are pinned by the VMAs that
547  * map parts of them.
548  */
549 struct vm_region {
550 	struct rb_node	vm_rb;		/* link in global region tree */
551 	vm_flags_t	vm_flags;	/* VMA vm_flags */
552 	unsigned long	vm_start;	/* start address of region */
553 	unsigned long	vm_end;		/* region initialised to here */
554 	unsigned long	vm_top;		/* region allocated to here */
555 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
556 	struct file	*vm_file;	/* the backing file or NULL */
557 
558 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
559 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
560 						* this region */
561 };
562 
563 #ifdef CONFIG_USERFAULTFD
564 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
565 struct vm_userfaultfd_ctx {
566 	struct userfaultfd_ctx *ctx;
567 };
568 #else /* CONFIG_USERFAULTFD */
569 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
570 struct vm_userfaultfd_ctx {};
571 #endif /* CONFIG_USERFAULTFD */
572 
573 struct anon_vma_name {
574 	struct kref kref;
575 	/* The name needs to be at the end because it is dynamically sized. */
576 	char name[];
577 };
578 
579 #ifdef CONFIG_ANON_VMA_NAME
580 /*
581  * mmap_lock should be read-locked when calling anon_vma_name(). Caller should
582  * either keep holding the lock while using the returned pointer or it should
583  * raise anon_vma_name refcount before releasing the lock.
584  */
585 struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma);
586 struct anon_vma_name *anon_vma_name_alloc(const char *name);
587 void anon_vma_name_free(struct kref *kref);
588 #else /* CONFIG_ANON_VMA_NAME */
589 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
590 {
591 	return NULL;
592 }
593 
594 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
595 {
596 	return NULL;
597 }
598 #endif
599 
600 struct vma_lock {
601 	struct rw_semaphore lock;
602 };
603 
604 struct vma_numab_state {
605 	/*
606 	 * Initialised as time in 'jiffies' after which VMA
607 	 * should be scanned.  Delays first scan of new VMA by at
608 	 * least sysctl_numa_balancing_scan_delay:
609 	 */
610 	unsigned long next_scan;
611 
612 	/*
613 	 * Time in jiffies when pids_active[] is reset to
614 	 * detect phase change behaviour:
615 	 */
616 	unsigned long pids_active_reset;
617 
618 	/*
619 	 * Approximate tracking of PIDs that trapped a NUMA hinting
620 	 * fault. May produce false positives due to hash collisions.
621 	 *
622 	 *   [0] Previous PID tracking
623 	 *   [1] Current PID tracking
624 	 *
625 	 * Window moves after next_pid_reset has expired approximately
626 	 * every VMA_PID_RESET_PERIOD jiffies:
627 	 */
628 	unsigned long pids_active[2];
629 
630 	/* MM scan sequence ID when scan first started after VMA creation */
631 	int start_scan_seq;
632 
633 	/*
634 	 * MM scan sequence ID when the VMA was last completely scanned.
635 	 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq
636 	 */
637 	int prev_scan_seq;
638 };
639 
640 /*
641  * This struct describes a virtual memory area. There is one of these
642  * per VM-area/task. A VM area is any part of the process virtual memory
643  * space that has a special rule for the page-fault handlers (ie a shared
644  * library, the executable area etc).
645  *
646  * Only explicitly marked struct members may be accessed by RCU readers before
647  * getting a stable reference.
648  */
649 struct vm_area_struct {
650 	/* The first cache line has the info for VMA tree walking. */
651 
652 	union {
653 		struct {
654 			/* VMA covers [vm_start; vm_end) addresses within mm */
655 			unsigned long vm_start;
656 			unsigned long vm_end;
657 		};
658 #ifdef CONFIG_PER_VMA_LOCK
659 		struct rcu_head vm_rcu;	/* Used for deferred freeing. */
660 #endif
661 	};
662 
663 	/*
664 	 * The address space we belong to.
665 	 * Unstable RCU readers are allowed to read this.
666 	 */
667 	struct mm_struct *vm_mm;
668 	pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
669 
670 	/*
671 	 * Flags, see mm.h.
672 	 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
673 	 */
674 	union {
675 		const vm_flags_t vm_flags;
676 		vm_flags_t __private __vm_flags;
677 	};
678 
679 #ifdef CONFIG_PER_VMA_LOCK
680 	/*
681 	 * Flag to indicate areas detached from the mm->mm_mt tree.
682 	 * Unstable RCU readers are allowed to read this.
683 	 */
684 	bool detached;
685 
686 	/*
687 	 * Can only be written (using WRITE_ONCE()) while holding both:
688 	 *  - mmap_lock (in write mode)
689 	 *  - vm_lock->lock (in write mode)
690 	 * Can be read reliably while holding one of:
691 	 *  - mmap_lock (in read or write mode)
692 	 *  - vm_lock->lock (in read or write mode)
693 	 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
694 	 * while holding nothing (except RCU to keep the VMA struct allocated).
695 	 *
696 	 * This sequence counter is explicitly allowed to overflow; sequence
697 	 * counter reuse can only lead to occasional unnecessary use of the
698 	 * slowpath.
699 	 */
700 	int vm_lock_seq;
701 	/* Unstable RCU readers are allowed to read this. */
702 	struct vma_lock *vm_lock;
703 #endif
704 
705 	/*
706 	 * For areas with an address space and backing store,
707 	 * linkage into the address_space->i_mmap interval tree.
708 	 *
709 	 */
710 	struct {
711 		struct rb_node rb;
712 		unsigned long rb_subtree_last;
713 	} shared;
714 
715 	/*
716 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
717 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
718 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
719 	 * or brk vma (with NULL file) can only be in an anon_vma list.
720 	 */
721 	struct list_head anon_vma_chain; /* Serialized by mmap_lock &
722 					  * page_table_lock */
723 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
724 
725 	/* Function pointers to deal with this struct. */
726 	const struct vm_operations_struct *vm_ops;
727 
728 	/* Information about our backing store: */
729 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
730 					   units */
731 	struct file * vm_file;		/* File we map to (can be NULL). */
732 	void * vm_private_data;		/* was vm_pte (shared mem) */
733 
734 #ifdef CONFIG_ANON_VMA_NAME
735 	/*
736 	 * For private and shared anonymous mappings, a pointer to a null
737 	 * terminated string containing the name given to the vma, or NULL if
738 	 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
739 	 */
740 	struct anon_vma_name *anon_name;
741 #endif
742 #ifdef CONFIG_SWAP
743 	atomic_long_t swap_readahead_info;
744 #endif
745 #ifndef CONFIG_MMU
746 	struct vm_region *vm_region;	/* NOMMU mapping region */
747 #endif
748 #ifdef CONFIG_NUMA
749 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
750 #endif
751 #ifdef CONFIG_NUMA_BALANCING
752 	struct vma_numab_state *numab_state;	/* NUMA Balancing state */
753 #endif
754 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
755 } __randomize_layout;
756 
757 #ifdef CONFIG_NUMA
758 #define vma_policy(vma) ((vma)->vm_policy)
759 #else
760 #define vma_policy(vma) NULL
761 #endif
762 
763 #ifdef CONFIG_SCHED_MM_CID
764 struct mm_cid {
765 	u64 time;
766 	int cid;
767 	int recent_cid;
768 };
769 #endif
770 
771 struct kioctx_table;
772 struct iommu_mm_data;
773 struct mm_struct {
774 	struct {
775 		/*
776 		 * Fields which are often written to are placed in a separate
777 		 * cache line.
778 		 */
779 		struct {
780 			/**
781 			 * @mm_count: The number of references to &struct
782 			 * mm_struct (@mm_users count as 1).
783 			 *
784 			 * Use mmgrab()/mmdrop() to modify. When this drops to
785 			 * 0, the &struct mm_struct is freed.
786 			 */
787 			atomic_t mm_count;
788 		} ____cacheline_aligned_in_smp;
789 
790 		struct maple_tree mm_mt;
791 
792 		unsigned long mmap_base;	/* base of mmap area */
793 		unsigned long mmap_legacy_base;	/* base of mmap area in bottom-up allocations */
794 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
795 		/* Base addresses for compatible mmap() */
796 		unsigned long mmap_compat_base;
797 		unsigned long mmap_compat_legacy_base;
798 #endif
799 		unsigned long task_size;	/* size of task vm space */
800 		pgd_t * pgd;
801 
802 #ifdef CONFIG_MEMBARRIER
803 		/**
804 		 * @membarrier_state: Flags controlling membarrier behavior.
805 		 *
806 		 * This field is close to @pgd to hopefully fit in the same
807 		 * cache-line, which needs to be touched by switch_mm().
808 		 */
809 		atomic_t membarrier_state;
810 #endif
811 
812 		/**
813 		 * @mm_users: The number of users including userspace.
814 		 *
815 		 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
816 		 * drops to 0 (i.e. when the task exits and there are no other
817 		 * temporary reference holders), we also release a reference on
818 		 * @mm_count (which may then free the &struct mm_struct if
819 		 * @mm_count also drops to 0).
820 		 */
821 		atomic_t mm_users;
822 
823 #ifdef CONFIG_SCHED_MM_CID
824 		/**
825 		 * @pcpu_cid: Per-cpu current cid.
826 		 *
827 		 * Keep track of the currently allocated mm_cid for each cpu.
828 		 * The per-cpu mm_cid values are serialized by their respective
829 		 * runqueue locks.
830 		 */
831 		struct mm_cid __percpu *pcpu_cid;
832 		/*
833 		 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
834 		 *
835 		 * When the next mm_cid scan is due (in jiffies).
836 		 */
837 		unsigned long mm_cid_next_scan;
838 		/**
839 		 * @nr_cpus_allowed: Number of CPUs allowed for mm.
840 		 *
841 		 * Number of CPUs allowed in the union of all mm's
842 		 * threads allowed CPUs.
843 		 */
844 		unsigned int nr_cpus_allowed;
845 		/**
846 		 * @max_nr_cid: Maximum number of concurrency IDs allocated.
847 		 *
848 		 * Track the highest number of concurrency IDs allocated for the
849 		 * mm.
850 		 */
851 		atomic_t max_nr_cid;
852 		/**
853 		 * @cpus_allowed_lock: Lock protecting mm cpus_allowed.
854 		 *
855 		 * Provide mutual exclusion for mm cpus_allowed and
856 		 * mm nr_cpus_allowed updates.
857 		 */
858 		raw_spinlock_t cpus_allowed_lock;
859 #endif
860 #ifdef CONFIG_MMU
861 		atomic_long_t pgtables_bytes;	/* size of all page tables */
862 #endif
863 		int map_count;			/* number of VMAs */
864 
865 		spinlock_t page_table_lock; /* Protects page tables and some
866 					     * counters
867 					     */
868 		/*
869 		 * With some kernel config, the current mmap_lock's offset
870 		 * inside 'mm_struct' is at 0x120, which is very optimal, as
871 		 * its two hot fields 'count' and 'owner' sit in 2 different
872 		 * cachelines,  and when mmap_lock is highly contended, both
873 		 * of the 2 fields will be accessed frequently, current layout
874 		 * will help to reduce cache bouncing.
875 		 *
876 		 * So please be careful with adding new fields before
877 		 * mmap_lock, which can easily push the 2 fields into one
878 		 * cacheline.
879 		 */
880 		struct rw_semaphore mmap_lock;
881 
882 		struct list_head mmlist; /* List of maybe swapped mm's.	These
883 					  * are globally strung together off
884 					  * init_mm.mmlist, and are protected
885 					  * by mmlist_lock
886 					  */
887 #ifdef CONFIG_PER_VMA_LOCK
888 		/*
889 		 * This field has lock-like semantics, meaning it is sometimes
890 		 * accessed with ACQUIRE/RELEASE semantics.
891 		 * Roughly speaking, incrementing the sequence number is
892 		 * equivalent to releasing locks on VMAs; reading the sequence
893 		 * number can be part of taking a read lock on a VMA.
894 		 *
895 		 * Can be modified under write mmap_lock using RELEASE
896 		 * semantics.
897 		 * Can be read with no other protection when holding write
898 		 * mmap_lock.
899 		 * Can be read with ACQUIRE semantics if not holding write
900 		 * mmap_lock.
901 		 */
902 		int mm_lock_seq;
903 #endif
904 
905 
906 		unsigned long hiwater_rss; /* High-watermark of RSS usage */
907 		unsigned long hiwater_vm;  /* High-water virtual memory usage */
908 
909 		unsigned long total_vm;	   /* Total pages mapped */
910 		unsigned long locked_vm;   /* Pages that have PG_mlocked set */
911 		atomic64_t    pinned_vm;   /* Refcount permanently increased */
912 		unsigned long data_vm;	   /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
913 		unsigned long exec_vm;	   /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
914 		unsigned long stack_vm;	   /* VM_STACK */
915 		unsigned long def_flags;
916 
917 		/**
918 		 * @write_protect_seq: Locked when any thread is write
919 		 * protecting pages mapped by this mm to enforce a later COW,
920 		 * for instance during page table copying for fork().
921 		 */
922 		seqcount_t write_protect_seq;
923 
924 		spinlock_t arg_lock; /* protect the below fields */
925 
926 		unsigned long start_code, end_code, start_data, end_data;
927 		unsigned long start_brk, brk, start_stack;
928 		unsigned long arg_start, arg_end, env_start, env_end;
929 
930 		unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
931 
932 		struct percpu_counter rss_stat[NR_MM_COUNTERS];
933 
934 		struct linux_binfmt *binfmt;
935 
936 		/* Architecture-specific MM context */
937 		mm_context_t context;
938 
939 		unsigned long flags; /* Must use atomic bitops to access */
940 
941 #ifdef CONFIG_AIO
942 		spinlock_t			ioctx_lock;
943 		struct kioctx_table __rcu	*ioctx_table;
944 #endif
945 #ifdef CONFIG_MEMCG
946 		/*
947 		 * "owner" points to a task that is regarded as the canonical
948 		 * user/owner of this mm. All of the following must be true in
949 		 * order for it to be changed:
950 		 *
951 		 * current == mm->owner
952 		 * current->mm != mm
953 		 * new_owner->mm == mm
954 		 * new_owner->alloc_lock is held
955 		 */
956 		struct task_struct __rcu *owner;
957 #endif
958 		struct user_namespace *user_ns;
959 
960 		/* store ref to file /proc/<pid>/exe symlink points to */
961 		struct file __rcu *exe_file;
962 #ifdef CONFIG_MMU_NOTIFIER
963 		struct mmu_notifier_subscriptions *notifier_subscriptions;
964 #endif
965 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
966 		pgtable_t pmd_huge_pte; /* protected by page_table_lock */
967 #endif
968 #ifdef CONFIG_NUMA_BALANCING
969 		/*
970 		 * numa_next_scan is the next time that PTEs will be remapped
971 		 * PROT_NONE to trigger NUMA hinting faults; such faults gather
972 		 * statistics and migrate pages to new nodes if necessary.
973 		 */
974 		unsigned long numa_next_scan;
975 
976 		/* Restart point for scanning and remapping PTEs. */
977 		unsigned long numa_scan_offset;
978 
979 		/* numa_scan_seq prevents two threads remapping PTEs. */
980 		int numa_scan_seq;
981 #endif
982 		/*
983 		 * An operation with batched TLB flushing is going on. Anything
984 		 * that can move process memory needs to flush the TLB when
985 		 * moving a PROT_NONE mapped page.
986 		 */
987 		atomic_t tlb_flush_pending;
988 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
989 		/* See flush_tlb_batched_pending() */
990 		atomic_t tlb_flush_batched;
991 #endif
992 		struct uprobes_state uprobes_state;
993 #ifdef CONFIG_PREEMPT_RT
994 		struct rcu_head delayed_drop;
995 #endif
996 #ifdef CONFIG_HUGETLB_PAGE
997 		atomic_long_t hugetlb_usage;
998 #endif
999 		struct work_struct async_put_work;
1000 
1001 #ifdef CONFIG_IOMMU_MM_DATA
1002 		struct iommu_mm_data *iommu_mm;
1003 #endif
1004 #ifdef CONFIG_KSM
1005 		/*
1006 		 * Represent how many pages of this process are involved in KSM
1007 		 * merging (not including ksm_zero_pages).
1008 		 */
1009 		unsigned long ksm_merging_pages;
1010 		/*
1011 		 * Represent how many pages are checked for ksm merging
1012 		 * including merged and not merged.
1013 		 */
1014 		unsigned long ksm_rmap_items;
1015 		/*
1016 		 * Represent how many empty pages are merged with kernel zero
1017 		 * pages when enabling KSM use_zero_pages.
1018 		 */
1019 		atomic_long_t ksm_zero_pages;
1020 #endif /* CONFIG_KSM */
1021 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1022 		struct {
1023 			/* this mm_struct is on lru_gen_mm_list */
1024 			struct list_head list;
1025 			/*
1026 			 * Set when switching to this mm_struct, as a hint of
1027 			 * whether it has been used since the last time per-node
1028 			 * page table walkers cleared the corresponding bits.
1029 			 */
1030 			unsigned long bitmap;
1031 #ifdef CONFIG_MEMCG
1032 			/* points to the memcg of "owner" above */
1033 			struct mem_cgroup *memcg;
1034 #endif
1035 		} lru_gen;
1036 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1037 	} __randomize_layout;
1038 
1039 	/*
1040 	 * The mm_cpumask needs to be at the end of mm_struct, because it
1041 	 * is dynamically sized based on nr_cpu_ids.
1042 	 */
1043 	unsigned long cpu_bitmap[];
1044 };
1045 
1046 #define MM_MT_FLAGS	(MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
1047 			 MT_FLAGS_USE_RCU)
1048 extern struct mm_struct init_mm;
1049 
1050 /* Pointer magic because the dynamic array size confuses some compilers. */
1051 static inline void mm_init_cpumask(struct mm_struct *mm)
1052 {
1053 	unsigned long cpu_bitmap = (unsigned long)mm;
1054 
1055 	cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1056 	cpumask_clear((struct cpumask *)cpu_bitmap);
1057 }
1058 
1059 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
1060 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
1061 {
1062 	return (struct cpumask *)&mm->cpu_bitmap;
1063 }
1064 
1065 #ifdef CONFIG_LRU_GEN
1066 
1067 struct lru_gen_mm_list {
1068 	/* mm_struct list for page table walkers */
1069 	struct list_head fifo;
1070 	/* protects the list above */
1071 	spinlock_t lock;
1072 };
1073 
1074 #endif /* CONFIG_LRU_GEN */
1075 
1076 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1077 
1078 void lru_gen_add_mm(struct mm_struct *mm);
1079 void lru_gen_del_mm(struct mm_struct *mm);
1080 void lru_gen_migrate_mm(struct mm_struct *mm);
1081 
1082 static inline void lru_gen_init_mm(struct mm_struct *mm)
1083 {
1084 	INIT_LIST_HEAD(&mm->lru_gen.list);
1085 	mm->lru_gen.bitmap = 0;
1086 #ifdef CONFIG_MEMCG
1087 	mm->lru_gen.memcg = NULL;
1088 #endif
1089 }
1090 
1091 static inline void lru_gen_use_mm(struct mm_struct *mm)
1092 {
1093 	/*
1094 	 * When the bitmap is set, page reclaim knows this mm_struct has been
1095 	 * used since the last time it cleared the bitmap. So it might be worth
1096 	 * walking the page tables of this mm_struct to clear the accessed bit.
1097 	 */
1098 	WRITE_ONCE(mm->lru_gen.bitmap, -1);
1099 }
1100 
1101 #else /* !CONFIG_LRU_GEN_WALKS_MMU */
1102 
1103 static inline void lru_gen_add_mm(struct mm_struct *mm)
1104 {
1105 }
1106 
1107 static inline void lru_gen_del_mm(struct mm_struct *mm)
1108 {
1109 }
1110 
1111 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
1112 {
1113 }
1114 
1115 static inline void lru_gen_init_mm(struct mm_struct *mm)
1116 {
1117 }
1118 
1119 static inline void lru_gen_use_mm(struct mm_struct *mm)
1120 {
1121 }
1122 
1123 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1124 
1125 struct vma_iterator {
1126 	struct ma_state mas;
1127 };
1128 
1129 #define VMA_ITERATOR(name, __mm, __addr)				\
1130 	struct vma_iterator name = {					\
1131 		.mas = {						\
1132 			.tree = &(__mm)->mm_mt,				\
1133 			.index = __addr,				\
1134 			.node = NULL,					\
1135 			.status = ma_start,				\
1136 		},							\
1137 	}
1138 
1139 static inline void vma_iter_init(struct vma_iterator *vmi,
1140 		struct mm_struct *mm, unsigned long addr)
1141 {
1142 	mas_init(&vmi->mas, &mm->mm_mt, addr);
1143 }
1144 
1145 #ifdef CONFIG_SCHED_MM_CID
1146 
1147 enum mm_cid_state {
1148 	MM_CID_UNSET = -1U,		/* Unset state has lazy_put flag set. */
1149 	MM_CID_LAZY_PUT = (1U << 31),
1150 };
1151 
1152 static inline bool mm_cid_is_unset(int cid)
1153 {
1154 	return cid == MM_CID_UNSET;
1155 }
1156 
1157 static inline bool mm_cid_is_lazy_put(int cid)
1158 {
1159 	return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1160 }
1161 
1162 static inline bool mm_cid_is_valid(int cid)
1163 {
1164 	return !(cid & MM_CID_LAZY_PUT);
1165 }
1166 
1167 static inline int mm_cid_set_lazy_put(int cid)
1168 {
1169 	return cid | MM_CID_LAZY_PUT;
1170 }
1171 
1172 static inline int mm_cid_clear_lazy_put(int cid)
1173 {
1174 	return cid & ~MM_CID_LAZY_PUT;
1175 }
1176 
1177 /*
1178  * mm_cpus_allowed: Union of all mm's threads allowed CPUs.
1179  */
1180 static inline cpumask_t *mm_cpus_allowed(struct mm_struct *mm)
1181 {
1182 	unsigned long bitmap = (unsigned long)mm;
1183 
1184 	bitmap += offsetof(struct mm_struct, cpu_bitmap);
1185 	/* Skip cpu_bitmap */
1186 	bitmap += cpumask_size();
1187 	return (struct cpumask *)bitmap;
1188 }
1189 
1190 /* Accessor for struct mm_struct's cidmask. */
1191 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1192 {
1193 	unsigned long cid_bitmap = (unsigned long)mm_cpus_allowed(mm);
1194 
1195 	/* Skip mm_cpus_allowed */
1196 	cid_bitmap += cpumask_size();
1197 	return (struct cpumask *)cid_bitmap;
1198 }
1199 
1200 static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p)
1201 {
1202 	int i;
1203 
1204 	for_each_possible_cpu(i) {
1205 		struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1206 
1207 		pcpu_cid->cid = MM_CID_UNSET;
1208 		pcpu_cid->recent_cid = MM_CID_UNSET;
1209 		pcpu_cid->time = 0;
1210 	}
1211 	mm->nr_cpus_allowed = p->nr_cpus_allowed;
1212 	atomic_set(&mm->max_nr_cid, 0);
1213 	raw_spin_lock_init(&mm->cpus_allowed_lock);
1214 	cpumask_copy(mm_cpus_allowed(mm), &p->cpus_mask);
1215 	cpumask_clear(mm_cidmask(mm));
1216 }
1217 
1218 static inline int mm_alloc_cid_noprof(struct mm_struct *mm, struct task_struct *p)
1219 {
1220 	mm->pcpu_cid = alloc_percpu_noprof(struct mm_cid);
1221 	if (!mm->pcpu_cid)
1222 		return -ENOMEM;
1223 	mm_init_cid(mm, p);
1224 	return 0;
1225 }
1226 #define mm_alloc_cid(...)	alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__))
1227 
1228 static inline void mm_destroy_cid(struct mm_struct *mm)
1229 {
1230 	free_percpu(mm->pcpu_cid);
1231 	mm->pcpu_cid = NULL;
1232 }
1233 
1234 static inline unsigned int mm_cid_size(void)
1235 {
1236 	return 2 * cpumask_size();	/* mm_cpus_allowed(), mm_cidmask(). */
1237 }
1238 
1239 static inline void mm_set_cpus_allowed(struct mm_struct *mm, const struct cpumask *cpumask)
1240 {
1241 	struct cpumask *mm_allowed = mm_cpus_allowed(mm);
1242 
1243 	if (!mm)
1244 		return;
1245 	/* The mm_cpus_allowed is the union of each thread allowed CPUs masks. */
1246 	raw_spin_lock(&mm->cpus_allowed_lock);
1247 	cpumask_or(mm_allowed, mm_allowed, cpumask);
1248 	WRITE_ONCE(mm->nr_cpus_allowed, cpumask_weight(mm_allowed));
1249 	raw_spin_unlock(&mm->cpus_allowed_lock);
1250 }
1251 #else /* CONFIG_SCHED_MM_CID */
1252 static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p) { }
1253 static inline int mm_alloc_cid(struct mm_struct *mm, struct task_struct *p) { return 0; }
1254 static inline void mm_destroy_cid(struct mm_struct *mm) { }
1255 
1256 static inline unsigned int mm_cid_size(void)
1257 {
1258 	return 0;
1259 }
1260 static inline void mm_set_cpus_allowed(struct mm_struct *mm, const struct cpumask *cpumask) { }
1261 #endif /* CONFIG_SCHED_MM_CID */
1262 
1263 struct mmu_gather;
1264 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1265 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1266 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1267 
1268 struct vm_fault;
1269 
1270 /**
1271  * typedef vm_fault_t - Return type for page fault handlers.
1272  *
1273  * Page fault handlers return a bitmask of %VM_FAULT values.
1274  */
1275 typedef __bitwise unsigned int vm_fault_t;
1276 
1277 /**
1278  * enum vm_fault_reason - Page fault handlers return a bitmask of
1279  * these values to tell the core VM what happened when handling the
1280  * fault. Used to decide whether a process gets delivered SIGBUS or
1281  * just gets major/minor fault counters bumped up.
1282  *
1283  * @VM_FAULT_OOM:		Out Of Memory
1284  * @VM_FAULT_SIGBUS:		Bad access
1285  * @VM_FAULT_MAJOR:		Page read from storage
1286  * @VM_FAULT_HWPOISON:		Hit poisoned small page
1287  * @VM_FAULT_HWPOISON_LARGE:	Hit poisoned large page. Index encoded
1288  *				in upper bits
1289  * @VM_FAULT_SIGSEGV:		segmentation fault
1290  * @VM_FAULT_NOPAGE:		->fault installed the pte, not return page
1291  * @VM_FAULT_LOCKED:		->fault locked the returned page
1292  * @VM_FAULT_RETRY:		->fault blocked, must retry
1293  * @VM_FAULT_FALLBACK:		huge page fault failed, fall back to small
1294  * @VM_FAULT_DONE_COW:		->fault has fully handled COW
1295  * @VM_FAULT_NEEDDSYNC:		->fault did not modify page tables and needs
1296  *				fsync() to complete (for synchronous page faults
1297  *				in DAX)
1298  * @VM_FAULT_COMPLETED:		->fault completed, meanwhile mmap lock released
1299  * @VM_FAULT_HINDEX_MASK:	mask HINDEX value
1300  *
1301  */
1302 enum vm_fault_reason {
1303 	VM_FAULT_OOM            = (__force vm_fault_t)0x000001,
1304 	VM_FAULT_SIGBUS         = (__force vm_fault_t)0x000002,
1305 	VM_FAULT_MAJOR          = (__force vm_fault_t)0x000004,
1306 	VM_FAULT_HWPOISON       = (__force vm_fault_t)0x000010,
1307 	VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1308 	VM_FAULT_SIGSEGV        = (__force vm_fault_t)0x000040,
1309 	VM_FAULT_NOPAGE         = (__force vm_fault_t)0x000100,
1310 	VM_FAULT_LOCKED         = (__force vm_fault_t)0x000200,
1311 	VM_FAULT_RETRY          = (__force vm_fault_t)0x000400,
1312 	VM_FAULT_FALLBACK       = (__force vm_fault_t)0x000800,
1313 	VM_FAULT_DONE_COW       = (__force vm_fault_t)0x001000,
1314 	VM_FAULT_NEEDDSYNC      = (__force vm_fault_t)0x002000,
1315 	VM_FAULT_COMPLETED      = (__force vm_fault_t)0x004000,
1316 	VM_FAULT_HINDEX_MASK    = (__force vm_fault_t)0x0f0000,
1317 };
1318 
1319 /* Encode hstate index for a hwpoisoned large page */
1320 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1321 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1322 
1323 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS |	\
1324 			VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON |	\
1325 			VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1326 
1327 #define VM_FAULT_RESULT_TRACE \
1328 	{ VM_FAULT_OOM,                 "OOM" },	\
1329 	{ VM_FAULT_SIGBUS,              "SIGBUS" },	\
1330 	{ VM_FAULT_MAJOR,               "MAJOR" },	\
1331 	{ VM_FAULT_HWPOISON,            "HWPOISON" },	\
1332 	{ VM_FAULT_HWPOISON_LARGE,      "HWPOISON_LARGE" },	\
1333 	{ VM_FAULT_SIGSEGV,             "SIGSEGV" },	\
1334 	{ VM_FAULT_NOPAGE,              "NOPAGE" },	\
1335 	{ VM_FAULT_LOCKED,              "LOCKED" },	\
1336 	{ VM_FAULT_RETRY,               "RETRY" },	\
1337 	{ VM_FAULT_FALLBACK,            "FALLBACK" },	\
1338 	{ VM_FAULT_DONE_COW,            "DONE_COW" },	\
1339 	{ VM_FAULT_NEEDDSYNC,           "NEEDDSYNC" },	\
1340 	{ VM_FAULT_COMPLETED,           "COMPLETED" }
1341 
1342 struct vm_special_mapping {
1343 	const char *name;	/* The name, e.g. "[vdso]". */
1344 
1345 	/*
1346 	 * If .fault is not provided, this points to a
1347 	 * NULL-terminated array of pages that back the special mapping.
1348 	 *
1349 	 * This must not be NULL unless .fault is provided.
1350 	 */
1351 	struct page **pages;
1352 
1353 	/*
1354 	 * If non-NULL, then this is called to resolve page faults
1355 	 * on the special mapping.  If used, .pages is not checked.
1356 	 */
1357 	vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1358 				struct vm_area_struct *vma,
1359 				struct vm_fault *vmf);
1360 
1361 	int (*mremap)(const struct vm_special_mapping *sm,
1362 		     struct vm_area_struct *new_vma);
1363 
1364 	void (*close)(const struct vm_special_mapping *sm,
1365 		      struct vm_area_struct *vma);
1366 };
1367 
1368 enum tlb_flush_reason {
1369 	TLB_FLUSH_ON_TASK_SWITCH,
1370 	TLB_REMOTE_SHOOTDOWN,
1371 	TLB_LOCAL_SHOOTDOWN,
1372 	TLB_LOCAL_MM_SHOOTDOWN,
1373 	TLB_REMOTE_SEND_IPI,
1374 	NR_TLB_FLUSH_REASONS,
1375 };
1376 
1377 /**
1378  * enum fault_flag - Fault flag definitions.
1379  * @FAULT_FLAG_WRITE: Fault was a write fault.
1380  * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1381  * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1382  * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1383  * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1384  * @FAULT_FLAG_TRIED: The fault has been tried once.
1385  * @FAULT_FLAG_USER: The fault originated in userspace.
1386  * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1387  * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1388  * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1389  * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1390  *                      COW mapping, making sure that an exclusive anon page is
1391  *                      mapped after the fault.
1392  * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1393  *                        We should only access orig_pte if this flag set.
1394  * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1395  *
1396  * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1397  * whether we would allow page faults to retry by specifying these two
1398  * fault flags correctly.  Currently there can be three legal combinations:
1399  *
1400  * (a) ALLOW_RETRY and !TRIED:  this means the page fault allows retry, and
1401  *                              this is the first try
1402  *
1403  * (b) ALLOW_RETRY and TRIED:   this means the page fault allows retry, and
1404  *                              we've already tried at least once
1405  *
1406  * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1407  *
1408  * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1409  * be used.  Note that page faults can be allowed to retry for multiple times,
1410  * in which case we'll have an initial fault with flags (a) then later on
1411  * continuous faults with flags (b).  We should always try to detect pending
1412  * signals before a retry to make sure the continuous page faults can still be
1413  * interrupted if necessary.
1414  *
1415  * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1416  * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1417  * applied to mappings that are not COW mappings.
1418  */
1419 enum fault_flag {
1420 	FAULT_FLAG_WRITE =		1 << 0,
1421 	FAULT_FLAG_MKWRITE =		1 << 1,
1422 	FAULT_FLAG_ALLOW_RETRY =	1 << 2,
1423 	FAULT_FLAG_RETRY_NOWAIT = 	1 << 3,
1424 	FAULT_FLAG_KILLABLE =		1 << 4,
1425 	FAULT_FLAG_TRIED = 		1 << 5,
1426 	FAULT_FLAG_USER =		1 << 6,
1427 	FAULT_FLAG_REMOTE =		1 << 7,
1428 	FAULT_FLAG_INSTRUCTION =	1 << 8,
1429 	FAULT_FLAG_INTERRUPTIBLE =	1 << 9,
1430 	FAULT_FLAG_UNSHARE =		1 << 10,
1431 	FAULT_FLAG_ORIG_PTE_VALID =	1 << 11,
1432 	FAULT_FLAG_VMA_LOCK =		1 << 12,
1433 };
1434 
1435 typedef unsigned int __bitwise zap_flags_t;
1436 
1437 /* Flags for clear_young_dirty_ptes(). */
1438 typedef int __bitwise cydp_t;
1439 
1440 /* Clear the access bit */
1441 #define CYDP_CLEAR_YOUNG		((__force cydp_t)BIT(0))
1442 
1443 /* Clear the dirty bit */
1444 #define CYDP_CLEAR_DIRTY		((__force cydp_t)BIT(1))
1445 
1446 /*
1447  * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1448  * other. Here is what they mean, and how to use them:
1449  *
1450  *
1451  * FIXME: For pages which are part of a filesystem, mappings are subject to the
1452  * lifetime enforced by the filesystem and we need guarantees that longterm
1453  * users like RDMA and V4L2 only establish mappings which coordinate usage with
1454  * the filesystem.  Ideas for this coordination include revoking the longterm
1455  * pin, delaying writeback, bounce buffer page writeback, etc.  As FS DAX was
1456  * added after the problem with filesystems was found FS DAX VMAs are
1457  * specifically failed.  Filesystem pages are still subject to bugs and use of
1458  * FOLL_LONGTERM should be avoided on those pages.
1459  *
1460  * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1461  * that region.  And so, CMA attempts to migrate the page before pinning, when
1462  * FOLL_LONGTERM is specified.
1463  *
1464  * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1465  * but an additional pin counting system) will be invoked. This is intended for
1466  * anything that gets a page reference and then touches page data (for example,
1467  * Direct IO). This lets the filesystem know that some non-file-system entity is
1468  * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1469  * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1470  * a call to unpin_user_page().
1471  *
1472  * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1473  * and separate refcounting mechanisms, however, and that means that each has
1474  * its own acquire and release mechanisms:
1475  *
1476  *     FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1477  *
1478  *     FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1479  *
1480  * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1481  * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1482  * calls applied to them, and that's perfectly OK. This is a constraint on the
1483  * callers, not on the pages.)
1484  *
1485  * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1486  * directly by the caller. That's in order to help avoid mismatches when
1487  * releasing pages: get_user_pages*() pages must be released via put_page(),
1488  * while pin_user_pages*() pages must be released via unpin_user_page().
1489  *
1490  * Please see Documentation/core-api/pin_user_pages.rst for more information.
1491  */
1492 
1493 enum {
1494 	/* check pte is writable */
1495 	FOLL_WRITE = 1 << 0,
1496 	/* do get_page on page */
1497 	FOLL_GET = 1 << 1,
1498 	/* give error on hole if it would be zero */
1499 	FOLL_DUMP = 1 << 2,
1500 	/* get_user_pages read/write w/o permission */
1501 	FOLL_FORCE = 1 << 3,
1502 	/*
1503 	 * if a disk transfer is needed, start the IO and return without waiting
1504 	 * upon it
1505 	 */
1506 	FOLL_NOWAIT = 1 << 4,
1507 	/* do not fault in pages */
1508 	FOLL_NOFAULT = 1 << 5,
1509 	/* check page is hwpoisoned */
1510 	FOLL_HWPOISON = 1 << 6,
1511 	/* don't do file mappings */
1512 	FOLL_ANON = 1 << 7,
1513 	/*
1514 	 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1515 	 * time period _often_ under userspace control.  This is in contrast to
1516 	 * iov_iter_get_pages(), whose usages are transient.
1517 	 */
1518 	FOLL_LONGTERM = 1 << 8,
1519 	/* split huge pmd before returning */
1520 	FOLL_SPLIT_PMD = 1 << 9,
1521 	/* allow returning PCI P2PDMA pages */
1522 	FOLL_PCI_P2PDMA = 1 << 10,
1523 	/* allow interrupts from generic signals */
1524 	FOLL_INTERRUPTIBLE = 1 << 11,
1525 	/*
1526 	 * Always honor (trigger) NUMA hinting faults.
1527 	 *
1528 	 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1529 	 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1530 	 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1531 	 * hinting faults.
1532 	 */
1533 	FOLL_HONOR_NUMA_FAULT = 1 << 12,
1534 
1535 	/* See also internal only FOLL flags in mm/internal.h */
1536 };
1537 
1538 /* mm flags */
1539 
1540 /*
1541  * The first two bits represent core dump modes for set-user-ID,
1542  * the modes are SUID_DUMP_* defined in linux/sched/coredump.h
1543  */
1544 #define MMF_DUMPABLE_BITS 2
1545 #define MMF_DUMPABLE_MASK ((1 << MMF_DUMPABLE_BITS) - 1)
1546 /* coredump filter bits */
1547 #define MMF_DUMP_ANON_PRIVATE	2
1548 #define MMF_DUMP_ANON_SHARED	3
1549 #define MMF_DUMP_MAPPED_PRIVATE	4
1550 #define MMF_DUMP_MAPPED_SHARED	5
1551 #define MMF_DUMP_ELF_HEADERS	6
1552 #define MMF_DUMP_HUGETLB_PRIVATE 7
1553 #define MMF_DUMP_HUGETLB_SHARED  8
1554 #define MMF_DUMP_DAX_PRIVATE	9
1555 #define MMF_DUMP_DAX_SHARED	10
1556 
1557 #define MMF_DUMP_FILTER_SHIFT	MMF_DUMPABLE_BITS
1558 #define MMF_DUMP_FILTER_BITS	9
1559 #define MMF_DUMP_FILTER_MASK \
1560 	(((1 << MMF_DUMP_FILTER_BITS) - 1) << MMF_DUMP_FILTER_SHIFT)
1561 #define MMF_DUMP_FILTER_DEFAULT \
1562 	((1 << MMF_DUMP_ANON_PRIVATE) |	(1 << MMF_DUMP_ANON_SHARED) |\
1563 	 (1 << MMF_DUMP_HUGETLB_PRIVATE) | MMF_DUMP_MASK_DEFAULT_ELF)
1564 
1565 #ifdef CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS
1566 # define MMF_DUMP_MASK_DEFAULT_ELF	(1 << MMF_DUMP_ELF_HEADERS)
1567 #else
1568 # define MMF_DUMP_MASK_DEFAULT_ELF	0
1569 #endif
1570 					/* leave room for more dump flags */
1571 #define MMF_VM_MERGEABLE	16	/* KSM may merge identical pages */
1572 #define MMF_VM_HUGEPAGE		17	/* set when mm is available for khugepaged */
1573 
1574 /*
1575  * This one-shot flag is dropped due to necessity of changing exe once again
1576  * on NFS restore
1577  */
1578 //#define MMF_EXE_FILE_CHANGED	18	/* see prctl_set_mm_exe_file() */
1579 
1580 #define MMF_HAS_UPROBES		19	/* has uprobes */
1581 #define MMF_RECALC_UPROBES	20	/* MMF_HAS_UPROBES can be wrong */
1582 #define MMF_OOM_SKIP		21	/* mm is of no interest for the OOM killer */
1583 #define MMF_UNSTABLE		22	/* mm is unstable for copy_from_user */
1584 #define MMF_HUGE_ZERO_PAGE	23      /* mm has ever used the global huge zero page */
1585 #define MMF_DISABLE_THP		24	/* disable THP for all VMAs */
1586 #define MMF_DISABLE_THP_MASK	(1 << MMF_DISABLE_THP)
1587 #define MMF_OOM_REAP_QUEUED	25	/* mm was queued for oom_reaper */
1588 #define MMF_MULTIPROCESS	26	/* mm is shared between processes */
1589 /*
1590  * MMF_HAS_PINNED: Whether this mm has pinned any pages.  This can be either
1591  * replaced in the future by mm.pinned_vm when it becomes stable, or grow into
1592  * a counter on its own. We're aggresive on this bit for now: even if the
1593  * pinned pages were unpinned later on, we'll still keep this bit set for the
1594  * lifecycle of this mm, just for simplicity.
1595  */
1596 #define MMF_HAS_PINNED		27	/* FOLL_PIN has run, never cleared */
1597 
1598 #define MMF_HAS_MDWE		28
1599 #define MMF_HAS_MDWE_MASK	(1 << MMF_HAS_MDWE)
1600 
1601 
1602 #define MMF_HAS_MDWE_NO_INHERIT	29
1603 
1604 #define MMF_VM_MERGE_ANY	30
1605 #define MMF_VM_MERGE_ANY_MASK	(1 << MMF_VM_MERGE_ANY)
1606 
1607 #define MMF_TOPDOWN		31	/* mm searches top down by default */
1608 #define MMF_TOPDOWN_MASK	(1 << MMF_TOPDOWN)
1609 
1610 #define MMF_INIT_MASK		(MMF_DUMPABLE_MASK | MMF_DUMP_FILTER_MASK |\
1611 				 MMF_DISABLE_THP_MASK | MMF_HAS_MDWE_MASK |\
1612 				 MMF_VM_MERGE_ANY_MASK | MMF_TOPDOWN_MASK)
1613 
1614 static inline unsigned long mmf_init_flags(unsigned long flags)
1615 {
1616 	if (flags & (1UL << MMF_HAS_MDWE_NO_INHERIT))
1617 		flags &= ~((1UL << MMF_HAS_MDWE) |
1618 			   (1UL << MMF_HAS_MDWE_NO_INHERIT));
1619 	return flags & MMF_INIT_MASK;
1620 }
1621 
1622 #endif /* _LINUX_MM_TYPES_H */
1623