1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
4
5 #include <linux/mm_types_task.h>
6
7 #include <linux/auxvec.h>
8 #include <linux/kref.h>
9 #include <linux/list.h>
10 #include <linux/spinlock.h>
11 #include <linux/rbtree.h>
12 #include <linux/maple_tree.h>
13 #include <linux/rwsem.h>
14 #include <linux/completion.h>
15 #include <linux/cpumask.h>
16 #include <linux/uprobes.h>
17 #include <linux/rcupdate.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/workqueue.h>
20 #include <linux/seqlock.h>
21 #include <linux/percpu_counter.h>
22
23 #include <asm/mmu.h>
24
25 #ifndef AT_VECTOR_SIZE_ARCH
26 #define AT_VECTOR_SIZE_ARCH 0
27 #endif
28 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
29
30 #define INIT_PASID 0
31
32 struct address_space;
33 struct mem_cgroup;
34
35 /*
36 * Each physical page in the system has a struct page associated with
37 * it to keep track of whatever it is we are using the page for at the
38 * moment. Note that we have no way to track which tasks are using
39 * a page, though if it is a pagecache page, rmap structures can tell us
40 * who is mapping it.
41 *
42 * If you allocate the page using alloc_pages(), you can use some of the
43 * space in struct page for your own purposes. The five words in the main
44 * union are available, except for bit 0 of the first word which must be
45 * kept clear. Many users use this word to store a pointer to an object
46 * which is guaranteed to be aligned. If you use the same storage as
47 * page->mapping, you must restore it to NULL before freeing the page.
48 *
49 * If your page will not be mapped to userspace, you can also use the four
50 * bytes in the mapcount union, but you must call page_mapcount_reset()
51 * before freeing it.
52 *
53 * If you want to use the refcount field, it must be used in such a way
54 * that other CPUs temporarily incrementing and then decrementing the
55 * refcount does not cause problems. On receiving the page from
56 * alloc_pages(), the refcount will be positive.
57 *
58 * If you allocate pages of order > 0, you can use some of the fields
59 * in each subpage, but you may need to restore some of their values
60 * afterwards.
61 *
62 * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
63 * That requires that freelist & counters in struct slab be adjacent and
64 * double-word aligned. Because struct slab currently just reinterprets the
65 * bits of struct page, we align all struct pages to double-word boundaries,
66 * and ensure that 'freelist' is aligned within struct slab.
67 */
68 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
69 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
70 #else
71 #define _struct_page_alignment __aligned(sizeof(unsigned long))
72 #endif
73
74 struct page {
75 unsigned long flags; /* Atomic flags, some possibly
76 * updated asynchronously */
77 /*
78 * Five words (20/40 bytes) are available in this union.
79 * WARNING: bit 0 of the first word is used for PageTail(). That
80 * means the other users of this union MUST NOT use the bit to
81 * avoid collision and false-positive PageTail().
82 */
83 union {
84 struct { /* Page cache and anonymous pages */
85 /**
86 * @lru: Pageout list, eg. active_list protected by
87 * lruvec->lru_lock. Sometimes used as a generic list
88 * by the page owner.
89 */
90 union {
91 struct list_head lru;
92
93 /* Or, for the Unevictable "LRU list" slot */
94 struct {
95 /* Always even, to negate PageTail */
96 void *__filler;
97 /* Count page's or folio's mlocks */
98 unsigned int mlock_count;
99 };
100
101 /* Or, free page */
102 struct list_head buddy_list;
103 struct list_head pcp_list;
104 };
105 /* See page-flags.h for PAGE_MAPPING_FLAGS */
106 struct address_space *mapping;
107 union {
108 pgoff_t index; /* Our offset within mapping. */
109 unsigned long share; /* share count for fsdax */
110 };
111 /**
112 * @private: Mapping-private opaque data.
113 * Usually used for buffer_heads if PagePrivate.
114 * Used for swp_entry_t if PageSwapCache.
115 * Indicates order in the buddy system if PageBuddy.
116 */
117 unsigned long private;
118 };
119 struct { /* page_pool used by netstack */
120 /**
121 * @pp_magic: magic value to avoid recycling non
122 * page_pool allocated pages.
123 */
124 unsigned long pp_magic;
125 struct page_pool *pp;
126 unsigned long _pp_mapping_pad;
127 unsigned long dma_addr;
128 atomic_long_t pp_ref_count;
129 };
130 struct { /* Tail pages of compound page */
131 unsigned long compound_head; /* Bit zero is set */
132 };
133 struct { /* ZONE_DEVICE pages */
134 /** @pgmap: Points to the hosting device page map. */
135 struct dev_pagemap *pgmap;
136 void *zone_device_data;
137 /*
138 * ZONE_DEVICE private pages are counted as being
139 * mapped so the next 3 words hold the mapping, index,
140 * and private fields from the source anonymous or
141 * page cache page while the page is migrated to device
142 * private memory.
143 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
144 * use the mapping, index, and private fields when
145 * pmem backed DAX files are mapped.
146 */
147 };
148
149 /** @rcu_head: You can use this to free a page by RCU. */
150 struct rcu_head rcu_head;
151 };
152
153 union { /* This union is 4 bytes in size. */
154 /*
155 * If the page can be mapped to userspace, encodes the number
156 * of times this page is referenced by a page table.
157 */
158 atomic_t _mapcount;
159
160 /*
161 * If the page is neither PageSlab nor mappable to userspace,
162 * the value stored here may help determine what this page
163 * is used for. See page-flags.h for a list of page types
164 * which are currently stored here.
165 */
166 unsigned int page_type;
167 };
168
169 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
170 atomic_t _refcount;
171
172 #ifdef CONFIG_SLAB_OBJ_EXT
173 unsigned long memcg_data;
174 #endif
175
176 /*
177 * On machines where all RAM is mapped into kernel address space,
178 * we can simply calculate the virtual address. On machines with
179 * highmem some memory is mapped into kernel virtual memory
180 * dynamically, so we need a place to store that address.
181 * Note that this field could be 16 bits on x86 ... ;)
182 *
183 * Architectures with slow multiplication can define
184 * WANT_PAGE_VIRTUAL in asm/page.h
185 */
186 #if defined(WANT_PAGE_VIRTUAL)
187 void *virtual; /* Kernel virtual address (NULL if
188 not kmapped, ie. highmem) */
189 #endif /* WANT_PAGE_VIRTUAL */
190
191 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
192 int _last_cpupid;
193 #endif
194
195 #ifdef CONFIG_KMSAN
196 /*
197 * KMSAN metadata for this page:
198 * - shadow page: every bit indicates whether the corresponding
199 * bit of the original page is initialized (0) or not (1);
200 * - origin page: every 4 bytes contain an id of the stack trace
201 * where the uninitialized value was created.
202 */
203 struct page *kmsan_shadow;
204 struct page *kmsan_origin;
205 #endif
206 } _struct_page_alignment;
207
208 /*
209 * struct encoded_page - a nonexistent type marking this pointer
210 *
211 * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
212 * with the low bits of the pointer indicating extra context-dependent
213 * information. Only used in mmu_gather handling, and this acts as a type
214 * system check on that use.
215 *
216 * We only really have two guaranteed bits in general, although you could
217 * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
218 * for more.
219 *
220 * Use the supplied helper functions to endcode/decode the pointer and bits.
221 */
222 struct encoded_page;
223
224 #define ENCODED_PAGE_BITS 3ul
225
226 /* Perform rmap removal after we have flushed the TLB. */
227 #define ENCODED_PAGE_BIT_DELAY_RMAP 1ul
228
229 /*
230 * The next item in an encoded_page array is the "nr_pages" argument, specifying
231 * the number of consecutive pages starting from this page, that all belong to
232 * the same folio. For example, "nr_pages" corresponds to the number of folio
233 * references that must be dropped. If this bit is not set, "nr_pages" is
234 * implicitly 1.
235 */
236 #define ENCODED_PAGE_BIT_NR_PAGES_NEXT 2ul
237
encode_page(struct page * page,unsigned long flags)238 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
239 {
240 BUILD_BUG_ON(flags > ENCODED_PAGE_BITS);
241 return (struct encoded_page *)(flags | (unsigned long)page);
242 }
243
encoded_page_flags(struct encoded_page * page)244 static inline unsigned long encoded_page_flags(struct encoded_page *page)
245 {
246 return ENCODED_PAGE_BITS & (unsigned long)page;
247 }
248
encoded_page_ptr(struct encoded_page * page)249 static inline struct page *encoded_page_ptr(struct encoded_page *page)
250 {
251 return (struct page *)(~ENCODED_PAGE_BITS & (unsigned long)page);
252 }
253
encode_nr_pages(unsigned long nr)254 static __always_inline struct encoded_page *encode_nr_pages(unsigned long nr)
255 {
256 VM_WARN_ON_ONCE((nr << 2) >> 2 != nr);
257 return (struct encoded_page *)(nr << 2);
258 }
259
encoded_nr_pages(struct encoded_page * page)260 static __always_inline unsigned long encoded_nr_pages(struct encoded_page *page)
261 {
262 return ((unsigned long)page) >> 2;
263 }
264
265 /*
266 * A swap entry has to fit into a "unsigned long", as the entry is hidden
267 * in the "index" field of the swapper address space.
268 */
269 typedef struct {
270 unsigned long val;
271 } swp_entry_t;
272
273 /**
274 * struct folio - Represents a contiguous set of bytes.
275 * @flags: Identical to the page flags.
276 * @lru: Least Recently Used list; tracks how recently this folio was used.
277 * @mlock_count: Number of times this folio has been pinned by mlock().
278 * @mapping: The file this page belongs to, or refers to the anon_vma for
279 * anonymous memory.
280 * @index: Offset within the file, in units of pages. For anonymous memory,
281 * this is the index from the beginning of the mmap.
282 * @private: Filesystem per-folio data (see folio_attach_private()).
283 * @swap: Used for swp_entry_t if folio_test_swapcache().
284 * @_mapcount: Do not access this member directly. Use folio_mapcount() to
285 * find out how many times this folio is mapped by userspace.
286 * @_refcount: Do not access this member directly. Use folio_ref_count()
287 * to find how many references there are to this folio.
288 * @memcg_data: Memory Control Group data.
289 * @virtual: Virtual address in the kernel direct map.
290 * @_last_cpupid: IDs of last CPU and last process that accessed the folio.
291 * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
292 * @_large_mapcount: Do not use directly, call folio_mapcount().
293 * @_nr_pages_mapped: Do not use outside of rmap and debug code.
294 * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
295 * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
296 * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
297 * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
298 * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
299 * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
300 * @_deferred_list: Folios to be split under memory pressure.
301 *
302 * A folio is a physically, virtually and logically contiguous set
303 * of bytes. It is a power-of-two in size, and it is aligned to that
304 * same power-of-two. It is at least as large as %PAGE_SIZE. If it is
305 * in the page cache, it is at a file offset which is a multiple of that
306 * power-of-two. It may be mapped into userspace at an address which is
307 * at an arbitrary page offset, but its kernel virtual address is aligned
308 * to its size.
309 */
310 struct folio {
311 /* private: don't document the anon union */
312 union {
313 struct {
314 /* public: */
315 unsigned long flags;
316 union {
317 struct list_head lru;
318 /* private: avoid cluttering the output */
319 struct {
320 void *__filler;
321 /* public: */
322 unsigned int mlock_count;
323 /* private: */
324 };
325 /* public: */
326 };
327 struct address_space *mapping;
328 pgoff_t index;
329 union {
330 void *private;
331 swp_entry_t swap;
332 };
333 atomic_t _mapcount;
334 atomic_t _refcount;
335 #ifdef CONFIG_SLAB_OBJ_EXT
336 unsigned long memcg_data;
337 #endif
338 #if defined(WANT_PAGE_VIRTUAL)
339 void *virtual;
340 #endif
341 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
342 int _last_cpupid;
343 #endif
344 /* private: the union with struct page is transitional */
345 };
346 struct page page;
347 };
348 union {
349 struct {
350 unsigned long _flags_1;
351 unsigned long _head_1;
352 /* public: */
353 atomic_t _large_mapcount;
354 atomic_t _entire_mapcount;
355 atomic_t _nr_pages_mapped;
356 atomic_t _pincount;
357 #ifdef CONFIG_64BIT
358 unsigned int _folio_nr_pages;
359 #endif
360 /* private: the union with struct page is transitional */
361 };
362 struct page __page_1;
363 };
364 union {
365 struct {
366 unsigned long _flags_2;
367 unsigned long _head_2;
368 /* public: */
369 void *_hugetlb_subpool;
370 void *_hugetlb_cgroup;
371 void *_hugetlb_cgroup_rsvd;
372 void *_hugetlb_hwpoison;
373 /* private: the union with struct page is transitional */
374 };
375 struct {
376 unsigned long _flags_2a;
377 unsigned long _head_2a;
378 /* public: */
379 struct list_head _deferred_list;
380 /* private: the union with struct page is transitional */
381 };
382 struct page __page_2;
383 };
384 };
385
386 #define FOLIO_MATCH(pg, fl) \
387 static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
388 FOLIO_MATCH(flags, flags);
389 FOLIO_MATCH(lru, lru);
390 FOLIO_MATCH(mapping, mapping);
391 FOLIO_MATCH(compound_head, lru);
392 FOLIO_MATCH(index, index);
393 FOLIO_MATCH(private, private);
394 FOLIO_MATCH(_mapcount, _mapcount);
395 FOLIO_MATCH(_refcount, _refcount);
396 #ifdef CONFIG_MEMCG
397 FOLIO_MATCH(memcg_data, memcg_data);
398 #endif
399 #if defined(WANT_PAGE_VIRTUAL)
400 FOLIO_MATCH(virtual, virtual);
401 #endif
402 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
403 FOLIO_MATCH(_last_cpupid, _last_cpupid);
404 #endif
405 #undef FOLIO_MATCH
406 #define FOLIO_MATCH(pg, fl) \
407 static_assert(offsetof(struct folio, fl) == \
408 offsetof(struct page, pg) + sizeof(struct page))
409 FOLIO_MATCH(flags, _flags_1);
410 FOLIO_MATCH(compound_head, _head_1);
411 #undef FOLIO_MATCH
412 #define FOLIO_MATCH(pg, fl) \
413 static_assert(offsetof(struct folio, fl) == \
414 offsetof(struct page, pg) + 2 * sizeof(struct page))
415 FOLIO_MATCH(flags, _flags_2);
416 FOLIO_MATCH(compound_head, _head_2);
417 FOLIO_MATCH(flags, _flags_2a);
418 FOLIO_MATCH(compound_head, _head_2a);
419 #undef FOLIO_MATCH
420
421 /**
422 * struct ptdesc - Memory descriptor for page tables.
423 * @__page_flags: Same as page flags. Powerpc only.
424 * @pt_rcu_head: For freeing page table pages.
425 * @pt_list: List of used page tables. Used for s390 and x86.
426 * @_pt_pad_1: Padding that aliases with page's compound head.
427 * @pmd_huge_pte: Protected by ptdesc->ptl, used for THPs.
428 * @__page_mapping: Aliases with page->mapping. Unused for page tables.
429 * @pt_index: Used for s390 gmap.
430 * @pt_mm: Used for x86 pgds.
431 * @pt_frag_refcount: For fragmented page table tracking. Powerpc only.
432 * @_pt_pad_2: Padding to ensure proper alignment.
433 * @ptl: Lock for the page table.
434 * @__page_type: Same as page->page_type. Unused for page tables.
435 * @__page_refcount: Same as page refcount.
436 * @pt_memcg_data: Memcg data. Tracked for page tables here.
437 *
438 * This struct overlays struct page for now. Do not modify without a good
439 * understanding of the issues.
440 */
441 struct ptdesc {
442 unsigned long __page_flags;
443
444 union {
445 struct rcu_head pt_rcu_head;
446 struct list_head pt_list;
447 struct {
448 unsigned long _pt_pad_1;
449 pgtable_t pmd_huge_pte;
450 };
451 };
452 unsigned long __page_mapping;
453
454 union {
455 pgoff_t pt_index;
456 struct mm_struct *pt_mm;
457 atomic_t pt_frag_refcount;
458 };
459
460 union {
461 unsigned long _pt_pad_2;
462 #if ALLOC_SPLIT_PTLOCKS
463 spinlock_t *ptl;
464 #else
465 spinlock_t ptl;
466 #endif
467 };
468 unsigned int __page_type;
469 atomic_t __page_refcount;
470 #ifdef CONFIG_MEMCG
471 unsigned long pt_memcg_data;
472 #endif
473 };
474
475 #define TABLE_MATCH(pg, pt) \
476 static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
477 TABLE_MATCH(flags, __page_flags);
478 TABLE_MATCH(compound_head, pt_list);
479 TABLE_MATCH(compound_head, _pt_pad_1);
480 TABLE_MATCH(mapping, __page_mapping);
481 TABLE_MATCH(index, pt_index);
482 TABLE_MATCH(rcu_head, pt_rcu_head);
483 TABLE_MATCH(page_type, __page_type);
484 TABLE_MATCH(_refcount, __page_refcount);
485 #ifdef CONFIG_MEMCG
486 TABLE_MATCH(memcg_data, pt_memcg_data);
487 #endif
488 #undef TABLE_MATCH
489 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
490
491 #define ptdesc_page(pt) (_Generic((pt), \
492 const struct ptdesc *: (const struct page *)(pt), \
493 struct ptdesc *: (struct page *)(pt)))
494
495 #define ptdesc_folio(pt) (_Generic((pt), \
496 const struct ptdesc *: (const struct folio *)(pt), \
497 struct ptdesc *: (struct folio *)(pt)))
498
499 #define page_ptdesc(p) (_Generic((p), \
500 const struct page *: (const struct ptdesc *)(p), \
501 struct page *: (struct ptdesc *)(p)))
502
503 /*
504 * Used for sizing the vmemmap region on some architectures
505 */
506 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
507
508 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
509 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
510
511 /*
512 * page_private can be used on tail pages. However, PagePrivate is only
513 * checked by the VM on the head page. So page_private on the tail pages
514 * should be used for data that's ancillary to the head page (eg attaching
515 * buffer heads to tail pages after attaching buffer heads to the head page)
516 */
517 #define page_private(page) ((page)->private)
518
set_page_private(struct page * page,unsigned long private)519 static inline void set_page_private(struct page *page, unsigned long private)
520 {
521 page->private = private;
522 }
523
folio_get_private(struct folio * folio)524 static inline void *folio_get_private(struct folio *folio)
525 {
526 return folio->private;
527 }
528
529 struct page_frag_cache {
530 void * va;
531 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
532 __u16 offset;
533 __u16 size;
534 #else
535 __u32 offset;
536 #endif
537 /* we maintain a pagecount bias, so that we dont dirty cache line
538 * containing page->_refcount every time we allocate a fragment.
539 */
540 unsigned int pagecnt_bias;
541 bool pfmemalloc;
542 };
543
544 typedef unsigned long vm_flags_t;
545
546 /*
547 * A region containing a mapping of a non-memory backed file under NOMMU
548 * conditions. These are held in a global tree and are pinned by the VMAs that
549 * map parts of them.
550 */
551 struct vm_region {
552 struct rb_node vm_rb; /* link in global region tree */
553 vm_flags_t vm_flags; /* VMA vm_flags */
554 unsigned long vm_start; /* start address of region */
555 unsigned long vm_end; /* region initialised to here */
556 unsigned long vm_top; /* region allocated to here */
557 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
558 struct file *vm_file; /* the backing file or NULL */
559
560 int vm_usage; /* region usage count (access under nommu_region_sem) */
561 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
562 * this region */
563 };
564
565 #ifdef CONFIG_USERFAULTFD
566 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
567 struct vm_userfaultfd_ctx {
568 struct userfaultfd_ctx *ctx;
569 };
570 #else /* CONFIG_USERFAULTFD */
571 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
572 struct vm_userfaultfd_ctx {};
573 #endif /* CONFIG_USERFAULTFD */
574
575 struct anon_vma_name {
576 struct kref kref;
577 /* The name needs to be at the end because it is dynamically sized. */
578 char name[];
579 };
580
581 #ifdef CONFIG_ANON_VMA_NAME
582 /*
583 * mmap_lock should be read-locked when calling anon_vma_name(). Caller should
584 * either keep holding the lock while using the returned pointer or it should
585 * raise anon_vma_name refcount before releasing the lock.
586 */
587 struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma);
588 struct anon_vma_name *anon_vma_name_alloc(const char *name);
589 void anon_vma_name_free(struct kref *kref);
590 #else /* CONFIG_ANON_VMA_NAME */
anon_vma_name(struct vm_area_struct * vma)591 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
592 {
593 return NULL;
594 }
595
anon_vma_name_alloc(const char * name)596 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
597 {
598 return NULL;
599 }
600 #endif
601
602 struct vma_lock {
603 struct rw_semaphore lock;
604 };
605
606 struct vma_numab_state {
607 /*
608 * Initialised as time in 'jiffies' after which VMA
609 * should be scanned. Delays first scan of new VMA by at
610 * least sysctl_numa_balancing_scan_delay:
611 */
612 unsigned long next_scan;
613
614 /*
615 * Time in jiffies when pids_active[] is reset to
616 * detect phase change behaviour:
617 */
618 unsigned long pids_active_reset;
619
620 /*
621 * Approximate tracking of PIDs that trapped a NUMA hinting
622 * fault. May produce false positives due to hash collisions.
623 *
624 * [0] Previous PID tracking
625 * [1] Current PID tracking
626 *
627 * Window moves after next_pid_reset has expired approximately
628 * every VMA_PID_RESET_PERIOD jiffies:
629 */
630 unsigned long pids_active[2];
631
632 /* MM scan sequence ID when scan first started after VMA creation */
633 int start_scan_seq;
634
635 /*
636 * MM scan sequence ID when the VMA was last completely scanned.
637 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq
638 */
639 int prev_scan_seq;
640 };
641
642 /*
643 * This struct describes a virtual memory area. There is one of these
644 * per VM-area/task. A VM area is any part of the process virtual memory
645 * space that has a special rule for the page-fault handlers (ie a shared
646 * library, the executable area etc).
647 */
648 struct vm_area_struct {
649 /* The first cache line has the info for VMA tree walking. */
650
651 union {
652 struct {
653 /* VMA covers [vm_start; vm_end) addresses within mm */
654 unsigned long vm_start;
655 unsigned long vm_end;
656 };
657 #ifdef CONFIG_PER_VMA_LOCK
658 struct rcu_head vm_rcu; /* Used for deferred freeing. */
659 #endif
660 };
661
662 struct mm_struct *vm_mm; /* The address space we belong to. */
663 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
664
665 /*
666 * Flags, see mm.h.
667 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
668 */
669 union {
670 const vm_flags_t vm_flags;
671 vm_flags_t __private __vm_flags;
672 };
673
674 #ifdef CONFIG_PER_VMA_LOCK
675 /* Flag to indicate areas detached from the mm->mm_mt tree */
676 bool detached;
677
678 /*
679 * Can only be written (using WRITE_ONCE()) while holding both:
680 * - mmap_lock (in write mode)
681 * - vm_lock->lock (in write mode)
682 * Can be read reliably while holding one of:
683 * - mmap_lock (in read or write mode)
684 * - vm_lock->lock (in read or write mode)
685 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
686 * while holding nothing (except RCU to keep the VMA struct allocated).
687 *
688 * This sequence counter is explicitly allowed to overflow; sequence
689 * counter reuse can only lead to occasional unnecessary use of the
690 * slowpath.
691 */
692 int vm_lock_seq;
693 struct vma_lock *vm_lock;
694 #endif
695
696 /*
697 * For areas with an address space and backing store,
698 * linkage into the address_space->i_mmap interval tree.
699 *
700 */
701 struct {
702 struct rb_node rb;
703 unsigned long rb_subtree_last;
704 } shared;
705
706 /*
707 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
708 * list, after a COW of one of the file pages. A MAP_SHARED vma
709 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
710 * or brk vma (with NULL file) can only be in an anon_vma list.
711 */
712 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
713 * page_table_lock */
714 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
715
716 /* Function pointers to deal with this struct. */
717 const struct vm_operations_struct *vm_ops;
718
719 /* Information about our backing store: */
720 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
721 units */
722 struct file * vm_file; /* File we map to (can be NULL). */
723 void * vm_private_data; /* was vm_pte (shared mem) */
724
725 #ifdef CONFIG_ANON_VMA_NAME
726 /*
727 * For private and shared anonymous mappings, a pointer to a null
728 * terminated string containing the name given to the vma, or NULL if
729 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
730 */
731 struct anon_vma_name *anon_name;
732 #endif
733 #ifdef CONFIG_SWAP
734 atomic_long_t swap_readahead_info;
735 #endif
736 #ifndef CONFIG_MMU
737 struct vm_region *vm_region; /* NOMMU mapping region */
738 #endif
739 #ifdef CONFIG_NUMA
740 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
741 #endif
742 #ifdef CONFIG_NUMA_BALANCING
743 struct vma_numab_state *numab_state; /* NUMA Balancing state */
744 #endif
745 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
746 } __randomize_layout;
747
748 #ifdef CONFIG_NUMA
749 #define vma_policy(vma) ((vma)->vm_policy)
750 #else
751 #define vma_policy(vma) NULL
752 #endif
753
754 #ifdef CONFIG_SCHED_MM_CID
755 struct mm_cid {
756 u64 time;
757 int cid;
758 };
759 #endif
760
761 struct kioctx_table;
762 struct iommu_mm_data;
763 struct mm_struct {
764 struct {
765 /*
766 * Fields which are often written to are placed in a separate
767 * cache line.
768 */
769 struct {
770 /**
771 * @mm_count: The number of references to &struct
772 * mm_struct (@mm_users count as 1).
773 *
774 * Use mmgrab()/mmdrop() to modify. When this drops to
775 * 0, the &struct mm_struct is freed.
776 */
777 atomic_t mm_count;
778 } ____cacheline_aligned_in_smp;
779
780 struct maple_tree mm_mt;
781
782 unsigned long mmap_base; /* base of mmap area */
783 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
784 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
785 /* Base addresses for compatible mmap() */
786 unsigned long mmap_compat_base;
787 unsigned long mmap_compat_legacy_base;
788 #endif
789 unsigned long task_size; /* size of task vm space */
790 pgd_t * pgd;
791
792 #ifdef CONFIG_MEMBARRIER
793 /**
794 * @membarrier_state: Flags controlling membarrier behavior.
795 *
796 * This field is close to @pgd to hopefully fit in the same
797 * cache-line, which needs to be touched by switch_mm().
798 */
799 atomic_t membarrier_state;
800 #endif
801
802 /**
803 * @mm_users: The number of users including userspace.
804 *
805 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
806 * drops to 0 (i.e. when the task exits and there are no other
807 * temporary reference holders), we also release a reference on
808 * @mm_count (which may then free the &struct mm_struct if
809 * @mm_count also drops to 0).
810 */
811 atomic_t mm_users;
812
813 #ifdef CONFIG_SCHED_MM_CID
814 /**
815 * @pcpu_cid: Per-cpu current cid.
816 *
817 * Keep track of the currently allocated mm_cid for each cpu.
818 * The per-cpu mm_cid values are serialized by their respective
819 * runqueue locks.
820 */
821 struct mm_cid __percpu *pcpu_cid;
822 /*
823 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
824 *
825 * When the next mm_cid scan is due (in jiffies).
826 */
827 unsigned long mm_cid_next_scan;
828 #endif
829 #ifdef CONFIG_MMU
830 atomic_long_t pgtables_bytes; /* size of all page tables */
831 #endif
832 int map_count; /* number of VMAs */
833
834 spinlock_t page_table_lock; /* Protects page tables and some
835 * counters
836 */
837 /*
838 * With some kernel config, the current mmap_lock's offset
839 * inside 'mm_struct' is at 0x120, which is very optimal, as
840 * its two hot fields 'count' and 'owner' sit in 2 different
841 * cachelines, and when mmap_lock is highly contended, both
842 * of the 2 fields will be accessed frequently, current layout
843 * will help to reduce cache bouncing.
844 *
845 * So please be careful with adding new fields before
846 * mmap_lock, which can easily push the 2 fields into one
847 * cacheline.
848 */
849 struct rw_semaphore mmap_lock;
850
851 struct list_head mmlist; /* List of maybe swapped mm's. These
852 * are globally strung together off
853 * init_mm.mmlist, and are protected
854 * by mmlist_lock
855 */
856 #ifdef CONFIG_PER_VMA_LOCK
857 /*
858 * This field has lock-like semantics, meaning it is sometimes
859 * accessed with ACQUIRE/RELEASE semantics.
860 * Roughly speaking, incrementing the sequence number is
861 * equivalent to releasing locks on VMAs; reading the sequence
862 * number can be part of taking a read lock on a VMA.
863 *
864 * Can be modified under write mmap_lock using RELEASE
865 * semantics.
866 * Can be read with no other protection when holding write
867 * mmap_lock.
868 * Can be read with ACQUIRE semantics if not holding write
869 * mmap_lock.
870 */
871 int mm_lock_seq;
872 #endif
873
874
875 unsigned long hiwater_rss; /* High-watermark of RSS usage */
876 unsigned long hiwater_vm; /* High-water virtual memory usage */
877
878 unsigned long total_vm; /* Total pages mapped */
879 unsigned long locked_vm; /* Pages that have PG_mlocked set */
880 atomic64_t pinned_vm; /* Refcount permanently increased */
881 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
882 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
883 unsigned long stack_vm; /* VM_STACK */
884 unsigned long def_flags;
885
886 /**
887 * @write_protect_seq: Locked when any thread is write
888 * protecting pages mapped by this mm to enforce a later COW,
889 * for instance during page table copying for fork().
890 */
891 seqcount_t write_protect_seq;
892
893 spinlock_t arg_lock; /* protect the below fields */
894
895 unsigned long start_code, end_code, start_data, end_data;
896 unsigned long start_brk, brk, start_stack;
897 unsigned long arg_start, arg_end, env_start, env_end;
898
899 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
900
901 struct percpu_counter rss_stat[NR_MM_COUNTERS];
902
903 struct linux_binfmt *binfmt;
904
905 /* Architecture-specific MM context */
906 mm_context_t context;
907
908 unsigned long flags; /* Must use atomic bitops to access */
909
910 #ifdef CONFIG_AIO
911 spinlock_t ioctx_lock;
912 struct kioctx_table __rcu *ioctx_table;
913 #endif
914 #ifdef CONFIG_MEMCG
915 /*
916 * "owner" points to a task that is regarded as the canonical
917 * user/owner of this mm. All of the following must be true in
918 * order for it to be changed:
919 *
920 * current == mm->owner
921 * current->mm != mm
922 * new_owner->mm == mm
923 * new_owner->alloc_lock is held
924 */
925 struct task_struct __rcu *owner;
926 #endif
927 struct user_namespace *user_ns;
928
929 /* store ref to file /proc/<pid>/exe symlink points to */
930 struct file __rcu *exe_file;
931 #ifdef CONFIG_MMU_NOTIFIER
932 struct mmu_notifier_subscriptions *notifier_subscriptions;
933 #endif
934 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
935 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
936 #endif
937 #ifdef CONFIG_NUMA_BALANCING
938 /*
939 * numa_next_scan is the next time that PTEs will be remapped
940 * PROT_NONE to trigger NUMA hinting faults; such faults gather
941 * statistics and migrate pages to new nodes if necessary.
942 */
943 unsigned long numa_next_scan;
944
945 /* Restart point for scanning and remapping PTEs. */
946 unsigned long numa_scan_offset;
947
948 /* numa_scan_seq prevents two threads remapping PTEs. */
949 int numa_scan_seq;
950 #endif
951 /*
952 * An operation with batched TLB flushing is going on. Anything
953 * that can move process memory needs to flush the TLB when
954 * moving a PROT_NONE mapped page.
955 */
956 atomic_t tlb_flush_pending;
957 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
958 /* See flush_tlb_batched_pending() */
959 atomic_t tlb_flush_batched;
960 #endif
961 struct uprobes_state uprobes_state;
962 #ifdef CONFIG_PREEMPT_RT
963 struct rcu_head delayed_drop;
964 #endif
965 #ifdef CONFIG_HUGETLB_PAGE
966 atomic_long_t hugetlb_usage;
967 #endif
968 struct work_struct async_put_work;
969
970 #ifdef CONFIG_IOMMU_MM_DATA
971 struct iommu_mm_data *iommu_mm;
972 #endif
973 #ifdef CONFIG_KSM
974 /*
975 * Represent how many pages of this process are involved in KSM
976 * merging (not including ksm_zero_pages).
977 */
978 unsigned long ksm_merging_pages;
979 /*
980 * Represent how many pages are checked for ksm merging
981 * including merged and not merged.
982 */
983 unsigned long ksm_rmap_items;
984 /*
985 * Represent how many empty pages are merged with kernel zero
986 * pages when enabling KSM use_zero_pages.
987 */
988 atomic_long_t ksm_zero_pages;
989 #endif /* CONFIG_KSM */
990 #ifdef CONFIG_LRU_GEN_WALKS_MMU
991 struct {
992 /* this mm_struct is on lru_gen_mm_list */
993 struct list_head list;
994 /*
995 * Set when switching to this mm_struct, as a hint of
996 * whether it has been used since the last time per-node
997 * page table walkers cleared the corresponding bits.
998 */
999 unsigned long bitmap;
1000 #ifdef CONFIG_MEMCG
1001 /* points to the memcg of "owner" above */
1002 struct mem_cgroup *memcg;
1003 #endif
1004 } lru_gen;
1005 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1006 } __randomize_layout;
1007
1008 /*
1009 * The mm_cpumask needs to be at the end of mm_struct, because it
1010 * is dynamically sized based on nr_cpu_ids.
1011 */
1012 unsigned long cpu_bitmap[];
1013 };
1014
1015 #define MM_MT_FLAGS (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
1016 MT_FLAGS_USE_RCU)
1017 extern struct mm_struct init_mm;
1018
1019 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)1020 static inline void mm_init_cpumask(struct mm_struct *mm)
1021 {
1022 unsigned long cpu_bitmap = (unsigned long)mm;
1023
1024 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1025 cpumask_clear((struct cpumask *)cpu_bitmap);
1026 }
1027
1028 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)1029 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
1030 {
1031 return (struct cpumask *)&mm->cpu_bitmap;
1032 }
1033
1034 #ifdef CONFIG_LRU_GEN
1035
1036 struct lru_gen_mm_list {
1037 /* mm_struct list for page table walkers */
1038 struct list_head fifo;
1039 /* protects the list above */
1040 spinlock_t lock;
1041 };
1042
1043 #endif /* CONFIG_LRU_GEN */
1044
1045 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1046
1047 void lru_gen_add_mm(struct mm_struct *mm);
1048 void lru_gen_del_mm(struct mm_struct *mm);
1049 void lru_gen_migrate_mm(struct mm_struct *mm);
1050
lru_gen_init_mm(struct mm_struct * mm)1051 static inline void lru_gen_init_mm(struct mm_struct *mm)
1052 {
1053 INIT_LIST_HEAD(&mm->lru_gen.list);
1054 mm->lru_gen.bitmap = 0;
1055 #ifdef CONFIG_MEMCG
1056 mm->lru_gen.memcg = NULL;
1057 #endif
1058 }
1059
lru_gen_use_mm(struct mm_struct * mm)1060 static inline void lru_gen_use_mm(struct mm_struct *mm)
1061 {
1062 /*
1063 * When the bitmap is set, page reclaim knows this mm_struct has been
1064 * used since the last time it cleared the bitmap. So it might be worth
1065 * walking the page tables of this mm_struct to clear the accessed bit.
1066 */
1067 WRITE_ONCE(mm->lru_gen.bitmap, -1);
1068 }
1069
1070 #else /* !CONFIG_LRU_GEN_WALKS_MMU */
1071
lru_gen_add_mm(struct mm_struct * mm)1072 static inline void lru_gen_add_mm(struct mm_struct *mm)
1073 {
1074 }
1075
lru_gen_del_mm(struct mm_struct * mm)1076 static inline void lru_gen_del_mm(struct mm_struct *mm)
1077 {
1078 }
1079
lru_gen_migrate_mm(struct mm_struct * mm)1080 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
1081 {
1082 }
1083
lru_gen_init_mm(struct mm_struct * mm)1084 static inline void lru_gen_init_mm(struct mm_struct *mm)
1085 {
1086 }
1087
lru_gen_use_mm(struct mm_struct * mm)1088 static inline void lru_gen_use_mm(struct mm_struct *mm)
1089 {
1090 }
1091
1092 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1093
1094 struct vma_iterator {
1095 struct ma_state mas;
1096 };
1097
1098 #define VMA_ITERATOR(name, __mm, __addr) \
1099 struct vma_iterator name = { \
1100 .mas = { \
1101 .tree = &(__mm)->mm_mt, \
1102 .index = __addr, \
1103 .node = NULL, \
1104 .status = ma_start, \
1105 }, \
1106 }
1107
vma_iter_init(struct vma_iterator * vmi,struct mm_struct * mm,unsigned long addr)1108 static inline void vma_iter_init(struct vma_iterator *vmi,
1109 struct mm_struct *mm, unsigned long addr)
1110 {
1111 mas_init(&vmi->mas, &mm->mm_mt, addr);
1112 }
1113
1114 #ifdef CONFIG_SCHED_MM_CID
1115
1116 enum mm_cid_state {
1117 MM_CID_UNSET = -1U, /* Unset state has lazy_put flag set. */
1118 MM_CID_LAZY_PUT = (1U << 31),
1119 };
1120
mm_cid_is_unset(int cid)1121 static inline bool mm_cid_is_unset(int cid)
1122 {
1123 return cid == MM_CID_UNSET;
1124 }
1125
mm_cid_is_lazy_put(int cid)1126 static inline bool mm_cid_is_lazy_put(int cid)
1127 {
1128 return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1129 }
1130
mm_cid_is_valid(int cid)1131 static inline bool mm_cid_is_valid(int cid)
1132 {
1133 return !(cid & MM_CID_LAZY_PUT);
1134 }
1135
mm_cid_set_lazy_put(int cid)1136 static inline int mm_cid_set_lazy_put(int cid)
1137 {
1138 return cid | MM_CID_LAZY_PUT;
1139 }
1140
mm_cid_clear_lazy_put(int cid)1141 static inline int mm_cid_clear_lazy_put(int cid)
1142 {
1143 return cid & ~MM_CID_LAZY_PUT;
1144 }
1145
1146 /* Accessor for struct mm_struct's cidmask. */
mm_cidmask(struct mm_struct * mm)1147 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1148 {
1149 unsigned long cid_bitmap = (unsigned long)mm;
1150
1151 cid_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1152 /* Skip cpu_bitmap */
1153 cid_bitmap += cpumask_size();
1154 return (struct cpumask *)cid_bitmap;
1155 }
1156
mm_init_cid(struct mm_struct * mm)1157 static inline void mm_init_cid(struct mm_struct *mm)
1158 {
1159 int i;
1160
1161 for_each_possible_cpu(i) {
1162 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1163
1164 pcpu_cid->cid = MM_CID_UNSET;
1165 pcpu_cid->time = 0;
1166 }
1167 cpumask_clear(mm_cidmask(mm));
1168 }
1169
mm_alloc_cid_noprof(struct mm_struct * mm)1170 static inline int mm_alloc_cid_noprof(struct mm_struct *mm)
1171 {
1172 mm->pcpu_cid = alloc_percpu_noprof(struct mm_cid);
1173 if (!mm->pcpu_cid)
1174 return -ENOMEM;
1175 mm_init_cid(mm);
1176 return 0;
1177 }
1178 #define mm_alloc_cid(...) alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__))
1179
mm_destroy_cid(struct mm_struct * mm)1180 static inline void mm_destroy_cid(struct mm_struct *mm)
1181 {
1182 free_percpu(mm->pcpu_cid);
1183 mm->pcpu_cid = NULL;
1184 }
1185
mm_cid_size(void)1186 static inline unsigned int mm_cid_size(void)
1187 {
1188 return cpumask_size();
1189 }
1190 #else /* CONFIG_SCHED_MM_CID */
mm_init_cid(struct mm_struct * mm)1191 static inline void mm_init_cid(struct mm_struct *mm) { }
mm_alloc_cid(struct mm_struct * mm)1192 static inline int mm_alloc_cid(struct mm_struct *mm) { return 0; }
mm_destroy_cid(struct mm_struct * mm)1193 static inline void mm_destroy_cid(struct mm_struct *mm) { }
mm_cid_size(void)1194 static inline unsigned int mm_cid_size(void)
1195 {
1196 return 0;
1197 }
1198 #endif /* CONFIG_SCHED_MM_CID */
1199
1200 struct mmu_gather;
1201 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1202 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1203 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1204
1205 struct vm_fault;
1206
1207 /**
1208 * typedef vm_fault_t - Return type for page fault handlers.
1209 *
1210 * Page fault handlers return a bitmask of %VM_FAULT values.
1211 */
1212 typedef __bitwise unsigned int vm_fault_t;
1213
1214 /**
1215 * enum vm_fault_reason - Page fault handlers return a bitmask of
1216 * these values to tell the core VM what happened when handling the
1217 * fault. Used to decide whether a process gets delivered SIGBUS or
1218 * just gets major/minor fault counters bumped up.
1219 *
1220 * @VM_FAULT_OOM: Out Of Memory
1221 * @VM_FAULT_SIGBUS: Bad access
1222 * @VM_FAULT_MAJOR: Page read from storage
1223 * @VM_FAULT_HWPOISON: Hit poisoned small page
1224 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
1225 * in upper bits
1226 * @VM_FAULT_SIGSEGV: segmentation fault
1227 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
1228 * @VM_FAULT_LOCKED: ->fault locked the returned page
1229 * @VM_FAULT_RETRY: ->fault blocked, must retry
1230 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
1231 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
1232 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
1233 * fsync() to complete (for synchronous page faults
1234 * in DAX)
1235 * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released
1236 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
1237 *
1238 */
1239 enum vm_fault_reason {
1240 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
1241 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
1242 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
1243 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
1244 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1245 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
1246 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
1247 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
1248 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
1249 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
1250 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
1251 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
1252 VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000,
1253 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
1254 };
1255
1256 /* Encode hstate index for a hwpoisoned large page */
1257 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1258 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1259
1260 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
1261 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
1262 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1263
1264 #define VM_FAULT_RESULT_TRACE \
1265 { VM_FAULT_OOM, "OOM" }, \
1266 { VM_FAULT_SIGBUS, "SIGBUS" }, \
1267 { VM_FAULT_MAJOR, "MAJOR" }, \
1268 { VM_FAULT_HWPOISON, "HWPOISON" }, \
1269 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
1270 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
1271 { VM_FAULT_NOPAGE, "NOPAGE" }, \
1272 { VM_FAULT_LOCKED, "LOCKED" }, \
1273 { VM_FAULT_RETRY, "RETRY" }, \
1274 { VM_FAULT_FALLBACK, "FALLBACK" }, \
1275 { VM_FAULT_DONE_COW, "DONE_COW" }, \
1276 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }, \
1277 { VM_FAULT_COMPLETED, "COMPLETED" }
1278
1279 struct vm_special_mapping {
1280 const char *name; /* The name, e.g. "[vdso]". */
1281
1282 /*
1283 * If .fault is not provided, this points to a
1284 * NULL-terminated array of pages that back the special mapping.
1285 *
1286 * This must not be NULL unless .fault is provided.
1287 */
1288 struct page **pages;
1289
1290 /*
1291 * If non-NULL, then this is called to resolve page faults
1292 * on the special mapping. If used, .pages is not checked.
1293 */
1294 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1295 struct vm_area_struct *vma,
1296 struct vm_fault *vmf);
1297
1298 int (*mremap)(const struct vm_special_mapping *sm,
1299 struct vm_area_struct *new_vma);
1300 };
1301
1302 enum tlb_flush_reason {
1303 TLB_FLUSH_ON_TASK_SWITCH,
1304 TLB_REMOTE_SHOOTDOWN,
1305 TLB_LOCAL_SHOOTDOWN,
1306 TLB_LOCAL_MM_SHOOTDOWN,
1307 TLB_REMOTE_SEND_IPI,
1308 NR_TLB_FLUSH_REASONS,
1309 };
1310
1311 /**
1312 * enum fault_flag - Fault flag definitions.
1313 * @FAULT_FLAG_WRITE: Fault was a write fault.
1314 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1315 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1316 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1317 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1318 * @FAULT_FLAG_TRIED: The fault has been tried once.
1319 * @FAULT_FLAG_USER: The fault originated in userspace.
1320 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1321 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1322 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1323 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1324 * COW mapping, making sure that an exclusive anon page is
1325 * mapped after the fault.
1326 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1327 * We should only access orig_pte if this flag set.
1328 * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1329 *
1330 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1331 * whether we would allow page faults to retry by specifying these two
1332 * fault flags correctly. Currently there can be three legal combinations:
1333 *
1334 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and
1335 * this is the first try
1336 *
1337 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and
1338 * we've already tried at least once
1339 *
1340 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1341 *
1342 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1343 * be used. Note that page faults can be allowed to retry for multiple times,
1344 * in which case we'll have an initial fault with flags (a) then later on
1345 * continuous faults with flags (b). We should always try to detect pending
1346 * signals before a retry to make sure the continuous page faults can still be
1347 * interrupted if necessary.
1348 *
1349 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1350 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1351 * applied to mappings that are not COW mappings.
1352 */
1353 enum fault_flag {
1354 FAULT_FLAG_WRITE = 1 << 0,
1355 FAULT_FLAG_MKWRITE = 1 << 1,
1356 FAULT_FLAG_ALLOW_RETRY = 1 << 2,
1357 FAULT_FLAG_RETRY_NOWAIT = 1 << 3,
1358 FAULT_FLAG_KILLABLE = 1 << 4,
1359 FAULT_FLAG_TRIED = 1 << 5,
1360 FAULT_FLAG_USER = 1 << 6,
1361 FAULT_FLAG_REMOTE = 1 << 7,
1362 FAULT_FLAG_INSTRUCTION = 1 << 8,
1363 FAULT_FLAG_INTERRUPTIBLE = 1 << 9,
1364 FAULT_FLAG_UNSHARE = 1 << 10,
1365 FAULT_FLAG_ORIG_PTE_VALID = 1 << 11,
1366 FAULT_FLAG_VMA_LOCK = 1 << 12,
1367 };
1368
1369 typedef unsigned int __bitwise zap_flags_t;
1370
1371 /* Flags for clear_young_dirty_ptes(). */
1372 typedef int __bitwise cydp_t;
1373
1374 /* Clear the access bit */
1375 #define CYDP_CLEAR_YOUNG ((__force cydp_t)BIT(0))
1376
1377 /* Clear the dirty bit */
1378 #define CYDP_CLEAR_DIRTY ((__force cydp_t)BIT(1))
1379
1380 /*
1381 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1382 * other. Here is what they mean, and how to use them:
1383 *
1384 *
1385 * FIXME: For pages which are part of a filesystem, mappings are subject to the
1386 * lifetime enforced by the filesystem and we need guarantees that longterm
1387 * users like RDMA and V4L2 only establish mappings which coordinate usage with
1388 * the filesystem. Ideas for this coordination include revoking the longterm
1389 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was
1390 * added after the problem with filesystems was found FS DAX VMAs are
1391 * specifically failed. Filesystem pages are still subject to bugs and use of
1392 * FOLL_LONGTERM should be avoided on those pages.
1393 *
1394 * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1395 * that region. And so, CMA attempts to migrate the page before pinning, when
1396 * FOLL_LONGTERM is specified.
1397 *
1398 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1399 * but an additional pin counting system) will be invoked. This is intended for
1400 * anything that gets a page reference and then touches page data (for example,
1401 * Direct IO). This lets the filesystem know that some non-file-system entity is
1402 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1403 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1404 * a call to unpin_user_page().
1405 *
1406 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1407 * and separate refcounting mechanisms, however, and that means that each has
1408 * its own acquire and release mechanisms:
1409 *
1410 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1411 *
1412 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1413 *
1414 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1415 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1416 * calls applied to them, and that's perfectly OK. This is a constraint on the
1417 * callers, not on the pages.)
1418 *
1419 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1420 * directly by the caller. That's in order to help avoid mismatches when
1421 * releasing pages: get_user_pages*() pages must be released via put_page(),
1422 * while pin_user_pages*() pages must be released via unpin_user_page().
1423 *
1424 * Please see Documentation/core-api/pin_user_pages.rst for more information.
1425 */
1426
1427 enum {
1428 /* check pte is writable */
1429 FOLL_WRITE = 1 << 0,
1430 /* do get_page on page */
1431 FOLL_GET = 1 << 1,
1432 /* give error on hole if it would be zero */
1433 FOLL_DUMP = 1 << 2,
1434 /* get_user_pages read/write w/o permission */
1435 FOLL_FORCE = 1 << 3,
1436 /*
1437 * if a disk transfer is needed, start the IO and return without waiting
1438 * upon it
1439 */
1440 FOLL_NOWAIT = 1 << 4,
1441 /* do not fault in pages */
1442 FOLL_NOFAULT = 1 << 5,
1443 /* check page is hwpoisoned */
1444 FOLL_HWPOISON = 1 << 6,
1445 /* don't do file mappings */
1446 FOLL_ANON = 1 << 7,
1447 /*
1448 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1449 * time period _often_ under userspace control. This is in contrast to
1450 * iov_iter_get_pages(), whose usages are transient.
1451 */
1452 FOLL_LONGTERM = 1 << 8,
1453 /* split huge pmd before returning */
1454 FOLL_SPLIT_PMD = 1 << 9,
1455 /* allow returning PCI P2PDMA pages */
1456 FOLL_PCI_P2PDMA = 1 << 10,
1457 /* allow interrupts from generic signals */
1458 FOLL_INTERRUPTIBLE = 1 << 11,
1459 /*
1460 * Always honor (trigger) NUMA hinting faults.
1461 *
1462 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1463 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1464 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1465 * hinting faults.
1466 */
1467 FOLL_HONOR_NUMA_FAULT = 1 << 12,
1468
1469 /* See also internal only FOLL flags in mm/internal.h */
1470 };
1471
1472 #endif /* _LINUX_MM_TYPES_H */
1473