xref: /linux-6.15/include/linux/mm_types.h (revision e120d1bc)
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 			/*
134 			 * The first word is used for compound_head or folio
135 			 * pgmap
136 			 */
137 			void *_unused_pgmap_compound_head;
138 			void *zone_device_data;
139 			/*
140 			 * ZONE_DEVICE private pages are counted as being
141 			 * mapped so the next 3 words hold the mapping, index,
142 			 * and private fields from the source anonymous or
143 			 * page cache page while the page is migrated to device
144 			 * private memory.
145 			 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
146 			 * use the mapping, index, and private fields when
147 			 * pmem backed DAX files are mapped.
148 			 */
149 		};
150 
151 		/** @rcu_head: You can use this to free a page by RCU. */
152 		struct rcu_head rcu_head;
153 	};
154 
155 	union {		/* This union is 4 bytes in size. */
156 		/*
157 		 * For head pages of typed folios, the value stored here
158 		 * allows for determining what this page is used for. The
159 		 * tail pages of typed folios will not store a type
160 		 * (page_type == _mapcount == -1).
161 		 *
162 		 * See page-flags.h for a list of page types which are currently
163 		 * stored here.
164 		 *
165 		 * Owners of typed folios may reuse the lower 16 bit of the
166 		 * head page page_type field after setting the page type,
167 		 * but must reset these 16 bit to -1 before clearing the
168 		 * page type.
169 		 */
170 		unsigned int page_type;
171 
172 		/*
173 		 * For pages that are part of non-typed folios for which mappings
174 		 * are tracked via the RMAP, encodes the number of times this page
175 		 * is directly referenced by a page table.
176 		 *
177 		 * Note that the mapcount is always initialized to -1, so that
178 		 * transitions both from it and to it can be tracked, using
179 		 * atomic_inc_and_test() and atomic_add_negative(-1).
180 		 */
181 		atomic_t _mapcount;
182 	};
183 
184 	/* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
185 	atomic_t _refcount;
186 
187 #ifdef CONFIG_MEMCG
188 	unsigned long memcg_data;
189 #elif defined(CONFIG_SLAB_OBJ_EXT)
190 	unsigned long _unused_slab_obj_exts;
191 #endif
192 
193 	/*
194 	 * On machines where all RAM is mapped into kernel address space,
195 	 * we can simply calculate the virtual address. On machines with
196 	 * highmem some memory is mapped into kernel virtual memory
197 	 * dynamically, so we need a place to store that address.
198 	 * Note that this field could be 16 bits on x86 ... ;)
199 	 *
200 	 * Architectures with slow multiplication can define
201 	 * WANT_PAGE_VIRTUAL in asm/page.h
202 	 */
203 #if defined(WANT_PAGE_VIRTUAL)
204 	void *virtual;			/* Kernel virtual address (NULL if
205 					   not kmapped, ie. highmem) */
206 #endif /* WANT_PAGE_VIRTUAL */
207 
208 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
209 	int _last_cpupid;
210 #endif
211 
212 #ifdef CONFIG_KMSAN
213 	/*
214 	 * KMSAN metadata for this page:
215 	 *  - shadow page: every bit indicates whether the corresponding
216 	 *    bit of the original page is initialized (0) or not (1);
217 	 *  - origin page: every 4 bytes contain an id of the stack trace
218 	 *    where the uninitialized value was created.
219 	 */
220 	struct page *kmsan_shadow;
221 	struct page *kmsan_origin;
222 #endif
223 } _struct_page_alignment;
224 
225 /*
226  * struct encoded_page - a nonexistent type marking this pointer
227  *
228  * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
229  * with the low bits of the pointer indicating extra context-dependent
230  * information. Only used in mmu_gather handling, and this acts as a type
231  * system check on that use.
232  *
233  * We only really have two guaranteed bits in general, although you could
234  * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
235  * for more.
236  *
237  * Use the supplied helper functions to endcode/decode the pointer and bits.
238  */
239 struct encoded_page;
240 
241 #define ENCODED_PAGE_BITS			3ul
242 
243 /* Perform rmap removal after we have flushed the TLB. */
244 #define ENCODED_PAGE_BIT_DELAY_RMAP		1ul
245 
246 /*
247  * The next item in an encoded_page array is the "nr_pages" argument, specifying
248  * the number of consecutive pages starting from this page, that all belong to
249  * the same folio. For example, "nr_pages" corresponds to the number of folio
250  * references that must be dropped. If this bit is not set, "nr_pages" is
251  * implicitly 1.
252  */
253 #define ENCODED_PAGE_BIT_NR_PAGES_NEXT		2ul
254 
255 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
256 {
257 	BUILD_BUG_ON(flags > ENCODED_PAGE_BITS);
258 	return (struct encoded_page *)(flags | (unsigned long)page);
259 }
260 
261 static inline unsigned long encoded_page_flags(struct encoded_page *page)
262 {
263 	return ENCODED_PAGE_BITS & (unsigned long)page;
264 }
265 
266 static inline struct page *encoded_page_ptr(struct encoded_page *page)
267 {
268 	return (struct page *)(~ENCODED_PAGE_BITS & (unsigned long)page);
269 }
270 
271 static __always_inline struct encoded_page *encode_nr_pages(unsigned long nr)
272 {
273 	VM_WARN_ON_ONCE((nr << 2) >> 2 != nr);
274 	return (struct encoded_page *)(nr << 2);
275 }
276 
277 static __always_inline unsigned long encoded_nr_pages(struct encoded_page *page)
278 {
279 	return ((unsigned long)page) >> 2;
280 }
281 
282 /*
283  * A swap entry has to fit into a "unsigned long", as the entry is hidden
284  * in the "index" field of the swapper address space.
285  */
286 typedef struct {
287 	unsigned long val;
288 } swp_entry_t;
289 
290 #if defined(CONFIG_MEMCG) || defined(CONFIG_SLAB_OBJ_EXT)
291 /* We have some extra room after the refcount in tail pages. */
292 #define NR_PAGES_IN_LARGE_FOLIO
293 #endif
294 
295 /*
296  * On 32bit, we can cut the required metadata in half, because:
297  * (a) PID_MAX_LIMIT implicitly limits the number of MMs we could ever have,
298  *     so we can limit MM IDs to 15 bit (32767).
299  * (b) We don't expect folios where even a single complete PTE mapping by
300  *     one MM would exceed 15 bits (order-15).
301  */
302 #ifdef CONFIG_64BIT
303 typedef int mm_id_mapcount_t;
304 #define MM_ID_MAPCOUNT_MAX		INT_MAX
305 typedef unsigned int mm_id_t;
306 #else /* !CONFIG_64BIT */
307 typedef short mm_id_mapcount_t;
308 #define MM_ID_MAPCOUNT_MAX		SHRT_MAX
309 typedef unsigned short mm_id_t;
310 #endif /* CONFIG_64BIT */
311 
312 /* We implicitly use the dummy ID for init-mm etc. where we never rmap pages. */
313 #define MM_ID_DUMMY			0
314 #define MM_ID_MIN			(MM_ID_DUMMY + 1)
315 
316 /*
317  * We leave the highest bit of each MM id unused, so we can store a flag
318  * in the highest bit of each folio->_mm_id[].
319  */
320 #define MM_ID_BITS			((sizeof(mm_id_t) * BITS_PER_BYTE) - 1)
321 #define MM_ID_MASK			((1U << MM_ID_BITS) - 1)
322 #define MM_ID_MAX			MM_ID_MASK
323 
324 /*
325  * In order to use bit_spin_lock(), which requires an unsigned long, we
326  * operate on folio->_mm_ids when working on flags.
327  */
328 #define FOLIO_MM_IDS_LOCK_BITNUM	MM_ID_BITS
329 #define FOLIO_MM_IDS_LOCK_BIT		BIT(FOLIO_MM_IDS_LOCK_BITNUM)
330 #define FOLIO_MM_IDS_SHARED_BITNUM	(2 * MM_ID_BITS + 1)
331 #define FOLIO_MM_IDS_SHARED_BIT		BIT(FOLIO_MM_IDS_SHARED_BITNUM)
332 
333 /**
334  * struct folio - Represents a contiguous set of bytes.
335  * @flags: Identical to the page flags.
336  * @lru: Least Recently Used list; tracks how recently this folio was used.
337  * @mlock_count: Number of times this folio has been pinned by mlock().
338  * @mapping: The file this page belongs to, or refers to the anon_vma for
339  *    anonymous memory.
340  * @index: Offset within the file, in units of pages.  For anonymous memory,
341  *    this is the index from the beginning of the mmap.
342  * @share: number of DAX mappings that reference this folio. See
343  *    dax_associate_entry.
344  * @private: Filesystem per-folio data (see folio_attach_private()).
345  * @swap: Used for swp_entry_t if folio_test_swapcache().
346  * @_mapcount: Do not access this member directly.  Use folio_mapcount() to
347  *    find out how many times this folio is mapped by userspace.
348  * @_refcount: Do not access this member directly.  Use folio_ref_count()
349  *    to find how many references there are to this folio.
350  * @memcg_data: Memory Control Group data.
351  * @pgmap: Metadata for ZONE_DEVICE mappings
352  * @virtual: Virtual address in the kernel direct map.
353  * @_last_cpupid: IDs of last CPU and last process that accessed the folio.
354  * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
355  * @_large_mapcount: Do not use directly, call folio_mapcount().
356  * @_nr_pages_mapped: Do not use outside of rmap and debug code.
357  * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
358  * @_nr_pages: Do not use directly, call folio_nr_pages().
359  * @_mm_id: Do not use outside of rmap code.
360  * @_mm_ids: Do not use outside of rmap code.
361  * @_mm_id_mapcount: Do not use outside of rmap code.
362  * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
363  * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
364  * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
365  * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
366  * @_deferred_list: Folios to be split under memory pressure.
367  * @_unused_slab_obj_exts: Placeholder to match obj_exts in struct slab.
368  *
369  * A folio is a physically, virtually and logically contiguous set
370  * of bytes.  It is a power-of-two in size, and it is aligned to that
371  * same power-of-two.  It is at least as large as %PAGE_SIZE.  If it is
372  * in the page cache, it is at a file offset which is a multiple of that
373  * power-of-two.  It may be mapped into userspace at an address which is
374  * at an arbitrary page offset, but its kernel virtual address is aligned
375  * to its size.
376  */
377 struct folio {
378 	/* private: don't document the anon union */
379 	union {
380 		struct {
381 	/* public: */
382 			unsigned long flags;
383 			union {
384 				struct list_head lru;
385 	/* private: avoid cluttering the output */
386 				struct {
387 					void *__filler;
388 	/* public: */
389 					unsigned int mlock_count;
390 	/* private: */
391 				};
392 	/* public: */
393 				struct dev_pagemap *pgmap;
394 			};
395 			struct address_space *mapping;
396 			union {
397 				pgoff_t index;
398 				unsigned long share;
399 			};
400 			union {
401 				void *private;
402 				swp_entry_t swap;
403 			};
404 			atomic_t _mapcount;
405 			atomic_t _refcount;
406 #ifdef CONFIG_MEMCG
407 			unsigned long memcg_data;
408 #elif defined(CONFIG_SLAB_OBJ_EXT)
409 			unsigned long _unused_slab_obj_exts;
410 #endif
411 #if defined(WANT_PAGE_VIRTUAL)
412 			void *virtual;
413 #endif
414 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
415 			int _last_cpupid;
416 #endif
417 	/* private: the union with struct page is transitional */
418 		};
419 		struct page page;
420 	};
421 	union {
422 		struct {
423 			unsigned long _flags_1;
424 			unsigned long _head_1;
425 			union {
426 				struct {
427 	/* public: */
428 					atomic_t _large_mapcount;
429 					atomic_t _nr_pages_mapped;
430 #ifdef CONFIG_64BIT
431 					atomic_t _entire_mapcount;
432 					atomic_t _pincount;
433 #endif /* CONFIG_64BIT */
434 					mm_id_mapcount_t _mm_id_mapcount[2];
435 					union {
436 						mm_id_t _mm_id[2];
437 						unsigned long _mm_ids;
438 					};
439 	/* private: the union with struct page is transitional */
440 				};
441 				unsigned long _usable_1[4];
442 			};
443 			atomic_t _mapcount_1;
444 			atomic_t _refcount_1;
445 	/* public: */
446 #ifdef NR_PAGES_IN_LARGE_FOLIO
447 			unsigned int _nr_pages;
448 #endif /* NR_PAGES_IN_LARGE_FOLIO */
449 	/* private: the union with struct page is transitional */
450 		};
451 		struct page __page_1;
452 	};
453 	union {
454 		struct {
455 			unsigned long _flags_2;
456 			unsigned long _head_2;
457 	/* public: */
458 			struct list_head _deferred_list;
459 #ifndef CONFIG_64BIT
460 			atomic_t _entire_mapcount;
461 			atomic_t _pincount;
462 #endif /* !CONFIG_64BIT */
463 	/* private: the union with struct page is transitional */
464 		};
465 		struct page __page_2;
466 	};
467 	union {
468 		struct {
469 			unsigned long _flags_3;
470 			unsigned long _head_3;
471 	/* public: */
472 			void *_hugetlb_subpool;
473 			void *_hugetlb_cgroup;
474 			void *_hugetlb_cgroup_rsvd;
475 			void *_hugetlb_hwpoison;
476 	/* private: the union with struct page is transitional */
477 		};
478 		struct page __page_3;
479 	};
480 };
481 
482 #define FOLIO_MATCH(pg, fl)						\
483 	static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
484 FOLIO_MATCH(flags, flags);
485 FOLIO_MATCH(lru, lru);
486 FOLIO_MATCH(mapping, mapping);
487 FOLIO_MATCH(compound_head, lru);
488 FOLIO_MATCH(index, index);
489 FOLIO_MATCH(private, private);
490 FOLIO_MATCH(_mapcount, _mapcount);
491 FOLIO_MATCH(_refcount, _refcount);
492 #ifdef CONFIG_MEMCG
493 FOLIO_MATCH(memcg_data, memcg_data);
494 #endif
495 #if defined(WANT_PAGE_VIRTUAL)
496 FOLIO_MATCH(virtual, virtual);
497 #endif
498 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
499 FOLIO_MATCH(_last_cpupid, _last_cpupid);
500 #endif
501 #undef FOLIO_MATCH
502 #define FOLIO_MATCH(pg, fl)						\
503 	static_assert(offsetof(struct folio, fl) ==			\
504 			offsetof(struct page, pg) + sizeof(struct page))
505 FOLIO_MATCH(flags, _flags_1);
506 FOLIO_MATCH(compound_head, _head_1);
507 FOLIO_MATCH(_mapcount, _mapcount_1);
508 FOLIO_MATCH(_refcount, _refcount_1);
509 #undef FOLIO_MATCH
510 #define FOLIO_MATCH(pg, fl)						\
511 	static_assert(offsetof(struct folio, fl) ==			\
512 			offsetof(struct page, pg) + 2 * sizeof(struct page))
513 FOLIO_MATCH(flags, _flags_2);
514 FOLIO_MATCH(compound_head, _head_2);
515 #undef FOLIO_MATCH
516 #define FOLIO_MATCH(pg, fl)						\
517 	static_assert(offsetof(struct folio, fl) ==			\
518 			offsetof(struct page, pg) + 3 * sizeof(struct page))
519 FOLIO_MATCH(flags, _flags_3);
520 FOLIO_MATCH(compound_head, _head_3);
521 #undef FOLIO_MATCH
522 
523 /**
524  * struct ptdesc -    Memory descriptor for page tables.
525  * @__page_flags:     Same as page flags. Powerpc only.
526  * @pt_rcu_head:      For freeing page table pages.
527  * @pt_list:          List of used page tables. Used for s390 gmap shadow pages
528  *                    (which are not linked into the user page tables) and x86
529  *                    pgds.
530  * @_pt_pad_1:        Padding that aliases with page's compound head.
531  * @pmd_huge_pte:     Protected by ptdesc->ptl, used for THPs.
532  * @__page_mapping:   Aliases with page->mapping. Unused for page tables.
533  * @pt_index:         Used for s390 gmap.
534  * @pt_mm:            Used for x86 pgds.
535  * @pt_frag_refcount: For fragmented page table tracking. Powerpc only.
536  * @pt_share_count:   Used for HugeTLB PMD page table share count.
537  * @_pt_pad_2:        Padding to ensure proper alignment.
538  * @ptl:              Lock for the page table.
539  * @__page_type:      Same as page->page_type. Unused for page tables.
540  * @__page_refcount:  Same as page refcount.
541  * @pt_memcg_data:    Memcg data. Tracked for page tables here.
542  *
543  * This struct overlays struct page for now. Do not modify without a good
544  * understanding of the issues.
545  */
546 struct ptdesc {
547 	unsigned long __page_flags;
548 
549 	union {
550 		struct rcu_head pt_rcu_head;
551 		struct list_head pt_list;
552 		struct {
553 			unsigned long _pt_pad_1;
554 			pgtable_t pmd_huge_pte;
555 		};
556 	};
557 	unsigned long __page_mapping;
558 
559 	union {
560 		pgoff_t pt_index;
561 		struct mm_struct *pt_mm;
562 		atomic_t pt_frag_refcount;
563 #ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
564 		atomic_t pt_share_count;
565 #endif
566 	};
567 
568 	union {
569 		unsigned long _pt_pad_2;
570 #if ALLOC_SPLIT_PTLOCKS
571 		spinlock_t *ptl;
572 #else
573 		spinlock_t ptl;
574 #endif
575 	};
576 	unsigned int __page_type;
577 	atomic_t __page_refcount;
578 #ifdef CONFIG_MEMCG
579 	unsigned long pt_memcg_data;
580 #endif
581 };
582 
583 #define TABLE_MATCH(pg, pt)						\
584 	static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
585 TABLE_MATCH(flags, __page_flags);
586 TABLE_MATCH(compound_head, pt_list);
587 TABLE_MATCH(compound_head, _pt_pad_1);
588 TABLE_MATCH(mapping, __page_mapping);
589 TABLE_MATCH(index, pt_index);
590 TABLE_MATCH(rcu_head, pt_rcu_head);
591 TABLE_MATCH(page_type, __page_type);
592 TABLE_MATCH(_refcount, __page_refcount);
593 #ifdef CONFIG_MEMCG
594 TABLE_MATCH(memcg_data, pt_memcg_data);
595 #endif
596 #undef TABLE_MATCH
597 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
598 
599 #define ptdesc_page(pt)			(_Generic((pt),			\
600 	const struct ptdesc *:		(const struct page *)(pt),	\
601 	struct ptdesc *:		(struct page *)(pt)))
602 
603 #define ptdesc_folio(pt)		(_Generic((pt),			\
604 	const struct ptdesc *:		(const struct folio *)(pt),	\
605 	struct ptdesc *:		(struct folio *)(pt)))
606 
607 #define page_ptdesc(p)			(_Generic((p),			\
608 	const struct page *:		(const struct ptdesc *)(p),	\
609 	struct page *:			(struct ptdesc *)(p)))
610 
611 #ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
612 static inline void ptdesc_pmd_pts_init(struct ptdesc *ptdesc)
613 {
614 	atomic_set(&ptdesc->pt_share_count, 0);
615 }
616 
617 static inline void ptdesc_pmd_pts_inc(struct ptdesc *ptdesc)
618 {
619 	atomic_inc(&ptdesc->pt_share_count);
620 }
621 
622 static inline void ptdesc_pmd_pts_dec(struct ptdesc *ptdesc)
623 {
624 	atomic_dec(&ptdesc->pt_share_count);
625 }
626 
627 static inline int ptdesc_pmd_pts_count(struct ptdesc *ptdesc)
628 {
629 	return atomic_read(&ptdesc->pt_share_count);
630 }
631 #else
632 static inline void ptdesc_pmd_pts_init(struct ptdesc *ptdesc)
633 {
634 }
635 #endif
636 
637 /*
638  * Used for sizing the vmemmap region on some architectures
639  */
640 #define STRUCT_PAGE_MAX_SHIFT	(order_base_2(sizeof(struct page)))
641 
642 /*
643  * page_private can be used on tail pages.  However, PagePrivate is only
644  * checked by the VM on the head page.  So page_private on the tail pages
645  * should be used for data that's ancillary to the head page (eg attaching
646  * buffer heads to tail pages after attaching buffer heads to the head page)
647  */
648 #define page_private(page)		((page)->private)
649 
650 static inline void set_page_private(struct page *page, unsigned long private)
651 {
652 	page->private = private;
653 }
654 
655 static inline void *folio_get_private(struct folio *folio)
656 {
657 	return folio->private;
658 }
659 
660 typedef unsigned long vm_flags_t;
661 
662 /*
663  * freeptr_t represents a SLUB freelist pointer, which might be encoded
664  * and not dereferenceable if CONFIG_SLAB_FREELIST_HARDENED is enabled.
665  */
666 typedef struct { unsigned long v; } freeptr_t;
667 
668 /*
669  * A region containing a mapping of a non-memory backed file under NOMMU
670  * conditions.  These are held in a global tree and are pinned by the VMAs that
671  * map parts of them.
672  */
673 struct vm_region {
674 	struct rb_node	vm_rb;		/* link in global region tree */
675 	vm_flags_t	vm_flags;	/* VMA vm_flags */
676 	unsigned long	vm_start;	/* start address of region */
677 	unsigned long	vm_end;		/* region initialised to here */
678 	unsigned long	vm_top;		/* region allocated to here */
679 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
680 	struct file	*vm_file;	/* the backing file or NULL */
681 
682 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
683 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
684 						* this region */
685 };
686 
687 #ifdef CONFIG_USERFAULTFD
688 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
689 struct vm_userfaultfd_ctx {
690 	struct userfaultfd_ctx *ctx;
691 };
692 #else /* CONFIG_USERFAULTFD */
693 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
694 struct vm_userfaultfd_ctx {};
695 #endif /* CONFIG_USERFAULTFD */
696 
697 struct anon_vma_name {
698 	struct kref kref;
699 	/* The name needs to be at the end because it is dynamically sized. */
700 	char name[];
701 };
702 
703 #ifdef CONFIG_ANON_VMA_NAME
704 /*
705  * mmap_lock should be read-locked when calling anon_vma_name(). Caller should
706  * either keep holding the lock while using the returned pointer or it should
707  * raise anon_vma_name refcount before releasing the lock.
708  */
709 struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma);
710 struct anon_vma_name *anon_vma_name_alloc(const char *name);
711 void anon_vma_name_free(struct kref *kref);
712 #else /* CONFIG_ANON_VMA_NAME */
713 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
714 {
715 	return NULL;
716 }
717 
718 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
719 {
720 	return NULL;
721 }
722 #endif
723 
724 #define VMA_LOCK_OFFSET	0x40000000
725 #define VMA_REF_LIMIT	(VMA_LOCK_OFFSET - 1)
726 
727 struct vma_numab_state {
728 	/*
729 	 * Initialised as time in 'jiffies' after which VMA
730 	 * should be scanned.  Delays first scan of new VMA by at
731 	 * least sysctl_numa_balancing_scan_delay:
732 	 */
733 	unsigned long next_scan;
734 
735 	/*
736 	 * Time in jiffies when pids_active[] is reset to
737 	 * detect phase change behaviour:
738 	 */
739 	unsigned long pids_active_reset;
740 
741 	/*
742 	 * Approximate tracking of PIDs that trapped a NUMA hinting
743 	 * fault. May produce false positives due to hash collisions.
744 	 *
745 	 *   [0] Previous PID tracking
746 	 *   [1] Current PID tracking
747 	 *
748 	 * Window moves after next_pid_reset has expired approximately
749 	 * every VMA_PID_RESET_PERIOD jiffies:
750 	 */
751 	unsigned long pids_active[2];
752 
753 	/* MM scan sequence ID when scan first started after VMA creation */
754 	int start_scan_seq;
755 
756 	/*
757 	 * MM scan sequence ID when the VMA was last completely scanned.
758 	 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq
759 	 */
760 	int prev_scan_seq;
761 };
762 
763 /*
764  * This struct describes a virtual memory area. There is one of these
765  * per VM-area/task. A VM area is any part of the process virtual memory
766  * space that has a special rule for the page-fault handlers (ie a shared
767  * library, the executable area etc).
768  *
769  * Only explicitly marked struct members may be accessed by RCU readers before
770  * getting a stable reference.
771  *
772  * WARNING: when adding new members, please update vm_area_init_from() to copy
773  * them during vm_area_struct content duplication.
774  */
775 struct vm_area_struct {
776 	/* The first cache line has the info for VMA tree walking. */
777 
778 	union {
779 		struct {
780 			/* VMA covers [vm_start; vm_end) addresses within mm */
781 			unsigned long vm_start;
782 			unsigned long vm_end;
783 		};
784 		freeptr_t vm_freeptr; /* Pointer used by SLAB_TYPESAFE_BY_RCU */
785 	};
786 
787 	/*
788 	 * The address space we belong to.
789 	 * Unstable RCU readers are allowed to read this.
790 	 */
791 	struct mm_struct *vm_mm;
792 	pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
793 
794 	/*
795 	 * Flags, see mm.h.
796 	 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
797 	 */
798 	union {
799 		const vm_flags_t vm_flags;
800 		vm_flags_t __private __vm_flags;
801 	};
802 
803 #ifdef CONFIG_PER_VMA_LOCK
804 	/*
805 	 * Can only be written (using WRITE_ONCE()) while holding both:
806 	 *  - mmap_lock (in write mode)
807 	 *  - vm_refcnt bit at VMA_LOCK_OFFSET is set
808 	 * Can be read reliably while holding one of:
809 	 *  - mmap_lock (in read or write mode)
810 	 *  - vm_refcnt bit at VMA_LOCK_OFFSET is set or vm_refcnt > 1
811 	 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
812 	 * while holding nothing (except RCU to keep the VMA struct allocated).
813 	 *
814 	 * This sequence counter is explicitly allowed to overflow; sequence
815 	 * counter reuse can only lead to occasional unnecessary use of the
816 	 * slowpath.
817 	 */
818 	unsigned int vm_lock_seq;
819 #endif
820 	/*
821 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
822 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
823 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
824 	 * or brk vma (with NULL file) can only be in an anon_vma list.
825 	 */
826 	struct list_head anon_vma_chain; /* Serialized by mmap_lock &
827 					  * page_table_lock */
828 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
829 
830 	/* Function pointers to deal with this struct. */
831 	const struct vm_operations_struct *vm_ops;
832 
833 	/* Information about our backing store: */
834 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
835 					   units */
836 	struct file * vm_file;		/* File we map to (can be NULL). */
837 	void * vm_private_data;		/* was vm_pte (shared mem) */
838 
839 #ifdef CONFIG_SWAP
840 	atomic_long_t swap_readahead_info;
841 #endif
842 #ifndef CONFIG_MMU
843 	struct vm_region *vm_region;	/* NOMMU mapping region */
844 #endif
845 #ifdef CONFIG_NUMA
846 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
847 #endif
848 #ifdef CONFIG_NUMA_BALANCING
849 	struct vma_numab_state *numab_state;	/* NUMA Balancing state */
850 #endif
851 #ifdef CONFIG_PER_VMA_LOCK
852 	/* Unstable RCU readers are allowed to read this. */
853 	refcount_t vm_refcnt ____cacheline_aligned_in_smp;
854 #ifdef CONFIG_DEBUG_LOCK_ALLOC
855 	struct lockdep_map vmlock_dep_map;
856 #endif
857 #endif
858 	/*
859 	 * For areas with an address space and backing store,
860 	 * linkage into the address_space->i_mmap interval tree.
861 	 *
862 	 */
863 	struct {
864 		struct rb_node rb;
865 		unsigned long rb_subtree_last;
866 	} shared;
867 #ifdef CONFIG_ANON_VMA_NAME
868 	/*
869 	 * For private and shared anonymous mappings, a pointer to a null
870 	 * terminated string containing the name given to the vma, or NULL if
871 	 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
872 	 */
873 	struct anon_vma_name *anon_name;
874 #endif
875 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
876 } __randomize_layout;
877 
878 #ifdef CONFIG_NUMA
879 #define vma_policy(vma) ((vma)->vm_policy)
880 #else
881 #define vma_policy(vma) NULL
882 #endif
883 
884 #ifdef CONFIG_SCHED_MM_CID
885 struct mm_cid {
886 	u64 time;
887 	int cid;
888 	int recent_cid;
889 };
890 #endif
891 
892 struct kioctx_table;
893 struct iommu_mm_data;
894 struct mm_struct {
895 	struct {
896 		/*
897 		 * Fields which are often written to are placed in a separate
898 		 * cache line.
899 		 */
900 		struct {
901 			/**
902 			 * @mm_count: The number of references to &struct
903 			 * mm_struct (@mm_users count as 1).
904 			 *
905 			 * Use mmgrab()/mmdrop() to modify. When this drops to
906 			 * 0, the &struct mm_struct is freed.
907 			 */
908 			atomic_t mm_count;
909 		} ____cacheline_aligned_in_smp;
910 
911 		struct maple_tree mm_mt;
912 
913 		unsigned long mmap_base;	/* base of mmap area */
914 		unsigned long mmap_legacy_base;	/* base of mmap area in bottom-up allocations */
915 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
916 		/* Base addresses for compatible mmap() */
917 		unsigned long mmap_compat_base;
918 		unsigned long mmap_compat_legacy_base;
919 #endif
920 		unsigned long task_size;	/* size of task vm space */
921 		pgd_t * pgd;
922 
923 #ifdef CONFIG_MEMBARRIER
924 		/**
925 		 * @membarrier_state: Flags controlling membarrier behavior.
926 		 *
927 		 * This field is close to @pgd to hopefully fit in the same
928 		 * cache-line, which needs to be touched by switch_mm().
929 		 */
930 		atomic_t membarrier_state;
931 #endif
932 
933 		/**
934 		 * @mm_users: The number of users including userspace.
935 		 *
936 		 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
937 		 * drops to 0 (i.e. when the task exits and there are no other
938 		 * temporary reference holders), we also release a reference on
939 		 * @mm_count (which may then free the &struct mm_struct if
940 		 * @mm_count also drops to 0).
941 		 */
942 		atomic_t mm_users;
943 
944 #ifdef CONFIG_SCHED_MM_CID
945 		/**
946 		 * @pcpu_cid: Per-cpu current cid.
947 		 *
948 		 * Keep track of the currently allocated mm_cid for each cpu.
949 		 * The per-cpu mm_cid values are serialized by their respective
950 		 * runqueue locks.
951 		 */
952 		struct mm_cid __percpu *pcpu_cid;
953 		/*
954 		 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
955 		 *
956 		 * When the next mm_cid scan is due (in jiffies).
957 		 */
958 		unsigned long mm_cid_next_scan;
959 		/**
960 		 * @nr_cpus_allowed: Number of CPUs allowed for mm.
961 		 *
962 		 * Number of CPUs allowed in the union of all mm's
963 		 * threads allowed CPUs.
964 		 */
965 		unsigned int nr_cpus_allowed;
966 		/**
967 		 * @max_nr_cid: Maximum number of allowed concurrency
968 		 *              IDs allocated.
969 		 *
970 		 * Track the highest number of allowed concurrency IDs
971 		 * allocated for the mm.
972 		 */
973 		atomic_t max_nr_cid;
974 		/**
975 		 * @cpus_allowed_lock: Lock protecting mm cpus_allowed.
976 		 *
977 		 * Provide mutual exclusion for mm cpus_allowed and
978 		 * mm nr_cpus_allowed updates.
979 		 */
980 		raw_spinlock_t cpus_allowed_lock;
981 #endif
982 #ifdef CONFIG_MMU
983 		atomic_long_t pgtables_bytes;	/* size of all page tables */
984 #endif
985 		int map_count;			/* number of VMAs */
986 
987 		spinlock_t page_table_lock; /* Protects page tables and some
988 					     * counters
989 					     */
990 		/*
991 		 * With some kernel config, the current mmap_lock's offset
992 		 * inside 'mm_struct' is at 0x120, which is very optimal, as
993 		 * its two hot fields 'count' and 'owner' sit in 2 different
994 		 * cachelines,  and when mmap_lock is highly contended, both
995 		 * of the 2 fields will be accessed frequently, current layout
996 		 * will help to reduce cache bouncing.
997 		 *
998 		 * So please be careful with adding new fields before
999 		 * mmap_lock, which can easily push the 2 fields into one
1000 		 * cacheline.
1001 		 */
1002 		struct rw_semaphore mmap_lock;
1003 
1004 		struct list_head mmlist; /* List of maybe swapped mm's.	These
1005 					  * are globally strung together off
1006 					  * init_mm.mmlist, and are protected
1007 					  * by mmlist_lock
1008 					  */
1009 #ifdef CONFIG_PER_VMA_LOCK
1010 		struct rcuwait vma_writer_wait;
1011 		/*
1012 		 * This field has lock-like semantics, meaning it is sometimes
1013 		 * accessed with ACQUIRE/RELEASE semantics.
1014 		 * Roughly speaking, incrementing the sequence number is
1015 		 * equivalent to releasing locks on VMAs; reading the sequence
1016 		 * number can be part of taking a read lock on a VMA.
1017 		 * Incremented every time mmap_lock is write-locked/unlocked.
1018 		 * Initialized to 0, therefore odd values indicate mmap_lock
1019 		 * is write-locked and even values that it's released.
1020 		 *
1021 		 * Can be modified under write mmap_lock using RELEASE
1022 		 * semantics.
1023 		 * Can be read with no other protection when holding write
1024 		 * mmap_lock.
1025 		 * Can be read with ACQUIRE semantics if not holding write
1026 		 * mmap_lock.
1027 		 */
1028 		seqcount_t mm_lock_seq;
1029 #endif
1030 
1031 
1032 		unsigned long hiwater_rss; /* High-watermark of RSS usage */
1033 		unsigned long hiwater_vm;  /* High-water virtual memory usage */
1034 
1035 		unsigned long total_vm;	   /* Total pages mapped */
1036 		unsigned long locked_vm;   /* Pages that have PG_mlocked set */
1037 		atomic64_t    pinned_vm;   /* Refcount permanently increased */
1038 		unsigned long data_vm;	   /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
1039 		unsigned long exec_vm;	   /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
1040 		unsigned long stack_vm;	   /* VM_STACK */
1041 		unsigned long def_flags;
1042 
1043 		/**
1044 		 * @write_protect_seq: Locked when any thread is write
1045 		 * protecting pages mapped by this mm to enforce a later COW,
1046 		 * for instance during page table copying for fork().
1047 		 */
1048 		seqcount_t write_protect_seq;
1049 
1050 		spinlock_t arg_lock; /* protect the below fields */
1051 
1052 		unsigned long start_code, end_code, start_data, end_data;
1053 		unsigned long start_brk, brk, start_stack;
1054 		unsigned long arg_start, arg_end, env_start, env_end;
1055 
1056 		unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
1057 
1058 		struct percpu_counter rss_stat[NR_MM_COUNTERS];
1059 
1060 		struct linux_binfmt *binfmt;
1061 
1062 		/* Architecture-specific MM context */
1063 		mm_context_t context;
1064 
1065 		unsigned long flags; /* Must use atomic bitops to access */
1066 
1067 #ifdef CONFIG_AIO
1068 		spinlock_t			ioctx_lock;
1069 		struct kioctx_table __rcu	*ioctx_table;
1070 #endif
1071 #ifdef CONFIG_MEMCG
1072 		/*
1073 		 * "owner" points to a task that is regarded as the canonical
1074 		 * user/owner of this mm. All of the following must be true in
1075 		 * order for it to be changed:
1076 		 *
1077 		 * current == mm->owner
1078 		 * current->mm != mm
1079 		 * new_owner->mm == mm
1080 		 * new_owner->alloc_lock is held
1081 		 */
1082 		struct task_struct __rcu *owner;
1083 #endif
1084 		struct user_namespace *user_ns;
1085 
1086 		/* store ref to file /proc/<pid>/exe symlink points to */
1087 		struct file __rcu *exe_file;
1088 #ifdef CONFIG_MMU_NOTIFIER
1089 		struct mmu_notifier_subscriptions *notifier_subscriptions;
1090 #endif
1091 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1092 		pgtable_t pmd_huge_pte; /* protected by page_table_lock */
1093 #endif
1094 #ifdef CONFIG_NUMA_BALANCING
1095 		/*
1096 		 * numa_next_scan is the next time that PTEs will be remapped
1097 		 * PROT_NONE to trigger NUMA hinting faults; such faults gather
1098 		 * statistics and migrate pages to new nodes if necessary.
1099 		 */
1100 		unsigned long numa_next_scan;
1101 
1102 		/* Restart point for scanning and remapping PTEs. */
1103 		unsigned long numa_scan_offset;
1104 
1105 		/* numa_scan_seq prevents two threads remapping PTEs. */
1106 		int numa_scan_seq;
1107 #endif
1108 		/*
1109 		 * An operation with batched TLB flushing is going on. Anything
1110 		 * that can move process memory needs to flush the TLB when
1111 		 * moving a PROT_NONE mapped page.
1112 		 */
1113 		atomic_t tlb_flush_pending;
1114 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1115 		/* See flush_tlb_batched_pending() */
1116 		atomic_t tlb_flush_batched;
1117 #endif
1118 		struct uprobes_state uprobes_state;
1119 #ifdef CONFIG_PREEMPT_RT
1120 		struct rcu_head delayed_drop;
1121 #endif
1122 #ifdef CONFIG_HUGETLB_PAGE
1123 		atomic_long_t hugetlb_usage;
1124 #endif
1125 		struct work_struct async_put_work;
1126 
1127 #ifdef CONFIG_IOMMU_MM_DATA
1128 		struct iommu_mm_data *iommu_mm;
1129 #endif
1130 #ifdef CONFIG_KSM
1131 		/*
1132 		 * Represent how many pages of this process are involved in KSM
1133 		 * merging (not including ksm_zero_pages).
1134 		 */
1135 		unsigned long ksm_merging_pages;
1136 		/*
1137 		 * Represent how many pages are checked for ksm merging
1138 		 * including merged and not merged.
1139 		 */
1140 		unsigned long ksm_rmap_items;
1141 		/*
1142 		 * Represent how many empty pages are merged with kernel zero
1143 		 * pages when enabling KSM use_zero_pages.
1144 		 */
1145 		atomic_long_t ksm_zero_pages;
1146 #endif /* CONFIG_KSM */
1147 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1148 		struct {
1149 			/* this mm_struct is on lru_gen_mm_list */
1150 			struct list_head list;
1151 			/*
1152 			 * Set when switching to this mm_struct, as a hint of
1153 			 * whether it has been used since the last time per-node
1154 			 * page table walkers cleared the corresponding bits.
1155 			 */
1156 			unsigned long bitmap;
1157 #ifdef CONFIG_MEMCG
1158 			/* points to the memcg of "owner" above */
1159 			struct mem_cgroup *memcg;
1160 #endif
1161 		} lru_gen;
1162 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1163 #ifdef CONFIG_MM_ID
1164 		mm_id_t mm_id;
1165 #endif /* CONFIG_MM_ID */
1166 	} __randomize_layout;
1167 
1168 	/*
1169 	 * The mm_cpumask needs to be at the end of mm_struct, because it
1170 	 * is dynamically sized based on nr_cpu_ids.
1171 	 */
1172 	unsigned long cpu_bitmap[];
1173 };
1174 
1175 #define MM_MT_FLAGS	(MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
1176 			 MT_FLAGS_USE_RCU)
1177 extern struct mm_struct init_mm;
1178 
1179 /* Pointer magic because the dynamic array size confuses some compilers. */
1180 static inline void mm_init_cpumask(struct mm_struct *mm)
1181 {
1182 	unsigned long cpu_bitmap = (unsigned long)mm;
1183 
1184 	cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1185 	cpumask_clear((struct cpumask *)cpu_bitmap);
1186 }
1187 
1188 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
1189 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
1190 {
1191 	return (struct cpumask *)&mm->cpu_bitmap;
1192 }
1193 
1194 #ifdef CONFIG_LRU_GEN
1195 
1196 struct lru_gen_mm_list {
1197 	/* mm_struct list for page table walkers */
1198 	struct list_head fifo;
1199 	/* protects the list above */
1200 	spinlock_t lock;
1201 };
1202 
1203 #endif /* CONFIG_LRU_GEN */
1204 
1205 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1206 
1207 void lru_gen_add_mm(struct mm_struct *mm);
1208 void lru_gen_del_mm(struct mm_struct *mm);
1209 void lru_gen_migrate_mm(struct mm_struct *mm);
1210 
1211 static inline void lru_gen_init_mm(struct mm_struct *mm)
1212 {
1213 	INIT_LIST_HEAD(&mm->lru_gen.list);
1214 	mm->lru_gen.bitmap = 0;
1215 #ifdef CONFIG_MEMCG
1216 	mm->lru_gen.memcg = NULL;
1217 #endif
1218 }
1219 
1220 static inline void lru_gen_use_mm(struct mm_struct *mm)
1221 {
1222 	/*
1223 	 * When the bitmap is set, page reclaim knows this mm_struct has been
1224 	 * used since the last time it cleared the bitmap. So it might be worth
1225 	 * walking the page tables of this mm_struct to clear the accessed bit.
1226 	 */
1227 	WRITE_ONCE(mm->lru_gen.bitmap, -1);
1228 }
1229 
1230 #else /* !CONFIG_LRU_GEN_WALKS_MMU */
1231 
1232 static inline void lru_gen_add_mm(struct mm_struct *mm)
1233 {
1234 }
1235 
1236 static inline void lru_gen_del_mm(struct mm_struct *mm)
1237 {
1238 }
1239 
1240 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
1241 {
1242 }
1243 
1244 static inline void lru_gen_init_mm(struct mm_struct *mm)
1245 {
1246 }
1247 
1248 static inline void lru_gen_use_mm(struct mm_struct *mm)
1249 {
1250 }
1251 
1252 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1253 
1254 struct vma_iterator {
1255 	struct ma_state mas;
1256 };
1257 
1258 #define VMA_ITERATOR(name, __mm, __addr)				\
1259 	struct vma_iterator name = {					\
1260 		.mas = {						\
1261 			.tree = &(__mm)->mm_mt,				\
1262 			.index = __addr,				\
1263 			.node = NULL,					\
1264 			.status = ma_start,				\
1265 		},							\
1266 	}
1267 
1268 static inline void vma_iter_init(struct vma_iterator *vmi,
1269 		struct mm_struct *mm, unsigned long addr)
1270 {
1271 	mas_init(&vmi->mas, &mm->mm_mt, addr);
1272 }
1273 
1274 #ifdef CONFIG_SCHED_MM_CID
1275 
1276 enum mm_cid_state {
1277 	MM_CID_UNSET = -1U,		/* Unset state has lazy_put flag set. */
1278 	MM_CID_LAZY_PUT = (1U << 31),
1279 };
1280 
1281 static inline bool mm_cid_is_unset(int cid)
1282 {
1283 	return cid == MM_CID_UNSET;
1284 }
1285 
1286 static inline bool mm_cid_is_lazy_put(int cid)
1287 {
1288 	return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1289 }
1290 
1291 static inline bool mm_cid_is_valid(int cid)
1292 {
1293 	return !(cid & MM_CID_LAZY_PUT);
1294 }
1295 
1296 static inline int mm_cid_set_lazy_put(int cid)
1297 {
1298 	return cid | MM_CID_LAZY_PUT;
1299 }
1300 
1301 static inline int mm_cid_clear_lazy_put(int cid)
1302 {
1303 	return cid & ~MM_CID_LAZY_PUT;
1304 }
1305 
1306 /*
1307  * mm_cpus_allowed: Union of all mm's threads allowed CPUs.
1308  */
1309 static inline cpumask_t *mm_cpus_allowed(struct mm_struct *mm)
1310 {
1311 	unsigned long bitmap = (unsigned long)mm;
1312 
1313 	bitmap += offsetof(struct mm_struct, cpu_bitmap);
1314 	/* Skip cpu_bitmap */
1315 	bitmap += cpumask_size();
1316 	return (struct cpumask *)bitmap;
1317 }
1318 
1319 /* Accessor for struct mm_struct's cidmask. */
1320 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1321 {
1322 	unsigned long cid_bitmap = (unsigned long)mm_cpus_allowed(mm);
1323 
1324 	/* Skip mm_cpus_allowed */
1325 	cid_bitmap += cpumask_size();
1326 	return (struct cpumask *)cid_bitmap;
1327 }
1328 
1329 static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p)
1330 {
1331 	int i;
1332 
1333 	for_each_possible_cpu(i) {
1334 		struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1335 
1336 		pcpu_cid->cid = MM_CID_UNSET;
1337 		pcpu_cid->recent_cid = MM_CID_UNSET;
1338 		pcpu_cid->time = 0;
1339 	}
1340 	mm->nr_cpus_allowed = p->nr_cpus_allowed;
1341 	atomic_set(&mm->max_nr_cid, 0);
1342 	raw_spin_lock_init(&mm->cpus_allowed_lock);
1343 	cpumask_copy(mm_cpus_allowed(mm), &p->cpus_mask);
1344 	cpumask_clear(mm_cidmask(mm));
1345 }
1346 
1347 static inline int mm_alloc_cid_noprof(struct mm_struct *mm, struct task_struct *p)
1348 {
1349 	mm->pcpu_cid = alloc_percpu_noprof(struct mm_cid);
1350 	if (!mm->pcpu_cid)
1351 		return -ENOMEM;
1352 	mm_init_cid(mm, p);
1353 	return 0;
1354 }
1355 #define mm_alloc_cid(...)	alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__))
1356 
1357 static inline void mm_destroy_cid(struct mm_struct *mm)
1358 {
1359 	free_percpu(mm->pcpu_cid);
1360 	mm->pcpu_cid = NULL;
1361 }
1362 
1363 static inline unsigned int mm_cid_size(void)
1364 {
1365 	return 2 * cpumask_size();	/* mm_cpus_allowed(), mm_cidmask(). */
1366 }
1367 
1368 static inline void mm_set_cpus_allowed(struct mm_struct *mm, const struct cpumask *cpumask)
1369 {
1370 	struct cpumask *mm_allowed = mm_cpus_allowed(mm);
1371 
1372 	if (!mm)
1373 		return;
1374 	/* The mm_cpus_allowed is the union of each thread allowed CPUs masks. */
1375 	raw_spin_lock(&mm->cpus_allowed_lock);
1376 	cpumask_or(mm_allowed, mm_allowed, cpumask);
1377 	WRITE_ONCE(mm->nr_cpus_allowed, cpumask_weight(mm_allowed));
1378 	raw_spin_unlock(&mm->cpus_allowed_lock);
1379 }
1380 #else /* CONFIG_SCHED_MM_CID */
1381 static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p) { }
1382 static inline int mm_alloc_cid(struct mm_struct *mm, struct task_struct *p) { return 0; }
1383 static inline void mm_destroy_cid(struct mm_struct *mm) { }
1384 
1385 static inline unsigned int mm_cid_size(void)
1386 {
1387 	return 0;
1388 }
1389 static inline void mm_set_cpus_allowed(struct mm_struct *mm, const struct cpumask *cpumask) { }
1390 #endif /* CONFIG_SCHED_MM_CID */
1391 
1392 struct mmu_gather;
1393 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1394 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1395 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1396 
1397 struct vm_fault;
1398 
1399 /**
1400  * typedef vm_fault_t - Return type for page fault handlers.
1401  *
1402  * Page fault handlers return a bitmask of %VM_FAULT values.
1403  */
1404 typedef __bitwise unsigned int vm_fault_t;
1405 
1406 /**
1407  * enum vm_fault_reason - Page fault handlers return a bitmask of
1408  * these values to tell the core VM what happened when handling the
1409  * fault. Used to decide whether a process gets delivered SIGBUS or
1410  * just gets major/minor fault counters bumped up.
1411  *
1412  * @VM_FAULT_OOM:		Out Of Memory
1413  * @VM_FAULT_SIGBUS:		Bad access
1414  * @VM_FAULT_MAJOR:		Page read from storage
1415  * @VM_FAULT_HWPOISON:		Hit poisoned small page
1416  * @VM_FAULT_HWPOISON_LARGE:	Hit poisoned large page. Index encoded
1417  *				in upper bits
1418  * @VM_FAULT_SIGSEGV:		segmentation fault
1419  * @VM_FAULT_NOPAGE:		->fault installed the pte, not return page
1420  * @VM_FAULT_LOCKED:		->fault locked the returned page
1421  * @VM_FAULT_RETRY:		->fault blocked, must retry
1422  * @VM_FAULT_FALLBACK:		huge page fault failed, fall back to small
1423  * @VM_FAULT_DONE_COW:		->fault has fully handled COW
1424  * @VM_FAULT_NEEDDSYNC:		->fault did not modify page tables and needs
1425  *				fsync() to complete (for synchronous page faults
1426  *				in DAX)
1427  * @VM_FAULT_COMPLETED:		->fault completed, meanwhile mmap lock released
1428  * @VM_FAULT_HINDEX_MASK:	mask HINDEX value
1429  *
1430  */
1431 enum vm_fault_reason {
1432 	VM_FAULT_OOM            = (__force vm_fault_t)0x000001,
1433 	VM_FAULT_SIGBUS         = (__force vm_fault_t)0x000002,
1434 	VM_FAULT_MAJOR          = (__force vm_fault_t)0x000004,
1435 	VM_FAULT_HWPOISON       = (__force vm_fault_t)0x000010,
1436 	VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1437 	VM_FAULT_SIGSEGV        = (__force vm_fault_t)0x000040,
1438 	VM_FAULT_NOPAGE         = (__force vm_fault_t)0x000100,
1439 	VM_FAULT_LOCKED         = (__force vm_fault_t)0x000200,
1440 	VM_FAULT_RETRY          = (__force vm_fault_t)0x000400,
1441 	VM_FAULT_FALLBACK       = (__force vm_fault_t)0x000800,
1442 	VM_FAULT_DONE_COW       = (__force vm_fault_t)0x001000,
1443 	VM_FAULT_NEEDDSYNC      = (__force vm_fault_t)0x002000,
1444 	VM_FAULT_COMPLETED      = (__force vm_fault_t)0x004000,
1445 	VM_FAULT_HINDEX_MASK    = (__force vm_fault_t)0x0f0000,
1446 };
1447 
1448 /* Encode hstate index for a hwpoisoned large page */
1449 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1450 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1451 
1452 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS |	\
1453 			VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON |	\
1454 			VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1455 
1456 #define VM_FAULT_RESULT_TRACE \
1457 	{ VM_FAULT_OOM,                 "OOM" },	\
1458 	{ VM_FAULT_SIGBUS,              "SIGBUS" },	\
1459 	{ VM_FAULT_MAJOR,               "MAJOR" },	\
1460 	{ VM_FAULT_HWPOISON,            "HWPOISON" },	\
1461 	{ VM_FAULT_HWPOISON_LARGE,      "HWPOISON_LARGE" },	\
1462 	{ VM_FAULT_SIGSEGV,             "SIGSEGV" },	\
1463 	{ VM_FAULT_NOPAGE,              "NOPAGE" },	\
1464 	{ VM_FAULT_LOCKED,              "LOCKED" },	\
1465 	{ VM_FAULT_RETRY,               "RETRY" },	\
1466 	{ VM_FAULT_FALLBACK,            "FALLBACK" },	\
1467 	{ VM_FAULT_DONE_COW,            "DONE_COW" },	\
1468 	{ VM_FAULT_NEEDDSYNC,           "NEEDDSYNC" },	\
1469 	{ VM_FAULT_COMPLETED,           "COMPLETED" }
1470 
1471 struct vm_special_mapping {
1472 	const char *name;	/* The name, e.g. "[vdso]". */
1473 
1474 	/*
1475 	 * If .fault is not provided, this points to a
1476 	 * NULL-terminated array of pages that back the special mapping.
1477 	 *
1478 	 * This must not be NULL unless .fault is provided.
1479 	 */
1480 	struct page **pages;
1481 
1482 	/*
1483 	 * If non-NULL, then this is called to resolve page faults
1484 	 * on the special mapping.  If used, .pages is not checked.
1485 	 */
1486 	vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1487 				struct vm_area_struct *vma,
1488 				struct vm_fault *vmf);
1489 
1490 	int (*mremap)(const struct vm_special_mapping *sm,
1491 		     struct vm_area_struct *new_vma);
1492 
1493 	void (*close)(const struct vm_special_mapping *sm,
1494 		      struct vm_area_struct *vma);
1495 };
1496 
1497 enum tlb_flush_reason {
1498 	TLB_FLUSH_ON_TASK_SWITCH,
1499 	TLB_REMOTE_SHOOTDOWN,
1500 	TLB_LOCAL_SHOOTDOWN,
1501 	TLB_LOCAL_MM_SHOOTDOWN,
1502 	TLB_REMOTE_SEND_IPI,
1503 	TLB_REMOTE_WRONG_CPU,
1504 	NR_TLB_FLUSH_REASONS,
1505 };
1506 
1507 /**
1508  * enum fault_flag - Fault flag definitions.
1509  * @FAULT_FLAG_WRITE: Fault was a write fault.
1510  * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1511  * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1512  * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1513  * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1514  * @FAULT_FLAG_TRIED: The fault has been tried once.
1515  * @FAULT_FLAG_USER: The fault originated in userspace.
1516  * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1517  * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1518  * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1519  * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1520  *                      COW mapping, making sure that an exclusive anon page is
1521  *                      mapped after the fault.
1522  * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1523  *                        We should only access orig_pte if this flag set.
1524  * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1525  *
1526  * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1527  * whether we would allow page faults to retry by specifying these two
1528  * fault flags correctly.  Currently there can be three legal combinations:
1529  *
1530  * (a) ALLOW_RETRY and !TRIED:  this means the page fault allows retry, and
1531  *                              this is the first try
1532  *
1533  * (b) ALLOW_RETRY and TRIED:   this means the page fault allows retry, and
1534  *                              we've already tried at least once
1535  *
1536  * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1537  *
1538  * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1539  * be used.  Note that page faults can be allowed to retry for multiple times,
1540  * in which case we'll have an initial fault with flags (a) then later on
1541  * continuous faults with flags (b).  We should always try to detect pending
1542  * signals before a retry to make sure the continuous page faults can still be
1543  * interrupted if necessary.
1544  *
1545  * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1546  * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1547  * applied to mappings that are not COW mappings.
1548  */
1549 enum fault_flag {
1550 	FAULT_FLAG_WRITE =		1 << 0,
1551 	FAULT_FLAG_MKWRITE =		1 << 1,
1552 	FAULT_FLAG_ALLOW_RETRY =	1 << 2,
1553 	FAULT_FLAG_RETRY_NOWAIT = 	1 << 3,
1554 	FAULT_FLAG_KILLABLE =		1 << 4,
1555 	FAULT_FLAG_TRIED = 		1 << 5,
1556 	FAULT_FLAG_USER =		1 << 6,
1557 	FAULT_FLAG_REMOTE =		1 << 7,
1558 	FAULT_FLAG_INSTRUCTION =	1 << 8,
1559 	FAULT_FLAG_INTERRUPTIBLE =	1 << 9,
1560 	FAULT_FLAG_UNSHARE =		1 << 10,
1561 	FAULT_FLAG_ORIG_PTE_VALID =	1 << 11,
1562 	FAULT_FLAG_VMA_LOCK =		1 << 12,
1563 };
1564 
1565 typedef unsigned int __bitwise zap_flags_t;
1566 
1567 /* Flags for clear_young_dirty_ptes(). */
1568 typedef int __bitwise cydp_t;
1569 
1570 /* Clear the access bit */
1571 #define CYDP_CLEAR_YOUNG		((__force cydp_t)BIT(0))
1572 
1573 /* Clear the dirty bit */
1574 #define CYDP_CLEAR_DIRTY		((__force cydp_t)BIT(1))
1575 
1576 /*
1577  * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1578  * other. Here is what they mean, and how to use them:
1579  *
1580  *
1581  * FIXME: For pages which are part of a filesystem, mappings are subject to the
1582  * lifetime enforced by the filesystem and we need guarantees that longterm
1583  * users like RDMA and V4L2 only establish mappings which coordinate usage with
1584  * the filesystem.  Ideas for this coordination include revoking the longterm
1585  * pin, delaying writeback, bounce buffer page writeback, etc.  As FS DAX was
1586  * added after the problem with filesystems was found FS DAX VMAs are
1587  * specifically failed.  Filesystem pages are still subject to bugs and use of
1588  * FOLL_LONGTERM should be avoided on those pages.
1589  *
1590  * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1591  * that region.  And so, CMA attempts to migrate the page before pinning, when
1592  * FOLL_LONGTERM is specified.
1593  *
1594  * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1595  * but an additional pin counting system) will be invoked. This is intended for
1596  * anything that gets a page reference and then touches page data (for example,
1597  * Direct IO). This lets the filesystem know that some non-file-system entity is
1598  * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1599  * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1600  * a call to unpin_user_page().
1601  *
1602  * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1603  * and separate refcounting mechanisms, however, and that means that each has
1604  * its own acquire and release mechanisms:
1605  *
1606  *     FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1607  *
1608  *     FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1609  *
1610  * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1611  * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1612  * calls applied to them, and that's perfectly OK. This is a constraint on the
1613  * callers, not on the pages.)
1614  *
1615  * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1616  * directly by the caller. That's in order to help avoid mismatches when
1617  * releasing pages: get_user_pages*() pages must be released via put_page(),
1618  * while pin_user_pages*() pages must be released via unpin_user_page().
1619  *
1620  * Please see Documentation/core-api/pin_user_pages.rst for more information.
1621  */
1622 
1623 enum {
1624 	/* check pte is writable */
1625 	FOLL_WRITE = 1 << 0,
1626 	/* do get_page on page */
1627 	FOLL_GET = 1 << 1,
1628 	/* give error on hole if it would be zero */
1629 	FOLL_DUMP = 1 << 2,
1630 	/* get_user_pages read/write w/o permission */
1631 	FOLL_FORCE = 1 << 3,
1632 	/*
1633 	 * if a disk transfer is needed, start the IO and return without waiting
1634 	 * upon it
1635 	 */
1636 	FOLL_NOWAIT = 1 << 4,
1637 	/* do not fault in pages */
1638 	FOLL_NOFAULT = 1 << 5,
1639 	/* check page is hwpoisoned */
1640 	FOLL_HWPOISON = 1 << 6,
1641 	/* don't do file mappings */
1642 	FOLL_ANON = 1 << 7,
1643 	/*
1644 	 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1645 	 * time period _often_ under userspace control.  This is in contrast to
1646 	 * iov_iter_get_pages(), whose usages are transient.
1647 	 */
1648 	FOLL_LONGTERM = 1 << 8,
1649 	/* split huge pmd before returning */
1650 	FOLL_SPLIT_PMD = 1 << 9,
1651 	/* allow returning PCI P2PDMA pages */
1652 	FOLL_PCI_P2PDMA = 1 << 10,
1653 	/* allow interrupts from generic signals */
1654 	FOLL_INTERRUPTIBLE = 1 << 11,
1655 	/*
1656 	 * Always honor (trigger) NUMA hinting faults.
1657 	 *
1658 	 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1659 	 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1660 	 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1661 	 * hinting faults.
1662 	 */
1663 	FOLL_HONOR_NUMA_FAULT = 1 << 12,
1664 
1665 	/* See also internal only FOLL flags in mm/internal.h */
1666 };
1667 
1668 /* mm flags */
1669 
1670 /*
1671  * The first two bits represent core dump modes for set-user-ID,
1672  * the modes are SUID_DUMP_* defined in linux/sched/coredump.h
1673  */
1674 #define MMF_DUMPABLE_BITS 2
1675 #define MMF_DUMPABLE_MASK ((1 << MMF_DUMPABLE_BITS) - 1)
1676 /* coredump filter bits */
1677 #define MMF_DUMP_ANON_PRIVATE	2
1678 #define MMF_DUMP_ANON_SHARED	3
1679 #define MMF_DUMP_MAPPED_PRIVATE	4
1680 #define MMF_DUMP_MAPPED_SHARED	5
1681 #define MMF_DUMP_ELF_HEADERS	6
1682 #define MMF_DUMP_HUGETLB_PRIVATE 7
1683 #define MMF_DUMP_HUGETLB_SHARED  8
1684 #define MMF_DUMP_DAX_PRIVATE	9
1685 #define MMF_DUMP_DAX_SHARED	10
1686 
1687 #define MMF_DUMP_FILTER_SHIFT	MMF_DUMPABLE_BITS
1688 #define MMF_DUMP_FILTER_BITS	9
1689 #define MMF_DUMP_FILTER_MASK \
1690 	(((1 << MMF_DUMP_FILTER_BITS) - 1) << MMF_DUMP_FILTER_SHIFT)
1691 #define MMF_DUMP_FILTER_DEFAULT \
1692 	((1 << MMF_DUMP_ANON_PRIVATE) |	(1 << MMF_DUMP_ANON_SHARED) |\
1693 	 (1 << MMF_DUMP_HUGETLB_PRIVATE) | MMF_DUMP_MASK_DEFAULT_ELF)
1694 
1695 #ifdef CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS
1696 # define MMF_DUMP_MASK_DEFAULT_ELF	(1 << MMF_DUMP_ELF_HEADERS)
1697 #else
1698 # define MMF_DUMP_MASK_DEFAULT_ELF	0
1699 #endif
1700 					/* leave room for more dump flags */
1701 #define MMF_VM_MERGEABLE	16	/* KSM may merge identical pages */
1702 #define MMF_VM_HUGEPAGE		17	/* set when mm is available for khugepaged */
1703 
1704 /*
1705  * This one-shot flag is dropped due to necessity of changing exe once again
1706  * on NFS restore
1707  */
1708 //#define MMF_EXE_FILE_CHANGED	18	/* see prctl_set_mm_exe_file() */
1709 
1710 #define MMF_HAS_UPROBES		19	/* has uprobes */
1711 #define MMF_RECALC_UPROBES	20	/* MMF_HAS_UPROBES can be wrong */
1712 #define MMF_OOM_SKIP		21	/* mm is of no interest for the OOM killer */
1713 #define MMF_UNSTABLE		22	/* mm is unstable for copy_from_user */
1714 #define MMF_HUGE_ZERO_PAGE	23      /* mm has ever used the global huge zero page */
1715 #define MMF_DISABLE_THP		24	/* disable THP for all VMAs */
1716 #define MMF_DISABLE_THP_MASK	(1 << MMF_DISABLE_THP)
1717 #define MMF_OOM_REAP_QUEUED	25	/* mm was queued for oom_reaper */
1718 #define MMF_MULTIPROCESS	26	/* mm is shared between processes */
1719 /*
1720  * MMF_HAS_PINNED: Whether this mm has pinned any pages.  This can be either
1721  * replaced in the future by mm.pinned_vm when it becomes stable, or grow into
1722  * a counter on its own. We're aggresive on this bit for now: even if the
1723  * pinned pages were unpinned later on, we'll still keep this bit set for the
1724  * lifecycle of this mm, just for simplicity.
1725  */
1726 #define MMF_HAS_PINNED		27	/* FOLL_PIN has run, never cleared */
1727 
1728 #define MMF_HAS_MDWE		28
1729 #define MMF_HAS_MDWE_MASK	(1 << MMF_HAS_MDWE)
1730 
1731 
1732 #define MMF_HAS_MDWE_NO_INHERIT	29
1733 
1734 #define MMF_VM_MERGE_ANY	30
1735 #define MMF_VM_MERGE_ANY_MASK	(1 << MMF_VM_MERGE_ANY)
1736 
1737 #define MMF_TOPDOWN		31	/* mm searches top down by default */
1738 #define MMF_TOPDOWN_MASK	(1 << MMF_TOPDOWN)
1739 
1740 #define MMF_INIT_MASK		(MMF_DUMPABLE_MASK | MMF_DUMP_FILTER_MASK |\
1741 				 MMF_DISABLE_THP_MASK | MMF_HAS_MDWE_MASK |\
1742 				 MMF_VM_MERGE_ANY_MASK | MMF_TOPDOWN_MASK)
1743 
1744 static inline unsigned long mmf_init_flags(unsigned long flags)
1745 {
1746 	if (flags & (1UL << MMF_HAS_MDWE_NO_INHERIT))
1747 		flags &= ~((1UL << MMF_HAS_MDWE) |
1748 			   (1UL << MMF_HAS_MDWE_NO_INHERIT));
1749 	return flags & MMF_INIT_MASK;
1750 }
1751 
1752 #endif /* _LINUX_MM_TYPES_H */
1753